0% found this document useful (0 votes)
39 views555 pages

MI Vol 4

The Right Half Convex Function Theorem provides an extension of Jensen's inequality for convex functions defined on an interval I and convex on a subinterval I≥s, where s is an interior point of I. It states that if f satisfies a certain condition involving x, y ∈ I with x ≤ s ≤ y, then f is convex when averaged over variables that sum to ns or ns1, where s1 > s is also an interior point of I. The proof uses properties of convex functions, including Jensen's inequality and Karamata's inequality. This theorem allows deriving inequalities for convex functions that are convex on only part of their domain.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
39 views555 pages

MI Vol 4

The Right Half Convex Function Theorem provides an extension of Jensen's inequality for convex functions defined on an interval I and convex on a subinterval I≥s, where s is an interior point of I. It states that if f satisfies a certain condition involving x, y ∈ I with x ≤ s ≤ y, then f is convex when averaged over variables that sum to ns or ns1, where s1 > s is also an interior point of I. The proof uses properties of convex functions, including Jensen's inequality and Karamata's inequality. This theorem allows deriving inequalities for convex functions that are convex on only part of their domain.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 555

        

              !
"   #    #   "  $  "
"  " %"  ##"        !
   !   " #" " " % &  ' 
(" )     *  ("   
+"      ,  -  +  $ .
"       " " % " " " % 
 "   " & ! " " /  " $    !
 0  1      2  
 3 1 4"    " -" %  (%  5
#"     "%    " %    
6  7) "        "   ."
#   6%       #"
 $      " !   &    
   %    %  "   
 "$   #"         %"# #"8 "
  %   $

 

 "   -" 8  )" "  


   
9# "  .  -"  -#"
"  :"  )   $ 6     
"    !! "% 
%     !  "%   "        
" %  #"%      $




Vasile Cîrtoaje

▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
MATHEMATICAL
INEQUALITIES
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀

Volume 4

EXTENSIONS AND REFINEMENTS


OF JENSEN’S INEQUALITY

LAP LAMBERT Academic Publishing

EDITURA UNIVERSITĂŢII PETROL-GAZE DIN PLOIEŞTI


2021
Contents

1 Half Convex Function Method 1


1.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Half Convex Function Method for Ordered Variables 147


2.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
2.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
2.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

3 Partially Convex Function Method 211


3.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
3.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
3.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

4 Partially Convex Function Method for Ordered Variables 297


4.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
4.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
4.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

5 EV Method for Nonnegative Variables 323


5.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
5.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
5.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349

6 EV Method for Real Variables 489


6.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
6.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
6.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503

A Glosar 539

B Bibliography 549

i
ii Vasile Cîrtoaje
Chapter 1

Half Convex Function Method

1.1 Theoretical Basis


Let I be a real interval, s an interior point of I and

I≥s = {u|u ∈ I, u ≥ s}, I≤s = {u|u ∈ I, u ≤ s}.

The following statement is known as the Right Half Convex Function Theorem
(RHCF-Theorem).

Right Half Convex Function Theorem (Vasile Cîrtoaje, 2004). Let f be a real
function defined on an interval I and convex on I≥s , where s ∈ int(I). If

f (x) + (n − 1) f ( y) ≥ n f (s)

for all x, y ∈ I so that x ≤ s ≤ y and x + (n − 1) y = ns, then the inequality


a + a + ··· + a 
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (1)
n
holds for all a1 , a2 , . . . , an ∈ I satisfying a1 + a2 + · · · + an = ns. In addition, the
inequality (1) holds for all a1 , a2 , . . . , an ∈ I satisfying a1 + a2 + · · · + an = ns1 , where
s1 ∈ int(I), s1 > s.

Proof. Assume that


a1 ≤ a2 ≤ · · · ≤ an .
If a1 ≥ s, then the required inequality is just Jensen’s inequality for convex func-
tions. Otherwise, if a1 < s, then there exists

k ∈ {1, 2, . . . , n − 1}

so that
a1 ≤ · · · ≤ ak < s ≤ ak+1 ≤ · · · ≤ an .

1
2 Vasile Cîrtoaje

Since f is convex on I≥s , we may apply Jensen’s inequality to get

f (ak+1 ) + · · · + f (an ) ≥ (n − k) f (z),

where
ak+1 + · · · + an
z= , z ∈ I.
n−k
Thus, it suffices to show that

f (a1 ) + · · · + f (ak ) + (n − k) f (z) ≥ n f (s). (2)

Let b1 , . . . , bk be defined by

ai + (n − 1)bi = ns, i = 1, . . . , k.

We claim that
z ≥ b1 ≥ · · · ≥ bk > s,
which involves
b1 , . . . , bk ∈ I≥s .
Indeed, we have
b1 ≥ · · · ≥ b k ,
s − ak
bk − s = > 0,
n−1
and
z ≥ b1
because

(n − 1)b1 = ns − a1 = (a2 + · · · + ak ) + ak+1 + · · · + an


≤ (k − 1)s + ak+1 + · · · + an
= (k − 1)s + (n − k)z ≤ (n − 1)z.

Since b1 , . . . , bk ∈ I≥s , by hypothesis we have

f (a1 ) + (n − 1) f (b1 ) ≥ n f (s),

···
f (ak ) + (n − 1) f (bk ) ≥ n f (s),
hence
f (a1 ) + · · · + f (ak ) + (n − 1)[ f (b1 ) + · · · + f (bk )] ≥ kn f (s),
f (a1 ) + · · · + f (ak ) ≥ kn f (s) − (n − 1)[ f (b1 ) + · · · + f (bk )].
According to this result, the inequality (2) is true if

kn f (s) − (n − 1)[ f (b1 ) + · · · + f (bk )] + (n − k) f (z) ≥ n f (s),


Half Convex Function Method 3

which is equivalent to
n−k
p f (z) + (k − p) f (s) ≥ f (b1 ) + · · · + f (bk ), p= ≤ 1.
n−1
By Jensen’s inequality, we have

p f (z) + (1 − p) f (s) ≥ f (w), w = pz + (1 − p)s ≥ s.

Thus, we only need to show that

f (w) + (k − 1) f (s) ≥ f (b1 ) + · · · + f (bk ).

Since the decreasingly ordered vector A~k = (w, s, . . . , s) majorizes the decreasingly
ordered vector B~k = (b1 , b2 , . . . , bk ), this inequality follows from Karamata’s in-
equality for convex functions.
According to this result, the inequality (1) holds for all a1 , a2 , . . . , an ∈ I satisfying
a1 + a2 + · · · + an = ns1 if f (x 1 ) + (n − 1) f ( y1 ) ≥ n f (s1 ) for all x 1 , y1 ∈ I so that
x 1 ≤ s1 ≤ y1 and x 1 + (n − 1) y1 = ns1 . Thus, we need to show that if

f (x) + (n − 1) f ( y) ≥ n f (s)

for all x, y ∈ I so that x ≤ s ≤ y and x + (n − 1) y = ns, then

f (x 1 ) + (n − 1) f ( y1 ) ≥ n f (s1 ) (3)

for all x 1 , y1 ∈ I so that x 1 ≤ s1 ≤ y1 and x 1 + (n − 1) y1 = ns1 . Since this is true for


x 1 ≥ s (by Jensen’s inequality), consider next x 1 < s. By hypothesis, we have

f (x 1 ) + (n − 1) f ( y2 ) ≥ n f (s),

where y2 ∈ I such that

x 1 + (n − 1) y2 = ns, y2 > s.

Thus, (3) is true if

n f (s) − (n − 1) f ( y2 ) + (n − 1) f ( y1 ) ≥ n f (s1 ),

that is
(n − 1) f ( y1 ) + n f (s) ≥ (n − 1) f ( y2 ) + n f (s1 ).
Since
(n − 1) y1 + ns = (n − 1) y2 + ns1
and the decreasingly ordered vector C2n−1 ~ = ( y1 , . . . , y1 , s, . . . , s) majorizes the vec-
~
tor D2n−1 = ( y2 , . . . , y2 , s1 , . . . , s1 ), this inequality follows from Karamata’s inequality
for convex functions.
Similarly, we can prove the Left Half Convex Function Theorem (LHCF-Theorem).
4 Vasile Cîrtoaje

Left Half Convex Function Theorem. Let f be a real function defined on an interval
I and convex on I≤s , where s ∈ int(I). If

f (x) + (n − 1) f ( y) ≥ n f (s)

for all x, y ∈ I so that x ≥ s ≥ y and x + (n − 1) y = ns, then the inequality


a + a + ··· + a 
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (4)
n
holds for all a1 , a2 , . . . , an ∈ I satisfying a1 + a2 + · · · + an = ns. In addition, the
inequality (4) holds for all a1 , a2 , . . . , an ∈ I satisfying a1 + a2 + · · · + an = ns1 , where
s1 ∈ int(I), s1 < s.

From the RHCF-Theorem and the LHCF-Theorem, we find the HCF-Theorem (Half
Convex Function Theorem).

Half Convex Function Theorem. Let f be a real function defined on an interval I


and convex on I≥s or I≤s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x + (n − 1) y = ns.

The following LCRCF-Theorem is also useful to prove some symmetric inequali-


ties.
Left Convex-Right Concave Function Theorem (Vasile Cîrtoaje, 2004). Let a ≤ c
be real numbers, let f be a continuous function defined on I = [a, ∞), strictly convex
on [a, c] and strictly concave on [c, ∞), and let

E(a1 , a2 , . . . , an ) = f (a1 ) + f (a2 ) + · · · + f (an ).

If a1 , a2 , . . . , an ∈ I so that

a1 + a2 + · · · + an = S = const ant,

then
(a) E is minimum for a1 = a2 = · · · = an−1 ≤ an ;
(b) E is maximum for either a1 = a or a < a1 ≤ a2 = · · · = an .
Half Convex Function Method 5

Proof. Without loss of generality, assume that a1 ≤ a2 ≤ · · · ≤ an . Since the sum


E(a1 , a2 , . . . , an ) is a continuous function on the compact set

Λ = {(a1 , a2 , . . . , an ) : a1 + a2 + · · · + an = S, a1 , a2 , . . . , an ∈ I},

E attains its minimum and maximum values.


(a) For the sake of contradiction, suppose that E is minimum at (b1 , b2 , . . . , bn )
with
b1 ≤ b2 ≤ · · · ≤ b n , b1 < bn−1 .
For bn−1 ≤ c, by Jensen’s inequality for strictly convex functions we have

b1 + bn−1
 ‹
f (b1 ) + f (bn−1 ) > 2 f ,
2

while for bn−1 > c, by Karamata’s inequality for strictly concave functions we have

f (bn−1 ) + f (bn ) > f (c) + f (bn−1 + bn − c).

The both results contradict the assumption that E is minimum at (b1 , b2 , . . . , bn ).


(b) For the sake of contradiction, suppose that E is maximum at (b1 , b2 , . . . , bn )
with
a < b1 ≤ b2 ≤ · · · ≤ b n , b2 < bn .
There are three cases to consider.
Case 1: b2 ≥ c. By Jensen’s inequality for strictly concave functions, we have

b2 + b n
 ‹
f (b2 ) + f (bn ) < 2 f .
2

Case 2: b2 < c and b1 + b2 − a ≤ c. By Karamata’s inequality for strictly convex


functions, we have

f (b1 ) + f (b2 ) < f (a) + f (b1 + b2 − a).

Case 3: b2 < c and b1 + b2 − c ≥ a. By Karamata’s inequality for strictly convex


functions, we have

f (b1 ) + f (b2 ) < f (b1 + b2 − c) + f (c).

Clearly, all these results contradict the assumption that E is maximum at (b1 , b2 , . . . , bn ).

Note 1. Let us denote


f (u) − f (s) g(x) − g( y)
g(u) = , h(x, y) = .
u−s x−y
6 Vasile Cîrtoaje

In many applications, it is useful to replace the hypothesis

f (x) + (n − 1) f ( y) ≥ n f (s)

in the RHCF-Theorem, the LHCF-Theorem and the HCF-Theorem by the equivalent


condition
h(x, y) ≥ 0 for all x, y ∈ I so that x + (n − 1) y = ns.

This equivalence is true because

f (x) + (n − 1) f ( y) − n f (s) = [ f (x) − f (s)] + (n − 1)[ f ( y) − f (s)]


= (x − s)g(x) + (n − 1)( y − s)g( y)
n−1
= (x − y)[g(x) − g( y)]
n
n−1
= (x − y)2 h(x, y).
n
Note 2. Assume that f is differentiable on I, and let
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
The desired inequality of Jensen’s type in the RHCF-Theorem, the LHCF-Theorem
and the HCF-Theorem holds true by replacing the hypothesis

f (x) + (n − 1) f ( y) ≥ n f (s)

with the more restrictive condition


H(x, y) ≥ 0 for all x, y ∈ I so that x + (n − 1) y = ns.

To prove this, we will show that the new condition H(x, y) ≥ 0 implies

f (x) + (n − 1) f ( y) ≥ n f (s)

for all x, y ∈ I so that x + (n − 1) y = ns. Write this inequality as

f1 (x) ≥ n f (s),

where  ns − x 
f1 (x) = f (x) + (n − 1) f ( y) = f (x) + (n − 1) f .
n−1
From
 ns − x 
f10 (x) = f 0 (x) − f 0
n−1
= f (x) − f ( y)
0 0

n
= (x − s)H(x, y),
n−1
Half Convex Function Method 7

it follows that f1 is decreasing on I≤s and increasing on I≥s ; therefore,


f1 (x) ≥ f1 (s) = n f (s).

Note 3. From the proof of the RHCF-Theorem, it follows that the RHCF-Theorem,
the LHCF-Theorem and the HCF-Theorem are also valid in the case when f is de-
fined on I \ {u0 }, where u0 ∈ I<s for the RHCF-Theorem, and u0 ∈ I>s for the LHCF-
Theorem.

Note 4. The desired inequalities in the RHCF-Theorem, the LHCF-Theorem and the
HCF-Theorem become equalities for
a1 = a2 = · · · = an = s.
In addition, if there exist x, y ∈ I so that
x + (n − 1) y = ns, f (x) + (n − 1) f ( y) = n f (s), x 6= y,
then the equality holds also for
a1 = x, a2 = · · · = an = y
(or any cyclic permutation). Notice that these equality conditions are equivalent to
x + (n − 1) y = ns, h(x, y) = 0
(x < y for the RHCF-Theorem, and x > y for the LHCF-Theorem).
Note 5. The part (a) in LCRCF-Theorem is also true in the case where I = (a, ∞)
and f (a+ ) = ∞.

Note 6. Similarly, we can extend the weighted Jensen’s inequality to right and left
half convex functions establishing the WRHCF-Theorem, the WLHCF-Theorem and
the WHCF-Theorem (Vasile Cîrtoaje, 2008).
WHCF-Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that
p1 + p2 + · · · + pn = 1, p = min{p1 , p2 , . . . , pn },
and let f be a real function defined on an interval I and convex on I≥s or I≤s , where
s ∈ int(I). The inequality
p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) ≥ f (p1 a1 + p2 a2 + · · · + pn an )
holds for all a1 , a2 , . . . , an ∈ I so that
p1 a1 + p2 a2 + · · · + pn an = s,
if and only if
p f (x) + (1 − p) f ( y) ≥ f (s)
for all x, y ∈ I satisfying
px + (1 − p) y = s.
8 Vasile Cîrtoaje
Half Convex Function Method 9

1.2 Applications

1.1. If a, b, c are real numbers so that a + b + c = 3, then

3(a4 + b4 + c 4 ) + a2 + b2 + c 2 + 6 ≥ 6(a3 + b3 + c 3 ).

1 − 2n
1.2. If a1 , a2 , . . . , an ≥ so that a1 + a2 + · · · + an = n, then
n−2

a13 + a23 + · · · + an3 ≥ n.

−n
1.3. If a1 , a2 , . . . , an ≥ so that a1 + a2 + · · · + an = n, then
n−2

a13 + a23 + · · · + an3 ≥ a12 + a22 + · · · + an2 .

1.4. If a1 , a2 , . . . , an are real numbers so that a1 + a2 + · · · + an = n, then

(n2 − 3n + 3)(a14 + a24 + · · · + an4 − n) ≥ 2(n2 − n + 1)(a12 + a22 + · · · + an2 − n).

1.5. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then

(n2 + n + 1)(a13 + a23 + · · · + an3 − n) ≥ (n + 1)(a14 + a24 + · · · + an4 − n).

1.6. If a, b, c are real numbers so that a + b + c = 3, then


p
(a) a4 + b4 + c 4 − 3 + 2(7 + 3 7)(a3 + b3 + c 3 − 3) ≥ 0;
p
(b) a4 + b4 + c 4 − 3 + 2(7 − 3 7)(a3 + b3 + c 3 − 3) ≥ 0.

1.7. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


k is a positive integer satisfying 3 ≤ k ≤ n + 1, then

a1k + a2k + · · · + ank − n n k−1


• ˜
≥ (n − 1) −1 .
a12 + a22 + · · · + an2 − n n−1
10 Vasile Cîrtoaje

1.8. Let k ≥ 3 be an integer number. If a1 , a2 , . . . , an are nonnegative real numbers


so that a1 + a2 + · · · + an = n, then

a1k + a2k + · · · + ank − n nk−1 − 1


≤ .
a12 + a22 + · · · + an2 − n n−1

1.9. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then

1 1 1
 ‹
n 2
+ + ··· + − n ≥ 4(n − 1)(a12 + a22 + · · · + an2 − n).
a1 a2 an

1.10. If a1 , a2 , . . . , a8 are positive real numbers so that a1 + a2 + · · · + a8 = 8, then

1 1 1
2
+ 2 + · · · + 2 ≥ a12 + a22 + · · · + a82 .
a1 a2 a8

1 1 1
1.11. If a1 , a2 , . . . , an are positive real numbers so that + +···+ = n, then
a1 a2 an
 p 
n−1
a12 + a22 + ··· + an2 −n≥2 1+ (a1 + a2 + · · · + an − n).
n

1.12. If a, b, c, d, e are positive real numbers so that a2 + b2 + c 2 + d 2 + e2 = 5, then


p
1 1 1 1 1 4(1 + 5)
+ + + + −5+ (a + b + c + d + e − 5) ≥ 0.
a b c d e 5

1.13. If a, b, c are nonnegative real numbers, no two of which are zero, then

1 1 1 2 1 1 1
 ‹
+ + ≤ + + .
3a + b + c 3b + c + a 3c + a + b 5 b + c c + a a + b

p
1.14. If a, b, c, d ≥ 3 − 7 so that a + b + c + d = 4, then
1 1 1 1 4
+ + + ≥ .
2 + a2 2 + b2 2 + c 2 2 + d 2 3
Half Convex Function Method 11

p
1.15. If a1 , a2 , . . . , an ∈ [− n, n − 2] so that a1 + a2 + · · · + an = n, then
1 1 1 n
+ + ··· + ≤ .
n + a1 n + a2
2 2
n + an
2 n+1

1.16. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

3−a 3− b 3−c 3
+ + ≥ .
9+a 2 9+ b 2 9+c 2 5

1.17. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


1 1 1 3
+ + ≥ .
1 − a + 2a 2 1 − b + 2b 2 1 − c + 2c 2 2

1.18. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


1 1 1 3
+ + ≥ .
5 + a + a2 5 + b + b2 5 + c + c 2 7

1.19. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then


1 1 1 1 1
+ + + ≤ .
10 + a + a 2 10 + b + b 2 10 + c + c 2 10 + d + d 2 3

1.20. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


If
1
k ≥1− ,
n
then
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2
2 2
1 + kan
2 1+k

1.21. Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an = n. If


n−1
0<k≤ ,
n2 −n+1
then
1 1 1 n
+ + ··· + ≤ .
1 + ka1 1 + ka2
2 2
1 + kan2 1+k
12 Vasile Cîrtoaje

1.22. Let a1 , a2 , . . . , an be nonnegative numbers so that a1 + a2 + · · · + an = n. If

n2
k≥ ,
4(n − 1)
then
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + · · · + ≥ 0.
a12 + k a22 + k an2 + k

1.23. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
a1 − 1 a2 − 1 an − 1
+ + · · · + ≥ 0.
(n − 2a1 )2 (n − 2a2 )2 (n − 2an )2

1.24. If a1 , a2 , . . . , an are nonnegative real numbers so that


n
a1 + a2 + · · · + an = n, a1 , a2 , . . . , an > −k, k ≥1+ p ,
n−1
then
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (an + k)2

2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


1.25. Let a1 , as
2n − 1
If 0 < k ≤ 1 + , then
n−1
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≤ 0.
(a1 + k)2 (a2 + k)2 (an + k)2

p
1.26. If a1 , a2 , . . . , an ≥ n − 1 − n2 − n + 1 so that a1 + a2 + · · · + an = n, then

a12 − 1 a22 − 1 an2 − 1


+ + ··· + ≤ 0.
(a1 + 2)2 (a2 + 2)2 (an + 2)2

1.27. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


(n − 1)(2n − 1)
If k ≥ , then
n2
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2
3 3
1 + kan3 1+k
Half Convex Function Method 13

1.28. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


n−1
If 0 < k ≤ 2 , then
n − 2n + 2
1 1 1 n
+ + ··· + ≤ .
1 + ka1 1 + ka2
3 3
1 + kan
3 1+k

1.29. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


n2
If k ≥ , then
n−1
v v v
t a1 t a2 t an n
+ + ··· + ≤p .
k − a1 k − a2 k − an k−1

1.30. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
2 2 2
n−a1 + n−a2 + · · · + n−an ≥ 1.

1.31. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(3a2 + 1)(3b2 + 1)(3c 2 + 1)(3d 2 + 1) ≤ 256.

1.32. If a, b, c, d, e ≥ −1 so that a + b + c + d + e = 5, then

(a2 + 1)(b2 + 1)(c 2 + 1)(d 2 + 1)(e2 + 1) ≥ (a + 1)(b + 1)(c + 1)(d + 1)(e + 1).

1.33. Let a1 , a2 , . . . , an be positive numbers so that a1 + a2 + · · · + an = n. If


p v p
2 n−1 2 n−1
t
k≤ +2 , k ≤ 3,
n n
then
p 1 1 1
k( a1 + a2 + · · · + an ) + p + p + · · · + p ≥ (k + 1)n.
p p
a1 a2 an

1.34. If a1 , a2 , . . . , an (n ≥ 3) are positive numbers so that a1 + a2 + · · · + an = 1,


then
1 n
     
1 1 1 p
‹
p p p
p − a1 p − a2 · · · p − an ≥ n− p .
a1 a2 an n
14 Vasile Cîrtoaje

1.35. Let a1 , a2 , . . . , an be positive real numbers so that a1 + a2 + · · · + an = n. If


 p 2
2 n−1
k ≤ 1+ ,
n

then
1 1 1
 ‹ ‹  ‹
ka1 + ka2 + · · · kan + ≥ (k + 1)n .
a1 a2 an

1.36. If a, b, c, d are nonzero real numbers so that


−1
a, b, c, d ≥ , a + b + c + d = 4,
2
then
1 1 1 1 1 1 1 1
 ‹
3 2 + 2 + 2 + 2 + + + + ≥ 16.
a b c d a b c d

1.37. If a1 , a2 , . . . , an are nonnegative real numbers so that a12 + a22 + · · · + an2 = n,


then s
n
a1 + a2 + · · · + an − n +
3 3 3
(a1 + a2 + · · · + an − n) ≥ 0.
n−1

1.38. If a, b, c, d, e are nonnegative real numbers so that a2 + b2 + c 2 + d 2 + e2 = 5,


then
1 1 1 1 1
+ + + + ≤ 1.
7 − 2a 7 − 2b 7 − 2c 7 − 2d 7 − 2e

1.39. Let 0 ≤ a1 , a2 , . . . , an < k so that a12 + a22 + · · · + an2 = n. If


s
n
1< k ≤1+ ,
n−1
then
1 1 1 n
+ + ··· + ≥ .
k − a1 k − a2 k − an k−1

1.40. If a, b, c are nonnegative real numbers, no two of which are zero, then
v v v
t 48a t 48b t 48c
1+ + 1+ + 1+ ≥ 15.
b+c c+a a+b
Half Convex Function Method 15

1.41. If a, b, c are nonnegative real numbers, then


v v v
t 3a2 t 3b2 t 3c 2
+ + ≤ 1.
7a2 + 5(b + c)2 7b2 + 5(c + a)2 7c 2 + 5(a + b)2

1.42. If a, b, c are nonnegative real numbers, then


v v v
t a2 t b2 t c2
+ + ≥ 1.
a2 + 2(b + c)2 b2 + 2(c + a)2 c 2 + 2(a + b)2

1.43. Let a, b, c be nonnegative real numbers, no two of which are zero. If

ln 3
k ≥ k0 , k0 = − 1 ≈ 0.585,
ln 2
then ‹k ‹k ‹k
2a 2b 2c
  
+ + ≥ 3.
b+c c+a a+b

p
1.44. If a, b, c ∈ [1, 7 + 4 3], then
v v v
t 2a t 2b t 2c
+ + ≥ 3.
b+c c+a a+b

1.45. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 2
0 < k ≤ k0 , k0 = ≈ 1.71,
ln 3 − ln 2
then
a k (b + c) + b k (c + a) + c k (a + b) ≤ 6.

1.46. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


‚v v Œ
p p p ta + b tb+c s
c+a
a+ b+ c − 3 ≥ 13 + + −3 .
2 2 2
16 Vasile Cîrtoaje

1.47. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k > 2, then

a+b k b+c k  c + a k
 ‹  ‹
a + b +c +3≥2
k k k
+2 +2 .
2 2 2

1.48. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
‚v v v Œ
p tn − a tn − a tn − a
1 2 n
a1 + a2 + · · · + an + n(k − 1) ≤ k + + ··· +
p p
,
n−1 n−1 n−1

where p p p
k = ( n − 1)( n + n − 1).

1.49. If a, b, c are the lengths of the sides of a triangle so that a + b + c = 3, then


1 1 1 p 2 2 2
 ‹
+ + − 3 ≥ 4(2 + 3) + + −3 .
a+b−c b+c−a c+a−b a+b b+c c+a

1.50. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≤ 5.


If p
29 + 761
k ≥ k0 , k0 = ≈ 5.66,
10
then X 1 5
≥ .
ka1 + a2 + a3 + a4 + a5
2
k+4

1.51. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≤ 5.


If p
11 − 101
0 < k ≤ k0 , k0 = ≈ 0.095,
10
then X 1 5
≥ .
ka1 + a2 + a3 + a4 + a5
2
k+4

1.52. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


If
1
0<k≤ ,
n+1
then
a1 a2 an n
+ + ··· + ≥ .
ka1 + a2 + · · · + an a1 + ka2 + · · · + an
2 2
a1 + a2 + · · · + kan2 k+n−1
Half Convex Function Method 17

7
1.53. If a1 , a2 , a3 , a4 , a5 ≤so that a1 + a2 + a3 + a4 + a5 = 5, then
2
a1 a2 a3 a4 a5
+ + + + ≤ 1.
a12 − a1 + 5 a22 − a2 + 5 a32 − a3 + 5 a42 − a4 + 5 a52 − a5 + 5

1.54. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If
1
0<k≤ 1
,
1 + 4(n−1)2

then
a12 a22 an2 n
+ + ··· + ≥ .
ka12 + a2 + · · · + an a1 + ka22 + · · · + an a1 + a2 + · · · + kan2 k+n−1

1.55. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


If k ≥ n − 1, then

a12 a22 an2 n


+ + ··· + ≤ .
ka12 + a2 + · · · + an a1 + ka22 + · · · + an a1 + a2 + · · · + kan2 k+n−1

1
1.56. Let a1 , a2 , . . . , an ∈ [0, n] so that a1 + a2 + · · · + an ≥ n. If 0 < k ≤ , then
n
a1 − 1 a2 − 1 an − 1
+ + ··· + ≥ 0.
ka12 + a2 + · · · + an a1 + ka2 + · · · + an
2
a1 + a2 + · · · + kan2

1.57. If a, b, c are positive real numbers so that abc = 1, then


p p p
a2 − a + 1 + b2 − b + 1 + c 2 − c + 1 ≥ a + b + c.

1
1.58. If a, b, c, d ≥ p so that abcd = 1, then
1+ 6
1 1 1 1 4
+ + + ≤ .
a+2 b+2 c+2 d +2 3

1.59. If a, b, c are positive real numbers so that abc = 1, then

a2 + b2 + c 2 − 3 ≥ 2(ab + bc + ca − a − b − c).
18 Vasile Cîrtoaje

1.60. If a, b, c are positive real numbers so that abc = 1, then

a2 + b2 + c 2 − 3 ≥ 18(a + b + c − ab − bc − ca).

1.61. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


p 1 1 1
 ‹
a1 + a2 + · · · + an − n ≥ 6 3 a1 + a2 + · · · + an −
2 2 2
− − ··· − .
a1 a2 an

1.62. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that a1 a2 · · · an = 1, then

(n − 1)(a12 + a22 + · · · + an2 ) + n(n + 3) ≥ (2n + 2)(a1 + a2 + · · · + an ).

1.63. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If p


and q are nonnegative real numbers so that p + q ≥ n − 1, then

1 1 1 n
+ + ··· + ≥ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2 1+p+q

1.64. Let a, b, c, d be positive real numbers so that abcd = 1. If p and q are non-
negative real numbers so that p + q = 3, then

1 1 1 1
+ + + ≥ 1.
1 + pa + qa3 1 + pb + qb3 1 + pc + qc 3 1 + pd + qd 3

1.65. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1 1
+ + ··· + ≥ 1.
1 + a1 + · · · + a1
n−1
1 + a2 + · · · + a2
n−1
1 + an + · · · + ann−1

1.66. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If

k ≥ n2 − 1,

then
1 1 1 n
+p + ··· + p ≥p .
1 + ka1 1 + ka2 1 + kan 1+k
p
Half Convex Function Method 19

1.67. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If p, q ≥ 0


1
so that 0 < p + q ≤ , then
n−1
1 1 1 n
+ + ··· + ≤ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q

1.68. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If


2n − 1
0<k≤ ,
(n − 1)2
then
1 1 1 n
+p + ··· + p ≤p .
1 + ka1 1 + ka2 1 + kan 1+k
p

1.69. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


v v v
t 2n − 1 t 4 2n − 1 t 2n − 1 1
a1 +
4
+ a + +· · ·+ a 4+ ≥ (a1 +a2 +· · ·+an )2 .
(n − 1)2 2
(n − 1)2 n
(n − 1)2 n−1

1.70. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1 1
 ‹
a1 + a2 + · · · + an + n(n − 2) ≥ (n − 1)
n−1 n−1 n−1
+ + ··· + .
a1 a2 an

1.71. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If k ≥ n,


then
1 1 1
 ‹
a1 + a2 + · · · + an + kn ≥ (k + 1)
k k k
+ + ··· + .
a1 a2 an

1.72. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 a1 1 a2 1 an
 ‹  ‹  ‹
1− + 1− + ··· + 1 − ≤ n − 1.
n n n

1.73. If a, b, c are positive real numbers so that abc = 1, then


1 1 1
p + p + p ≤ 1.
1 + 1 + 3a 1 + 1 + 3b 1 + 1 + 3c
20 Vasile Cîrtoaje

1.74. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


1 1 1 1
+ + ··· + ≥ .
1+ 1 + 4n(n − 1)a1 1+ 1 + 4n(n − 1)a2 1+ 1 + 4n(n − 1)an
p p p
2

1.75. If a, b, c are positive real numbers so that abc = 1, then

a6 b6 c6
+ + ≥ 1.
1 + 2a5 1 + 2b5 1 + 2c 5

1.76. If a, b, c are positive real numbers so that abc = 1, then


p p p
25a2 + 144 + 25b2 + 144 + 25c 2 + 144 ≤ 5(a + b + c) + 24.

1.77. If a, b, c are positive real numbers so that abc = 1, then


p p p
16a2 + 9 + 16b2 + 9 + 16c 2 + 9 ≥ 4(a + b + c) + 3.

1.78. If ABC is a triangle, then


 ‹  ‹  ‹
A B C
sin A 2 sin − 1 + sin B 2 sin − 1 + sin C 2 sin − 1 ≥ 0.
2 2 2

1.79. If ABC is an acute or right triangle, then


 ‹  ‹  ‹
A B C
sin 2A 1 − 2 sin + sin 2B 1 − 2 sin + sin 2C 1 − 2 sin ≥ 0.
2 2 2

1.80. If a, b, c, d are real numbers so that a + b + c + d = 4, then


a b c d
+ + + ≤ 1.
a2 − a + 4 b2 − b + 4 c 2 − c + 4 d 2 − d + 4

1.81. Let a, b, c be nonnegative real numbers so that a + b + c = 2. If


ln 2
k0 ≤ k ≤ 3, k0 = ≈ 1.71,
ln 3 − ln 2
then
a k (b + c) + b k (c + a) + c k (a + b) ≤ 2.
Half Convex Function Method 21

1.82. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then


1 1 1
 ‹
(n + 1) 2
+ + ··· + ≥ 4(n + 2)(a12 + a22 + · · · + an2 ) + n(n2 − 3n − 6).
a1 a2 an

1.83. If a, b, c, d, e are positive real numbers such that a + b + c + d + e = 5, then


1 1 1 1 1
27( + + + + ) ≥ 4(a3 + b3 + c 3 + d 3 + e3 ) + 115.
a b c d e

1.84. If a, b, c are nonnegative real numbers so that a + b + c = 12, then

(a2 + 10)(b2 + 10)(c 2 + 10) ≥ 13310.

1.85. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
(n2 − 2n + 2)n
(a12 + 1)(a22 + 1) · · · (an2 + 1) ≥ .
(n − 1)2n−2

1.86. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

(a2 + 2)(b2 + 2)(c 2 + 2) ≤ 44.

1.87. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


169
(a2 + 1)(b2 + 1)(c 2 + 1) ≤ .
16

1.88. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


121
(2a2 + 1)(2b2 + 1)(2c 2 + 1) ≤ .
4

1.89. If a, b, c are nonnegative real numbers so that a + b + c ≥ k0 , where


3
Æ p
k0 = 66 + 10 105 ≈ 4.867,
8
then ‹2
a+b+c
Æ 
3
(a2 + 1)(b2 + 1)(c 2 + 1) ≤ + 1.
3
22 Vasile Cîrtoaje

1.90. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + 3)(b2 + 3)(c 2 + 3)(d 2 + 3) ≤ 513.

1.91. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + 2)(b2 + 2)(c 2 + 2)(d 2 + 2) ≤ 144.

1.92. If a, b, c, d are nonnegative real numbers such that

a + b + c + d = 4,

then
a b c d 4
+ 3 + 3 + 3 ≤ .
3a3 + 2 3b + 2 3c + 2 3d + 2 5

1.93. If a1 , a2 , ..., an are nonnegative real numbers such that a1 + a2 + · · · + an = 1,


then
1
a13 + a23 + · · · + an3 ≤ + a14 + a24 + · · · + an4 .
8
Half Convex Function Method 23

1.3 Solutions

P 1.1. If a, b, c are real numbers so that a + b + c = 3, then

3(a4 + b4 + c 4 ) + a2 + b2 + c 2 + 6 ≥ 6(a3 + b3 + c 3 ).

(Vasile C., 2006)

Solution. Write the inequality as

a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
f (u) = 3u4 − 6u3 + u2 , u ∈ R.
From
f 00 (u) = 2(18u2 − 18u + 1),
it follows that f 00 (u) > 0 for u ≥ 1, hence f is convex on [s, ∞). By the RHCF-
Theorem, it suffices to show that f (x) + 2 f ( y) ≥ 3 f (1) for all real x, y so that
x + 2 y = 3. Let
E = f (x) + 2 f ( y) − 3 f (1).
We have

E = [ f (x) − f (1)] + 2[ f ( y) − f (1)]


= (3x 4 − 6x 3 + x 2 + 2) + 2(3 y 4 − 6 y 3 + y 2 + 2)
= (x − 1)(3x 3 − 3x 2 − 2x − 2) + 2( y − 1)(3 y 3 − 3 y 2 − 2 y − 2)
= (x − 1)[(3x 3 − 3x 2 − 2x − 2) − (3 y 3 − 3 y 2 − 2 y − 2)]
= (x − 1)[3(x 3 − y 3 ) − 3(x 2 − y 2 ) − 2(x − y)]
= (x − 1)(x − y)[3(x 2 + x y + y 2 ) − 3(x + y) − 2]
(x − 1)2 [27(x 2 + x y + y 2 ) − 9(x + y)(x + 2 y) − 2(x + 2 y)2 ]
=
6
(x − 1) (4x − y)
2 2
= ≥ 0.
6
1 4
The equality holds for a = b = c = 1, and also for a = and b = c = (or any
3 3
cyclic permutation).
Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an are real numbers so that a1 + a2 + · · · + an = n, then

n−1
(a12 − a1 )2 + (a22 − a2 )2 + · · · + (an2 − an )2 ≥ (a12 + a22 + · · · + an2 − n),
n2 − 3n + 3
24 Vasile Cîrtoaje

with equality for a1 = a2 = · · · = an = 1, and also for


1 n−2
a1 = , a2 = a3 = · · · = an = 1 +
n2 − 3n + 3 n2 − 3n + 3
(or any cyclic permutation).

1 − 2n
P 1.2. If a1 , a2 , . . . , an ≥ so that a1 + a2 + · · · + an = n, then
n−2
a13 + a23 + · · · + an3 ≥ n.

(Vasile C., 2000)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1 − 2n
f (u) = u3 , u≥.
n−2
From f 00 (u) = 6u, it follows that f is convex on [s, ∞). By the RHCF-Theorem and
1 − 2n
Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ≥ so that x+(n−1) y =
n−2
n. We have
f (u) − f (1)
g(u) = = u2 + u + 1,
u−1
g(x) − g( y) (n − 2)x + 2n − 1
h(x, y) = = x + y +1= ≥ 0.
x−y n−1
From x + (n − 1) y = n and h(x, y) = 0, we get
1 − 2n n+1
x= , y= .
n−2 n−2
Therefore, according to Note 4, the equality holds for a1 = a2 = · · · = an = 1, and
also for
1 − 2n n+1
a1 = , a2 = a3 = · · · = an =
n−2 n−2
(or any cyclic permutation).

−n
P 1.3. If a1 , a2 , . . . , an ≥ so that a1 + a2 + · · · + an = n, then
n−2
a13 + a23 + · · · + an3 ≥ a12 + a22 + · · · + an2 .
Half Convex Function Method 25

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−n
f (u) = u3 − u2 , u≥
.
n−2
From f 00 (u) = 6u − 2, it follows that f is convex on [s, ∞). According to the RHCF-
−n
Theorem and Note 1, it suffices to show that h(x, y) ≥ 0 for x, y ≥ so that
n−2
x + (n − 1) y = n. We have
f (u) − f (1)
g(u) = = u2 ,
u−1
g(x) − g( y) (n − 2)x + n
h(x, y) = =x+y= ≥ 0.
x−y n−1
From x + (n − 1) y = n and h(x, y) = 0, we get
−n n
x= , y= .
n−2 n−2
Therefore, in accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1,
and also for
−n n
a1 = , a2 = a3 = · · · = a n =
n−2 n−2
(or any cyclic permutation).

P 1.4. If a1 , a2 , . . . , an are real numbers so that a1 + a2 + · · · + an = n, then

(n2 − 3n + 3)(a14 + a24 + · · · + an4 − n) ≥ 2(n2 − n + 1)(a12 + a22 + · · · + an2 − n).

(Vasile C., 2009)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = (n2 − 3n + 3)u4 − 2(n2 − n + 1)u2 , u ∈ I = R.
For u ≥ s = 1, we have
1 00
f (u) = 3(n2 − 3n + 3)u2 − (n2 − n + 1)
4
≥ 3(n2 − 3n + 3) − (n2 − n + 1) = 2(n − 2)2 ≥ 0;
26 Vasile Cîrtoaje

therefore, f is convex on I≥s . By the RHCF-Theorem and Note 1, it suffices to show


that h(x, y) ≥ 0 for x, y ∈ R so that x + (n − 1) y = n, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
We have

g(u) = (n2 − 3n + 3)(u3 + u2 + u + 1) − 2(n2 − n + 1)(u + 1)

and

h(x, y) = (n2 − 3n + 3)(x 2 + x y + y 2 + x + y + 1) − 2(n2 − n + 1)


= [(n2 − 3n + 3) y − n2 + n + 1]2 ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1, and also for


2 2n − 4
a1 = −1 + , a2 = a3 = · · · = an = 1 +
n2 − 3n + 3 n2 − 3n + 3
(or any cyclic permutation).

P 1.5. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then

(n2 + n + 1)(a13 + a23 + · · · + an3 − n) ≥ (n + 1)(a14 + a24 + · · · + an4 − n).

(Vasile C., 2009)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = (n2 + n + 1)u3 − (n + 1)u4 , u ∈ I = [0, n].
The function f is convex on I≤s because

f 00 (u) = 6u[n2 + n + 1 − 2(n + 1)u] ≥ 6u[n2 + n + 1 − 2(n + 1)]


= 6(n2 − n − 1)u ≥ 0.

By the LHCF-Theorem and Note 1, it suffices to show that h(x, y) ≥ 0 for x, y ≥ 0


so that x + (n − 1) y = n, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
Half Convex Function Method 27

We have

g(u) = (n2 + n + 1)(u2 + u + 1) − (n + 1)(u3 + u2 + u + 1)


= −(n + 1)u3 + n2 (u2 + u + 1)

and

h(x, y) = −(n + 1)(x 2 + x y + y 2 ) + n2 (x + y + 1)


= −(n + 1)(x 2 + x y + y 2 ) + n(x + y)[x + (n − 1) y] + [x + (n − 1) y]2
= (n2 + n − 3)x y + 2n(n − 2) y 2 ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = n, a2 = a3 = · · · = an = 0

(or any cyclic permutation).

P 1.6. Let a, b, c be real numbers so that a + b + c = 3. If


p p
−14 − 6 7 ≤ k ≤ −14 + 6 7,

then
a4 + b4 + c 4 − 3 ≥ k(a3 + b3 + c 3 − 3).

(Vasile C., 2009)

Solution. Write the desired inequalities as


a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
f (u) = u4 − ku3 , u ∈ R.
From
f 00 (u) = 6u(2u2 − k),
it follows that f 00 (u) > 0 for u ≥ 1, hence f is convex on [s, ∞). By the RHCF-
Theorem, it suffices to show that f (x) + 2 f ( y) ≥ 3 f (1) for all real x, y so that
x + 2 y = 3. Using Note 1, we only need to show that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
We have

g(u) = u3 + u2 + u + 1 − k(u2 + u + 1) + u + 1 = u3 + (1 − k)(u2 + u + 1),


28 Vasile Cîrtoaje

h(x, y) = x 2 + x y + y 2 + (1 − k)(x + y + 1) = 3 y 2 − (10 − k) y + 13 − 4k


p p
10 − k 2 (6 7 + 14 + k)(6 7 − 14 − k)
 ‹
=3 y− + ≥ 0.
6 12
p
The equality holds for a = b = c = 1. If k = −14 − 6 7, then the equality holds
also for
p p
a = −5 − 2 7, b = c =4+ 7
p
(or any cyclic permutation). If k = −14 + 6 7, then the equality holds also for
p p
a = −5 + 2 7, b = c =4− 7

(or any cyclic permutation).


Remark. Similarly, we can prove the following generalization:
• Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an = n. If k1 ≤ k ≤ k2 ,
where p
−2(n2 − n + 1) − 2 3(n2 − n + 1)(n2 − 3n + 3)
k1 = ,
(n − 2)2
p
−2(n2 − n + 1) + 2 3(n2 − n + 1)(n2 − 3n + 3)
k2 = ,
(n − 2)2
then
a14 + a24 + · · · + an4 − n ≥ k(a13 + a23 + · · · + an3 − n).
The equality holds for a1 = a2 = · · · = an = 1. If k ∈ {k1 , k2 }, then the equality
holds also for
−2(n2 − 3n + 1) + (n − 1)(n − 2)k
a1 = ,
2(n2 − 3n + 3)
2(n2 − n − 1) − (n − 2)k
a2 = a3 = · · · = an =
2(n2 − 3n + 3)
(or any cyclic permutation).

P 1.7. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


k is a positive integer satisfying 3 ≤ k ≤ n + 1, then

a1k + a2k + · · · + ank − n n k−1


• ˜
≥ (n − 1) −1 .
a12 + a22 + · · · + an2 − n n−1

(Vasile C., 2012)


Half Convex Function Method 29

Solution. Denote
n k−1 n k−2  n k−3
• ˜ 
m = (n − 1) −1 = + + · · · + 1,
n−1 n−1 n−1
and write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = uk − mu2 , u ∈ [0, n].
We will show that f is convex on [1, n]. Since

f 00 (u) = k(k − 1)uk−2 − 2m ≥ k(k − 1) − 2m,

we need to show that


k(k − 1)  n k−2  n k−3
≥ + + · · · + 1.
2 n−1 n−1
Since n ≥ k − 1, this inequality is true if

k(k − 1) k − 1 k−2 k − 1 k−3


 ‹  ‹
≥ + + · · · + 1.
2 k−2 k−2
By Bernoulli’s inequality, we have
‹j
k−1 1 1 k−1

= j ≤ j
= , j = 0, 1, . . . , k − 2.
k−2 1− 1 1− k− j−1
k−1 k−1

Therefore, it suffices to show that

k(k − 1) 1 1
 ‹
≥ (k − 1) 1 + + · · · + .
2 2 k−1

This is true if
k 1 1
≥ 1 + + ··· + ,
2 2 k−1
which can be easily proved by induction. According to the RHCF-Theorem and Note
1, we only need to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n,
where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have

(uk − 1) − m(u2 − 1)
g(u) = = (uk−1 + uk−2 + · · · + 1) − m(u + 1),
u−1
30 Vasile Cîrtoaje
 
x k−1 − y k−1 x k−2 − y k−1
h(x, y) = + + ··· + 1 − m
x−y x−y
k−2  j+1 
X x − y j+1  n  j
= − .
j=1
x−y n−1
h n i
It suffices to show that f j ( y) ≥ 0 for y ∈ 0, and j = 1, 2, . . . , k − 2, where
n−1
 n j
f j ( y) = x j + x j−1 y + · · · + x y j−1 + y j − , x = n − (n − 1) y.
n−1
For j = 1, we have

n (n − 2)x
f1 ( y) = x + y − = ≥ 0.
n−1 n−1
For j ≥ 2, from x 0 = −(n − 1) and n − 1 ≥ k − 2 ≥ j, we get

f j0 ( y) = −(n − 1)[ j x j−1 + ( j − 1)x j−2 y + · · · + y j−1 ] + x j−1 + 2x j−2 y + · · · + j y j−1


≤ − j[ j x j−1 + ( j − 1)x j−2 y + · · · + y j−1 ] + x j−1 + 2x j−2 y + · · · + j y j−1
= −( j · j − 1)x j−1 − [ j · ( j − 1) − 2]x j−2 y − · · · − ( j · 2 − j + 1)x y j−2 ≤ 0.
n
As a consequence, f j is decreasing, hence it is minimum for y = (when
n−1
x = 0):  n 
f j ( y) ≥ f j = 0.
n−1
From x + (n − 1) y = n and h(x, y) = 0, we get
n
x = 0, y= .
n−1
Therefore, the equality holds for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
Remark. For k = 3 and k = 4, we get the following statements (Vasile C. , 2002):
• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then

(n − 1)(a13 + a23 + · · · + an3 − n) ≥ (2n − 1)(a12 + a22 + · · · + an2 − n),

which is equivalent to

3 X 3n − 1 X
ai a j ak + n2 ≥ ai a j ,
n − 2 1≤i< j<k≤n n − 1 1≤i< j≤n
Half Convex Function Method 31

with equality for a1 = a2 = · · · = an = 1, and also for


n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

• If a1 , a2 , . . . , an (n ≥ 3) are nonnegative real numbers so that

a1 + a2 + · · · + an = n,

then

(n − 1)2 (a14 + a24 + · · · + an4 − n) ≥ (3n2 − 3n + 1)(a12 + a22 + · · · + an2 − n),

with equality for a1 = a2 = · · · = an = 1, and also for


n
a1 = 0, a2 = a3 = · · · = a n =
n−1
(or any cyclic permutation).

P 1.8. Let k ≥ 3 be an integer number. If a1 , a2 , . . . , an are nonnegative real numbers


so that a1 + a2 + · · · + an = n, then

a1k + a2k + · · · + ank − n nk−1 − 1


≤ .
a12 + a22 + · · · + an2 − n n−1

(Vasile C., 2012)

Solution. Denote
nk−1 − 1
m= = nk−2 + nk−3 + · · · + 1,
n−1
and write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = mu2 − uk , u ∈ [0, n].
We will show that f is convex on [0, 1]. Since

f 00 (u) = 2m − k(k − 1)uk−2 ≥ 2m − k(k − 1),

we need to show that


k(k − 1)
nk−2 + nk−3 + · · · + 1 ≥ .
2
32 Vasile Cîrtoaje

This is true if
k(k − 1)
2k−2 + 2k−3 + · · · + 1 ≥ ,
2
which is equivalent to
k(k − 1)
2k−1 − 1 ≥ ,
2
2k ≥ k2 − k + 2.
Since
 ‹  ‹  ‹
k k k
2 = (1 + 1) ≥ 1 +
k k
+ +
1 2 3
k(k − 1) k(k − 1)(k − 2)
=1+k+ + ,
2 6
it suffices to show that
k(k − 1) k(k − 1)(k − 2)
1+k+ + ≥ k2 − k + 2,
2 6
which reduces to
(k − 1)(k − 2)(k − 3) ≥ 0.
According to the LHCF-Theorem and Note 1, we only need to show that h(x, y) ≥ 0
for x, y ≥ 0 so that x + (n − 1) y = n, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

We have

m(u2 − 1) − (uk − 1)
g(u) = = m(u + 1) − (uk−1 + uk−2 + · · · + 1)
u−1
and
x k−1 − y k−1 x k−2 − y k−1
h(x, y) = m − − − ··· − 1
x−y x−y
 k−1 k−1
  k−2 k−2
  2 2

x − y x − y x − y
= nk−2 − + nk−3 − + ··· + n − .
x−y x−y x−y
It suffices to show that

x j+1 − y j+1
nj ≥ , j = 1, 2, . . . , k − 2.
x−y

We will show that


x j+1 − y j+1
n j ≥ (x + y) j ≥ .
x−y
Half Convex Function Method 33

The left inequality is true since

n − (x + y) = x + (n − 1) y − (x + y) = (n − 2) y ≥ 0.

The right inequality is also true since


 ‹  ‹
j j−1 j
(x + y) = x +
j j
x y + ··· + x y j−1 + y j
1 j−1

and
x j+1 − y j+1
= x j + x j−1 y + · · · + x y j−1 + y j .
x−y
The equality holds for a1 = n and a2 = a3 = · · · = an = 0 (or any cyclic permuta-
tion).
Remark. For k = 3 and k = 4, we get the following statements (Vasile C. , 2002):
• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then

a13 + a23 + · · · + an3 − n ≤ (n + 1)(a12 + a22 + · · · + an2 − n),

with equality for a1 = a2 = · · · = an = 1, and also for

a1 = n, a2 = a3 = · · · = an = 0

(or any cyclic permutation).

• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then

a14 + a24 + · · · + an4 − n ≤ (n2 + n + 1)(a12 + a22 + · · · + an2 − n),

with equality for a1 = a2 = · · · = an = 1, and also for

a1 = n, a2 = a3 = · · · = an = 0

(or any cyclic permutation).

P 1.9. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then

1 1 1
 ‹
n 2
+ + ··· + − n ≥ 4(n − 1)(a12 + a22 + · · · + an2 − n).
a1 a2 an

(Vasile C., 2004)


34 Vasile Cîrtoaje

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
n2
f (u) = − 4(n − 1)u2 , u ∈ I = (0, n).
u
For u ∈ (0, 1], we have

2n2
f 00 (u) = − 8(n − 1) ≥ 2n2 − 8(n − 1) = 2(n − 2)2 ≥ 0.
u3
Thus, f is convex on I≤s . By the LHCF-Theorem and Note 1, it suffices to show that
h(x, y) ≥ 0 for x, y > 0 so that x + (n − 1) y = n, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
We have
−n2
g(u) = − 4(n − 1)(u + 1)
u
and
n2 [x + (n − 1) y]2 [x − (n − 1) y]2
h(x, y) = − 4(n − 1) = − 4(n − 1 = .
xy xy xy
In accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1, and also
for
n n
a1 = , a2 = a3 = · · · = an =
2 2n − 2
(or any cyclic permutation).

P 1.10. If a1 , a2 , . . . , a8 are positive real numbers so that a1 + a2 + · · · + a8 = 8, then


1 1 1
2
+ 2 + · · · + 2 ≥ a12 + a22 + · · · + a82 .
a1 a2 a8

(Vasile C., 2007)

Solution. Write the inequality as


a1 + a2 + · · · + a8
f (a1 ) + f (a2 ) + · · · + f (a8 ) ≥ 8 f (s), s= = 1,
8
where
1
f (u) = − u2 , u ∈ (0, 8).
u2
Half Convex Function Method 35

For u ∈ (0, 1], we have

6
f 00 (u) = − 2 ≥ 6 − 2 > 0.
u4
Thus, f is convex on (0, s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y > 0 so that x + 7 y = 8, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

We have
1 1
g(u) = −u − 1 − − 2
u u
and
1 x+y
h(x, y) = −1 + + 2 2.
xy x y
From 8 = x + 7 y ≥ 2 7x y, we get x y ≤ 16/7. Therefore,
p

1 7(x + y) 112 y 2 − 170 y + 72


h(x, y) ≥ −1 + + =
xy 16x y 16x y
112 y − 176 y + 72 14 y − 22 y + 9
2 2
> = > 0.
16x y 2x y

The equality holds for a1 = a2 = · · · = a8 = 1.


Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that a1 + a2 + · · · + an = n,
then
1 1 1 8 2 
2
+ 2 + ··· + 2 + 8 − n ≥ a1 + a22 + · · · + an2 .
a1 a2 an n

1 1 1
P 1.11. If a1 , a2 , . . . , an are positive real numbers so that + +···+ = n, then
a1 a2 an
 p 
n−1
a12 + a22 + ··· + an2 −n≥2 1+ (a1 + a2 + · · · + an − n).
n

(Vasile C., 2006)

Solution. Replacing each ai by 1/ai , we need to prove that


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
36 Vasile Cîrtoaje

where p
1 2k n−1
f (u) = 2 − , k =1+ , u ∈ (0, n).
u u n
For u ∈ (0, 1], we have
p
6 − 4ku 6 − 4k 2( n − 1 − 1)2
f (u) =
00
≥ = ≥ 0.
u4 u4 nu4
Thus, f is convex on (0, s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y > 0 so that x + (n − 1) y = n, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
−1 2k − 1
g(u) = +
u2 u
and
1 1 1
 ‹
h(x, y) = + + 1 − 2k .
xy x y
We only need to show that
1 1
+ ≥ 2k − 1.
x y
Indeed, using the Cauchy-Schwarz inequality, we get
p p
1 1 (1 + n − 1)2 (1 + n − 1)2
+ ≥ = = 2k − 1,
x y x + (n − 1) y n
p
with equality for x = n − 1 y. From x + (n − 1) y = n and h(x, y) = 0, we get
n n
x= p , y= p .
1+ n−1 n−1+ n−1
In accordance with Note 4, the original equality holds for a1 = a2 = · · · = an = 1,
and also for
p p
1+ n−1 n−1+ n−1
a1 = , a2 = a3 = · · · = an =
n n
(or any cyclic permutation).

P 1.12. If a, b, c, d, e are positive real numbers so that a2 + b2 + c 2 + d 2 + e2 = 5, then


p
1 1 1 1 1 4(1 + 5)
+ + + + −5+ (a + b + c + d + e − 5) ≥ 0.
a b c d e 5
(Vasile C., 2006)
Half Convex Function Method 37

p p p p p
Solution. Replacing a, b, c, d, e by a, b, c, d, e, respectively, we need to
prove that

a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where p
1 p 4(1 + 5)
f (u) = p + k u, k= ≈ 2.59, u ∈ (0, 5).
u 5
For u ∈ (0, 1], we have
3 − ku
f 00 (u) = p > 0;
4u2 u
therefore, f is convex on (0, s]. By the LHCF-Theorem and Note 1, it suffices to
show that h(x, y) ≥ 0 for x, y > 0 so that x + 4 y = 5. We have
p
f (u) − f (1) k u − 1
g(u) = = p
u−1 u+ u

and p p p
g(x) − g( y) x + y +1−k xy
h(x, y) = =p p p p p .
x−y x y( x + y)( x + 1)( y + 1)
Thus, we only need to show that
p p p
x+ y + 1 − k x y ≥ 0,

which is true if
p p
2 4 x y + 1 − k x y ≥ 0.
Let
p
t= 4
x y.
From
p
5 = x + 4 y ≥ 4 x y = 4t 2 ,
we get p
5
t≤ .
2
Thus,
p p
2 4 x y + 1 − k x y = 2t + 1 − kt 2
2 1
 ‹•  ‹ ˜
= 1 − p t 1 + 2 1 + p t ≥ 0.
5 5
The equality holds for a = b = c = d = e = 1.
38 Vasile Cîrtoaje

P 1.13. If a, b, c are nonnegative real numbers, no two of which are zero, then
1 1 1 2 1 1 1
 ‹
+ + ≤ + + .
3a + b + c 3b + c + a 3c + a + b 5 b + c c + a a + b

(Vasile C., 2006)

Solution. Due to homogeneity, we may assume that a + b + c = 3. So, we need to


show that
a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s = = 1,
3
where
2 5
f (u) = − , u ∈ [0, 3).
3 − u 2u + 3
For u ∈ [1, 3), we have

4 40 36[2u3 + 3u2 + 9(u − 1)(3 − u)]


f 00 (u) = − = > 0;
(3 − u)3 (2u + 3)3 (3 − u)3 (2u + 3)3
therefore, f is convex on [s, 3). By the RHCF-Theorem and Note 1, it suffices to
show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 2 y = 3, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
1 2
g(u) = +
3 − u 2u + 3
and
1 4
h(x, y) = −
(3 − x)(3 − y) (2x + 3)(2 y + 3)
9(2x + 2 y − 3)
=
(3 − x)(3 − y)(2x + 3)(2 y + 3)
9x
= ≥ 0.
(3 − x)(3 − y)(2x + 3)(2 y + 3)
The equality holds for a = b = c, and also for a = 0 and b = c (or any cyclic
permutation).

p
P 1.14. If a, b, c, d ≥ 3 − 7 so that a + b + c + d = 4, then
1 1 1 1 4
+ + + ≥ .
2+a 2 2+ b 2 2+c 2 2+d 2 3

(Vasile C., 2008)


Half Convex Function Method 39

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
1 p
f (u) = , u ≥ 3 − 7.
2 + u2
For u ≥ s = 1, f (u) is convex because

3(3u2 − 2)
f (u) =
00
> 0.
(2 + u2 )3

p RHCF-Theorem and Note 1, it suffices to show that h(x, y) ≥ 0 for x, y ≥


By the
3 − 7 so that x + 3 y = 4. We have

f (u) − f (1) −1 − u
g(u) = =
u−1 3(2 + u2 )
and
g(x) − g( y) xy + x + y −2
h(x, y) = = ,
x−y 3(2 + x 2 )(2 + y 2 )
where
p p
−x 2 + 6x − 2 (3 + 7 − x)(x − 3 + 7)
xy + x + y −2= =
3p p 3
(−1 + 7 + 3 y)(x − 3 + 7)
= ≥ 0.
3
In accordance with Note 4, the equality holds for a = b = c = d = 1, and also for
p
p 1+ 7
a = 3 − 7, b=c=d=
3
(or any cyclic permutation).
Remark. Similarly, we can prove the following generalization:
p
• If a1 , a2 , . . . , an ≥ n − 1 − n2 − 3n + 3 so that a1 + a2 + · · · + an = n, then
1 1 1 n
+ + ··· + ≥ ,
2 + a1 2 + a2
2 2
2 + an
2 3

with equality for a1 = a2 = · · · = an = 1, and also for


p
p 1+ n2 − 3n + 3
a1 = n − 1 − n2 − 3n + 3, a2 = a3 = · · · = a n =
n−1
(or any cyclic permutation).
40 Vasile Cîrtoaje

p
P 1.15. If a1 , a2 , . . . , an ∈ [− n, n − 2] so that a1 + a2 + · · · + an = n, then

1 1 1 n
+ + ··· + ≤ .
n + a1 n + a2
2 2
n + an
2 n+1

(Vasile C., 2008)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−1 p
f (u) = , n ≥ 3, u ∈ [− n, n − 2].
n + u2
p
For u ∈ [− n, 1], we have

2(n − u2 )
f 00 (u) = ≥ 0,
(n + u2 )3
p
hence f is convex on [− n, s]. By the LHCF-Theorem and Note 1, it suffices to
p
show that h(x, y) ≥ 0 for x, y ∈ [− n, n − 2] so that x + (n − 1) y = n. We have

f (u) − f (1) u+1


g(u) = =
u−1 (n + 1)(n + u2 )

and
g(x) − g( y) n− x − y − xy
h(x, y) = =
x−y (n + 1)(n + x 2 )(n + y 2 )
(n − x)(n − 2 − x)
= 2 ≥ 0.
(n − 1)(n + x 2 )(n + y 2 )

The equality holds for a1 = a2 = · · · = an = 1, and also for

2
a1 = n − 2, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

P 1.16. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

3−a 3− b 3−c 3
+ + ≥ .
9+a 2 9+ b 2 9+c 2 5

(Vasile C., 2013)


Half Convex Function Method 41

Solution. Write the inequality as


a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
3−u
f (u) = , u ∈ [0, 3].
9 + u2
For u ∈ [1, 3], we have
1 00 u2 (9 − u) + 27(u − 1)
f (u) = > 0.
2 (9 + u2 )3
Thus, f is convex on [s, 3]. By the RHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 2 y = 3, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
−(6 + u)
g(u) =
5(9 + u2 )
and
x y + 6x + 6 y − 9 x(9 − x)
h(x, y) = = ≥ 0.
5(9 + x )(9 + y ) 10(9 + x 2 )(9 + y 2 )
2 2

3
The equality holds for a = b = c = 1, and also for a = 0 and b = c = (or any
2
cyclic permutation).

Remark. In the same manner, we can prove the following generalization:

• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then


n − a1 n − a2 n − an n
+ 2 +···+ 2 ≥ ,
n2 + (n2− 3n + 1)a1 n + (n − 3n + 1)a2
2 2 2
n + (n − 3n + 1)an
2 2 2n − 1
with equality for a1 = a2 = · · · = an = 1, and also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

P 1.17. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


1 1 1 3
+ + ≥ .
1 − a + 2a 2 1 − b + 2b 2 1 − c + 2c 2 2
(Vasile C., 2012)
42 Vasile Cîrtoaje

Solution. Write the inequality as


a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
1
f (u) = , u ∈ [0, 3].
1 − u + 2u2
For u ∈ [1, 3], we have

1 00 12u2 − 6u − 1
f (u) = > 0.
2 (1 − u + 2u2 )3
Thus, f is convex on [s, 3]. By the RHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 2 y = 3, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
We have
−(1 + 2u)
g(u) =
2(1 − u + 2u2 )
and
4x y + 2x + 2 y − 3 x(1 + 4 y)
h(x, y) = = ≥ 0.
2(1 − x + 2x )(1 − y + 2 y ) 2(1 − x + 2x 2 )(1 − y + 2 y 2 )
2 2

3
The equality holds for a = b = c = 1, and also for a = 0 and b = c = (or any
2
cyclic permutation).

Remark. In the same manner, we can prove the following generalization:

• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


p
3n − 2 + 5n2 − 8n + 4
k ≥ k1 , k1 = ,
2n
then
1 1 1 n
+ + · · · + ≥ ,
1 − a1 + ka12 1 − a2 + ka22 1 − an + kan2 k
with equality for a1 = a2 = · · · = an = 1. If k = k1 , then the equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
Half Convex Function Method 43

P 1.18. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


1 1 1 3
+ + ≥ .
5+a+a 2 5+ b+ b 2 5+c+c 2 7
(Vasile C., 2008)
Solution. Write the inequality as
a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
1
f (u) = , u ∈ [0, 3].
5 + u + u2
For u ≥ 1, from
2(3u2 + 3u − 4)
f 00 (u) = > 0,
(5 + u + u2 )3
it follows that f is convex on [s,3]. By the RHCF-Theorem and Note 1, it suffices
to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 2 y = 3. We have
f (u) − f (1) −2 − u
g(u) = =
u−1 7(5 + u + u2 )
and
g(x) − g( y) x y + 2(x + y) − 3
h(x, y) = =
x−y 7(5 + x + x 2 )(5 + y + y 2 )
x(5 − x)
= ≥ 0.
14(5 + x + x 2 )(5 + y + y 2 )
According to Note 4, the equality holds for a = b = c = 1, and also for a = 0 and
3
b = c = (or any cyclic permutation).
2

Remark. Similarly, we can prove the following generalization:

• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


2(2n − 1)
0 < k ≤ k1 , k1 = ,
n−1
then
1 1 1 n
+ + ··· + ≥ ,
k + a1 + a1 k + a2 + a2
2 2
k + an + an2 k+2
with equality for a1 = a2 = · · · = an = 1. If k = k1 , then the equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
44 Vasile Cîrtoaje

P 1.19. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then


1 1 1 1 1
+ + + ≤ .
10 + a + a 2 10 + b + b 2 10 + c + c 2 10 + d + d 2 3
(Vasile C., 2008)
Solution. Write the inequality as
a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
−1
f (u) = , u ∈ [0, 4].
10 + u + u2
For u ∈ [0, 1], we have
6(3 − u − u2 )
f (u) =
00
> 0.
(10 + u + u2 )3
Thus, f is convex on [0,s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 3 y = 4. We have
f (u) − f (1) 2+u
g(u) = =
u−1 12(10 + u + u2 )
and
g(x) − g( y) 8 − 2(x + y) − x y
h(x, y) = =
x−y 12(10 + x + x 2 )(10 + y + y 2 )
3 y2
= ≥ 0.
12(10 + x + x 2 )(10 + y + y 2 )
The equality holds for a = b = c = d = 1, and also for a = 4 and b = c = d = 0
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:

• Let a1 , a2 , . . . , an (n ≥ 4) be nonnegative real numbers so that


a1 + a2 + · · · + an = n.
If k ≥ 2n + 2, then
1 1 1 n
+ + ··· + ≤ ,
k + a1 + a1 k + a2 + a2
2 2
k + an + an2 k+2
with equality for a1 = a2 = · · · = an = 1. If k = 2n + 2, then the equality holds also
for
a1 = n, a2 = a3 = · · · = an = 0
(or any cyclic permutation).
Half Convex Function Method 45

P 1.20. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


If
1
k ≥1− ,
n
then
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2
2 2
1 + kan
2 1+k
(Vasile C., 2005)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1
f (u) = , u ∈ [0, n].
1 + ku2
For u ∈ [1, n], we have

2k(3ku2 − 1) 2k(3k − 1)
f 00 (u) = ≥ > 0.
(1 + ku2 )3 (1 + ku2 )3

Thus, f is convex on [s, n]. By the RHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n. We have

f (u) − f (1) −k(u + 1)


g(u) = =
u−1 (1 + k)(1 + ku2 )

and
g(x) − g( y) k2 (x + y + x y) − k
h(x, y) = = .
x−y (1 + k)(1 + kx 2 )(1 + k y 2 )
We need to show that
k(x + y + x y) − 1 ≥ 0.
Indeed, we have

1 x(2n − 2 − x)
 ‹
k(x + y + x y) − 1 ≥ 1 − (x + y + x y) − 1 = ≥ 0.
n n
1
The equality holds for a1 = a2 = · · · = an = 1. If k = 1 − , then the equality
n
holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
46 Vasile Cîrtoaje

P 1.21. Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an = n. If

n−1
0<k≤ ,
n2 − n + 1

then
1 1 1 n
+ + ··· + ≤ .
1 + ka1 1 + ka2
2 2
1 + kan
2 1+k

(Vasile C., 2005)

Solution. Replacing all negative numbers ai by −ai , we need to show the same
inequality for
a1 , a2 , . . . , an ≥ 0, a1 + a2 + · · · + an ≥ n.
Since the left side of the desired inequality is decreasing with respect to each ai , is
sufficient to consider that a1 + a2 + · · · + an = n. Write this inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−1
f (u) = , u ∈ [0, n].
1 + ku2
For u ∈ [0, 1], we have
2k(1 − 3ku2 )
f 00 (u) = ≥ 0,
(1 + ku2 )3
since
3(n − 1) (n − 2)2
1 − 3ku2 ≥ 1 − 3k ≥ 1 − = ≥ 0.
n2 − n + 1 n2 − n + 1
Thus, f is convex on [0, s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n. We have

f (u) − f (1) k(u + 1)


g(u) = =
u−1 (1 + k)(1 + ku2 )

and
g(x) − g( y) k − k2 (x + y + x y)
h(x, y) = = .
x−y (1 + k)(1 + kx 2 )(1 + k y 2 )
It suffices to show that
1 − k(x + y + x y) ≥ 0.
Indeed, we have

n−1 (x − n + 1)2
1 − k(x + y + x y) ≥ 1 − (x + y + x y) = ≥ 0.
n2 − n + 1 n2 − n + 1
Half Convex Function Method 47

n−1
The equality holds for a1 = a2 = · · · = an = 1. If k = , then the equality
n2 −n+1
holds also for
1
a1 = n − 1, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

P 1.22. Let a1 , a2 , . . . , an be nonnegative numbers so that a1 + a2 + · · · + an = n. If


n2
k≥ , then
4(n − 1)
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + ··· + ≥ 0.
a1 + k
2
a2 + k
2
an2 + k

(Vasile C., 2012)


Solution. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
u(u − 1)
f (u) = , u ∈ [0, n].
u2 + k
From
u2 + 2ku − k 2(k2 − u3 ) + 6ku(1 − u)
f 0 (u) = , f 00
(u) = ,
(u2 + k)2 (u2 + k)3
it follows that f is convex on [0, 1]. By the LHCF-Theorem and Note 1, it suffices
to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
u
g(u) =
u2 +k
and
k−xy n2 − 4(n − 1)x y
h(x, y) = ≥
(x 2 + k)( y 2 + k) 4(n − 1)(x 2 + k)( y 2 + k)
[x + (n − 1) y]2 − 4(n − 1)x y [x − (n − 1) y]2
= = ≥ 0.
4(n − 1)(x 2 + k)( y 2 + k) 4(n − 1)(x 2 + k)( y 2 + k)
The equality holds for a1 = a2 = · · · = an = 1, and also for
a1 = n/2, a2 = a3 = · · · = an = n/(2n − 2)
(or any cyclic permutation).
48 Vasile Cîrtoaje

P 1.23. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
a1 − 1 a2 − 1 an − 1
+ + ··· + ≥ 0.
(n − 2a1 ) 2 (n − 2a2 )2 (n − 2an )2
(Vasile C., 2012)

Solution. For n = 2, the inequality is an identity. Consider further n ≥ 3 and write


the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
u−1
f (u) = , u ∈ I = [0, n] \ {n/2}.
(n − 2u)2
From
2u + n − 4 8(u + n − 3)
f 0 (u) = , f 00 (u) = ,
(n − 2u)3 (n − 2u)4
it follows that f is convex on I≤s . By the LHCF-Theorem, Note 1 and Note 3, it
suffices to show that h(x, y) ≥ 0 for x, y ∈ I so that x + (n − 1) y = n. We have

f (u) − f (1) 1
g(u) = =
u−1 (n − 2u)2

and
g(x) − g( y) 4(n − x − y) 4(n − 2) y
h(x, y) = = = ≥ 0.
x−y (n − 2x)2 (n − 2 y)2 (n − 2x)2 (n − 2 y)2

In accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1, and


also for
a1 = n, a2 = a3 = · · · = an = 0
(or any cyclic permutation).

P 1.24. If a1 , a2 , . . . , an are nonnegative real numbers so that


n
a1 + a2 + · · · + an = n, a1 , a2 , . . . , an > −k, k ≥1+ p ,
n−1
then
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (an + k)2
(Vasile C., 2008)
Half Convex Function Method 49

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
u2 − 1
f (u) = , u > −k.
(u + k)2
For u ∈ (−k, 1], we have
2(k2 − 3 − 2ku) 2(k2 − 2k − 3) 2(k + 1)(k − 3)
f 00 (u) = ≥ = ≥ 0.
(u + k)4 (u + k)4 (u + k)4
Thus, f is convex on (−k, s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y > −k so that x + (n − 1) y = n. We have
f (u) − f (1) u+1
g(u) = =
u−1 (u + k)2
and
g(x) − g( y) (k − 1)2 − (1 + x)(1 + y)
h(x, y) = = .
x−y (x + k)2 ( y + k)2
Since
n2
(k − 1)2 ≥ ,
n−1
we need to show that
n2 ≥ (n − 1)(1 + x)(1 + y).
Indeed,

n2 − (n − 1)(1 + x)(1 + y) = n2 − (1 + x)(2n − 1 − x) = (x − n + 1)2 ≥ 0.


n
The equality holds for a1 = a2 = · · · = an = 1. If k = 1 + p , then the
n−1
equality holds also for
1
a1 = n − 1, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


P 1.25. Let a1 ,s
2n − 1
If 0 < k ≤ 1 + , then
n−1
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≤ 0.
(a1 + k)2 (a2 + k)2 (an + k)2

(Vasile C., 2008)


50 Vasile Cîrtoaje

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1 − u2
f (u) = , u ∈ [0, n].
(u + k)2
For u ≥ 1, we have

2(2ku − k2 + 3) 2(2k − k2 + 3) 2(1 + k)(3 − k)


f 00 (u) = ≥ = > 0.
(u + k)4 (u + k)4 (u + k)4
Thus, f is convex on [s, n]. By the RHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n. We have
f (u) − f (1) −u − 1
g(u) = =
u−1 (u + k)2
and
g(x) − g( y) 2k − k2 + x + y + x y 2k − k2 + x + y
h(x, y) = = ≥ .
x−y (x + k)2 ( y + k)2 (x + k)2 ( y + k)2
Since
x + (n − 1) y n
x+y≥ = ,
n−1 n−1
we get
n 2n − 1
2k − k2 + x + y ≥ 2k − k2 + = −(k − 1)2 + ≥ 0,
n−1 n−1
hence h(x, y) ≥ 0. s
2n − 1
The equality holds for a1 = a2 = · · · = an = 1. If k = 1 + , then the
n−1
equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

p
P 1.26. If a1 , a2 , . . . , an ≥ n − 1 − n2 − n + 1 so that a1 + a2 + · · · + an = n, then

a12 − 1 a22 − 1 an2 − 1


+ + ··· + ≤ 0.
(a1 + 2)2 (a2 + 2)2 (an + 2)2

(Vasile C., 2008)


Half Convex Function Method 51

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1 − u2 p
f (u) = , u ≥ n − 1 − n2 − n + 1.
(u + 2)2
For u ≥ 1, we have
2(4u − 1)
f 00 (u) = > 0.
(u + 2)4
Thus, f (u) is convex for u ≥ s. By the RHCF-Theorem and Note 1, it suffices to
show that h(x, y) ≥ 0 for
p
n − 1 − n2 − n + 1 ≤ x ≤ 1 ≤ y, x + (n − 1) y = n.

Since
f (u) − f (1) −u − 1
g(u) = = ,
u−1 (u + 2)2
g(x) − g( y) x+ y+xy −x 2 + 2(n − 1)x + n
h(x, y) = = = ,
x−y (x + 2)2 ( y + 2)2 (n − 1)(x + 2)2 ( y + 2)2
we need to show that
p p
n−1− n2 − n + 1 ≤ x ≤ n − 1 + n2 − n + 1.

This is true because


p p
n−1− n2 − n + 1 ≤ x ≤ 1 < n − 1 + n2 − n + 1.

The equality holds for a1 = a2 = · · · = an = 1, and also for


p
p 1+ n2 − n + 1
a1 = n − 1 − n2 − n + 1, a2 = a3 = · · · = a n =
n−1
(or any cyclic permutation).

P 1.27. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


(n − 1)(2n − 1)
If k ≥ , then
n2
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2
3 3
1 + kan
3 1+k

(Vasile C., 2008)


52 Vasile Cîrtoaje

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1
f (u) = , u ∈ [0, n].
1 + ku3
For u ∈ [1, n], we have
6ku(2ku3 − 1) 6ku(2k − 1)
f 00 (u) = ≥ > 0.
(1 + ku3 )3 (1 + ku3 )3
Thus, f is convex on [s, n]. By the RHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
−k(u2 + u + 1)
g(u) =
(1 + k)(1 + ku3 )
and
h(x, y) x 2 y 2 + x y(x + y − 1) + (x + y)2 − (x + y + 1)/k
= .
k2 (1 + k)(1 + kx 3 )(1 + k y 3 )
Since
x + (n − 1) y n
x+y≥ = > 1,
n−1 n−1
it suffices to show that
x + y +1
(x + y)2 ≥ .
k
n
From x + y ≥ , we get
n−1
2n − 1
k(x + y) ≥ ,
n
hence
2n − 1
 ‹
n
k(x + y) − x − y = (x + y)[k(x + y) − 1] ≥
2
− 1 = 1.
n−1 n
(n − 1)(2n − 1)
The equality holds for a1 = a2 = · · · = an = 1. If k = , then the
n2
equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
Half Convex Function Method 53

P 1.28. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


n−1
If 0 < k ≤ 2 , then
n − 2n + 2
1 1 1 n
+ + ··· + ≤ .
1 + ka1 1 + ka2
3 3
1 + kan
3 1+k
(Vasile C., 2008)
Solution. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−1
f (u) = , u ∈ [0, n].
1 + ku3
For u ∈ [0, 1], we have
6ku(1 − 2ku3 ) 6ku(1 − 2k)
f 00 (u) = ≥ ≥ 0.
(1 + ku3 )3 (1 + ku3 )3
Thus, f is convex on [0, s]. By the LHCF-Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
k(u2 + u + 1)
g(u) =
(1 + k)(1 + ku3 )
and
h(x, y) (x + y + 1)/k − x 2 y 2 − x y(x + y − 1) − (x + y)2
= .
k2 (1 + k)(1 + kx 3 )(1 + k y 3 )
It suffices to show that
(n2 − 2n + 2)(x + y + 1)
− x 2 y 2 − x y(x + y − 1) − (x + y)2 ≥ 0,
n−1
which is equivalent to
[2 + n y − (n − 1) y 2 ][1 − (n − 1) y]2 ≥ 0.
This is true because
2 + n y − (n − 1) y 2 = 2 + y[n − (n − 1) y] = 2 + x y > 0.
n−1
The equality holds for a1 = a2 = · · · = an = 1. If k = , then the
n2 − 2n + 2
equality holds also for
1
a1 = n − 1, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
54 Vasile Cîrtoaje

P 1.29. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


n2
If k ≥ , then
n−1
v v v
t a1 t a2 t an n
+ + ··· + ≤p .
k − a1 k − a2 k − an k−1

(Vasile C., 2008)


Solution. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where s
u
f (u) = − , u ∈ [0, n].
k−u
For u ∈ [0, 1], we have
k(k − 4u) k(k − 4)
f 00 (u) = ≥ 3/2 ≥ 0.
4u3/2 (k− u)5/2 4u (k − u)5/2
Thus, f is convex on [0, s]. By the LHCF-Theorem, it suffices to prove that

f (x) + (n − 1) f ( y) ≥ n f (1)

for x ≥ 1 ≥ y ≥ 0 so that x + (n − 1) y = n. We write the inequality as


v v
t (k − 1)x t (k − 1) y
+ (n − 1) ≤ n,
k−x k− y
v – v ™
t (n − 1)k(1 − y) t (k − 1) y
1+ ≤ 1 + (n − 1) 1 − .
(n − 1) y + k − n k− y
Let v
t (k − 1) y
z= , z ≤ 1,
k− y
which yields
kz 2
y= ,
z2 + k − 1
(k − 1)(1 − z 2 ) (k − 1)(nz 2 + k − n)
1− y = , (n − 1) y + k − n = .
z2 + k − 1 z2 + k − 1
Since
k(1 − y) k(1 − z 2 ) 1 − z2
= =
(n − 1) y + k − n k − n(1 − z 2 ) 1 − n(1 − z 2 )/k
1 − z2 n(1 − z 2 )
≤ = ,
1 − (1 − z 2 )(n − 1)/n (n − 1)z 2 + 1
Half Convex Function Method 55

it suffices to show that


v
t n(n − 1)(1 − z 2 )
1+ ≤ 1 + (n − 1)(1 − z).
(n − 1)z 2 + 1
By squaring, we get the obvious inequality

(z − 1)2 [(n − 1)z − 1]2 ≥ 0.

n2
The equality holds for a1 = a2 = · · · = an = 1. If k = , then the equality
n−1
holds also for
n(n − 1)2 n
a1 = , a2 = a3 = · · · = an =
n2 − 2n + 2 (n − 1)(n2 − 2n + 2)
(or any cyclic permutation).

P 1.30. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
2 2 2
n−a1 + n−a2 + · · · + n−an ≥ 1.

(Vasile C., 2006)

Solution. Let k = ln n. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
2
f (u) = n−u , u ∈ [0, n].
For u ≥ 1, we have
2 2 2
f 00 (u) = 2kn−u (2ku2 − 1) ≥ 2kn−u (2k − 1) ≥ 2kn−u (2 ln 2 − 1) > 0;

therefore, f is convex on [s, n]. By the RHCF-Theorem, it suffices to show that

f (x) + (n − 1) f ( y) ≥ n f (1)

for 0 ≤ x ≤ 1 ≤ y and x + (n − 1) y = n. The desired inequality is equivalent to


g(x) ≥ 0, where
2 2 n− x
g(x) = n−x + (n − 1)n− y − 1, y= , 0 ≤ x ≤ 1.
n−1
Since y 0 = −1/(n − 1), we get
2 2 2 2
g 0 (x) = −2x kn−x − 2(n − 1)k y y 0 n− y = 2k( y n− y − x n−x ).
56 Vasile Cîrtoaje

The derivative g 0 (x) has the same sign as g1 (x), where


2 2
g1 (x) = ln( y n− y ) − ln(x n−x ) = ln y − ln x + k(x 2 − y 2 ),

y0 1 −1 2k(1 + nx − 2x)
• ˜
g10 (x) = − + 2k(x − y y 0 ) = n + .
y x x(n − x) (n − 1)2
For 0 < x ≤ 1, g10 (x) has the same sign as

−(n − 1)2
h(x) = + x(n − x)(1 + nx − 2x).
2k
Since

h0 (x) = n + 2(n2 − 2n − 1)x − 3(n − 2)x 2


≥ nx + 2(n2 − 2n − 1)x − 3(n − 2)x
= 2(n − 1)(n − 2)x ≥ 0,

h is strictly increasing on [0, 1]. From

−(n − 1)2 1
 ‹
h(0) = < 0, h(1) = (n − 1) 1 −2
> 0,
2k 2k

it follows that there is x 1 ∈ (0, 1) so that h(x 1 ) = 0, h(x) < 0 for x ∈ [0, x 1 ) and
h(x) > 0 for x ∈ (x 1 , 1]. Therefore, g1 is strictly decreasing on (0, x 1 ] and strictly
increasing on [x 1 , 1]. Since g1 (0+ ) = ∞ and g1 (1) = 0, there is x 2 ∈ (0, x 1 ) so that
g1 (x 2 ) = 0, g1 (x) > 0 for x ∈ (0, x 2 ) and g1 (x) < 0 for x ∈ (x 2 , 1). Consequently, g
is strictly increasing on [0, x 2 ] and strictly decreasing on [x 2 , 1]. Because g(0) > 0
and g(1) = 0, it follows that g(x) ≥ 0 for x ∈ [0, 1]. The proof is completed.
The equality holds for a1 = a2 = · · · = an = 1.

P 1.31. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(3a2 + 1)(3b2 + 1)(3c 2 + 1)(3d 2 + 1) ≤ 256.

(Vasile C., 2006)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
f (u) = − ln(3u2 + 1), u ∈ [0, 4].
Half Convex Function Method 57

For u ∈ [1, 4], we have


6(3u2 − 1)
f 00 (u) = > 0.
(3u2 + 1)2
Therefore, f is convex on [s, 4]. By the RHCF-Theorem, we only need to show that

f (x) + 3 f ( y) ≥ 4 f (1)

for 0 ≤ x ≤ 1 ≤ y so that x + 3 y = 4; that is, to show that g(x) ≥ 0 for x ∈ [0, 1],
where
4− x
g(x) = f (x) + 3 f ( y) − 4 f (1), y= .
3
Since y 0 (x) = −1/3, we have

−6x 6y
g 0 (x) = f 0 (x) + 3 y 0 f 0 ( y) = +
3x + 1 3 y 2 + 1
2

6(x − y)(3x y − 1) 8(1 − x)(x 2 − 4x + 1)


= = ≥ 0.
(3x 2 + 1)(3 y 2 + 1) (3x 2 + 1)(3 y 2 + 1)
p p
Since g is increasing on [0, 2 − 3] and decreasing on [2 − 3, 1], it suffices to
show that g(0) ≥ 0 and g(1) ≥ 0. The inequality g(0) ≥ 0 is true if the original
inequality holds for a = 0 and b = c = d = 4/3. This reduces to 193 ≤ 27 · 256,
which is true because 27 · 256 − 193 = 53 > 0. The inequality g(1) ≥ 0 is also true
because g(1) = 0.
The equality holds for a = b = c = d = 1.

P 1.32. If a, b, c, d, e ≥ −1 so that a + b + c + d + e = 5, then

(a2 + 1)(b2 + 1)(c 2 + 1)(d 2 + 1)(e2 + 1) ≥ (a + 1)(b + 1)(c + 1)(d + 1)(e + 1).

(Vasile C., 2007)

Solution. Consider the nontrivial case a, b, c, d, e > −1, and write the inequality
as
a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ n f (s), s= = 1,
5
where
f (u) = ln(u2 + 1) − ln(u + 1), u > −1.
For u ∈ (−1, 1], we have

2(1 − u2 ) 1
f 00 (u) = + > 0.
(u2 + 1)2 (u + 1)2
58 Vasile Cîrtoaje

Therefore, f is convex on (−1, s]. By the LHCF-Theorem and Note 2, it suffices to


show that H(x, y) ≥ 0 for x, y > −1 so that x + 4 y = 5, where

f 0 (x) − f 0 ( y) 2(1 − x y) 1
H(x, y) = = 2 + ;
x−y (x + 1)( y + 1) (x + 1)( y + 1)
2

thus, we need to show that

(x 2 + 1)( y 2 + 1)
2(1 − x y) + ≥ 0.
(x + 1)( y + 1)

Since
x2 + 1 x +1 y2 + 1 y +1
≥ , ≥ ,
x +1 2 y +1 2
it suffices to prove that

(x + 1)( y + 1)
2(1 − x y) + ≥ 0,
4
which is equivalent to
x + y + 9 − 7x y ≥ 0,
28x 2 − 38x + 14 ≥ 0,
(28x − 19)2 + 31 ≥ 0.
The equality holds for a = b = c = d = e = 1.

P 1.33. Let a1 , a2 , . . . , an be positive numbers so that a1 + a2 + · · · + an = n. If


p v p
2 n−1 2 n−1
t
k≤ +2 , k ≤ 3,
n n
then
p 1 1 1
k( a1 + a2 + · · · + an ) + p + p + · · · + p ≥ (k + 1)n.
p p
a1 a2 an

(Vasile C., 2006)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
k p
f (u) = p + u, u ∈ (0, n).
u
Half Convex Function Method 59

From
3 − ku
f 00 (u) =
,
4u5/2
it follows that f is convex on (0, 1]. Thus, according to the LHCF-Theorem and Note
1, it suffices to show that h(x, y) ≥ 0 for x ≥ 1 ≥ y > 0 such that x + (n − 1) y = n,
where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
k 1
g(u) = p − p
u+1 u+ u
and p p
p p p p x + y +1
( x + y)( x + 1)( y + 1)h(x, y) = −k + p .
xy
So, we need to show that p p
x + y +1
p ≥ k.
xy
Since p p p
x+ y ≥ 2 4 x y,
it suffices to show that p
2 4 xy +1
p ≥ k,
xy
which is equivalent to
1 2
p +p ≥ k.
xy 4
xy
From Æ
n = x + (n − 1) y ≥ 2 (n − 1)x y,
we get p
1 2 n−1
p ≥ ,
xy n
hence
p v p
1 2 2 n−1 2 n−1
t
p +p ≥ +2 ≥ k.
xy 4
xy n n
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1.
Remark. Since
p v p
2 n−1 2 n−1
t
1< +2
n n
for n ≤ 134, the following inequality holds for a1 , a2 , . . . , a134 > 0 such that a1 +
a2 + · · · + a134 = 134:
p 1 1 1
a1 + a2 + · · · + a134 + p + p + · · · + p
p p
≥ 268.
a1 a2 a134
60 Vasile Cîrtoaje

Since
p v p
2 n−1 2 n−1
t
2< +2
n n
for n ≤ 12, the following inequality holds for a1 , a2 , . . . , a12 > 0 such that a1 + a2 +
· · · + a12 = 12:

p 1 1 1
2( a1 + a2 + · · · + a12 ) + p + p + · · · + p
p p
≥ 36.
a1 a2 a12

P 1.34. If a1 , a2 , . . . , an (n ≥ 3) are positive numbers so that a1 + a2 + · · · + an = 1,


then

1 n
     
1 1 1 p
‹
p p p
p − a1 p − a2 · · · p − an ≥ n− p .
a1 a2 an n

(Vasile C., 2006)

Solution. Write the inequality as

a1 + a2 + · · · + an 1
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = ,
n n
where
1 p 1
 ‹
f (u) = ln p − u = ln(1 − u) − ln u, u ∈ (0, 1).
u 2
From
−1 1 1 − 2u − u2
f 0 (u) = − , f 00 (u) = 2 ,
1 − u 2u 2u (1 − u)2
p
it follows that f 00 (u) ≥ 0 for u ∈ (0, 2 − 1]. Since

1 1 p
s= ≤ < 2 − 1,
n 3
f is convex on (0, s]. Thus, we can apply the LHCF-Theorem.
First Solution. By the LHCF-Theorem, it suffices to show that

1
 ‹
f (x) + (n − 1) f ( y) ≥ n f
n

for all x, y > 0 so that x + (n − 1) y = 1; that is, to show that


‹ n−1 ‹n
1 p 1 p 1
 
p
p − x p − y ≥ n− p .
x y n
Half Convex Function Method 61

Write this inequality as

nn/2 (1 − y)n−1 ≥ (n − 1)n−1 x 1/2 y (n−3)/2 .

By squaring, this inequality becomes as follows:

nn (1 − y)2n−2 ≥ (n − 1)2n−2 x y n−3 ,

(2n − 2)2n−2 n−3


(2 − 2 y)2n−2 ≥ xy ,
nn
˜2n−2
1 1
•
n· + x + (n − 3) y ≥ [n + 1 + (n − 3)]n+1+(n−3) · n
· x · y n−3 .
n n
The last inequality follows from the AM-GM inequality. The proof is completed.
The equality holds for a1 = a2 = · · · = an = 1/n.
Second Solution. By the LHCF-Theorem and Note 2, it suffices to prove that H(x, y) ≥
0 for x, y > 0 so that x + (n − 1) y = 1, where

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y

We have
1− x − y − xy n( y + 1) − y − 3
H(x, y) = =
2x y(1 − x)(1 − y) 2x(1 − x)(1 − y)
3( y + 1) − y − 3 y
≥ = > 0.
2x(1 − x)(1 − y) x(1 − x)(1 − y)

Remark 1. We may write the inequality in P 1.34 in the form


n  n
1 n
 Y
1 p
Y  ‹
(1 + ai ) ≥
p
p −1 · n− p .
i=1
ai i=1
n

On the other hand, by the AM-GM inequality and the Cauchy-Schwarz inequality,
we have
‚ Œn v !n
n n n
1 Xp t1 1 n
Y u X  ‹
(1 + ai ) ≤ 1 + ai ≤ 1 + ai = 1 + p
p
.
i=1
n i=1
n i=1
n

Thus, the following statement follows:


• If a1 , a2 , . . . , an (n ≥ 3) are positive real numbers so that a1 + a2 + · · · + an = 1,
then     
1 1 1 p
p − 1 · · · p − 1 ≥ ( n − 1) ,
n
p −1
a1 a2 an
with equality for a1 = a2 = · · · = an = 1/n.
62 Vasile Cîrtoaje

Remark 2. By squaring, the inequality in P 1.34 becomes


n
Y (1 − a )2 i (n − 1)2n
≥ .
i=1
ai nn

1+ x
On the other hand, since the function f (x) = ln is convex on (0, 1), by
1− x
Jensen’s inequality we have
a1 + a2 + · · · + an n
 
n  1+
1 + ai n+1 n
‹  ‹
n
Y
≥ = .
 
1 − ai a1 + a2 + · · · + an  n−1
i=1 1−
n
Multiplying these inequalities yields the following result (Kee-Wai Lau, 2000):
• If a1 , a2 , . . . , an (n ≥ 3) are positive real numbers so that a1 + a2 + · · · + an = 1,
then
1 1 1 1 n
 ‹ ‹  ‹  ‹
− a1 − a2 · · · − an ≥ n − ,
a1 a2 an n
with equality for a1 = a2 = · · · = an = 1/n.

P 1.35. Let a1 , a2 , . . . , an be positive real numbers so that a1 + a2 + · · · + an = n. If


 p 2
2 n−1
0< k ≤ 1+ ,
n

then
1 1 1
 ‹ ‹  ‹
ka1 + ka2 + · · · kan + ≥ (k + 1)n .
a1 a2 an
(Vasile C., 2006)
Solution. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1
 ‹
f (u) = ln ku + , u ∈ (0, n).
u
We have
ku2 − 1 1 + 4ku2 − k2 u4
f 0 (u) = , f 00 (u) = .
u(ku2 + 1) u2 (ku2 + 1)2
For u ∈ (0, 1], we get f 00 (u) > 0 since

1 + 4ku2 − k2 u4 > ku2 (4 − ku2 ) ≥ ku2 (4 − k) ≥ 0.


Half Convex Function Method 63

Therefore, f is convex on (0, s]. By the LHCF-Theorem and Note 2, it suffices to


prove that H(x, y) ≥ 0 for x, y > 0 so that x + (n − 1) y = n, where

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y

Since
1 + k(x + y)2 − k2 x 2 y 2 k[(x + y)2 − kx 2 y 2 ]
H(x, y) = > ,
x y(kx 2 + 1)(k y 2 + 1) x y(kx 2 + 1)(k y 2 + 1)
it suffices to show that p
x+y≥ k x y.
Indeed, by the Cauchy-Schwarz inequality, we have
p
(x + y)[(n − 1) y + x] ≥ ( n − 1 + 1)2 x y,

hence p
1 p
 
2 n − 1 p
x + y ≥ ( n − 1 + 1)2 x y = 1 + x y ≥ k x y.
n n
The equality holds for a1 = a2 = · · · = an = 1.

P 1.36. If a, b, c, d are nonzero real numbers so that

−1
a, b, c, d ≥ , a + b + c + d = 4,
2
then
1 1 1 1 1 1 1 1
 ‹
3 2 + 2 + 2 + 2 + + + + ≥ 16.
a b c d a b c d
Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
3 1 −1 11
• ˜
f (u) = 2 + , u ∈ I = , \ {0},
u u 2 2
is convex on I≥s (because 3/u2 and 1/u are convex). By the RHCF-Theorem, Note
1 and Note 3, it suffices to prove that h(x, y) ≥ 0 for x, y ∈ I so that

x + 3 y = 4,

where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
64 Vasile Cîrtoaje

Indeed, we have
4 3
g(u) = − − 2 ,
u u
4x y + 3x + 3 y 2(1 + 2x)(6 − x)
h(x, y) = = ≥ 0.
x2 y2 3x 2 y 2
In accordance with Note 4, the equality holds for a = b = c = d = 1, and also
for
−1 3
a= , b=c=d=
2 2
(or any cyclic permutation).

P 1.37. If a1 , a2 , . . . , an are nonnegative real numbers so that a12 + a22 + · · · + an2 = n,


then s
n
a1 + a2 + · · · + an − n +
3 3 3
(a1 + a2 + · · · + an − n) ≥ 0.
n−1
(Vasile C., 2007)
p
Solution. Replacing each ai by ai , we have to prove that

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s),

where
a1 + a2 + · · · + an
s= =1
n
and
p p
s
n
f (u) = u u + k u, k= , u ∈ [0, n].
n−1
For u ≥ 1, we have
3u − k 3−k
f 00 (u) = p ≥ p > 0.
4u u 4u u
Therefore, f is convex on [s, n]. According to the RHCF-Theorem and Note 1, it
suffices to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + (n − 1) y = n. Since

f (u) − f (1) u+k


g(u) = =1+ p
u−1 u+1

and p p p
g(x) − g( y) x + y+ xy−k
h(x, y) = = p p p p ,
x−y ( x + y)( x + 1)( y + 1)
we need to show that
p p p
x+ y+ x y ≥ k.
Half Convex Function Method 65

Since p p
p p p
x+ y+ x+ x + y,
p
xy ≥ y≥
it suffices to show that
x + y ≥ k2 .
Indeed, we have
x n
x+y≥ +y= = k2 .
n−1 n−1
In accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1, and
also for s
n
a1 = 0, a2 = · · · = an =
n−1
(or any cyclic permutation).

P 1.38. If a, b, c, d, e are nonnegative real numbers so that a2 + b2 + c 2 + d 2 + e2 = 5,


then
1 1 1 1 1
+ + + + ≤ 1.
7 − 2a 7 − 2b 7 − 2c 7 − 2d 7 − 2e
(Vasile C., 2010)
p p p p p
Solution. Replacing a, b, c, d, e by a, b, c, d, e, we have to prove that

f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s),

where
a+b+c+d+e
s= =1
5
and
1
f (u) = p , u ∈ [0, 5].
2 u−7
For u ∈ [0, 1], we have
p
7−6 u
f (u) = p
00
p > 0.
2u u(7 − 2 u)3
Therefore, f is convex on [0, s]. According to the LHCF-Theorem and Note 1, it
suffices to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 4 y = 5. Since

f (u) − f (1) −2
g(u) = = p p
u−1 5(7 − 2 u)(1 + u)
and
p p
g(x) − g( y) 2(5 − 2 x − 2 y)
h(x, y) = = p p p p p p ,
x−y ( x + y)(1 + x)(1 + y)(7 − 2 x)(7 − 2 y)
66 Vasile Cîrtoaje

we need to show that


p 5 p
y≤ . x+
2
Indeed, by the Cauchy-Schwarz inequality, we have
p 1 25
 ‹
p 2
( x + y) ≤ 1 + (x + 4 y) = .
4 4
The proof is completed. The equality holds for a = b = c = d = e = 1, and also for
1
a = 2, b=c=d=e=
2
(or any cyclic permutation).
Remark In the same manner, we can prove the following generalization:
• Let a1 , a2 , . . . , an be nonnegative real numbers so that a12 + a22 + · · · + an2 = n. If
n
k ≥1+ p , then
n−1
1 1 1 n
+ + ··· + ≤ ,
k − a1 k − a2 k − an k−1
n
with equality for a1 = a2 = · · · = an = 1. If k = 1 + p , then the equality holds
n−1
also for
p 1
a1 = n − 1, a2 = · · · = an = p
n−1
(or any cyclic permutation).

P 1.39. Let 0 ≤ a1 , a2 , . . . , an < k so that a12 + a22 + · · · + an2 = n. If


s
n
1< k ≤1+ ,
n−1
then
1 1 1 n
+ + ··· + ≥ .
k − a1 k − a2 k − an k−1
(Vasile C., 2010)
p p p
Solution. Replacing a1 , a2 , . . . , an by a1 , a2 , . . . , an , we have to prove that

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s),

where
a1 + a2 + · · · + an
s= =1
n
Half Convex Function Method 67

and
1
f (u) = p , u ∈ [0, k2 ).
k− u
From p
3 u−k
f (u) = p
00
p ,
4u u(k − u)3
it follows that f is convex on [s, k2 ). According to the RHCF-Theorem and Note 1,
it suffices to show that h(x, y) ≥ 0 for all x, y ∈ [0, k2 ) so that x + (n − 1) y = n.
Since
f (u) − f (1) 1
g(u) = = p p
u−1 (k − 1)(k − u)(1 + u)
and
p p
g(x) − g( y) x + y +1−k
h(x, y) = = p p p p p p ,
x−y (k − 1)( x + y)(1 + x)(1 + y)(k − x)(k − y)

we need to show that


p p
x+ y ≥ k − 1.
Indeed,
p
s s
p x n
x+ x+y≥ +y=
p
y≥ ≥ k − 1.
n−1 n−1
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1, and also for
s
n
a1 = 0, a2 = · · · = an =
n−1
(or any cyclic permutation).

P 1.40. If a, b, c are nonnegative real numbers, no two of which are zero, then
v v v
t 48a t 48b t 48c
1+ + 1+ + 1+ ≥ 15.
b+c c+a a+b

(Vasile C., 2005)

Solution. Due to homogeneity, we may assume that a + b + c = 1. Thus, we need


to show that
f (a) + f (b) + f (c) ≥ 3 f (s),
where
a+b+c 1
s= =
3 3
68 Vasile Cîrtoaje

and v
t 1 + 47u
f (u) = , u ∈ [0, 1).
1−u
From
48(47u − 11)
f 00 (u) = p ,
(1 − u)5 (1 + 47u)3
it follows that f is convex on [s, 1). By the RHCF-Theorem, it suffices to show that

1
 ‹
f (x) + 2 f ( y) ≥ 3 f
3

for x, y ≥ 0 so that x + 2 y = 1; that is,


v v
t 1 + 47x t 49 − 47x
+2 ≥ 15.
1− x 1+ x

Setting
v
t 49 − 47x
t= , 1 < t ≤ 7,
1+ x
the inequality turns into
v
t 1175 − 23t 2
≥ 15 − 2t.
t2 − 1

By squaring, this inequality becomes

350 − 15t − 61t 2 + 15t 3 − t 4 ≥ 0,

(5 − t)2 (2 + t)(7 − t) ≥ 0.

The original inequality is an equality for a = b = c, and also for a = 0 and b = c


(or any cyclic permutation).

P 1.41. If a, b, c are nonnegative real numbers, then


v v v
t 3a2 t 3b2 t 3c 2
+ + ≤ 1.
7a2 + 5(b + c)2 7b2 + 5(c + a)2 7c 2 + 5(a + b)2

(Vasile C., 2008)


Half Convex Function Method 69

Solution. Due to homogeneity, we may assume that a + b + c = 3. Thus, we need


to show that
f (a) + f (b) + f (c) ≥ 3 f (s),
where
a+b+c
s= =1
3
and v
t 3u2 −u
f (u) = − =p , u ∈ [0, 3].
7u + 5(3 − u)
2 2
4u2 − 10u + 15
From

5(−8u2 + 41u − 30) 5(−8u2 + 38u − 30) 10(u − 1)(15 − 4u)


f 00 (u) = ≥ = ,
(4u2 − 10u + 15)5/2 (4u2 − 10u + 15)5/2 (4u2 − 10u + 15)5/2

it follows that f is convex on [s, 3]. By the RHCF-Theorem, it suffices to prove the
original homogeneous inequality for b = c = 0 and b = c = 1. For the nontrivial
case b = c = 1, we need to show that
v v
t 3a2 t 3
+ 2 ≤ 1.
7a + 20
2 5a + 10a + 12
2

By squaring two times, the inequality becomes


Æ
a(5a3 + 10a2 + 16a + 50) ≥ 3a (7a2 + 20)(5a2 + 10a + 12),

a2 (5a6 + 20a5 − 11a4 + 38a3 − 80a2 − 40a + 68) ≥ 0,


a2 (a − 1)2 (5a4 + 30a3 + 44a2 + 96a + 68) ≥ 0.
The last inequality is clearly true.
The equality holds for a = b = c, and also for a = 0 and b = c (or any cyclic
permutation).

P 1.42. If a, b, c are nonnegative real numbers, then


v v v
t a2 t b2 t c2
+ + ≥ 1.
a2 + 2(b + c)2 b2 + 2(c + a)2 c 2 + 2(a + b)2

(Vasile C., 2008)

Solution. Due to homogeneity, we may assume that a + b + c = 3. Thus, we need


to show that
f (a) + f (b) + f (c) ≥ 3 f (s),
70 Vasile Cîrtoaje

where
a+b+c
s= =1
3
and v
t 3u2 u
f (u) = =p , u ∈ [0, 3].
u + 2(3 − u)
2 2
u2 − 4u + 6
From
2(2u2 − 11u + 12) 2(−11u + 12)
f 00 (u) = ≥ 2 ,
(u − 4u + 6)
2 5/2 (u − 4u + 6)5/2
it follows that f is convex on [0, s]. By the LHCF-Theorem, it suffices to prove the
original homogeneous inequality for b = c = 0 and b = c = 1. For the nontrivial
case b = c = 1, the inequality has the form

a 2
p +p ≥ 1.
a2 + 8 2a2 + 4a + 3

By squaring, the inequality becomes


Æ
a (a2 + 8)(2a2 + 4a + 3) ≥ 3a2 + 8a − 2.

For the nontrivial case 3a2 + 8a − 2 > 0, by squaring both sides we get

a6 + 2a5 + 5a4 − 8a3 − 14a2 + 16a − 2 ≥ 0,

(a − 1)2 [a4 + 4a3 + 9a2 + 4a + (3a2 + 8a − 2)] ≥ 0.


The equality holds for a = b = c, and also for b = c = 0 (or any cyclic permutation).

P 1.43. Let a, b, c be nonnegative real numbers, no two of which are zero. If

ln 3
k ≥ k0 , k0 = − 1 ≈ 0.585,
ln 2
then ‹k ‹k ‹k
2a 2b 2c
  
+ + ≥ 3.
b+c c+a a+b
(Vasile C., 2005)

Solution. For k = 1, the inequality is just the well known Nesbitt’s inequality

2a 2b 2c
+ + ≥ 3.
b+c c+a a+b
Half Convex Function Method 71

For k ≥ 1, the inequality follows from Nesbitt’s inequality and Jensens’s inequality
applied to the convex function f (u) = uk :

2a 2b 2c
Œk
+ +
‹k ‹k ‹k ‚
2a 2b 2c
  
b+c c+a a+b
+ + ≥3 ≥ 3.
b+c c+a a+b 3

Consider now that


k0 ≤ k < 1.
Due to homogeneity, we may assume that a + b + c = 1. Thus, we need to show
that
f (a) + f (b) + f (c) ≥ 3 f (s),
where
a+b+c 1
s= =
3 3
and ‹k
2u

f (u) = , u ∈ [0, 1).
1−u
From ‹k−2
4k 2u

f (u) =
00
(2u + k − 1),
(1 − u)4 1−u
it follows that f is convex on [s, 1) (because u ≥ s = 1/3 involves 2u + k − 1 ≥
2/3 + k − 1 = k − 1/3 > 0). By the RHCF-Theorem, it suffices to prove the original
homogeneous inequality for b = c = 1 and a ∈ [0, 1]; that is, to show that h(a) ≥ 3,
where ‹k
2

h(a) = a + 2
k
, a ∈ [0, 1].
a+1
For a ∈ (0, 1], the derivative
‹k+1
2

h (a) = ka
0 k−1
−k
a+1

has the same sign as

2
g(a) = (k − 1) ln a − (k + 1) ln .
a+1
From
2ka + k − 1
g 0 (a) = ,
a(a + 1)
it follows that g 0 (a0 ) = 0 for a0 = (1 − k)/(2k) < 1, g 0 (a) < 0 for a ∈ (0, a0 ) and
g 0 (a) > 0 for a ∈ (a0 , 1]. Consequently, g is strictly decreasing on (0, a0 ] and strictly
increasing on (a0 , 1]. Since g(0+ ) = ∞ and g(1) = 0, there exists a1 ∈ (0, a0 ) so
72 Vasile Cîrtoaje

that g(a1 ) = 0, g(a) > 0 for a ∈ (0, a1 ) and g(a) < 0 for a ∈ (a1 , 1); therefore, h(a)
is strictly increasing on [0, a1 ] and strictly decreasing on [a1 , 1]. As a result,

h(a) ≥ min{h(0), h(1)}.

Since h(0) = 2k+1 ≥ 3 and h(1) = 3, we get h(a) ≥ 3. The proof is completed. The
equality holds for a = b = c. If k = k0 , then the equality holds also for a = 0 and
b = c (or any cyclic permutation).
Remark. For k = 2/3, we can give the following solution (based on the AM-GM
inequality):
X  2a ‹2/3 X 2a
=
b+c
p
2a · (b + c) · (b + c)
3

X 6a
≥ = 3.
2a + (b + c) + (b + c)

p
P 1.44. If a, b, c ∈ [1, 7 + 4 3], then
v v v
t 2a t 2b t 2c
+ + ≥ 3.
b+c c+a a+b

(Vasile C., 2007)

Solution. Denoting
a+b+c p
s= , 1 ≤ s ≤ 7 + 4 3,
3
we need to show that
f (a) + f (b) + f (c) ≥ 3 f (s),
where v
t 2u
f (u) = , 1 ≤ u < 3s.
3s − u
For u ≥ s, we have
‹3/2
3s − u 4u − 3s

f (u) = 3s
00
> 0.
2u (3s − u)4
Therefore, f (u) is convex for u ≥ s. By the RHCF-Theorem, it suffices to prove the
original inequality for b = c; that is,
s v
a t 2b
+2 ≥ 3.
b a+b
Half Convex Function Method 73

v
tb p
Putting t = , the condition a, b ∈ [1, 7 + 4 3] involves
a
p p
2− 3 ≤ t ≤ 2 + 3.

We need to show that v


t 2t 2 1
2 ≥3− .
t +1
2 t
This is true if
8t 2 1 2
 ‹
≥ 3− ,
t2 + 1 t
which is equivalent to the obvious inequality
p p
(t − 1)2 (t − 2 + 3 )(t − 2 − 3 ) ≤ 0.
p
The equality holds for a = b = c, and also for a = 1, and b = c = 7 + 4 3 (or
any cyclic permutation).

P 1.45. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 2
0 < k ≤ k0 , k0 = ≈ 1.71,
ln 3 − ln 2
then
a k (b + c) + b k (c + a) + c k (a + b) ≤ 6.

Solution. For 0 < k ≤ 1, the inequality follows from Jensens’s inequality applied
to the convex function f (u) = −uk :
˜k
(b + c)a + (c + a)b + (a + b)c
•
(b + c)a + (c + a)b + (a + b)c ≤ 2(a + b + c)
k k k
2(a + b + c)
ab + bc + ca a + b + c 2k
 ‹k  ‹
=6 ≤6 = 6.
3 3
Consider now that
1 < k ≤ k0 ,
and write the inequality as

f (a) + f (b) + f (c) ≥ 3 f (s),

where
a+b+c
s= =1
3
74 Vasile Cîrtoaje

and
f (u) = uk (u − 3), u ∈ [0, 3].
For u ≥ 1, we have

f 00 (u) = kuk−2 [(k + 1)u − 3k + 3] ≥ kuk−2 [(k + 1) − 3k + 3] = 2k(2 − k)uk−2 > 0;

therefore, f is convex on [1, s]. By the RHCF-Theorem, it suffices to consider the


case a ≤ b = c. So, we only need to prove the homogeneous inequality

a+b+c
 ‹k+1
a k (b + c) + b k (c + a) + c k (a + b) ≤ 6
3
for b = c = 1 and a ∈ [0, 1]; that is, to show that g(a) ≥ 0 for a ≥ 0, where

a+2
 ‹k+1
g(a) = 3 − a k − a − 1.
3
We have

a+2 k + 1 a + 2 k−1 k − 1
‹k
1 00
  ‹
g (a) = (k + 1)
0
− ka k−1
− 1, g (a) = − 2−k .
3 k 3 3 a
Since g 00 is strictly increasing, g 00 (0+ ) = −∞ and g 00 (1) = 2k(2 − k)/3 > 0, there
exists a1 ∈ (0, 1) so that g 00 (a1 ) = 0, g 00 (a) < 0 for a ∈ (0, a1 ), g 00 (a) > 0 for
a ∈ (a1 , 1]. Therefore, g 0 is strictly decreasing on [0, a1 ] and strictly increasing on
[a1 , 1]. Since

k+1 k−1
g 0 (0) = (k + 1)(2/3)k − 1 ≥ (k + 1)(2/3)k0 − 1 = −1= > 0,
2 2
g 0 (1) = 0,
there exists a2 ∈ (0, a1 ) so that g 0 (a2 ) = 0, g 0 (a) > 0 for a ∈ [0, a2 ), g 0 (a) < 0
for a ∈ (a2 , 1]. Thus, g is strictly increasing on [0, a2 ] and strictly decreasing on
[a2 , 1]; consequently,
g(a) ≥ min{g(0), g(1)}.
From

g(0) = 3(2/3)k+1 − 1 ≥ 3(2/3)k0 +1 − 1 = 1 − 1 = 0, g(1) = 0,

we get g(a) ≥ 0. This completes the proof. The equality holds for a = b = c = 1.
If k = k0 , then the equality holds also for a = 0 and b = c = 3/2 (or any cyclic
permutation).
Remark 1. Using the Cauchy-Schwarz inequality and the inequality in P 1.45, we
get
X a (a + b + c)2 9 3
≥ =P ≥ .
b +c a(b + c ) a (b + c) 2
P
k k k k k
Half Convex Function Method 75

Thus, the following statement holds.


• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 2
0 < k ≤ k0 , k0 = ≈ 1.71,
ln 3 − ln 2
then
a b c 3
+ + ≥ ,
bk + c k c k + ak ak + bk 2

with equality for a = b = c = 1. If k = k0 , then the equality holds also for a = 0 and
b = c = 3/2 (or any cyclic permutation).
Remark 2. Also, the following statement holds:
• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 9 − ln 8
k ≥ k1 , k1 = ≈ 0.2905,
ln 3 − ln 2
then
ak bk ck 3
+ + ≥ ,
b+c c+a a+b 2

with equality for a = b = c = 1. If k = k1 , then the equality holds also for a = 0 and
b = c = 3/2 (or any cyclic permutation).

For k1 ≤ k ≤ 2, the inequality can be proved using the Cauchy-Schwarz inequality


and the inequality in P 1.45, as follows:
X ak (a + b + c)2 9 3
≥P =P ≥ .
b+c a2−k (b + c) a2−k (b + c) 2

For k ≥ 2, the inequality can be deduced from the Cauchy-Schwarz inequality and
Bernoulli’s inequality, as follows:
P k/2 2 P k/2 2
X ak a a
≥ P = ,
b+c (b + c) 6
X X• k
˜
a k/2
≥ 1 + (a − 1) = 3.
2
76 Vasile Cîrtoaje

P 1.46. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


‚v v Œ
p p p ta + b tb+c s
c+a
a+ b+ c − 3 ≥ 13 + + −3 .
2 2 2

(Vasile C., 2008)

Solution. Write the inequality as

a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where v
p t3 − u
f (u) = u − 13 , u ∈ [0, 3].
2
For u ∈ [1, 3), we have

13 3 − u −3/2 13
 ‹
4 f (u) = −u
00 −3/2
+ ≥ −1 + > 0.
4 2 4

Therefore, f is convex on [s, 3]. By the RHCF-Theorem, it suffices to consider only


the case a ≤ b = c. Write the original inequality in the homogeneous form
v ‚v v v Œ
p p p ta + b + c ta + b tb+c s
c+a ta + b + c
a+ b+ c−3 ≥ 13 + + −3 .
3 2 2 2 3

Due to homogeneity, we may assume that b = c = 1. Moreover, it is convenient


p
to use the notation a = x. Thus, we need to show that g(x) ≥ 0 for x ∈ [0, 1],
where v v
t x2 + 2 t x2 + 1
g(x) = x − 11 + 36 − 26 .
3 2
We have v v
t 3 t 2
g 0 (x) = 1 + 12x − 13x ,
x2 + 2 x2 + 1
‹3/2 – ‹3/2 ™
13 2 2
+ 1

x
g 00 (x) = m· 2 −1 ,
2 x2 + 1 x +2
where p
3
6 52
m= ≈ 1.72.
13
Clearly, g 00 (x) has the same sign as h(x), where

x2 + 1
h(x) = m · − 1.
x2 + 2
Half Convex Function Method 77

Since h is strictly increasing,


m 2m
h(0) = − 1 < 0, h(1) = − 1 > 0,
2 3
there is x 1 ∈ (0, 1) so that h(x 1 ) = 0, h(x) < 0 for x ∈ [0, x 1 ) and h(x) > 0 for
x ∈ (x 1 , 1]. Therefore, g 0 is strictly decreasing on [0, x 1 ] and strictly increasing on
[x 1 , 1]. Since g 0 (0) = 1 and g 0 (1) = 0, there is x 2 ∈ (0, x 1 ) so that g 0 (x 2 ) = 0,
g 0 (x) > 0 for x ∈ (0, x 2 ) and g 0 (x) < 0 for x ∈ (x 2 , 1). Thus, g(x) is strictly
increasing on [0, x 2 ] and strictly decreasing on [x 2 , 1]. From
p p
g(0) = −11 + 12 6 − 13 2 > 0

and g(1) = 0, it follows that g(x) ≥ 0 for x ∈ [0, 1]. This completes the proof. The
equality holds for a = b = c = 1.
Remark. Similarly, we can prove the following generalizations:
• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k ≥ k0 , where
p
6−2 p p
k0 = p p = (2 + 2)(2 + 3) ≈ 12.74 ,
6− 2−1
then ‚v v Œ
p p p ta + b tb+c s
c+a
a+ b+ c−3≥ k + + −3 ,
2 2 2
with equality for a = b = c = 1. If k = k0 , then the equality holds also for a = 0 and
b = c = 3/2 (or any cyclic permutation).

• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


k ≥ k0 , where p
p
n− n−1
k0 = p p 1
,
n − n − 2 − pn−1
then
‚v v v Œ
p tn − a tn − a tn − a
1 2 n
a1 + a2 + · · · + + + ··· +
p p
an − n ≥ k −n ,
n−1 n−1 n−1

with equality for a1 = a2 = · · · = an = 1. If k = k0 , then the equality holds also for


n
a1 = 0 and a2 = a3 = · · · = an = (or any cyclic permutation).
n−1

P 1.47. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k > 2, then

a+b k b+c k  c + a k
 ‹  ‹
a + b +c +3≥2
k k k
+2 +2 .
2 2 2
78 Vasile Cîrtoaje

Solution. Write the inequality as


a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where ‹k
3−u

f (u) = u − 2
k
, u ∈ [0, 3].
2
For u ≥ 1, we have

f 00 (u) 1 3 − u k−2 1
 ‹
=u −
k−2
≥ 1 − > 0.
k(k − 1) 2 2 2
Therefore, f is convex on [s, 3]. By the RHCF-Theorem, it suffices to consider only
the case a ≤ b = c. Write the original inequality in the homogeneous form

a+b+c a+b b+c  c + a k


 ‹k  ‹k  ‹k
a + b +c +3
k k k
≥2 +2 +2 .
3 2 2 2
Due to homogeneity, we may assume that b = c = 1. Thus, we need to prove that

a+2 a+1
 ‹k  ‹k
a +3
k
≥4
3 2
for a ∈ [0, 1]. Substituting

a k = t, t ∈ [0, 1],

we need to show that g(t) ≥ 0, where


k k
t 1/k + 2 t 1/k + 1
 
g(t) = t + 3 −4 .
3 2

We have
k−1 k−1
t 1/k + 2 t 1/k + 1
 
g 0 (t) = 1 + t 1/k−1 − 2t 1/k−1 ,
3 2
k−2 k−2
t 1/k + 1 t 1/k + 2
 
kt 2−1/k 00 2
g (t) = − .
k−1 2 3 3
Setting
1
2
 ‹ k−2
m= , 0 < m < 1,
3
we see that g 00 (t) has the same sign as h(t), where

t +1 t 1/k + 2
 1/k 
h(t) = 6 −m = (3 − 2m)t 1/k + 3 − 4m
2 3
Half Convex Function Method 79

is strictly increasing. There are two cases to consider: 0 < m ≤ 3/4 and 3/4 < m <
1.
Case 1: 0 < m ≤ 3/4. Since h(0) = 3 − 4m ≥ 0, we have h(t) > 0 for t ∈ (0, 1],
hence g 0 is strictly increasing on (0, 1]. From g 0 (1) = 0, it follows that g 0 (t) < 0 for
t ∈ (0, 1), hence g is strictly decreasing on [0, 1]. Since g(1) = 0, we get g(t) > 0
for t ∈ [0, 1).
Case 2: 3/4 < m < 1. From m > 3/4, we get
22k−3 > 3k−1 .
Since h(0) = 3 − 4m < 0 and h(1) = 3(1 − m) > 0, there is t 1 ∈ (0, 1) so that
h(t 1 ) = 0, h(t) < 0 for t ∈ [0, t 1 ) and h(t) > 0 for t ∈ (t 1 , 1]. Thus, g 0 (t) is strictly
decreasing on (0, t1] and strictly increasing on [t 1 , 1]. Since g 0 (0+ ) = +∞ and
g 0 (1) = 0, there exists t 2 ∈ (0, t 1 ) so that g 0 (t 2 ) = 0, g 0 (t) > 0 for t ∈ (0, t 2 ) and
g 0 (t) < 0 for t ∈ (t 2 , 1). Therefore, g(t) is strictly increasing on [0, t 2 ] and strictly
decreasing on [t 2 , 1]. Since
22k−2 − 3k−1
g(0) = >0
2k 3k−1
and g(1) = 0, we have g(t) ≥ 0 for t ∈ [0, 1].
The equality holds for a = b = c = 1.
Remark 1. The inequality in P 1.47 is Popoviciu’s inequality
a+b+c a+b b+c c + a
 ‹  ‹  ‹
f (a) + f (b) + f (c) + 3 f ≥ 2f + 2f + 2f
3 2 2 2
applied to the convex function f (x) = x k defined on [0, ∞).
Remark 2. In the same manner, we can prove the following refinements (Vasile C.,
2008):
• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k > 2 and
m ≤ m0 , where
2k (3k−1 − 2k−1 )
m0 = k−1 > 2,
6 + 3k−1 − 22k−1
then
a+b k b + c k  c + a k
 ‹  ‹ 
a + b +c −3≥ m
k k k
+ + −3 ,
2 2 2
with equality for a = b = c = 1. If m = m0 , then the equality holds also for a = 0
and b = c = 3/2 (or any cyclic permutation).

• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If


k > 2 and m ≤ m1 , where
1 1
(n−1)k−1
− nk−1
m1 = (n−2)k
> n − 1,
1 1
(n−1)k
+ (n−1)2k−1 − nk−1
80 Vasile Cîrtoaje

then
n − a1 k  n − a2 k  n − a k
• ˜
n
a1k + a2k + ··· + ank −n≥m + + ··· + −n ,
n−1 n−1 n−1
with equality for a1 = a2 = · · · = an = 1. If m = m1 , then the equality holds also for
n
a1 = 0 and a2 = a3 = · · · = an = (or any cyclic permutation).
n−1

P 1.48. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
‚v v v Œ
p tn − a tn − a tn − a
1 2 n
a1 + a2 + · · · + an + n(k − 1) ≤ k + + ··· +
p p
,
n−1 n−1 n−1

where p p p
k = ( n − 1)( n + n − 1).

(Vasile C., 2008)

Solution. For n = 2, the inequality is an identity. Consider further that n ≥ 3. We


will show first that
n − 1 < k < 2(n − 1).
The left inequality reduces to
p p
( n − 1)( n − 1 − 1) > 0,

while the right inequality is equivalent to


p p p
( n − 1)( n − n − 1 + 2) > 0.

Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where s
p n−u
f (u) = − u + k , u ∈ [0, n].
n−1
For u ≤ 1, we have

k k
4 f 00 (u) = u−3/2 − p (n − u)−3/2 ≥ 1 − p (n − 1)−3/2
n−1 n−1
k k
=1− ≥1− > 0.
(n − 1)2 2(n − 1)
Half Convex Function Method 81

Therefore, f is convex on [0, s]. By the LHCF-Theorem, it suffices to consider the


case
a1 ≥ a2 = · · · = an .
Write the original inequality in the homogeneous form
v v
Xp ta + a + ··· + a Xta + ··· + a
1 2 n 2 n
a1 + n(k − 1) ≤k .
n n−1

Do to homogeneity, we need to prove this inequality for a2 = · · · = an = 1 and


p
a1 = x ≥ 1; that is, to show that g(x) ≤ 0 for x ≥ 1, where
Æ Æ
g(x) = x + n − 1 − k + (k − 1) n(x 2 + n − 1) − k (n − 1)(x 2 + n − 2).

We have v v
t nx 2 t (n − 1)x 2
g (x) = 1 + (k − 1)
0
−k ,
x2 + n − 1 x2 + n − 2
p – ‹3/2 ™
k(n − 2) n − 1 x 2
+ n − 2
g 00 (x) = m· 2 −1 ,
(x 2 + n − 2)3/2 x +n−1
where v
3 (k − 1) n(n − 1)
t 2
m= .
k2 (n − 2)2
Clearly, g 00 (x) has the same sign as h(x), where

m(x 2 + n − 2) 1
 ‹
h(x) = −1= m 1− 2 − 1.
x2 + n − 1 x +n−1

We have
m(n − 1)
h(1) = − 1, lim h(x) = m − 1.
n x→∞

We will show that h(1) < 0 and lim x→∞ h(x) > 0; that is, to show that
n
1<m< .
n−1
The inequality m > 1 is equivalent to

1 n−2
1− >p ,
k n(n − 1)

which is true since


1 1 n−2 n−2
1− >1− = >p .
k n−1 n−1 n(n − 1)
82 Vasile Cîrtoaje

n
The inequality m < is equivalent to
n−1
1 n(n − 2)
1− < ,
k (n − 1)2
which is also true because
1 1 2n − 3 n(n − 2)
1− <1− = ≤ .
k 2(n − 1) 2(n − 1) (n − 1)2
Since h is strictly increasing on [1, ∞), h(1) < 0 and lim x→∞ h(x) > 0, there
is x 1 ∈ (1, ∞) so that h(x 1 ) = 0, h(x) < 0 for x ∈ [1, x 1 ) and h(x) > 0 for
x ∈ (x 1 , ∞). Therefore, g 0 is strictly decreasing on [1, x 1 ] and strictly increasing
on [x 1 , ∞). Since g 0 (1) = 0 and lim x→∞ g 0 (x) = 0, it follows that g 0 (x) < 0 for
x ∈ (1, ∞). Thus, g(x) is strictly decreasing on [1, ∞), hence g(x) ≤ g(1) = 0.
The equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = n, a2 = a3 = · · · = an = 0

(or any cyclic permutation).


Remark. Since k > n − 1 for n ≥ 3, the inequality in P 1.48 is sharper than Popovi-
p
ciu’s inequality applied to the convex function f (x) = − x, x ≥ 0:
‚v v v Œ
p tn − a tn − a tn − a
1 2 n
a1 + a2 +· · ·+ an +n(n−2) ≤ (n−1) + + ··· +
p p
.
n−1 n−1 n−1

P 1.49. If a, b, c are the lengths of the sides of a triangle so that a + b + c = 3, then


1 1 1 p 2 2 2
 ‹
+ + − 3 ≥ 4(2 + 3) + + −3 .
a+b−c b+c−a c+a−b a+b b+c c+a

(Vasile C., 2008)


Solution. Write the inequality as
a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
1 4k p
f (u) = − , k = 2(2 + 3) ≈ 7.464, u ∈ [0, 3/2).
3 − 2u 3 − u
For u ≥ 1, we have
 ‹3 ‹3 
8 8k 1 2

f 00 (u) = − >8 − .
(3 − 2u)3 (3 − u)3 3 − 2u 3−u
Half Convex Function Method 83

Since
1 2
≥ , u ∈ [1, 3/2),
3 − 2u 3 − u
it follows that f is convex on [s,3/2). By the RHCF-Theorem and Note 1, it suffices
to show that h(x, y) ≥ 0 for x, y ∈ [0, 3/2) so that x + 2 y = 3. We have

f (u) − f (1) 2 2k
g(u) = = −
u−1 3 − 2u 3 − u
and
g(x) − g( y) 2 k
h(x, y) = = −
x−y (3 − 2x)(3 − 2 y) (3 − x)(3 − y)
2 k
= −
(2 y − x)x 2 y(x + y)
kx 2 − 2(k − 2)x y + 4 y 2
=
2x y(x + y)(2 y − x)
p
[( 3 + 1)x − 2 y]2
= ≥ 0.
2x y(x + y)(2 y − x)

According to Note 4, the equality holds for a = b = c = 1, and also for


p
p 3( 3 − 1)
a = 3(2 − 3), b=c=
2
(or anu cyclic permutation).

P 1.50. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≤ 5.


If p
29 + 761
k ≥ k0 , k0 = ≈ 5.66,
10
then X 1 5
≥ .
ka1 + a2 + a3 + a4 + a5
2
k+4
(Vasile C., 2006)

Solution. Since each term of the left hand side of the inequality decreases by in-
creasing any number ai , it suffices to consider the case

a1 + a2 + a3 + a4 + a5 = 5,

when the desired inequality can be written as

f (a1 ) + f (a2 ) + f (a3 ) + f (a4 ) + f (a4 ) ≥ 5 f (s),


84 Vasile Cîrtoaje

where
a1 + a2 + a3 + a4 + a5
s= =1
5
and
1
f (u) = , u ∈ [0, 5].
ku2 − u + 5
For u ≥ 1, we have
2[3ku(ku − 1) − 5k + 1]
f 00 (u) =
(ku2 − u + 5)3
2[3k(k − 1) − 5k + 1]

(ku2 − u + 5)3
2[k(3k − 8) + 1]
= > 0;
(ku2 − u + 5)3
therefore, f is convex on [s, 5]. By the RHCF-Theorem, it suffices to show that
1 4 5
+ ≥
kx 2 − x +5 ky − y +5 k +4
2

for
0 ≤ x ≤ 1 ≤ y, x + 4 y = 5.
Write this inequality as follows:
1 1 1 1
• ˜
− +4 − ≥ 0,
kx 2 − x + 5 k + 4 k y2 − y + 5 k + 4
(x − 1)(1 − k − kx) 4( y − 1)(1 − k − k y)
+ ≥ 0.
kx 2 − x + 5 k y2 − y + 5
Since
4( y − 1) = 1 − x,
the inequality is equivalent to
1 − k − kx 1− k − ky
 ‹
(x − 1) − ≥ 0,
kx 2 − x + 5 k y 2 − y + 5

5(x − 1)2 g(x, y, k)


≥ 0,
4(kx 2 − x + 5)(k y 2 − y + 5)
where
g(x, y, k) = k2 x y + k(k − 1)(x + y) − 6k + 1.
For fixed x and y, let h(k) = g(x, y, k). Since

h0 (k) = 2kx y + (2k − 1)(x + y) − 6 ≥ (2k − 1)(x + y) − 6


 y 10k − 29
≥ (2k − 1) x + −6= > 0,
4 4
Half Convex Function Method 85

it suffices to show that g(x, y, k0 ) ≥ 0. We have


g(x, y, k0 ) = k02 x y + k0 (k0 − 1)(x + y) − 6k0 + 1
= −4k02 y 2 + k0 (2k0 + 3) y + 5k02 − 11k0 + 1.
Since
5k02 − 29k0 + 4 = 0,
we get
11k0 − 1 11k0 − 1
 ‹  ‹
g(x, y, k0 ) = (5 − 4 y) k02 y + k02 − = x k0 y + k0 −
2 2
.
5 5
It suffices to show that
11k0 − 1
k02 − ≥ 0.
5
Indeed,
11k0 − 1 k0 (5k0 − 11) + 1
k02 − = > 0.
5 5
The equality holds for a1 = a2 = a3 = a4 = a5 = 1. If k = k0 , then the equality
holds also for
5
a1 = 0, a2 = a3 = a4 = a5 =
4
(or any cyclic permutation).
Remark. In the same manner, we can prove the following statement:
• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n. If
p
n2 + n − 1 + n4 + 2n3 − 5n2 + 2n + 1
k ≥ k0 , k0 = ,
2n
then X 1 n
≥ ,
ka1 + a2 + · · · + an
2
k+n−1
with equality for a1 = a2 = · · · = an = 1. If k = k0 , then the equality holds also for
n
a1 = 0, a2 = · · · = a n =
n−1
(or any cyclic permutation).

P 1.51. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≤ 5.


If p
11 − 101
0 < k ≤ k0 , k0 = ≈ 0.095,
10
then X 1 5
≥ .
ka1 + a2 + a3 + a4 + a5
2
k+4
(Vasile C., 2006)
86 Vasile Cîrtoaje

Solution. As shown at the preceding P 1.50, it suffices to consider the case

a1 + a2 + a3 + a4 + a5 = 5,

when the desired inequality can be written as

f (a1 ) + f (a2 ) + f (a3 ) + f (a4 ) + f (a4 ) ≥ 5 f (s),

where
a1 + a2 + a3 + a4 + a5
s= = 1,
5
and
1
f (u) = , u ∈ [0, 5].
ku2 −u+5
For u ∈ [0, 1], we have

u(ku − 1) − (k − 1) = (1 − u)(1 − ku) ≥ 0,

hence
2[3ku(ku − 1) − 5k + 1]
f 00 (u) =
(ku2 − u + 5)3
2[3k(k − 1) − 5k + 1]

(ku2 − u + 5)3
2[(1 − 8k) + 3k2 ]
= > 0;
(ku2 − u + 5)3
therefore, f is convex on [0, s]. By the LHCF-Theorem, it suffices to show that
1 4 5
+ ≥
kx 2 − x +5 ky − y +5 k +4
2

for
x ≥ 1 ≥ y ≥ 0, x + 4 y = 5.
Write this inequality as follows:
1 1 1 1
• ˜
− +4 − ≥ 0,
kx 2 − x + 5 k + 4 k y2 − y + 5 k + 4

(x − 1)(1 − k − kx) 4( y − 1)(1 − k − k y)


+ ≥ 0.
kx 2 − x + 5 k y2 − y + 5
Since
4( y − 1) = 1 − x,
the inequality is equivalent to
1 − k − kx 1− k − ky
 ‹
(x − 1) − ≥ 0,
kx 2 − x + 5 k y 2 − y + 5
Half Convex Function Method 87

5(x − 1)2 g(x, y, k)


≥ 0,
4(kx 2 − x + 5)(k y 2 − y + 5)
where
g(x, y, k) = k2 x y − k(1 − k)(x + y) − 6k + 1.
For fixed x and y, let h(k) = g(x, y, k). Since

h0 (k) = 2kx y − (1 − 2k)(x + y) − 6 ≤ 2kx y − 6


k(x + 4 y)2 25k
≤ −6= − 6 < 0,
8 8
it suffices to show that g(x, y, k0 ) ≥ 0. We have

g(x, y, k0 ) = k02 x y + k0 (k0 − 1)(x + y) − 6k + 1


= −4k02 y 2 + k0 (2k0 + 3) y + 5k02 − 11k0 + 1.

Since
5k02 − 11k0 + 1 = 0,
we get

g(x, y, k0 ) = k0 y(−4k0 y + 2k0 + 3) ≥ k0 y(−4k0 + 2k0 + 3) = k0 (3 − 2k0 ) y ≥ 0.

The equality holds for a1 = a2 = a3 = a4 = a5 = 1. If k = k0 , then the equality


holds also for
a1 = 5, a2 = a3 = a4 = a5 = 0
(or any cyclic permutation).
Remark. Similarly, we can prove the following statement:
• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n. If
p
2n + 1 − 4n2 + 1
0 ≤ k ≤ k0 , k0 = ,
2n
then X 1 n
≥ ,
ka12 + a2 + · · · + an k+n−1
with equality for a1 = a2 = · · · = an = 1. If k = k0 , then the equality holds also for

a1 = n, a2 = · · · = an = 0

(or any cyclic permutation).


88 Vasile Cîrtoaje

P 1.52. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


If
1
0<k≤ ,
n+1
then
a1 a2 an n
+ + ··· + ≥ .
ka12 + a2 + · · · + an a1 + ka2 + · · · + an
2
a1 + a2 + · · · + kan
2 k+n−1

(Vasile C., 2006)

Solution. Using the notation


a1 a an
x1 = , x2 = 2 , . . . , x n = ,
s s s
where
a1 + a2 + · · · + an
s= ≤ 1,
n
we need to show that x 1 + x 2 + · · · + x n = n involves
x1 xn n
+ ··· + ≥ .
ksx 12 + x2 + · · · + x n x 1 + x 2 + · · · + ksx n2 k+n−1

Since s ≤ 1, it suffices to prove the inequality for s = 1; that is, to show that
a1 a2 an n
+ + · · · + ≥
ka12 − a1 + n ka22 − a2 + n kan2 − an + n k + n − 1

for
a1 + a2 + · · · + an = n.
Write the desired inequality as

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s),

where
a1 + a2 + · · · + an
s= =1
n
and
u
f (u) = , u ∈ [0, n].
u2 − u + n
We have
n − ku2 f1 (u)
f 0 (u) = , f 00 (u) = ,
(ku2 − u + n)2 (u2 − u + n)3
where
f1 (u) = k2 u3 − 3knu + n.
Half Convex Function Method 89

For u ∈ [0, 1], we have

f1 (u) ≥ −3knu + n ≥ −3kn + n


3n n(n − 2)
≥− +n= ≥ 0.
n+1 n+1
Since f 00 (u) > 0, it follows that f is convex on [0, s]. By the LHCF-Theorem, we
only need to show that

x (n − 1) y n
+ ≥
kx 2 − x + n ky − y + n k + n−1
2

for all nonnegative x, y which satisfy x + (n − 1) y = n. Write this inequality as


follows:
1 1
• ˜
x y
− + (n − 1) − ≥ 0,
kx 2 − x + n k + n − 1 k y2 − y + n k + n − 1

n − kx n− ky
 ‹
(x − 1) − ≥ 0,
kx 2 − x + n k y 2 − y + n
(x − 1)2 h(x, y)
≥ 0,
(kx 2 − x + n)(k y 2 − y + n)
where
h(x, y) = k2 x y − kn(x + y) + n − nk.
We need to show that h(x, y) ≥ 0. Indeed,

h(x, y) = k y[n(k + n − 2) − k(n − 1) y] + n[1 − k(n + 1)]


= k y[n(n − 2) + kx] + n[1 − k(n + 1)] ≥ 0.

1
The equality holds for a1 = a2 = · · · = an = 1. If k = , then the equality holds
n+1
also for
a1 = n, a2 = a3 = · · · = an = 0
(or any cyclic permutation).

7
P 1.53. If a1 , a2 , a3 , a4 , a5 ≤ so that a1 + a2 + a3 + a4 + a5 = 5, then
2
a1 a2 a3 a4 a5
+ + + + ≤ 1.
a12 − a1 + 5 a22 − a2 + 5 a32 − a3 + 5 a42 − a4 + 5 a52 − a5 + 5

(Vasile C., 2006)


90 Vasile Cîrtoaje

Solution. Write the desired inequality as

f (a1 ) + f (a2 ) + f (a3 ) + f (a4 ) + f (a5 ) ≥ 5 f (s),

where
a1 + a2 + a3 + a4 + a5
s= =1
5
and
−u 7
f (u) = , u≤ .
u2 −u+5 2
7
• ˜
For u ∈ 1, , we have
2

−u3 + 15u − 5
f 00 (u) =
(u2 − u + 5)3
(2u + 9)(u − 1)(7 − 2u) + 43 − 7u
= > 0.
4(u2 − u + 5)3

7
• ˜
Thus, f is convex on s, . By the RHCF-Theorem, it suffices to show that
2
x 4y
+ ≤1
x2 − x + 5 y2 − y + 5

7
for all nonnegative x, y ≤ which satisfy x + 4 y = 5. Write this inequality as
2
follows:
1 1
 ‹
x y
− +4 − ] ≤ 0,
x2 − x + 5 5 y2 − y + 5 5
5− x 5− y
 ‹
(x − 1) − ≤ 0,
x2 − x + 5 y2 − y + 5
(x − 1)2 [5(x + y) − x y]
≥ 0,
(x 2 − x + 5)( y 2 − y + 5)
(x − 1)2 [(x + 4 y)(x + y) − x y]
≥ 0,
(x 2 − x + 5)( y 2 − y + 5)
(x − 1)2 (x + 2 y)2
≥ 0.
(x 2 − x + 5)( y 2 − y + 5)
The equality holds for a1 = a2 = a3 = a4 = a5 = 1, and also for

5
a1 = −5, a2 = a3 = a4 = a5 =
2
(or any cyclic permutation).
Remark. In the same manner, we can prove the following generalization:
Half Convex Function Method 91

p
• Let a1 , a2 , . . . , an ≤ 3 so that a1 + a2 + · · · + an ≤ n. If
p
n2 + 2n − 2 − 2 (n − 1)(2n2 − 1)
k= ,
n
then
a1 a an n
+ 2 2 + ··· + 2 ≤ ,
ka12 − a1 + n ka2 − a2 + n kan − an + n k − 1 + n
with equality for a1 = a2 = · · · = an = 1, and also for
n(k − n + 2) n(k + n − 2)
a1 = , a2 = · · · = a n =
2k 2k(n − 1)
(or any cyclic permutation).

P 1.54. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If
1
0<k≤ 1
,
1 + 4(n−1)2

then
a12 a22 an2 n
+ + ··· + ≥ .
ka12 + a2 + · · · + an a1 + ka22 + · · · + an a1 + a2 + · · · + kan2 k+n−1

(Vasile C., 2006)

Solution. Using the substitution


a1 a an
x1 = , x2 = 2 , . . . , x n = ,
s s s
where
a1 + a2 + · · · + an
s= ≥ 1,
n
we need to show that x 1 + x 2 + · · · + x n = n involves

x 12 x n2 n
+ ··· + ≥ .
kx 12 + (x 2 + · · · + x n )/s (x 1 + · · · + x n−1 )/s + kx n2 k+n−1

Since s ≥ 1, it suffices to prove the inequality for s = 1; that is, to show that

a12 a22 an2 n


+ + ··· + ≥
ka12 − a1 + n ka22 − a2 + n kan2 − an + n k+n−1

for
a1 + a2 + · · · + an = n.
92 Vasile Cîrtoaje

Write the desired inequality as

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s),

where
a1 + a2 + · · · + an
s= =1
n
and
u2
f (u) = , u ∈ [0, n].
u2 − u + n
We have
u(2n − u) 2 f1 (u)
f 0 (u) = , f 00 (u) = ,
(ku2 − u + n)2 (u2 − u + n)3
where
f1 (u) = ku3 − 3knu2 + n2 .
For u ∈ [0, 1] and n ≥ 3, we have

f1 (u) ≥ −3knu2 + n2 ≥ −3kn + n2 > −3n + n2 ≥ 0.

Also, for u ∈ [0, 1] and n = 2, we have


4
f1 (u) = 4 − ku2 (6 − u) ≥ 4 − u2 (6 − u)
5
4 4(1 − u)(5 − u)
≥ 4 − u(6 − u) = ≥ 0.
5 5
Since f 00 (u) ≥ 0 for u ∈ [0, 1], it follows that f is convex on [0, s]. By the LHCF-
Theorem, we need to show that

x2 (n − 1) y 2 n
+ ≥
kx − x + n k y − y + n k + n − 1
2 2

for all nonnegative x, y which satisfy x + (n − 1) y = n. Write this inequality as


follows:
 
x2 1 y2 1
− + (n − 1) − ≥ 0,
kx 2 − x + n k + n − 1 k y2 − y + n k + n − 1

(x − 1)(nx − x + n) 4( y − 1)(n y − y + n)
+ ≥ 0,
kx 2 − x + 5 k y2 − y + 5
nx − x + n ny − y + n
 ‹
(x − 1) − ≥ 0,
kx 2 − x + n k y 2 − y + n
(x − 1)2 h(x, y)
≥ 0,
(kx 2 − x + n)(k y 2 − y + n)
where
h(x, y) = n2 − kn(x + y) − k(n − 1)x y.
Half Convex Function Method 93

Since
1
0 < k ≤ k0 , k0 = 1
,
1 + 4(n−1)2

we have

h(x, y) ≥ n2 − k0 n(x + y) − k0 (n − 1)x y


= (n − 1)2 k0 y 2 − nk0 y + n2 (1 − k0 )
• ˜2
n
= k0 (n − 1) y − ≥ 0.
2(n − 1)
The equality holds for a1 = a2 = · · · = an = 1. If k = k0 , then the equality holds
also for
n(2n − 3) n
a1 = , a2 = a3 = · · · = a n =
2(n − 1) 2(n − 1)2
(or any cyclic permutation).

P 1.55. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


If k ≥ n − 1, then

a12 a22 an2 n


+ + ··· + ≤ .
ka12 + a2 + · · · + an a1 + ka22 + · · · + an a1 + a2 + · · · + kan2 k+n−1

(Vasile C., 2006)

Solution. Using the notation


a1 a an
x1 = , x2 = 2 , . . . , x n = ,
s s s
where
a1 + a2 + · · · + an
s= ≤ 1,
n
we need to show that x 1 + x 2 + · · · + x n = n involves

x 12 x n2 n
+ ··· + ≤ .
kx 12 + (x 2 + · · · + x n )/s (x 1 + · · · + x n−1 )/s + kx n2 k+n−1

Since s ≤ 1, it suffices to prove the inequality for s = 1; that is, to show that

a12 a22 an2 n


+ + ··· + ≤
ka12 − a1 + n ka22 − a2 + n kan2 − an + n k+n−1

for
a1 + a2 + · · · + an = n.
94 Vasile Cîrtoaje

Write the desired inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−u2
f (u) = 2 , u ∈ [0, n].
u −u+n
We have
u(u − 2n) 2 f1 (u)
f 0 (u) = , f 00 (u) = ,
(ku2 − u + n)2 (u2 − u + n)3
where
f1 (u) = −ku3 + 3knu2 − n2 .
For u ∈ [1, n], we have

f1 (u) ≥ −knu2 + 3knu2 − n2 = 2knu2 − n2


≥ 2kn − n2 ≥ 2(n − 1)n − n2 = n(n − 2) ≥ 0.

Since f 00 (u) ≥ 0 for u ∈ [1, n], it follows that f is convex on [s, n]. By the RHCF-
Theorem, it suffices to show that

x2 (n − 1) y 2 n
+ ≤
kx 2 − x + n k y 2 − y + n k + n − 1

for all nonnegative x, y which satisfy x + (n − 1) y = n. As shown in the proof of


the preceding P 1.54, we only need to show that h(x, y) ≥ 0, where

h(x, y) = kn(x + y) + k(n − 1)x y − n2 .

Since k ≥ n − 1, we have

h(x, y) ≥ n(n − 1)(x + y) + (n − 1)2 x y − n2


= −(n − 1)3 y 2 + n(n − 1) y + n2 (n − 2)
= [n − (n − 1) y][n(n − 2) + (n − 1)2 y]
= x[n(n − 2) + (n − 1)2 y] ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1. If k = n − 1, then the equality holds


also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
Half Convex Function Method 95

1
P 1.56. Let a1 , a2 , . . . , an ∈ [0, n] so that a1 + a2 + · · · + an ≥ n. If 0 < k ≤ , then
n
a1 − 1 a2 − 1 an − 1
+ + · · · + ≥ 0.
ka12 + a2 + · · · + an a1 + ka22 + · · · + an a1 + a2 + · · · + kan2

(Vasile C., 2006)

Solution. Let
a1 + a2 + · · · + an
s= , s ≥ 1.
n
Case 1: s > 1 Without loss of generality, assume that

a1 ≥ · · · ≥ a j > 1 ≥ a j+1 · · · ≥ an , j ∈ {1, 2, . . . , n}.

Clearly, there are b1 , b2 , . . . , bn so that b1 + b2 + · · · + bn = n and

a1 ≥ b1 ≥ 1, . . . , a j ≥ b j ≥ 1, b j+1 = a j+1 , . . . , bn = an .

Write the desired inequality as

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ 0,

where
u−1
f (u) = , u ∈ [0, n],
ku2 − u + ns
f1 (u)
f 0 (u) , f1 (u) = k(−u2 + 2u) + ns − 1.
(ku2 − u + ns)2
For u ∈ [1, n), we have

f1 (u) ≥ k(−nu + 2u) + ns − 1 = −k(n − 2)u + ns − 1


≥ −k(n − 2)n + ns − 1 ≥ −(n − 2) + ns − 1 = n(s − 1) + 1 > 0.

Consequently, f is strictly increasing on [1, n] and

f (b1 ) ≤ f (a1 ), . . . , f (b j ) ≤ f (a j ), f (b j+1 ) = f (a j+1 ), . . . , f (bn ) = f (an ).

Since
f (b1 ) + f (b2 ) + · · · + f (bn ) ≤ f (a1 ) + f (a2 ) + · · · + f (an ),
it suffices to show that f (b1 ) + f (b2 ) + · · · + f (bn ) ≥ 0 for b1 + b2 + · · · + bn = n.
This inequality is proved at Case 2.
Case 2: s = 1. Write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
96 Vasile Cîrtoaje

where
u−1
f (u) = , u ∈ [0, n],
ku2 − u + n
2g(u)
f 00 (u) = , g(u) = k2 u3 − 3k2 u2 − 3k(n − 1)u + kn + n − 1.
(ku2− u + n)3
We will show that f 00 (u) ≥ 0 for u ∈ [0, 1]. From

g 0 (u) = 3k2 u(u − 2) − 3k(n − 1),

it follows that g 0 (u) < 0, g is decreasing, hence

g(u) ≥ g(1) = −2k2 − (2n − 3)k + n − 1


−2 2n − 3
≥ 2 − +n−1
n n
(n − 1)3 − 1
= ≥ 0.
n2
Thus, f is convex on [0, s]. By the LHCF-Theorem, it suffices to show that

x −1 (n − 1)( y − 1)
+ ≥0
kx 2 − x + n k y2 − y + n

for all nonnegative real x, y so that x + (n − 1) y = n. Since (n − 1)( y − 1) = 1 − x,


we have
x −1 (n − 1)( y − 1) 1 1
 ‹
+ = (x − 1) −
kx 2 − x + n k y2 − y + n kx 2 − x + n k y 2 − y + n
(x − 1)(x − y)(1 − kx − k y)
=
(kx 2 − x + n)(k y 2 − y + n)
n(x − 1)2 (1 − kx − k y)
=
(n − 1)(kx 2 − x + n)(k y 2 − y + n)
x+ y
n(x − 1)2 (1 − n )

(n − 1)(kx 2 − x + n)(k y 2 − y + n)
(n − 2) y(x − 1)2
= ≥ 0.
(n − 1)(kx 2 − x + n)(k y 2 − y + n)

1
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1. If k = ,
n
then the equality holds also for

a1 = n, a2 = a3 = · · · = an = 0.
Half Convex Function Method 97

P 1.57. If a, b, c are positive real numbers so that abc = 1, then


p p p
a2 − a + 1 + b2 − b + 1 + c 2 − c + 1 ≥ a + b + c.

Solution. Using the substitution

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where p
f (u) = e2u − eu + 1 − eu , u ∈ I = R.
We claim that f is convex on I≥s . Since

4e3u − 6e2u + 9eu − 2


e−u f 00 (u) = − 1,
4(e2u − eu + 1)3/2

we need to show that 4x 3 − 6x 2 + 9x − 2 > 0 and

(4x 3 − 6x 2 + 9x − 2)2 ≥ 16(x 2 − x + 1)3 ,

where x = eu ≥ 1. Indeed,

4x 3 − 6x 2 + 9x − 2 = x(x − 3)2 + (3x 3 − 2) > 0

and

(4x 3 − 6x 2 + 9x − 2)2 − 16(x 2 − x + 1)3 = 12x 3 (x − 1) + 9x 2 + 12(x − 1) > 0.

By the RHCF-Theorem, it suffices to prove the original inequality for

b = c := t, a = 1/t 2 , t > 0;

that is, p
t4 − t2 + 1 p 1
2
+ 2 t 2 − t + 1 ≥ 2 + 2t,
t t
2
t −1 2(1 − t)
p +p ≥ 0.
t4 − t2 + 1 + 1 t2 − t + 1 + t
Since
t2 − 1 t2 − 1
p ≥ ,
t4 − t2 + 1 t2 + 1
it suffices to show that
t2 − 1 2(1 − t)
+ p ≥ 0,
t2 + 1 t2 − t + 1 + t
98 Vasile Cîrtoaje

which is equivalent to

t +1 2
• ˜
(t − 1) 2 −p ≥ 0,
t +1 t2 − t + 1 + t
” p —
(t − 1) (t + 1) t 2 − t + 1 − t 2 + t − 2 ≥ 0,

(t − 1)2 (3t 2 − 2t + 3)
p ≥ 0.
(t + 1) t 2 − t + 1 + t 2 − t + 2
The equality holds for a = b = c = 1.

1
P 1.58. If a, b, c, d ≥ p so that abcd = 1, then
1+ 6
1 1 1 1 4
+ + + ≤ .
a+2 b+2 c+2 d +2 3

(Vasile C., 2005)

Solution. Using the notation

a = ex , b = ey, c = ez , d = ew,

we need to show that


x + y +z+w
f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s), s= = 0,
4
where
−1
f (u) = , u ∈ I = R.
eu + 2
For u ≤ 0, we have
eu (2 − eu )
f (u) = u
00
> 0,
(e + 2)3
hence f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the original
inequality for
1
b = c = d := t, a = 1/t 3 , t ≥ p ;
1+ 6
that is,
t3 3 4
+ ≤ ,
2t + 1 t + 2 3
3

which is equivalent to the obvious inequality

(t − 1)2 (5t 2 + 2t − 1) ≥ 0.
Half Convex Function Method 99

According to Note 4, the equality holds for a = b = c = d = 1, and also for


p 1
a = 19 + 9 6, b=c=d= p
1+ 6
(or any cyclic permutation).

P 1.59. If a, b, c are positive real numbers so that abc = 1, then

a2 + b2 + c 2 − 3 ≥ 2(ab + bc + ca − a − b − c).

Solution. Using the substitution

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where
f (u) = e2u − 1 + 2(eu − e−u ), u ∈ R = R.
For u ≥ 0, we have
f 00 (u) = 4e2u + 2(eu − e−u ) > 0,
hence f is convex on I≥s . By the RHCF-Theorem, it suffices to prove the original
inequality for b = c := t and a = 1/t 2 , where t > 0; that is, to show that

4t 5 − 3t 4 − 4t 3 + 2t 2 + 1 ≥ 0,

which is equivalent to

(t − 1)2 (4t 3 + 5t 2 + 2t + 1) ≥ 0.

The equality holds for a = b = c = 1.

P 1.60. If a, b, c are positive real numbers so that abc = 1, then

a2 + b2 + c 2 − 3 ≥ 18(a + b + c − ab − bc − ca).
100 Vasile Cîrtoaje

Solution. Using the substitution

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where
f (u) = e2u − 1 − 18(eu − e−u ), u ∈ R.
For u ≤ 0, we have
f 00 (u) = 4e2u + 18(e−u − eu ) > 0,
hence f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the original
inequality for b = c := t and a = 1/t 2 , where t > 0. Since

1 (t 2 − 1)2 (2t 2 + 1)
a2 + b2 + c 2 − 3 = + 2t 2
− 3 =
t4 t4
and

−(t 4 − 2t 3 + 2t − 1) −(t − 1)3 (t + 1)


a + b + c − ab − bc − ca = = ,
t2 t2
we get

(t − 1)2 (2t − 1)2 (t + 1)(5t + 1)


a2 + b2 + c 2 −3−18(a + b + c − ab − bc − ca) = ≥ 0.
t4
The equality holds for a = b = c = 1, and also for a = 4 and b = c = 1/2 (or any
cyclic permutation).

P 1.61. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


p 1 1 1
 ‹
a1 + a2 + · · · + an − n ≥ 6 3 a1 + a2 + · · · + an −
2 2 2
− − ··· − .
a1 a2 an

Solution. Using the notation ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where p
f (u) = e2u − 1 − 6 3 (eu − e−u ), u ∈ I = R.
Half Convex Function Method 101

For u ≤ 0, we have p
f 00 (u) = 4e2u + 6 3(e−u − eu ) > 0,
hence f is convex on I≤s . By the LHCF-Theorem and Note 2, it suffices to show that
H(x, y) ≥ 0 for x, y ∈ R so that x + (n − 1) y = 0, where

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y

From p
f 0 (u) = 2e2u − 6 3 (eu + e−u ),
we get
2(e x − e y ) x p p 
H(x, y) = e + e y − 3 3 + 3 3 e−x− y .
x−y
Since (e x − e y )/(x − y) > 0, we need to prove that
p p
e x + e y + 3 3 e−x− y ≥ 3 3.

Indeed, by the AM-GM inequality, we have


p Æ
3 p p
e x + e y + 3 3 e−x− y ≥ 3 e x · e y · 3 3 e−x− y = 3 3.

The proof is completed. The equality holds for a1 = a2 = · · · = an = 1.

P 1.62. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that a1 a2 · · · an = 1, then

(n − 1)(a12 + a22 + · · · + an2 ) + n(n + 3) ≥ (2n + 2)(a1 + a2 + · · · + an ).

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that

x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
f (u) = (n − 1)e2u − (2n + 2)eu , u ∈ I = R.
For u ≥ 0, we have

f 00 (u) = 4(n − 1)e2u − (2n + 2)eu


= 2eu [2(n − 1)eu − n − 1]
≥ 2eu [2(n − 1) − n − 1] = 2(n − 3)eu > 0.
102 Vasile Cîrtoaje

Therefore, f is convex on I≥s . By the RHCF-Theorem and Note 2, it suffices to show


that H(x, y) ≥ 0 for x, y ∈ R so that x + (n − 1) y = 0, where
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
From
f 0 (u) = 2(n − 1)e2u − (2n + 2)eu ,
we get
2(e x − e y )
H(x, y) = [(n − 1)(e x + e y ) − (n + 1)] .
x−y
Since (e x − e y )/(x − y) > 0, we need to prove that (n − 1)(e x + e y ) ≥ n + 1. Using
the AM-GM inequality, we have
(n − 1)(e x + e y ) = (n − 1)e x + e y + e y + · · · + e y
Æ
≥ n (n − 1)e x · e y · e y · · · e y
n

Æ pn
= n (n − 1)e x+(n−1) y = n n − 1.
n

Thus, it suffices to show that


p
n
n n − 1 ≥ n + 1,
which is equivalent to
1 n
 ‹
n−1≥ 1+ .
n
This is true for n ≥ 4, since
1 n
 ‹
n−1≥3> 1+ .
n
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1.
Remark. From the proof above, the following sharper inequality follows (Gabriel
Dospinescu and Calin Popa):
• If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then
pn
2n n − 1
a1 + a2 + · · · + an − n ≥
2 2 2
(a1 + a2 + · · · + an − n).
n−1

P 1.63. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If


p, q ≥ 0 so that p + q ≥ n − 1, then
1 1 1 n
+ + ··· + ≥ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q

(Vasile C., 2007)


Half Convex Function Method 103

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
1
f (u) = , u ∈ I = R.
1+ peu + qe2u
For u ≥ 0, we have

eu [4q2 e3u + 3pqe2u + (p2 − 4q)eu − p]


f (u) =
00
(1 + peu + qe2u )3
e2u [4q2 + 3pq + (p2 − 4q) − p]

(1 + peu + qe2u )3
e2u [(p + 2q)(p + q − 2) + 2q2 + p]
= > 0,
(1 + peu + qe2u )3

therefore f is convex on I≥s . By the RHCF-Theorem, it suffices to prove the original


inequality for
a1 = 1/t n−1 , a2 = · · · = an = t, t > 0.
Write this inequality as

t 2n−2 n−1 n
+ ≥ .
t 2n−2 + pt n−1 + q 1 + pt + qt 2 1+p+q

Applying the Cauchy-Schwarz inequality, it suffices to prove that

(t n−1 + n − 1)2 n
≥ ,
(t 2n−2 + pt n−1 + q) + (n − 1)(1 + pt + qt ) 1 + p + q
2

which is equivalent to
pB + qC ≥ A,
where
A = (n − 1)(t n−1 − 1)2 ≥ 0,
A
B = (t n−1 − 1)2 + nE = + nE, E = t n−1 + n − 2 − (n − 1)t,
n−1
A
C = (t n−1 − 1)2 + nF = + nF, F = 2t n−1 + n − 3 − (n − 1)t 2 .
n−1
By the AM-GM inequality applied to n − 1 positive numbers, we have E ≥ 0 and
F ≥ 0 for n ≥ 3. Since A ≥ 0 and p + q ≥ n − 1, we have

(p + q)A
pB + qC − A ≥ pB + qC − = n(pE + qF ) ≥ 0.
n−1
104 Vasile Cîrtoaje

The equality holds for a1 = a2 = · · · = an = 1.

Remark 1. For p = 2k and q = k2 , we get the following result:


• Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If
p
k ≥ n − 1, then

1 1 1 n
+ + ··· + ≥ ,
(1 + ka1 )2 (1 + ka2 )2 (1 + kan )2 (1 + k)2

with equality for a1 = a2 = · · · = an = 1.

In addition, for n = 4 and k = 1, we get the known inequality (Vasile C., 1999):

1 1 1 1
+ + + ≥ 1,
(1 + a)2 (1 + b)2 (1 + c)2 (1 + d)2

where a, b, c, d > 0 so that abcd = 1.

Remark 2. For p + q = n − 1 (n ≥ 3), we get the beautiful inequality

1 1 1
+ + ··· + ≥ 1,
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

which is a generalization of the following inequalities:

1 1 1
+ + ··· + ≥ 1,
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an

1 1 1
p + p + ··· + p ≥ 1,
[1 + ( n − 1)a1 ] 2 [1 + ( n − 1)a1 ] 2 [1 + ( n − 1)a1 ]2
1 1 1 1
+ + ··· + ≥ .
2 + (n − 1)(a1 + a1 ) 2 + (n − 1)(a2 + a2 )
2 2
2 + (n − 1)(an + an ) 2
2

P 1.64. Let a, b, c, d be positive real numbers so that abcd = 1. If p and q are


nonnegative real numbers so that p + q = 3, then

1 1 1 1
+ + + ≥ 1.
1 + pa + qa3 1 + pb + qb3 1 + pc + qc 3 1 + pd + qd 3

(Vasile C., 2007)


Half Convex Function Method 105

Solution. Using the notation

a = ex , b = ey, c = ez , d = ew,

we need to show that


x + y +z+w
f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s), s= = 0,
4
where
1
f (u) = , u ∈ I = R.
1+ peu + qe3u
We will show that f 00 (u) > 0 for u ≥ 0, hence f is convex on I≥s . Since

th(t)
f 00 (u) = ,
(1 + pt + qt 3 )3
where
h(t) = 9q2 t 5 + 2pqt 3 − 9qt 2 + p2 t − p, t = eu ,
we need to show that h(t) ≥ 0 for t ≥ 1. Indeed, we have

h(t) ≥ 9q2 t 3 + 2pqt 3 − 9qt 2 + p2 t − pt = t g(t),

where

g(t) = (9q2 + 2pq)t 2 − 9qt + p2 − p


≥ (9q2 + 2pq)(2t − 1) − 9qt + p2 − p
= q(18q + 4p − 9)t − 9q2 − 2pq + p2 − p
≥ q(18q + 4p − 9) − 9q2 − 2pq + p2 − p
= p2 + 2pq + 9q2 − p − 9q
(p + 9q)(p + q)
= p2 + 2pq + 9q2 −
3
2(p − q) + 16q
2 2
= ≥ 0.
3
By the RHCF-Theorem, it suffices to prove the original inequality for

b = c = d = t, a = 1/t 3 , t > 0;

that is,
t9 3
+ ≥ 1,
t 9 + pt 6 + q 1 + pt + qt 3
3 pt 6 + q
≥ ,
1 + pt + qt 3 t 9 + pt 6 + q
(3 − pq)t 9 − p2 t 7 + 2pt 6 − q2 t 3 − pqt + 2q ≥ 0,
106 Vasile Cîrtoaje

[(p + q)2 − 3pq]t 9 − 3p2 t 7 + 2p(p + q)t 6 − 3q2 t 3 − 3pqt + 2q(p + q) ≥ 0,

Ap2 + Bq2 ≥ C pq,


where
A = t 9 − 3t 7 + 2t 6 = t 6 (t − 1)2 (t + 2) ≥ 0,

B = t 9 − 3t 3 + 2 = (t 3 − 1)2 (t 3 + 2) ≥ 0,

C = t 9 − 2t 6 + 3t − 2.
Since A ≥ 0 and B ≥ 0, it suffices to consider the case C ≥ 0. Since
p
Ap2 + Bq2 ≥ 2 ABpq,

we only need to show that 4AB ≥ C 2 . From

t 3 − 3t + 2 = (t − 1)2 (t + 2) ≥ 0,

we get 3t − 2 ≤ t 3 . Therefore

C ≤ t 9 − 2t 6 + t 3 = t 3 (t 3 − 1)2 ,

hence

4AB − C 2 ≥ 4AB − t 6 (t 3 − 1)4


= t 6 (t − 1)2 (t 3 − 1)2 [4(t + 2)(t 3 + 2) − (t 2 + t + 1)2 ]
= t 6 (t − 1)2 (t 3 − 1)2 (3t 4 + 6t 3 − 3t 2 + 6t + 15) ≥ 0.

The proof is completed. The inequality holds for a = b = c = d = 1.

Remark 1. For p = 1 and p = 2, we get the following nice inequalities:

1 1 1 1
+ + + ≥ 1,
1 + a + 2a 3 1 + b + 2b 3 1 + c + 2c 3 1 + d + 2d 3

1 1 1 1
+ + + ≥ 1.
1 + 2a + a 3 1 + 2b + b 3 1 + 2c + c 3 1 + 2d + d 3

Remark 2. Similarly, we can prove the following generalizations:


• Let a, b, c, d be positive real numbers so that abcd = 1. If p and q are nonnegative
real numbers so that p + q ≥ 3, then

1 1 1 1 4
+ + + ≥ .
1 + pa + qa 3 1 + pb + qb 3 1 + pc + qc 3 1 + pd + qd 3 1+p+q
Half Convex Function Method 107

• Let a1 , a2 , . . . , an (n ≥ 4) be positive real numbers so that a1 a2 · · · an = 1. If


p, q, r ≥ 0 so that p + q + r ≥ n − 1, then
n
X 1 n
≥ .
i=1
1 + pai + qai + r ai
2 3
1+p+q+r

For n = 4 and p + q + r = 3, we get the beautiful inequality


4
X 1
≥ 1.
i=1
1 + pai + qai2 + r ai3

Since
ai + ai3
ai2 ≤
,
2
the best inequality with respect to q if for q = 0:
4
X 1
≥ 1, p + r = 3.
i=1
1 + pai + r ai3

P 1.65. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


1 1 1
+ + ··· + ≥ 1.
1 + a1 + · · · + a1
n−1
1 + a2 + · · · + a2
n−1
1 + an + · · · + ann−1

(Vasile C., 2007)

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
1
f (u) = , u ∈ I = R.
1 + + · · · + e(n−1)u
eu
We will show by induction on n that f is convex on I≥s . Setting t = eu , the condition
f 00 (u) ≥ 0 for u ≥ 0 (t ≥ 1) is equivalent to

2A2 ≥ B(1 + C),

where
A = t + 2t 2 + · · · + (n − 1)t n−1 ,
B = t + 4t 2 + · · · + (n − 1)2 t n−1 ,
C = t + t 2 + · · · + t n−1 .
108 Vasile Cîrtoaje

For n = 2, the inequality becomes t(t − 1) ≥ 0. Assume now that the inequality is
true for n and prove it for n + 1, n ≥ 2. So, we need to show that 2A2 ≥ B(1 + C)
involves
2(A + nt n )2 ≥ (B + n2 t n )(1 + C + t n ),
which is equivalent to

2A2 − B(1 + C) + t n [n2 (t n − 1) + D] ≥ 0,

where
n−1
X
D = 4nA − B − n C = 2
bi t i , bi = 3n2 − (2n − i)2 .
i=1

Since 2A2 − B(1 + C) ≥ 0 (by the induction hypothesis), it suffices to show that
D ≥ 0. Since
b1 < b2 < · · · < bn−1 , t ≤ t 2 ≤ · · · ≤ t n−1 ,
we may apply Chebyshev’s inequality to get

1
D≥ (b1 + b2 + · · · + bn−1 )(t + t 2 + · · · + t n−1 ).
n
Thus, it suffices to show that b1 + b2 + · · · + bn−1 ≥ 0. Indeed,
n−1
X n(n − 1)(4n + 1)
b1 + b2 + · · · + bn−1 = [3n2 − (2n − i)2 ] = > 0.
i=1
6

By the RHCF-Theorem, it suffices to prove the original inequality for

a1 = 1/t n−1 , a2 = · · · = an = t, t ≥ 1,

Setting k = n − 1 (k ≥ 1), we need to show that


2
tk k
+ ≥ 1.
1 + t + ··· + t
k k 2
1 + t + · · · + tk
For the nontrivial case t > 1, this inequality is equivalent to each of the following
inequalities:
k 1 + t k + · · · + t (k−1)k
≥ ,
1 + t + · · · + tk 1 + t k + · · · + t k2
2
k(t − 1) tk − 1 tk − 1
≥ · ,
t k+1 − 1 t k − 1 t (k+1)k − 1
2
k(t − 1) tk − 1
≥ ,
t k+1 − 1 t (k+1)k − 1
2
t (k+1)k − 1 tk − 1
k k+1 ≥ ,
t −1 t −1
Half Convex Function Method 109

k 1 + t k+1 + t 2(k+1) + · · · + t (k−1)(k+1) ≥ 1 + t + t 2 + · · · + t (k−1)(k+1) ,


 

k 1 · 1 + t · t k + · · · + t k−1 · t (k−1)k ≥ 1 + t + · · · + t k−1 1 + t k + · · · + t (k−1)k .


   

Since 1 < t < · · · < t k−1 and 1 < t k < · · · < t (k−1)k , the last inequality follows from
Chebyshev’s inequality.
The equality holds for a1 = a2 = · · · = an = 1.

Remark. Actually, the following generalization holds:


• Let a1 , a2 , . . . , an be positive numbers so that a1 a2 · · · an = 1, and let k1 , k2 , . . . , km ≥
0 so that k1 + k2 + · · · + km ≥ n − 1. If m ≤ n − 1, then
n
X 1 n
≥ .
i=1
1 + k1 ai + k2 ai + · · · + km ai
2 m
1 + k1 + k2 + · · · + km

In addition, since

(m − k)ai + (k − 1)aim
aik ≤ , k = 2, 3, . . . , m − 1
m−1
(by the AM-GM inequality applied to m − 1 positive numbers), the best inequality
with respect to k2 , . . . , km−1 is for k2 = 0, . . . , km−1 = 0; that is,
n
X 1 n
≥ , k1 + km ≥ n − 1, 1 ≤ m ≤ n − 1.
i=1
1 + k1 a i + k m a i
m
1 + k1 + km

If k1 + km = n − 1, then
n
X 1
≥ 1, 1 ≤ m ≤ n − 1,
i=1
1 + k1 ai + km aim

therefore
n
X 1
≥ 1, k1 + kn−1 = n − 1.
i=1
1 + k1 ai + kn−1 ain−1
For k1 = 1 and k1 = n − 2, we get the following strong inequalities:
n
X 1
≥ 1,
i=1
1 + ai + (n − 2)ain−1

n
X 1
≥ 1.
i=1
1 + (n − 2)ai + ain−1
110 Vasile Cîrtoaje

P 1.66. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If

k ≥ n2 − 1,

then
1 1 1 n
+p + ··· + p ≥p .
1 + ka1 1 + ka2 1 + kan 1+k
p

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
1
f (u) = p , u ∈ I = R.
1 + keu
For u ≥ 0, we have
keu (keu − 2) keu (k − 2)
f 00 (u) = ≥ > 0.
4(1 + keu )5/2 4(1 + keu )5/2
Therefore, f is convex on I≥s . By the RHCF-Theorem, it suffices to prove the original
inequality for
a1 = 1/t n−1 , a2 = · · · = an = t, t ≥ 1.
Write this inequality as h(t) ≥ 0, where
v
t t n−1 n−1 n
h(t) = +p −p .
t n−1 +k 1 + kt 1+k
The derivative
(n − 1)kt (n−3)/2 (n − 1)k
h0 (t) = −
2(t n−1 + k)3/2 2(kt + 1)3/2
has the same sign as

h1 (t) = t n/3−1 (kt + 1) − t n−1 − k.

Denoting m = n/3 (m ≥ 2/3), we see that

h1 (t) = kt m + t m−1 − t 3m−1 − k = k(t m − 1) − t m−1 (t 2m − 1) = (t m − 1)h2 (t),

where
h2 (t) = k − t m−1 − t 2m−1 .
For t > 1, we have

h02 (t) = t m−2 [−m + 1 − (2m − 1)t m ] < t m−2 [−m + 1 − (2m − 1)]
= −(3m − 2)t m−2 ≤ 0,
Half Convex Function Method 111

hence h2 (t) is strictly decreasing for t ≥ 1. Since

h2 (1) = k − 2 > 0, lim h2 (t) = −∞,


t→∞

there exists t 1 > 1 so that h2 (t 1 ) = 0, h2 (t) > 0 for t ∈ [1, t 1 ), h2 (t) < 0 for
t ∈ (t 1 , ∞). Since h2 (t), h1 (t) and h0 (t) has the same sign for t > 1, h(t) is strictly
increasing for t ∈ [1, t 1 ] and strictly decreasing for t ∈ [t 1 , ∞); this yields

h(t) ≥ min{h(1), h(∞)}.


n
From h(1) = 0 and h(∞) = 1 − p ≥ 0, it follows that h(t) ≥ 0 for all t ≥ 1.
1+k
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1.
Remark. The following generalization holds (Vasile C., 2005):
• Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If k and m are
positive numbers so that

m ≤ n − 1, k ≥ n1/m − 1,

then
1 1 1 n
+ + ··· + ≥ ,
(1 + ka1 ) m (1 + ka2 ) m (1 + kan ) m (1 + k)m
with equality for a1 = a2 = · · · = an = 1.

For 0 < m ≤ n − 1 and k = n1/m − 1, we get the beautiful inequality


1 1 1
+ + ··· + ≥ 1.
(1 + ka1 ) m (1 + ka2 ) m (1 + kan )m

P 1.67. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If p, q ≥ 0


1
so that 0 < p + q ≤ , then
n−1
1 1 1 n
+ + ··· + ≤ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q

(Vasile C., 2007)


Solution. Using the notation ai = e x i for i = 1, 2, . . . , n, we need to show that
x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
−1
f (u) = , u ∈ I = R.
1 + peu + qe2u
112 Vasile Cîrtoaje

For u ≤ 0, we have

eu [−4q2 e3u − 3pqe2u + (4q − p2 )eu + p]


f 00 (u) =
(1 + peu + qe2u )3
e2u [−4q2 e2u − 3pqeu + (4q − p2 ) + pe−u ]
=
(1 + peu + qe2u )3
e2u [−4q2 − 3pq + (4q − p2 ) + p]

(1 + peu + qe2u )3
e [(p + 4q)(1 − p − q) + 2pq]
2u
= ≥ 0,
(1 + peu + qe2u )3
therefore f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the original
inequality for
a1 = 1/t n−1 , a2 = · · · = an = t, t > 0.
Write this inequality as

t 2n−2 n−1 n
+ ≤ ,
t 2n−2 + pt n−1 + q 1 + pt + qt 2 1+p+q

p2 A + q2 B + pqC ≤ pD + qE,
where
A = t n−1 (t n − nt + n − 1), B = t 2n − nt 2 + n − 1,
C = t 2n−1 + t 2n − nt n+1 + (n − 1)t n−1 − nt + n − 1,
D = t n−1 [(n − 1)t n + 1 − nt n−1 ], E = (n − 1)t 2n + 1 − nt 2n−2 .
Applying the AM-GM inequality to n positive numbers yields D ≥ 0 and E ≥ 0.
Since (n − 1)(p + q) ≤ 1 involves pD + qE ≥ (n − 1)(p + q)(pD + qE), it suffices to
show that
p2 A + q2 B + pqC ≤ (n − 1)(p + q)(pD + qE).
Write this inequality as
p2 A1 + q2 B1 + pqC1 ≥ 0,
where
A1 = (n − 1)D − A = nt n [(n − 2)t n−1 + 1 − (n − 1)t n−2 ],
B1 = (n − 1)E − B = nt 2 [(n − 2)t 2n−2 + 1 − (n − 1)t 2n−4 ],
C1 = (n − 1)(D + E) − C = nt[(n − 2)(t 2n−1 + t 2n−2 ) − 2(n − 1)t 2n−3 + t n + 1].
Applying the AM-GM inequality to n − 1 nonnegative numbers yields A1 ≥ 0 and
B1 ≥ 0. So, it suffices to show that C1 ≥ 0. Indeed, we have

(n − 2)(t 2n−1 + t 2n−2 ) − 2(n − 1)t 2n−3 + t n + 1 = A2 + B2 + C2 ,

where
A2 = (n − 2)t 2n−1 + t − (n − 1)t 2n−3 ≥ 0,
Half Convex Function Method 113

B2 = (n − 2)t 2n−2 + t n−1 − (n − 1)t 2n−3 ≥ 0,


C2 = t n − t n−1 − t + 1 = (t − 1)(t n−1 − 1) ≥ 0.
The inequalities A2 ≥ 0 and B2 ≥ 0 follow by applying the AM-GM inequality to
n − 1 nonnegative numbers.
The equality holds for a1 = a2 = · · · = an = 1.
1
Remark 1. For p + q = , we get the inequality
n−1
1 1 1
+ + ··· + ≤ n − 1,
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

which is a generalization of the following inequalities:

1 1 1
+ + ··· + ≤ 1,
n − 1 + a1 n − 1 + a2 n − 1 + an

1 1 1 1
+ + ··· + ≤ .
2n − 2 + a1 + a1 2n − 2 + a2 + a2
2 2
2n − 2 + an + an2 2

Remark 2. For
4n − 3 1
p= , q= ,
2(n − 1)(2n − 1) 2(n − 1)(2n − 1)

we get the inequality

1 1 1
+ ··· + ≤ ,
(a1 + 2n − 2)(a1 + 2n − 1) (an + 2n − 2)(an + 2n − 1) 4n − 2

which is equivalent to

1 1 1 1 1
+ ··· + ≤ + + ··· + .
a1 + 2n − 2 an + 2n − 2 4n − 2 a1 + 2n − 1 an + 2n − 1

Remark 3. For p = 2k and q = k2 , we get the following statement:


• Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If
s
n
0<k≤ − 1,
n−1
then
1 1 1 n
+ + ··· + ≤ ,
(1 + ka1 )2 (1 + ka2 )2 (1 + kan )2 (1 + k)2
with equality for a1 = a2 = · · · = an = 1.
114 Vasile Cîrtoaje

P 1.68. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If


2n − 1
0<k≤ ,
(n − 1)2

then
1 1 1 n
+p + ··· + p ≤p .
1 + ka1 1 + ka2 1 + kan 1+k
p

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
−1
f (u) = p , u ∈ I = R.
1 + keu
For u ≤ 0, we have

keu (2 − keu ) keu (2 − k)


f 00 (u) = ≥ > 0.
4(1 + keu )5/2 4(1 + keu )5/2

Therefore, f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the original


inequality for

a1 = 1/t n−1 , a2 = · · · = an = t. 0 < t ≤ 1.

Write this inequality as h(t) ≤ 0, where


v
t t n−1 n−1 n
h(t) = +p −p .
t n−1 +k 1 + kt 1+k
The derivative
(n − 1)kt (n−3)/2 (n − 1)k
h0 (t) = −
2(t n−1 + k)3/2 2(kt + 1)3/2
has the same sign as

h1 (t) = t n/3−1 (kt + 1) − t n−1 − k.

Denoting m = n/3, m ≥ 1, we see that

h1 (t) = kt m + t m−1 − t 3m−1 − k = −k(1 − t m ) + t m−1 (1 − t 2m ) = (1 − t m )h2 (t),

where
h2 (t) = t m−1 + t 2m−1 − k
is strictly increasing for t ∈ [0, 1]. There are two possible cases: h2 (0) ≥ 0 and
h2 (0) < 0.
Half Convex Function Method 115

Case 1: h2 (0) ≥ 0. This case is possible only for m = 1 and k ≤ 1, when h2 (t) =
t + 1 − k > 0 for t ∈ (0, 1]. Also, we have h1 (t) > 0 and h0 (t) > 0 for t ∈ (0, 1).
Therefore, h is strictly increasing on [0, 1], hence h(t) ≤ h(1) = 0.
Case 2: h2 (0) < 0. This case is possible for either m = 1 (n = 3) and 1 < k ≤ 5/4,
or m > 1 (n ≥ 4). Since h2 (1) = 2−k > 0, there exists t 1 ∈ (0, 1) so that h2 (t 1 ) = 0,
h2 (t) < 0 for t ∈ (0, t 1 ), and h2 (t) > 0 for t ∈ (t 1 , 1). Since h0 has the same sign as
h2 on (0, 1), it follows that h is strictly decreasing on [0, t 1 ] and strictly increasing
n
on [t 1 , 1]. Therefore, h(t) ≤ max{h(0), h(1)}. Since h(0) = n − 1 − p ≤ 0 and
1+k
h(1) = 0, we have h(t) ≤ 0 for all t ∈ (0, 1].
The equality holds for a1 = a2 = · · · = an = 1.
Remark. The following generalization holds (Vasile C., 2005):
• Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If k
and m are positive numbers so that

1  n 1/m
m≥ , k≤ − 1,
n−1 n−1

then
1 1 1 n
+ + ··· + ≤ ,
(1 + ka1 ) m (1 + ka2 ) m (1 + kan ) m (1 + k)m
with equality for a1 = a2 = · · · = an = 1.

1  n 1/m
For n ≥ 3, m ≥ and k = − 1, we get the beautiful inequality
n−1 n−1
1 1 1
+ + ··· + ≤ n − 1.
(1 + ka1 ) m (1 + ka2 ) m (1 + kan )m

P 1.69. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


v v v
t 2n − 1 t 4 2n − 1 t 2n − 1 1
a14 + + a + +· · ·+ a 4+ ≥ (a1 +a2 +· · ·+an )2 .
(n − 1)2 2
(n − 1)2 n
(n − 1)2 n−1

(Vasile C., 2006)

Solution. According to the preceding P 1.68, the following inequality holds


X 1
q ≤ n − 1.
2n−1 −4
1 + (n−1)2 a1
116 Vasile Cîrtoaje

On the other hand, by the Cauchy-Schwarz inequality


! v 
X 1 X t 2n − 1 −4 €X Š2
a1 1 +
2
a ≥ a1 .
(n − 1)2 1
q
1 + 2n−12 a−4
(n−1) 1

From these inequalities, we get


X v 
t 2n − 1 −4 €X Š2
(n − 1) a1 1 +
2
a ≥ a1 ,
(n − 1)2 1
which is the desired inequality.
The equality holds for a1 = a2 = · · · = an = 1.

P 1.70. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


1 1 1
 ‹
a1 + a2 + · · · + an + n(n − 2) ≥ (n − 1)
n−1 n−1 n−1
+ + ··· + .
a1 a2 an

Solution. Using the notation ai = e x i for i = 1, 2, . . . , n, we need to show that


x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
f (u) = e(n−1)u − (n − 1)e−u , u ∈ I = R.
For u ≥ 0, we have
f 00 (u) = (n − 1)2 e(n−1)u − (n − 1)e−u = (n − 1)e−u [(n − 1)e nu − 1] ≥ 0;
therefore, f is convex on I≥s . By the RHCF-Theorem and Note 2, it suffices to show
that H(x, y) ≥ 0 for x, y ∈ R so that x + (n − 1) y = 0, where
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
From
f 0 (u) = (n − 1)[e(n−1)u + e−u ],
we get
(n − 1)(e x − e y )  (n−2)x
+ e(n−3)x+ y + · · · + e x+(n−3) y + e(n−2) y − e−x− y

H(x, y) = e
x−y
(n − 1)(e x − e y )  (n−2)x
+ e(n−3)x+ y + · · · + e x+(n−3) y ) .

= e
x−y
Since (e x − e y )/(x − y) > 0, we have H(x, y) > 0.
The equality holds for a1 = a2 = · · · = an = 1.
Half Convex Function Method 117

P 1.71. Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If k ≥ n,


then
1 1 1
 ‹
a1 + a2 + · · · + an + kn ≥ (k + 1)
k k k
+ + ··· + .
a1 a2 an
(Vasile C., 2006)
Solution. Using the notations ai = e x i for i = 1, 2, . . . , n, we need to show that
x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
f (u) = e ku − (k + 1)e−u , u ∈ I = R.
For u ≥ 0, we have

f 00 (u) = k2 e ku − (k + 1)e−u = e−u k2 e(k+1)u − k − 1 ≥ e−u (k2 − k − 1) > 0;


 

therefore, f is convex on I≥s . By the RHCF-Theorem, it suffices to to prove the


original inequality for a1 ≤ 1 ≤ a2 = · · · = an ; that is, to show that
k + 1 (k + 1)(n − 1)
a k + (n − 1)b k − − + kn ≥ 0
a b
for
ab n−1 = 1, 0 < a ≤ 1 ≤ b.
By the weighted AM-GM inequality, we have
k k(n − 1)
a k + (kn − k − 1) ≥ [1 + (kn − k − 1)]a 1+(kn−k−1) = .
b
Thus, we still have to show that
1 1
 ‹  ‹
(n − 1) b −
k
− (k + 1) − 1 ≥ 0,
b a
which is equivalent to h(b) ≥ 0 for b ≥ 1, where

h(b) = (n − 1)(b k+1 − 1) − (k + 1)(b n − b).

Since
h0 (b)
= (n − 1)b k − nb n−1 + 1 ≥ (n − 1)b n − nb n−1 + 1
k+1
= nb n−1 (b − 1) − (b n − 1)
 
= (b − 1) (b n−1 − b n−2 ) + (b n−1 − b n−3 ) + · · · + (b n−1 − 1) ≥ 0,

h is increasing on [1, ∞), hence h(b) ≥ h(1) = 0. The proof is completed. The
equality holds for a1 = a2 = · · · = an = 1.
118 Vasile Cîrtoaje

P 1.72. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 a1 1 a2 1 an
 ‹  ‹  ‹
1− + 1− + ··· + 1 − ≤ n − 1.
n n n

(Vasile C., 2006)

Solution. Let
n
k= , k > 1,
n−1
and
m = ln k, 0 < m ≤ ln 2 < 1.
Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that
x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where
u
f (u) = −k−e , u ∈ I = R.
From
u
f 00 (u) = meu k−e (1 − meu ),
it follows that f 00 (u) > 0 for u ≤ 0, since

1 − meu ≥ 1 − m ≥ 1 − ln 2 > 0.

Therefore, f is convex on I≤s . By the LHCF-Theorem and Note 5, it suffices to prove


the original inequality for

a2 = · · · = an := t, a1 = t −n+1 , 0 < t ≤ 1.

Write this inequality as


h(t) ≤ n − 1,
where
−n+1
h(t) = k−t + (n − 1)k−t , t ∈ (0, 1].
We have
−n+1 −n+1
h0 (t) = (n − 1)mt −n k−t h1 (t), h1 (t) = 1 − t n k t −t
,
−n+1
h01 (t) = k t −t
h2 (t), h2 (t) = m(n − 1 + t n ) − nt n−1 .
Since

h02 (t) = nt n−2 (mt − n + 1) ≤ nt n−2 (m − n + 1) ≤ nt n−2 (m − 1) < 0,

h2 is strictly decreasing on [0, 1]. From

h2 (0) = (n − 1)m > 0, h2 (1) = n(m − 1) < 0,


Half Convex Function Method 119

it follows that there is t 1 ∈ (0, 1) so that h2 (t 1 ) = 0, h2 (t) > 0 for t ∈ [0, t 1 ) and
h2 (t) < 0 for t ∈ (t 1 , 1]. Therefore, h1 is strictly increasing on (0, t 1 ] and strictly
decreasing on [t 1 , 1]. Since h1 (0+ ) = −∞ and h1 (1) = 0, there is t 2 ∈ (0, t 1 ) so
that h1 (t 2 ) = 0, h1 (t) < 0 for t ∈ (0, t 2 ), h1 (t) > 0 for t ∈ (t 2 , 1). Thus, h is strictly
decreasing on (0, t 2 ] and strictly increasing on [t 2 , 1]. Since h(0+ ) = n − 1 and
h(1) = n − 1, we have h(t) ≤ n − 1 for all t ∈ (0, 1]. This completes the proof. The
equality holds for a1 = a2 = · · · = an = 1.

P 1.73. If a, b, c are positive real numbers so that abc = 1, then

1 1 1
p + p + p ≤ 1.
1 + 1 + 3a 1 + 1 + 3b 1 + 1 + 3c

(Vasile C., 2008)

Solution. Write the inequality as


p p p
1 + 3a − 1 1 + 3b − 1 1 + 3c − 1
+ + ≤ 1,
3a 3b 3c
v v v
1 1 1 t1 3 t1 3 t1 3
+ + +3≥ + + + + + .
a b c a2 a b2 b c2 c
Replacing a, b, c by 1/a, 1/b, 1/c, respectively, we need to prove that abc = 1 in-
volves
p p p
a + b + c + 3 ≥ a2 + 3a + b2 + 3b + c 2 + 3c. (*)

Using the notation


a = ex , b = ey, c = ez ,
we need to show that
x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where p
f (u) = eu − e2u + 3eu , u ∈ I = R.
We have
4t 2 + 18t + 9
 
f (u) = t 1 −
00
p , t = eu ≥ 1.
4(t + 3) t(t + 3)
For u ≥ 0, which involves t ≥ 1, from

16t(t + 3)3 − (4t 2 + 18t + 9)2 = 9(4t 2 + 12t − 9) > 0,


120 Vasile Cîrtoaje

it follows that f 00 > 0, hence f is convex on I≥s . By the RHCF-Theorem, it suffices


to prove the inequality (*) for b = c. Thus, we need to show that
p p
a− a2 + 3a + 2(b − b2 + 3b ) + 3 ≥ 0

for a b2 = 1. Write this inequality as


p p
2b3 + 3b2 + 1 ≥ 3b2 + 1 + 2b2 b2 + 3b.

Squaring and dividing by b2 , the inequality becomes


Æ
9b2 + 4b + 3 ≥ 4 (b2 + 3b)(3b2 + 1).

Since
Æ
2 (b2 + 3b)(3b2 + 1) ≤ (b2 + 3b) + (3b2 + 1) = 4b2 + 3b + 1,

it suffices to show that

9b2 + 4b + 3 ≥ 2(4b2 + 3b + 1),

which is equivalent to (b − 1)2 ≥ 0. The equality holds for a = b = c = 1.


Remark. In the same manner, we can prove the following generalization:
• Let a1 , a2 , . . . , an be positive real numbers so that a1 a2 · · · an = 1. If

4n
0<k≤ ,
(n − 1)2

then

1 1 1 n
+ + ··· + ≤ p .
1 + 1 + ka1 1 + 1 + ka2 1 + 1 + kan 1+ 1+k
p p p

P 1.74. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1 1 1
+ + ··· + ≥ .
1+ 1 + 4n(n − 1)a1 1+ 1 + 4n(n − 1)a2 1+ 1 + 4n(n − 1)an
p p p
2

(Vasile C., 2008)


Half Convex Function Method 121

Solution. Denote
k = 4n(n − 1), k ≥ 8,
and write the inequality as follows:

1 + ka1 − 1 1 + ka2 − 1 1 + kan − 1


p p p
1
+ + ··· + ≥ ,
ka1 ka2 kan 2
v v v
1 k 1 k t1 k 1 1 1 k
u u u
+ + + + · · · + + ≥ + + ··· + + .
t t
2 2 2
a1 a1 a2 a2 a1 a1 a1 a2 an 2

Replacing a1 , a2 , . . . , an by 1/a1 , 1/a2 , . . . , 1/an , we need to prove that a1 a2 · · · an =


1 implies
q q Æ k
a12 + ka1 + a22 + ka2 + · · · + an2 + kan ≥ a1 + a2 + · · · + an + . (*)
2
Using the substitutions ai = e x i for i = 1, 2, . . . , n, we need to show that
x1 + x2 + · · · + x n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s), s= = 0,
n
where p
f (u) = e2u + keu − eu , u ∈ I = R.
We will show that f 00 (u) > 0 for u ≤ 0. Indeed, denoting t = eu , t ∈ (0, 1], we have

4t 2 + 6kt + k2
 
f (u) = t
00
p −1 >0
4(t + k) t(t + k)

because

(4t 2 + 6kt + k2 )2 − 16t(t + k)3 = k2 (k2 − 4kt − 4t 2 ) ≥ k2 (k2 − 4k − 4) > 0.

Thus, f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the inequality


(*) for a2 = a3 = · · · = an ; that is, to show that
p €p Š €p Š
a2 + ka − a + (n − 1) b2 + kb − b ≥ n 1+k−1 ,

for all positive a, b satisfying ab n−1 = 1. Write this inequality as


p p
kb n−1 + 1 + (n − 1) kb2n−1 + b2n ≥ (n − 1)b n + 2n(n − 1)b n−1 + 1.

By Minkowski’s inequality, we have


p p
kb n−1 + 1 + (n − 1) kb2n−1 + b2n ≥
Æ
≥ kb n−1 [1 + (n − 1)b n/2 ]2 + [1 + (n − 1)b n ]2 .
122 Vasile Cîrtoaje

Thus, it suffices to show that

k b n−1 [1 + (n − 1)b n/2 ]2 + [1 + (n − 1)b n ]2 ≥ [(n − 1)b n + 2n(n − 1)b n−1 + 1]2 ,

which is equivalent to
3n−2
” n n−2
—
4n(n − 1)2 b 2 2 + (n − 2)b 2 − nb 2 ≥ 0.

This inequality follows immediately by the AM-GM inequality applied to n positive


numbers.
The equality holds for a1 = a2 = · · · = an = 1.

P 1.75. If a, b, c are positive real numbers so that abc = 1, then

a6 b6 c6
+ + ≥ 1.
1 + 2a5 1 + 2b5 1 + 2c 5

(Vasile C., 2008)

Solution. Using the substitution

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where
e6u
f (u) = , u ∈ I = R.
1 + 2e5u
For u ≤ 0, which involves w = eu ∈ (0, 1], we have

2w6 (2 − w5 )(9 − 2w5 )


f 00 (u) = > 0.
(1 + 2w5 )3

Therefore, f is convex on I≤s . By the LHCF-Theorem, it suffices to prove the original


inequality for b = c and ab2 = 1; that is,

1 2b6
+ ≥ 1.
b2 (b10 + 2) 1 + 2b5

Since
1 + 2b5 ≤ 1 + b4 + b6 ,
Half Convex Function Method 123

it suffices to show that


1 2x 3 p
+ ≥ 1, x= b.
x(x 5 + 2) 1 + x 2 + x 3

This inequality can be written as follows:

x 3 (x 6 − x 5 − x 3 + 2x − 1) + (x − 1)2 ≥ 0,

x 3 (x − 1)2 (x 4 + x 3 + x 2 − 1) + (x − 1)2 ≥ 0,
(x − 1)2 [x 7 + x 5 + (x 6 − x 3 + 1)] ≥ 0.
The equality holds for a = b = c = 1.

P 1.76. If a, b, c are positive real numbers so that abc = 1, then


p p p
25a2 + 144 + 25b2 + 144 + 25c 2 + 144 ≤ 5(a + b + c) + 24.

(Vasile C., 2008)

Solution. Using the notation

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where p
f (u) = 5eu − 25e2u + 144, u ∈ R.
We will show that f (u) is convex for u ≤ 0. From

5w(25w2 + 288)
• ˜
f (u) = 5w 1 −
00
, w = eu ∈ (0, 1],
(25w2 + 144)3/2

we need to show that

(25w2 + 144)3 ≥ 25w2 (25w2 + 288)2 .

25
 ˜
Setting 25w2 = 144z, we have z ∈ 0, and
144

(25w2 + 144)3 − 25w2 (25w2 + 288)2 = 1443 (z + 1)3 − 1443 z(z + 2)2
= 1443 (1 − z − z 2 ) > 0.
124 Vasile Cîrtoaje

By the LHCF-Theorem, it suffices to prove the original inequality for

a = t 2, b = c = 1/t, t > 0;

that is, p p
5t 3 + 24t + 10 ≥ 25t 6 + 144t 2 + 2 25 + 144t 2 .
Squaring and dividing by 4t give
Æ
60t 3 + 25t 2 − 36t + 120 ≥ (25t 4 + 144)(144t 2 + 25).

Squaring again and dividing by 120, the inequality becomes

25t 5 − 36t 4 + 105t 3 − 112t 2 − 72t + 90 ≥ 0,

(t − 1)2 (25t 3 + 14t 2 + 108t + 90) ≥ 0.


The equality holds for a = b = c = 1.

P 1.77. If a, b, c are positive real numbers so that abc = 1, then


p p p
16a2 + 9 + 16b2 + 9 + 16c 2 + 9 ≥ 4(a + b + c) + 3.

(Vasile C., 2008)

Solution. Using the substitution

a = ex , b = ey, c = ez ,

we need to show that


x + y +z
f (x) + f ( y) + f (z) ≥ 3 f (s), s= = 0,
3
where p
f (u) = 16e2u + 9 − 4eu , u ∈ R.
We will show that f (u) is convex for u ≥ 0. From

4w(16w2 + 18)
• ˜
f (u) = 4w
00
−1 , w = eu ≥ 1,
(16w2 + 9)3/2

we need to show that

16w2 (16w2 + 18)2 ≥ (16w2 + 9)3 .


Half Convex Function Method 125

16
Setting 16w2 = 9z, we have z ≥ and
9

16w2 (16w2 + 18)2 − (16w2 + 9)3 = 729z(z + 2)2 − 729(z + 1)3


= 729(z 2 + z − 1) > 0.

By the RHCF-Theorem, it suffices to prove the original inequality for

a = t 2, b = c = 1/t, t > 0;

that is, p p
16t 6 + 9t 2 + 2 16 + 9t 2 ≥ 4t 3 + 3t + 8.
Squaring and dividing by 4t give
Æ
(16t 4 + 9)(9t 2 + 16) ≥ 6t 3 + 16t 2 − 9t + 12.

Squaring again and dividing by 12t, the inequality becomes

9t 5 − 16t 4 + 9t 3 + 12t 2 − 32t + 18 ≥ 0,

(t − 1)2 (9t 3 + 2t 2 + 4t + 18) ≥ 0.


The equality holds for a = b = c = 1.

P 1.78. If ABC is a triangle, then


 ‹  ‹  ‹
A B C
sin A 2 sin − 1 + sin B 2 sin − 1 + sin C 2 sin − 1 ≥ 0.
2 2 2
(Lorian Saceanu, 2015)

Solution. Write the inequality as


A+ B + C π
f (A) + f (B) + f (C) ≥ 3 f (s), s= = ,
3 3
where
 u  u 3u
f (u) = sin u 2 sin − 1 = cos − cos − sin u, u ∈ I = [0, π].
2 2 2
We will show that f is convex on I≤s . Indeed, for u ∈ [0, π/3], we have
u u 2 u
 u u 2 u

f (u) = cos
00
2 + 2 sin − 9 sin ≥ cos 2 + 2 sin − 12 sin
2  2  2 2 2 2
u u u 
= 2 cos 1 + 3 sin 1 − 2 sin ≥ 0.
2 2 2
126 Vasile Cîrtoaje

By the LHCF-Theorem, it suffices to prove the original inequality for B = C, when


it transforms into
 ‹
B
sin 2B(2 cos B − 1) + 2 sin B 2 sin − 1 ≥ 0,
2
 ‹ ‹2
B B B
sin B sin sin + 1 2 sin − 1 ≥ 0.
2 2 2
The equality occurs for an equilateral triangle, and for a degenerate triangle with
A = π and B = C = 0 (or any cyclic permutation).
Remark. Based on this inequality, we can prove the following statement:
• If ABC is a triangle, then

sin 2A(2 cos A − 1) + sin 2B(2 cos B − 1) + sin 2C(2 cos C − 1) ≥ 0,

with equality for an equilateral triangle, for a degenerate triangle with A = 0 and
B = C = π/2 (or any cyclic permutation), and for a degenerate triangle with A = π
and B = C = 0 (or any cyclic permutation).
If ABC is an acute or right triangle, then this inequality follows by replacing A,
B and C with π − 2A, π − 2B and π − 2C in the inequality from P 1.78. Consider
now that
π
A > > B ≥ C ≥ 0.
2
The inequality is true for B ≤ π/3, because

sin 2A(2 cos A − 1) ≥ 0, sin 2B(2 cos B − 1) ≥ 0, sin 2C(2 cos C − 1) ≥ 0.

Consider further that


2π π π
> A > > B > > C ≥ 0.
3 2 3
From
1 − 2 cos A > 1 − 2 cos B,
it follows that

(− sin 2A)(1 − 2 cos A) > (− sin 2A)(1 − 2 cos B).

Therefore it suffices to

(− sin 2A)(1 − 2 cos B) + sin 2B(2 cos B − 1) + sin 2C(2 cos C − 1) ≥ 0,

which is equivalent to

(sin 2A + sin 2B)(2 cos B − 1) + sin 2C(2 cos C − 1) ≥ 0,


Half Convex Function Method 127

2 sin C cos(A − B)(2 cos B − 1) + 2 sin C cos C(2 cos C − 1) ≥ 0.


This inequality is true if

cos(A − B)(2 cos B − 1) + cos C(2 cos C − 1) ≥ 0,

which can be written as

cos C(2 cos C − 1) ≥ cos(A − B)(1 − 2 cos B).

Since
2π π π
C < A− B < − = ,
3 3 3
we have cos C > cos(A − B). Therefore, it suffices to show that

2 cos C − 1 ≥ 1 − 2 cos B,

which is equivalent to
cos B + cos C ≥ 1.
From B + C < π/2, we get cos B > cos(π/2 − C) = sin C, hence
p
cos B + cos C > sin C + cos C = 1 + sin 2C ≥ 1.

P 1.79. If ABC is an acute or right triangle, then


 ‹  ‹  ‹
A B C
sin 2A 1 − 2 sin + sin 2B 1 − 2 sin + sin 2C 1 − 2 sin ≥ 0.
2 2 2

(Vasile C., 2015)

Solution. Write the inequality as


A+ B + C π
f (A) + f (B) + f (C) ≥ 3 f (s), s= = ,
3 3
where
 u 3u 5u
f (u) = sin 2u 1 − 2 sin = sin 2u − cos + cos , u ∈ I = [0, π/2].
2 2 2
We will show that f is convex on [s, π/2]. From
9 3u 25 5u
f 00 (u) = −4 sin 2u + cos − cos
4 2 4 2
and
3u 5u u
cos − cos = 2 sin sin 2u ≥ 0,
2 2 2
128 Vasile Cîrtoaje

we get
9 5u 25 5u
f 00 (u) ≥ −4 sin 2u + cos − cos
4 2 ˜4 2
π − 5u π−u 5π − 9u
•
= −4 sin 2u + sin = 8 sin cos .
2 4 4
For π/3 ≤ u ≤ π/2, we have
π 5π − 9u π
≤ ≤ ,
8 4 2
hence f 00 (u) ≥ 0. By the RHCF-Theorem, it suffices to prove the original inequality
for B = C, 0 ≤ B ≤ π/2, when it becomes
 ‹
B
− sin 4B(1 − 2 cos B) + 2 sin 2B 1 − 2 sin ≥ 0,
2
• ˜
B
2 sin 2B cos 2B(2 cos B − 1) + 1 − sin ≥ 0.
2
We need to show that
B
cos 2B(2 cos B − 1) + 1 − sin ≥ 0,
2
which is equivalent to g(t) ≥ 0, where
B 1
g(t) = (1 − 8t 2 + 8t 4 )(1 − 4t 2 ) + 1 − 2t, t = sin , 0≤t ≤ p .
2 2
Indeed, we have

g(t) = 2(1 − t)2 (1 + 3t + 2t 2 − 4t 3 − 4t 4 ) ≥ 0

because

1 + 3t + 2t 2 − 4t 3 − 4t 4 ≥ 1 + 3t + 2t 2 − 2t − 2t 2 = 1 + t > 0.

The equality occurs for an equilateral triangle, for a degenerate triangle with
A = 0 and and B = C = π/2 (or any cyclic permutation), and for a degenerate
triangle with A = π and B = C = 0 (or any cyclic permutation).
Remark 1. Actually, the inequality holds also for an obtuse triangle ABC. To prove
this, consider that
π
A > > B ≥ C ≥ 0.
2
The inequality is true for B ≤ π/3, because
 ‹  ‹  ‹
A B C
sin 2A 1 − 2 sin ≥ 0, sin 2B 1 − 2 sin ≥ 0, sin 2C 1 − 2 sin ≥ 0.
2 2 2
Half Convex Function Method 129

Consider further that


2π π π
> A > > B > > C ≥ 0.
3 2 3
From
A B
2 sin − 1 > 2 sin − 1,
2 2
it follows that
 ‹  ‹
A B
(− sin 2A) 2 sin − 1 > (− sin 2A) 2 sin − 1 .
2 2
Therefore it suffices to
 ‹  ‹  ‹
B B C
(− sin 2A) 2 sin − 1 + sin 2B 1 − 2 sin + sin 2C 1 − 2 sin ≥ 0,
2 2 2
which is equivalent to
 ‹  ‹
B C
(sin 2A + sin 2B) 1 − 2 sin + sin 2C 1 − 2 sin ≥ 0,
2 2
 ‹  ‹
B C
2 sin C cos(A − B) 1 − 2 sin + 2 sin C cos C 1 − 2 sin ≥ 0.
2 2
This inequality is true if
 ‹  ‹
B C
cos(A − B) 1 − 2 sin + cos C 1 − 2 sin ≥ 0,
2 2
which can be written as
 ‹  ‹
C B
cos C 1 − 2 sin ≥ cos(A − B) 2 sin − 1 .
2 2
Since
2π π π
C < A− B < − = ,
3 3 3
we have cos C > cos(A − B). Therefore, it suffices to show that
C B
1 − 2 sin ≥ 2 sin − 1,
2 2
which is equivalent to
B C
sin + sin ≤ 1,
2 2
B+C B−C
2 sin cos ≤ 1.
4 4
This is true since
B+C π B−C
2 sin < 2 sin < 1, cos < 1.
4 8 4
130 Vasile Cîrtoaje

Remark 2. Replacing A, B and C in P 1.79 by π−2A, π−2B and π−2C, respectively,


we get the following inequality for an acute or right triangle ABC:

sin 4A(2 cos A − 1) + sin 4B(2 cos B − 1) + sin 4C(2 cos C − 1) ≥ 0,

with equality for an equilateral triangle, for a triangle with A = π/2 and B = C =
π/4 (or any cyclic permutation), and for a degenerate triangle with A = 0 and and
B = C = π/2 (or any cyclic permutation).

P 1.80. If a, b, c, d are real numbers so that a + b + c + d = 4, then


a b c d
+ + + ≤ 1.
a2 − a + 4 b2 − b + 4 c 2 − c + 4 d 2 − d + 4
(Sqing, 2015)
Solution. Write the inequality as
a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
−u
f (u) = , u ∈ R.
u2 − u + 4
We see that
(u − 2)2
f (u) − f (2) = ≥ 0.
3(u2 − u + 4
From
2(−u3 + 12u − 4)
f 00 (u) = ,
(u2 − u + 4)3
it follows that f is convex on [1, 2]. Define the function

 f (u), u ≤ 2

f0 (u) = .
 f (2), u > 2

Since f0 (u) ≤ f (u) for u ∈ R and f0 (1) = f (1), it suffices to show that

f0 (a) + f0 (b) + f0 (c) + f0 (d) ≥ 4 f0 (s).

The function f0 is convex on [1, ∞) because it is differentiable on [1, ∞) and its


derivative
 f (u), u ≤ 2
 0

f0 (u) =
0
 0, u>2
Half Convex Function Method 131

is continuous and increasing on [1, ∞). Therefore, by the RHCF-Theorem, we only


need to show that f0 (x) + 3 f0 ( y) ≥ 4 f0 (1) for all x, y ∈ R so that x ≤ 1 ≤ y and
x + 3 y = 4. There are two cases to consider: y ≤ 2 and y > 2.
Case 1: y ≤ 2. The inequality f0 (x) + 3 f0 ( y) ≥ 4 f0 (1) is equivalent to f (x) +
3 f ( y) ≥ 4 f (1). According to Note 1, this is true if h(x, y) ≥ 0 for x + 3 y = 4. We
have
f (u) − f (1) u−4
g(u) = = ,
u−1 4(u − u + 4)
2

g(x) − g( y) 4(x + y) − x y
h(x, y) = =
x−y 4(x − x + 4)( y 2 − y + 4)
2

3( y − 2)2 + 4
= > 0.
4(x 2 − x + 4)( y 2 − y + 4)

Case 2: y > 2. From y > 2 and x + 3 y = 4, we get x < −2 and


−x
f0 (x) + 3 f0 ( y) − 4 f0 (1) = f (x) + 3 f (2) − 4 f (1) = > 0.
x2 − x +4
The equality holds for a = b = c = d = 1.

P 1.81. Let a, b, c be nonnegative real numbers so that a + b + c = 2. If

ln 2
k0 ≤ k ≤ 3, k0 = ≈ 1.71,
ln 3 − ln 2
then
a k (b + c) + b k (c + a) + c k (a + b) ≤ 2.

Solution. Write the inequality as

f (a) + f (b) + f (c) ≤ 2,

where
f (u) = uk (2 − u), u ∈ [0, ∞).
From
f 00 (u) = kuk−2 [2k − 2 − (k + 1)u],
2k − 2 2k − 2
• ˜ • ˜
it follows that f is convex on 0, and concave on , 2 . According
k+1 k+1
to LCRCF-Theorem, the sum f (a) + f (b) + f (c) is maximum when either a = 0 or
0 < a ≤ b = c.
132 Vasile Cîrtoaje

Case 1: a = 0. We need to show that

bc(b k−1 + c k−1 ) ≤ 2

for b + c = 2. Since 0 < (k − 1)/2 ≤ 1, Bernoulli’s inequality gives

k−1 2 k−1 2
b k−1 + c k−1 = (b2 )(k−1)/2 + (c 2 )(k−1)/2 ≤ 1 + (b − 1) + 1 + (c − 1)
2 2
k−1 2
=3−k+ (b + c 2 ).
2
Thus, it suffices to show that
k−1
(3 − k)bc + bc(b2 + c 2 ) ≤ 2.
2
Since ‹2
b+c

bc ≤ = 1,
2
we only need to show that

k−1
3−k+ bc(b2 + c 2 ) ≤ 2,
2
which is equivalent to
bc(b2 + c 2 ) ≤ 2.
Indeed, we have

8[2 − bc(b2 + c 2 )] = (b + c)4 − 8bc(b2 + c 2 ) = (b − c)4 ≥ 0.

Case 2: 0 < a ≤ b = c. We only need to prove the homogeneous inequality

a+b+c
 ‹k+1
a (b + c) + b (c + a) + c (a + b) ≤ 2
k k k
2
for b = c = 1 and 0 < a ≤ 1; that is,
 a k+1
1+ − a k − a − 1 ≥ 0.
2
 a k+1
Since 1 + is increasing and a k is decreasing with respect to k, it suffices
2
consider the case k = k0 ; that is, to prove that g(a) ≥ 0, where
 a k0 +1
g(a) = 1 + − a k0 − a − 1, 0 < a ≤ 1.
2
We have
k0 + 1  a  k0
g (a) =
0
1+ − k0 a k0 −1 − 1,
2 2
Half Convex Function Method 133

1 00 k0 + 1  a k0 −1 k0 − 1
g (a) = 1+ − 2−k .
k0 4 2 a 0
Since g 00 is increasing on (0, 1], g 00 (0+ ) = −∞ and

1 00 k0 + 1 3 k0 −1 k0 + 1 2(2 − k0 )
 ‹
g (1) = − k0 + 1 = − k0 + 1 = > 0,
k0 4 2 3 3
there exists a1 ∈ (0, 1) so that g 00 (a1 ) = 0, g 00 (a) < 0 for a ∈ (0, a1 ), g 00 (a) > 0 for
a ∈ (a1 , 1]. Therefore, g 0 is strictly decreasing on [0, a1 ] and strictly increasing on
[a1 , 1]. Since
k0 − 1 k0 + 1  
g 0 (0) = > 0, g 0 (1) = (3/2)k0 − 2 = 0,
2 2
there exists a2 ∈ (0, a1 ) so that g 0 (a2 ) = 0, g 0 (a) > 0 for a ∈ [0, a2 ), g 0 (a) < 0
for a ∈ (a2 , 1). Thus, g is strictly increasing on [0, a2 ] and strictly decreasing on
[a2 , 1]. Consequently,
g(a) ≥ min{g(0), g(1)},
and from
g(0) = 0, g(1) = (3/2)k0 +1 − 3 = 0,
we get g(a) ≥ 0.
The equality holds for a = 0 and b = c (or any cyclic permutation). If k = k0 ,
then the equality holds also for a = b = c.

P 1.82. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then


1 1 1
 ‹
(n + 1) 2
+ + ··· + ≥ 4(n + 2)(a12 + a22 + · · · + an2 ) + n(n2 − 3n − 6).
a1 a2 an

(Vasile C., 2006)


Solution. Write the inequality as

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n(n2 − 3n − 6),

where
(n + 1)2
f (u) = − 4(n + 2)u2 , u ∈ (0, ∞).
u
From
2(n + 1)2
f 00 (u) = − 8(n + 2),
u3
it follows that f is strictly convex on (0, c] and strictly concave on [c, ∞), where
v
3 (n + 1)
t 2
c= .
4(n + 2)
134 Vasile Cîrtoaje

According to LCRCF-Theorem and Note 5, it suffices to consider the case

a1 = a2 = · · · = an−1 = x, an = n − (n − 1)x, 0 < x ≤ 1,

when the inequality becomes as follows:

2 n−1 1
 ‹
(n + 1) + ≥ 4(n + 2)[(n − 1)x 2 + an2 ) + n(n2 − 3n − 6),
x an

n(n − 1)(2x − 1)2 [(n + 2)(n − 1)x 2 − (n + 2)(2n − 1)x + (n + 1)2 ] ≥ 0.


The last inequality is true since

(n + 1)2 2n − 1 2 3(n − 2)
 ‹
(n − 1)x − (2n − 1)x +
2
= (n − 1) x − + ≥ 0.
n+2 2n − 2 4(n − 1)(n + 2)
The equality holds for
1 n+1
a1 = a2 = · · · = an−1 = , an =
2 2
(or any cyclic permutation).

P 1.83. If a, b, c, d, e are positive real numbers such that a + b + c + d + e = 5, then


1 1 1 1 1
27( + + + + ) ≥ 4(a3 + b3 + c 3 + d 3 + e3 ) + 115.
a b c d e
(Vasile Cîrtoaje)

Proof. Write the inequality as


a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
27
f (u) = − 4u3 , 0 < u < 5.
u
From
6(9 − 4u4 )
f 00 (u) = ,
u3
it follows that f is convex on (0, 1]. According to LHCF-Theorem, it suffices to
prove that
f (x) + 4 f ( y) ≥ 5 f (1)
for x ≥ 1 ≥ y > 0 and x + 4 y = 5. This occurs if h(x, y) ≥ 0, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
Half Convex Function Method 135

Since
27
g(u) = − − 4(u2 + u + 1),
u
A(x, y)
h(x, y) = , A(x, y) = 27 − 4x y(x + y + 1),
xy
we need show that A(x, y) ≥ 0. Indeed,

1
A(x, y) = 9 − 4 y(4 y − 5)( y − 2) = 9 − 40 y + 52 y 2 − 16 y 3
3
= (1 − 2 y)2 (9 − 4 y) ≥ 0.
The equality holds for a = b = c = d = e = 1, and for a = 3 and b = c = d = e =
1/2 (or any cyclic permutation).
Generalization. If a1 , a2 , ..., an are positive real numbers such that

a1 + a2 + · · · + an = n,

then
1 1 1
(n + 1)2 (2n − 1)( + + ··· + − n) ≥ 27(n − 1)2 (a13 + a23 + · · · + an3 − n),
a1 a2 an

with equality for a1 = a2 = · · · = an = 1, and for

2n − 1 n+1
a1 = , a2 = · · · = an =
3 3(n − 1)

(or any cyclic permutation).

P 1.84. If a, b, c are nonnegative real numbers so that a + b + c = 12, then

(a2 + 10)(b2 + 10)(c 2 + 10) ≥ 13310.

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) ≥ 2 ln 11 + ln 110,

where
f (u) = ln(u2 + 10), u ∈ [0, 12].
From
2(10 − u2 )
f 00 (u) = ,
(u2 + 10)2
136 Vasile Cîrtoaje

p p
it follows that f is convex on [0, 10] and concave on [ 10, 12]. According to
LCRCF-Theorem, the sum f (a) + f (b) + f (c) is minimum when a = b ≤ c. There-
fore, it suffices to prove that g(a) ≥ 0, where

g(a) = 2 f (a) + f (c) − 2 ln 11 − ln 110, c = 12 − 2a, a ∈ [0, 4].

Since c 0 (a) = −2, we have


a  c 
g (a) = 2 f (a) − 2 f (c) = 4 2
0 0 0

a + 10 c 2 + 10
4(a − c)(10 − ac) 24(4 − a)(5 − a)(a − 1)
= 2 = .
(a + 10)(c 2 + 10) (a2 + 10)(c 2 + 10)

Therefore, g 0 (a) < 0 for a ∈ [0, 1) and g 0 (a) > 0 for a ∈ (1, 4), hence g is strictly
decreasing on [0, 1] and strictly increasing on [1, 4]. Thus, we have

g(a) ≥ g(1) = 0.

The equality holds for a = b = 1 and c = 10 (or any cyclic permutation).


Remark. Similarly, we can prove the following generalization:
• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 +a2 +· · ·+an = 2n(n−1).
If k = (n − 1)(2n − 1), then

(a12 + k)(a22 + k) · · · (an2 + k) ≥ k(k + 1)n ,

with equality for a1 = k and a2 = · · · = an = 1 (or any cyclic permutation).

P 1.85. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
(n2 − 2n + 2)n
(a12 + 1)(a22 + 1) · · · (an2 + 1) ≥ .
(n − 1)2n−2
(Vasile C., 2006)

Solution. Write the inequality as

(n2 − 2n + 2)n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ ln k, k= ,
(n − 1)2n−2

where
f (u) = ln(u2 + 1), u ∈ [0, n].
From
2(1 − u2 )
f 00 (u) = ,
(u2 + 1)2
Half Convex Function Method 137

it follows that f is strictly convex on [0, 1] and strictly concave on [1, n]. According
to LCRCF-Theorem, it suffices to consider the case a1 = a2 = · · · = an−1 ≤ an ; that
is, to show that g(x) ≥ 0, where

g(x) = (n − 1) f (x) + f ( y) − ln k, y = n − (n − 1)x, x ∈ [0, 1].

Since y 0 (x) = −(n − 1), we get

g 0 (x) = (n − 1) f 0 (x) − (n − 1) f 0 ( y) = (n − 1)[ f 0 (x) − f 0 ( y)]


2(n − 1)(x − y)(1 − x y)
 ‹
x y
= 2(n − 1) − =
x2 + 1 y2 + 1 (x 2 + 1)( y 2 + 1)
2n(n − 1)(x − 1) [(n − 1)x − 1]
2
= .
(x 2 + 1)( y 2 + 1)
1 1
• ˜ • ˜
Therefore, g (x) ≤ 0 for x ∈ 0,
0
and g (x) ≥ 0 for x ∈
0
, n , hence g
n−1 n− 1 ‹
1 1 1
• ˜ • ˜
is decreasing on 0, and increasing on , 1 . Since g = 0, the
n−1 n−1 n−1
conclusion follows.
1
The equality holds for a1 = a2 = · · · = an−1 = and an = n − 1 (or any cyclic
n−1
permutation).

P 1.86. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

(a2 + 2)(b2 + 2)(c 2 + 2) ≤ 44.

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) ≤ ln 44,

where
f (u) = ln(u2 + 2), u ∈ [0, 3].
From
2(2 − u2 )
f 00 (u) = ,
(u2 + 2)2
p p
it follows that f is strictly convex on [0, 2] and strictly concave on [ 2, 3]. Ac-
cording to LCRCF-Theorem, the sum f (a)+ f (b)+ f (c) is maximum for either a = 0
or 0 < a ≤ b = c.
Case 1: a = 0. We need to show that b + c = 3 involves

(b2 + 2)(c 2 + 2) ≤ 22,


138 Vasile Cîrtoaje

which is equivalent to
bc(bc − 4) ≤ 0.
This is true because
b+c 2 9
 ‹
bc ≤ = < 4.
2 4
Case 2: 0 < a ≤ b = c. We need to show that a + 2b = 3 (0 < a ≤ 1) involves

(a2 + 2)(b2 + 2)2 ≤ 44,

which is equivalent to g(a) ≤ 0, where


3−a
g(a) = ln(a2 + 2) + 2 ln(b2 + 2) − ln 44, b= , a ∈ (0, 1].
2
Since b0 (a) = −1/2, we have
2a 2b 2(a − b)(2 − ab)
g 0 (a) = − 2 = 2
+2 b +2
a2 (a + 2)(b2 + 2)
3(a − 1)(a2 − 3a + 4)
= .
2(a2 + 2)(b2 + 2)
Because
a2 − 3a + 4 = (a − 2)2 + a > 0,
we have g 0 (a) < 0 for a ∈ (0, 1), g is strictly decreasing on [0, 1], hence it suffices
to show that g(0) ≤ 0. This reduces to 16 · 22 ≥ 172 , which is true because

16 · 22 − 172 = 63 > 0.

The equality holds for a = b = 0 and c = 3 (or any cyclic permutation).


Remark. In the same manner, we can prove the following generalization:
9
• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k ≥ , then
8
(a2 + k)(b2 + k)(c 2 + k) ≤ k2 (k + 9),

with equality for a = b = 0 and c = 3 (or any cyclic permutation). If k = 9/8, then
the equality holds also for a = 0 and b = c = 3/2 (or any cyclic permutation).

P 1.87. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


169
(a2 + 1)(b2 + 1)(c 2 + 1) ≤ .
16

(Vasile C., 2006)


Half Convex Function Method 139

Solution. Write the inequality as

f (a) + f (b) + f (c) ≤ ln 169 − ln 16,

where
f (u) = ln(u2 + 1), u ∈ [0, 3].
From
2(1 − u2 )
f (u) = 2
00
,
(u + 1)2
it follows that f is strictly convex on [0, 1] and strictly concave on [1, 3]. According
to LCRCF-Theorem, it suffices to consider the cases a = 0 and 0 < a ≤ b = c.
Case 1: a = 0. We need to show that b + c = 3 involves
169
(b2 + 1)(c 2 + 1) ≤ ,
16
which is equivalent to
(4bc + 1)(4bc − 9) ≤ 0.
This is true because
4bc ≤ (b + c)2 = 9.
Case 2: 0 < a ≤ b = c. We need to show that a + 2b = 3 (0 < a ≤ 1) involves

169
(a2 + 1)(b2 + 1)2 ≤ ,
16
which is equivalent to g(a) ≤ 0, where

3−a
g(a) = ln(a2 + 1) + 2 ln(b2 + 1) − ln 169 + ln 16, b= , a ∈ (0, 1].
2
Since b0 (a) = −1/2, we have

2a 2b 2(a − b)(1 − ab)


g 0 (a) = − 2 = 2
+1 b +1
a2 (a + 1)(b2 + 1)
3(a − 1)2 (a − 2)
= ≤ 0,
2(a2 + 1)(b2 + 1)

hence g is strictly decreasing. Consequently, we have

g(a) < g(0) = 0.

The equality holds for a = 0 and b = c = 3/2 (or any cyclic permutation).
140 Vasile Cîrtoaje

P 1.88. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


121
(2a2 + 1)(2b2 + 1)(2c 2 + 1) ≤ .
4

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) ≤ ln 121 − ln 4,

where
f (u) = ln(2u2 + 1), u ∈ [0, 3].
From
4(1 − 2u2 )
f 00 (u) = ,
(2u2 + 1)2
p p
it follows that f is strictly convex on [0, 1/ 2] and strictly concave on [1/ 2, 3].
By LCRCF-Theorem, it suffices to consider the cases a = 0 and 0 < a ≤ b = c.
Case 1: a = 0. We need to show that b + c = 3 involves
121
(2b2 + 1)(2c 2 + 1) ≤ ,
4
which is equivalent to
(4bc + 5)(4bc − 9) ≤ 0.
This is true because
4bc ≤ (b + c)2 = 9.
Case 2: 0 < a ≤ b = c. We need to show that a + 2b = 3 (0 < a ≤ 1) involves
121
(2a2 + 1)(2b2 + 1)2 ≤ ,
4
which is equivalent to g(a) ≤ 0, where
3−a
g(a) = ln(2a2 + 1) + 2 ln(2b2 + 1) − ln 121 + ln 4, b= , a ∈ (0, 1].
2
Since b0 (a) = −1/2, we have
4a 4b 4(a − b)(1 − 2ab)
g 0 (a) = − 2 =
2a + 1 2b + 1 (2a2 + 1)(2b2 + 1)
2

6(a − 1)(a2 − 3a + 1)
=
(2a2 + 1)(2b2 + 1)
p p
3(1 − a)(3 + 5 − 2a)(2a − 3 + 5)
= ,
2(2a2 + 1)(2b2 + 1)
Half Convex Function Method 141
 p   p   p 
3 − 5 3 − 5 3 − 5
hence g 0 = 0, g 0 (a) < 0 for a ∈ 0, , g 0 (a) > 0 for a ∈ ,1 .
2 2  2 p
 p  
3− 5 3− 5
Therefore, g is strictly decreasing on 0, and strictly increasing on ,1 .
2 2
Since g(0) = 0, it suffices to show that g(1) ≤ 0, which reduces to 27 · 4 ≤ 121.
The equality holds for a = 0 and b = c = 3/2 (or any cyclic permutation).

P 1.89. If a, b, c are nonnegative real numbers so that a + b + c ≥ k0 , where

3
Æ p
k0 = 66 + 10 105 ≈ 4.867,
8
then ‹2
a+b+c
Æ 
3
(a2 + 1)(b2 + 1)(c 2 + 1) ≤ + 1.
3
(Vasile C., 2018)

Solution. Consider first the case a + b + c = k0 , and write the inequality as

a+b+c k0
f (a) + f (b) + f (c) ≥ 3 f (s), s= = ,
3 3
where
f (u) = − ln(u2 + 1), u ∈ [0, k0 ].
For u ∈ [s, k0 ], f (u) is convex because

6(3u2 − 1)
f 00 (u) = > 0.
(3u2 + 1)2

By the RHCF-Theorem, we only need to show that

f (x) + 2 f ( y) ≥ 3 f (s)

for 0 ≤ x ≤ s ≤ y so that x + 2 y = 3s; that is, to show that g(x) ≥ 0 for x ∈ [0, s],
where
k0 − x
g(x) = f (x) + 2 f ( y) − 3 f (s), y= .
2
Since y 0 (x) = −1/3, we have

−2x 2y
g 0 (x) = f 0 (x) + 2 y 0 f 0 ( y) = +
x2 + 1 y2 + 1
2(x − y)(x y − 1) 3(s − x)(x 2 − k0 x + 2)
= = .
(x 2 + 1)( y 2 + 1) 2(x 2 + 1)( y 2 + 1)
142 Vasile Cîrtoaje

Æ
k0 − k02 − 8
Since g is increasing on [0, s1 ] and decreasing on [s1 , s], where s1 = ,
2
it suffices to show that g(0) ≥ 0 and g(s) ≥ 0. These inequalities are true because
g(0) = 0 and g(s) = 0. The equality g(0) = 0 is equivalent to
‹2
2y
Æ 
( y2 + 1)2 = + 1,
3

3
k0
where y = .
2
According to RHCF-Theorem, if the inequality

a+b+c
 ‹
f (a) + f (b) + f (c) ≥ 3 f
3

holds for a + b + c = k0 , then it holds for a + b + c > k0 , too.


The equality holds for a = b = c. In addition, for a + b + c = k0 , the equality
occurs again for a = 0 and b = c = k0 /2 (or any cyclic permutation).

P 1.90. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + 3)(b2 + 3)(c 2 + 3)(d 2 + 3) ≤ 513.

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) + f (d) ≤ ln 513,

where
f (u) = ln(u2 + 3), u ∈ [0, 4].
From
2(3 − u2 )
f 00 (u) = ,
(u2 + 3)2
p p
it follows that f is strictly convex on [0, 3] and strictly concave on [ 3, 4]. By
LCRCF-Theorem, it suffices to consider the cases a = 0 and 0 < a ≤ b = c.
Case 1: a = 0. We need to show that b + c + d = 4 involves

(b2 + 3)(c 2 + 3)(d 2 + 3) ≤ 171.

Substituting b, c, d by 4b/3, 4c/3, 4d/3, respectively, we need to show that b + c +


d = 3 involves
(b2 + k)(c 2 + k)(d 2 + k) ≤ k2 (k + 9),
Half Convex Function Method 143

where k = 27/16. According to Remark from the proof of P 1.86, this inequality
holds for all k ≥ 9/8.
Case 2: 0 < a ≤ b = c = d. We need to show that a + 3b = 4 (0 < a ≤ 1) involves

(a2 + 3)(b2 + 3)3 ≤ 513,

which is equivalent to g(a) ≤ 0, where

4−a
g(a) = ln(a2 + 3) + 3 ln(b2 + 3) − ln 513, b= , a ∈ (0, 1].
3
Since b0 (a) = −1/3, we have

2a 2b 2(a − b)(3 − ab)


g 0 (a) = − 2 = 2
+3 b +3
a2 (a + 3)(b2 + 3)
8(a − 1)(a2 − 4a + 9)
= .
9(a2 + 3)(b2 + 3)

Because
a2 − 4a + 9 = (a − 2)2 + 5 > 0,
we have g 0 (a) > 0 for a ∈ [0, 1), g is strictly decreasing on [0, 1], hence it suffices
to show that g(0) ≤ 0. This reduces to show that the original inequality holds for
a = 0 and b = c = d = 4/3, which follows immediately from the case 1.
The equality holds for a = b = c = 0 and d = 4 (or any cyclic permutation).

P 1.91. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + 2)(b2 + 2)(c 2 + 2)(d 2 + 2) ≤ 144.

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) + f (d) ≤ ln 144,

where
f (u) = ln(u2 + 2), u ∈ [0, 4].
From
2(2 − u2 )
f 00 (u) = ,
(u2 + 2)2
p p
it follows that f is strictly convex on [0, 2] and strictly concave on [ 2, 4]. By
LCRCF-Theorem, it suffices to consider the cases a = 0 and 0 < a ≤ b = c.
144 Vasile Cîrtoaje

Case 1: a = 0. We need to show that b + c + d = 4 involves

(b2 + 2)(c 2 + 2)(d 2 + 2) ≤ 72.

Substituting b, c, d by 4b/3, 4c/3, 4d/3, respectively, we need to show that b + c +


d = 3 involves
(8b2 + 9)(8c 2 + 9)(8d 2 + 9) ≤ 94 .
This is true according to Remark from the proof of P 1.86.
Case 2: 0 < a ≤ b = c = d. We need to show that a + 3b = 4 (0 < a ≤ 1) involves

(a2 + 2)(b2 + 2)3 ≤ 144,

which is equivalent to g(a) ≤ 0, where

4−a
g(a) = ln(a2 + 2) + 3 ln(b2 + 2) − ln 144, b= , a ∈ (0, 1].
3

Since b0 (a) = −1/3, we have

2a 2b 2(a − b)(2 − ab)


g 0 (a) = − 2 = 2
+2 b +2
a2 (a + 2)(b2 + 2)
8(a − 1)(a2 − 4a + 6)
= .
9(a2 + 2)(b2 + 2)

Because
a2 − 4a + 6 = (a − 2)2 + 2 > 0,
we have g 0 (a) > 0 for a ∈ [0, 1), g is strictly decreasing on [0, 1], hence it suffices
to show that g(0) ≤ 0. This reduces to show that the original inequality holds for
a = 0 and b = c = d = 4/3, which follows immediately from the case 1.
The equality holds for a = b = c = 0 and d = 4 (or any cyclic permutation), and
also for a = b = 0 and c = d = 2 (or any permutation).

P 1.92. If a, b, c, d are nonnegative real numbers such that

a + b + c + d = 4,

then
a b c d 4
+ 3 + 3 + 3 ≤ .
3a3 + 2 3b + 2 3c + 2 3d + 2 5

(Vasile Cîrtoaje, 2019)


Half Convex Function Method 145

Solution. Consider the function


−u
f (u) = : I = [0, 4].
3u3 + 2
Since
18u2 (4 − 3u3 )
f 00 (u) =
(3u3 + 2)3
is positive for u ∈ [0, 1], f is left convex on I≤1 . According to LHCF-Theorem and
Note 1, it is enough to show that h(x, y) ≥ 0 for x, y ∈ [0, 4] such that x + 3 y = 4.
We have
f (u) − f (1) 3u2 + 3u − 2
g(u) = = ,
u−1 3u3 + 2
g(x) − g( y) 2F (x, y)
h(x, y) = = ,
x−y (3x + 2)(3 y 3 + 2)
3

where

F (x, y) = 2(x 2 + x y + y 2 ) + 2(x + y) + 2 − 3x 2 y 2 − 3x y(x + y).

From p
4 = x + 3 y ≥ 2 3x y,
we get 3x y ≤ 4. Thus, we have

F (x, y) ≥ 2(x 2 + x y + y 2 ) + 2(x + y) + 2 − 4x y − 4(x + y) = 26( y − 1)2 ≥ 0.

The proof is completed. The equality occurs for a = b = c = d = 1.

P 1.93. If a1 , a2 , ..., an are nonnegative real numbers such that a1 + a2 + · · · + an = 1,


then
1
a13 + a23 + · · · + an3 ≤ + a14 + a24 + · · · + an4 .
8
(Vasile C., 2018)

Solution. We use the induction method. For n = 2, denoting

a1 a2 = p, p ≤ 1/4,

we have
a13 + a23 = (a1 + a2 )3 − 3a1 a2 (a1 + a2 ) = 1 − 3p,
a14 + a24 = (a12 + a22 )2 − 2a12 a22 = 2p2 − 4p + 1,
and the inequality is equivalent to

(4p − 1)2 ≥ 0.
146 Vasile Cîrtoaje

Consider further that n ≥ 3, a1 ≤ a2 ≤ · · · ≤ an , and write the inequality as

1
f (a1 ) + f (a2 ) + · · · + f (an ) ≤ ,
8
where
f (u) = u3 − u4 , u ∈ [0, 1].
From
f 00 (u) = 6u(1 − 2u),
it follows that f is strictly convex on [0, 1/2] and strictly concave on [1/2, 1]. By
LCRCF-Theorem, it suffices to consider the cases a1 = 0 and 0 < a1 ≤ a2 = · · · = an .
Case 1: a1 = 0. The inequality follows by the induction hypothesis.
Case 2: 0 < a1 ≤ a2 = · · · = an . We only need to prove the homogeneous inequality

8(a14 + a24 + · · · + an4 ) + (a1 + a2 + · · · + an )4 ≥ 8(a1 + a2 + · · · + an )(a13 + a23 + · · · + an3 )

for a1 = x and a2 = · · · = an−1 = 1, that is

8(x 4 + n − 1) + (x + n − 1)4 ≥ 8(x + n − 1)(x 3 + n − 1),

x 4 − 4(n − 1)x 3 + 6(n − 1)2 x 2 + 4(n − 1)(n2 − 2n − 1)x + (n − 3)(n − 1)(n2 − 5) ≥ 0,


x 2 (x − 2n + 2)2 + 2(n − 1)2 x 2 + 4(n − 1)(n2 − 2n − 1)x + (n − 3)(n − 1)(n2 − 5) ≥ 0.
The equality holds for a1 = · · · = an−2 = 0 and an−1 = an = 1/2 (or any permuta-
tion).
Remark. The inequality can be also proved by using EV-method (see Corollary 5
from section 5, case k = 3 and m = 4): If

a1 + a2 + · · · + an = 1, a13 + a23 + · · · + an3 = const ant,

then the sum


Sn = a14 + a24 + · · · + an4
is minimum for either a1 = 0 or 0 < a1 ≤ a2 = · · · = an .
Chapter 2

Half Convex Function Method for


Ordered Variables

2.1 Theoretical Basis


The following statement is known as the Right Half Convex Function Theorem for
Ordered Variables (RHCF-OV Theorem).
RHCF-OV Theorem (Vasile Cîrtoaje, 2008). Let f be a real function defined on an
interval I and convex on I≥s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≤ a2 ≤ · · · ≤ am ≤ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so that

x ≤ s ≤ y, x + (n − m) y = (1 + n − m)s.

Proof. For
a1 = x, a2 = · · · = am = s, am+1 = · · · = an = y,
the inequality
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s)
becomes
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s);

147
148 Vasile Cîrtoaje

thus, the necessity is proved. To prove the sufficiency, we assume that

a1 ≤ a2 ≤ · · · ≤ an .

From a1 ≤ a2 ≤ · · · ≤ am ≤ s, it follows that there is an integer

k ∈ {m, m + 1, . . . , n − 1}

so that
a1 ≤ · · · ≤ ak ≤ s ≤ ak+1 ≤ · · · ≤ an .
Since f is convex on I≥s , we may apply Jensen’s inequality to get

f (ak+1 ) + · · · + f (an ) ≥ (n − k) f (z),

where
ak+1 + · · · + an
z= , z ∈ I.
n−k
Therefore, to prove the desired inequality

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ f (s),

it suffices to show that

f (a1 ) + · · · + f (ak ) + (n − k) f (z) ≥ n f (s). (*)

Let b1 , . . . , bk be defined by

ai + (n − m)bi = (1 + n − m)s, i = 1, . . . , k.

We claim that
z ≥ b1 ≥ · · · ≥ bk ≥ s, b1 , . . . , bk ∈ I.
Indeed, we have
b1 ≥ · · · ≥ b k ,
s − ak
bk − s = ≥ 0,
n−m
and
z ≥ b1
because

(n − m)b1 = (1 + n − m)s − a1
= −(m − 1)s + (a2 + · · · + ak ) + (ak+1 + · · · + an )
≤ −(m − 1)s + (k − 1)s + (ak+1 + · · · + an ) =
= (k − m)s + (n − k)z ≤ (n − m)z.
HCF Method for Ordered Variables 149

Since b1 , . . . , bk ∈ I≥s , by hypothesis we have

f (a1 ) + (n − m) f (b1 ) ≥ (1 + n − m) f (s),

···
f (ak ) + (n − m) f (bk ) ≥ (1 + n − m) f (s),
hence

f (a1 ) + · · · + f (ak ) + (n − m)[ f (b1 ) + · · · + f (bk )] ≥ k(1 + n − m) f (s),

f (a1 ) + · · · + f (ak ) ≥ k(1 + n − m) f (s) − (n − m)[ f (b1 ) + · · · + f (bk )].


According to this result, the inequality (*) is true if

k(1 + n − m) f (s) − (n − m)[ f (b1 ) + · · · + f (bk )] + (n − k) f (z) ≥ n f (s),

which is equivalent to

n−k
p f (z) + (k − p) f (s) ≥ f (b1 ) + · · · + f (bk ), p= ≤ 1.
n−m
By Jensen’s inequality, we have

p f (z) + (1 − p) f (s) ≥ f (w), w = pz + (1 − p)s ≥ s.

Thus, we only need to show that

f (w) + (k − 1) f (s) ≥ f (b1 ) + · · · + f (bk ).

Since the decreasingly ordered vector A~k = (w, s, . . . , s) majorizes the decreasingly
ordered vector B~k = (b1 , b2 , . . . , bk ), this inequality follows from Karamata’s in-
equality for convex functions.
Similarly, we can prove the Left Half Convex Function Theorem for Ordered Vari-
ables (LHCF-OV Theorem).

LHCF-OV Theorem. Let f be a real function defined on an interval I and convex on


I≤s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≥ a2 ≥ · · · ≥ am ≥ s, m ∈ {1, 2, . . . , n − 1},
150 Vasile Cîrtoaje

if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so tht

x ≥ s ≥ y, x + (n − m) y = (1 + n − m)s.

From the RHCF-OV Theorem and the LHCF-OV Theorem, we find the HCF-OV
Theorem (Half Convex Function Theorem for Ordered Variables).
HCF-OV Theorem. Let f be a real function defined on an interval I and convex on
I≥s (or I≤s ), where s ∈ int(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 so that

a1 + a2 + · · · + an = ns

and at least m of a1 , a2 , . . . , an are smaller (greater) than s, where m ∈ {1, 2, . . . , n−1},


if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I satisfying x + (n − m) y = (1 + n − m)s.
The RHCF-OV Theorem, the LHCF-OV Theorem and the HCF-OV Theorem are
respectively generalizations of the RHCF-Theorem, the LHCF Theorem and the HCF-
Theorem, because the last theorems can be obtained from the first theorems for
m = 1.
Note 1. Let us denote
f (u) − f (s) g(x) − g( y)
g(u) = , h(x, y) = .
u−s x−y
In many applications, it is useful to replace the hypothesis

f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)

in the RHCF-OV Theorem and the LHCF-OV Theorem by the equivalent condition
h(x, y) ≥ 0 for all x, y ∈ I so that x + (n − m) y = (1 + n − m)s.

This equivalence is true since

f (x) + (n − m) f ( y) − (1 + n − m) f (s) = [ f (x) − f (s)] + (n − m)[ f ( y) − f (s)]


= (x − s)g(x) + (n − m)( y − s)g( y)
n−m
= (x − y)[g(x) − g( y)]
1+n−m
n−m
= (x − y)2 h(x, y).
1+n−m
HCF Method for Ordered Variables 151

Note 2. Assume that f is differentiable on I, and let

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
The desired inequality of Jensen’s type in the RHCF-OV Theorem and the LHCF-OV
Theorem holds true by replacing the hypothesis

f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)

with the more restrictive condition


H(x, y) ≥ 0 for all x, y ∈ I so that x + (n − m) y = (1 + n − m)s.

To prove this, we will show that the new condition implies

f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)

for all x, y ∈ I so that x + (n − m) y = (1 + n − m)s. Write this inequality as

f1 (x) ≥ (1 + n − m) f (s),

where
(1 + n − m)s − x
 ‹
f1 (x) = f (x) + (n − m) f .
n−m
From
(1 + n − m)s − x
 ‹
f10 (x) = f (x) − f
0 0
n−m
= f (x) − f ( y)
0 0

1+n−m
= (x − s)H(x, y),
n−m
it follows that f1 is decreasing on I≤s and increasing on I≥s ; therefore,

f1 (x) ≥ f1 (s) = (1 + n − m) f (s).

Note 3. The RHCF-OV Theorem and the LHCF-OV Theorem are also valid in the
case when f is defined on I \ {u0 }, where u0 ∈ I<s for the RHCF-OV Theorem, and
u0 ∈ I>s for the LHCF-OV Theorem.

Note 4. The desired inequalities in the RHCF-OV Theorem and the LHCF-OV The-
orem become equalities for

a1 = a2 = · · · = an = s.

In addition, if there exist x, y ∈ I so that

x + (n − m) y = (1 + n − m)s, f (x) + (n − m) f ( y) = (1 + n − m) f (s), x 6= y,


152 Vasile Cîrtoaje

then the equality holds also for

a1 = x, a2 = · · · = am = s, am+1 = · · · = an = y

Notice that these equality conditions are equivalent to

x + (n − m) y = (1 + n − m)s, h(x, y) = 0

(x < y for the RHCF-OV Theorem, and x > y for the LHCF-OV Theorem).

Note 5. The WRHCF-OV Theorem and the WLHCF-OV Theorem are extensions of
the weighted Jensen’s inequality to right and left half convex functions with ordered
variables (Vasile Cirtoaje, 2008).
WRHCF-OV Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1,

and let f be a real function defined on an interval I and convex on I≥s , where s ∈ int(I).
The inequality

p1 f (x 1 ) + p2 f (x 2 ) + · · · + pn f (x n ) ≥ f (p1 x 1 + p2 x 2 + · · · + pn x n )

holds for all x 1 , x 2 , . . . , x n ∈ I so that p1 x 1 + p2 x 2 + · · · + pn x n = s and

x1 ≤ x2 ≤ · · · ≤ x n, x m ≤ s, m ∈ {1, 2, . . . , n − 1},

if and only if
f (x) + k f ( y) ≥ (1 + k) f (s)
for all x, y ∈ I satisfying

x ≤ s ≤ y, x + k y = (1 + k)s,

where
pm+1 + pm+2 + · · · + pn
k= .
p1

WLHCF-OV Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1,

and let f be a real function defined on an interval I and convex on I≤s , where s ∈ int(I).
The inequality

p1 f (x 1 ) + p2 f (x 2 ) + · · · + pn f (x n ) ≥ f (p1 x 1 + p2 x 2 + · · · + pn x n )

holds for all x 1 , x 2 , . . . , x n ∈ I so that p1 x 1 + p2 x 2 + · · · + pn x n = s and

x1 ≥ x2 ≥ · · · ≥ x n, x m ≥ s, m ∈ {1, 2, . . . , n − 1},
HCF Method for Ordered Variables 153

if and only if
f (x) + k f ( y) ≥ (1 + k) f (s)
for all x, y ∈ I satisfying

x ≥ s ≥ y, x + k y = (1 + k)s,

where
pm+1 + pm+2 + · · · + pn
k= .
p1
154 Vasile Cîrtoaje
HCF Method for Ordered Variables 155

2.2 Applications

2.1. If a, b, c, d are real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, a + b + c + d = 4,

then

(3a2 − 2)(a − 1)2 + (3b2 − 2)(b − 1)2 + (3c 2 − 2)(c − 1)2 + (3d 2 − 2)(d − 1)2 ≥ 0.

2.2. If a, b, c, d are nonnegative real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, a + b + c + d = 4,

then
1 1 1 1 4
+ 3 + 3 + 3 ≤ .
2a3 + 5 2b + 5 2c + 5 2d + 5 7

2.3. If
−2n − 1
≤ a1 ≤ · · · ≤ an ≤ 1 ≤ an+1 ≤ · · · ≤ a2n , a1 + a2 + · · · + a2n = 2n,
n−1
then
a13 + a23 + · · · + a2n
3
≥ 2n.

2.4. Let a1 , a2 , . . . , an (n ≥ 3) be real numbers so that a1 + a2 + · · · + an = n. Prove


that
(a) if −3 ≤ a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then

a13 + a23 + · · · + an3 ≥ a12 + a22 + · · · + an2 ;


n−1
(b) if − ≤ a1 ≤ a2 ≤ 1 ≤ · · · ≤ an , then
n−3
a13 + a23 + · · · + an3 + n ≥ 2(a12 + a22 + · · · + an2 ).

2.5. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n


and let m ∈ {1, 2, . . . , n − 1}. Prove that
(a) if a1 ≤ a2 ≤ · · · ≤ am ≤ 1, then

(n − m)(a13 + a23 + · · · + an3 − n) ≥ (2n − 2m + 1)(a12 + a22 + · · · + an2 − n);

(b) if a1 ≥ a2 ≥ · · · ≥ am ≥ 1, then

a13 + a23 + · · · + an3 − n ≤ (n − m + 2)(a12 + a22 + · · · + an2 − n).


156 Vasile Cîrtoaje

2.6. Let a1 , a2 , . . . , an (n ≥ 3) be real numbers so that a1 + a2 + · · · + an = n. Prove


that
(a) if a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , then

a14 + a24 + · · · + an4 − n ≥ 6(a12 + a22 + · · · + an2 − n);

(b) if a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then


14 2
a14 + a24 + · · · + an4 − n ≥ (a1 + a22 + · · · + an2 − n);
3
(c) if a1 ≤ a2 ≤ 1 ≤ a3 ≤ · · · ≤ an , then

2(n2 − 3n + 3) 2
a14 + a24 + · · · + an4 − n ≥ (a1 + a22 + · · · + an2 − n).
n2 − 5n + 7

2.7. Let a, b, c, d, e be nonnegative real numbers so that a + b + c + d + e = 5. Prove


that
(a) if a ≥ b ≥ 1 ≥ c ≥ d ≥ e, then

21(a2 + b2 + c 2 + d 2 + e2 ) ≥ a4 + b4 + c 4 + d 4 + e4 + 100;

(b) if a ≥ b ≥ c ≥ 1 ≥ d ≥ e, then

13(a2 + b2 + c 2 + d 2 + e2 ) ≥ a4 + b4 + c 4 + d 4 + e4 + 60.

2.8. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 + a2 + · · · + an = n.


Prove that
(a) if a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , then

7(a13 + a23 + · · · + an3 ) ≥ 3(a14 + a24 + · · · + an4 ) + 4n;

(b) if a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , then

13(a13 + a23 + · · · + an3 ) ≥ 4(a14 + a24 + · · · + an4 ) + 9n.

2.9. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n and

a1 ≥ · · · ≥ am ≥ 1 ≥ am+1 ≥ · · · ≥ an , m ∈ {1, 2, . . . , n − 1},

then
1 1 1
 ‹
(n − m + 1) 2
+ + ··· + − n ≥ 4(n − m)(a12 + a22 + · · · + an2 − n).
a1 a2 an
HCF Method for Ordered Variables 157

1 1 1
2.10. If a1 , a2 , . . . , an are positive real numbers so that + + ··· + = n and
a1 a2 an

a1 ≤ · · · ≤ am ≤ 1 ≤ am+1 ≤ · · · ≤ an , m ∈ {1, 2, . . . , n − 1},

then
 p 
n−m
a12 + a22 + ··· + an2 −n≥2 1+ (a1 + a2 + · · · + an − n).
n−m+1

2.11. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 +a2 +· · ·+an = n.


Prove that
(a) if a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , then

1 1 1 n
+ 2 + ··· + 2 ≥ ;
a12 + 2 a2 + 2 an + 2 3

(b) if a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then

1 1 1 n
+ 2 + ··· + 2 ≥ .
2a12 + 3 2a2 + 3 2an + 3 5

2.12. If a1 , a2 , . . . , a2n are nonnegative real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n,

then
1 1 1 2n
+ 2 + ··· + 2 ≤ .
na12 + n + n + 1 na2 + n + n + 1
2 2 na2n + n + n + 1 (n + 1)2
2

2.13. If a, b, c, d, e, f are nonnegative real numbers so that

a ≥ b ≥ c ≥ 1 ≥ d ≥ e ≥ f, a + b + c + d + e + f = 6,

then
3a + 4 3b + 4 3c + 4 3d + 4 3e + 4 3f + 4
+ 2 + 2 + 2 + 2 + 2 ≤ 6.
3a + 4 3b + 4 3c + 4 3d + 4 3e + 4 3 f + 4
2
158 Vasile Cîrtoaje

2.14. If a, b, c, d, e, f are nonnegative real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d ≥ e ≥ f, a + b + c + d + e + f = 6,

then
a2 − 1 b2 − 1 c2 − 1 d2 − 1 e2 − 1 f2−1
+ + + + + ≥ 0.
(2a + 7)2 (2b + 7)2 (2c + 7)2 (2d + 7)2 (2e + 7)2 (2 f + 7)2

2.15. If a, b, c, d, e, f are nonnegative real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d ≤ e ≤ f, a + b + c + d + e + f = 6,

then
a2 − 1 b2 − 1 c2 − 1 d2 − 1 e2 − 1 f2−1
+ + + + + ≤ 0.
(2a + 5)2 (2b + 5)2 (2c + 5)2 (2d + 5)2 (2e + 5)2 (2 f + 5)2

2.16. If a, b, c are nonnegative real numbers so that

a ≤ b ≤ 1 ≤ c, a + b + c = 3,

then v v v
t 2a t 2b t 2c
+ + ≥ 3.
b+c c+a a+b

2.17. If a1 , a2 , . . . , a8 are nonnegative real numbers so that

a1 ≥ a2 ≥ a3 ≥ a4 ≥ 1 ≥ a5 ≥ a6 ≥ a7 ≥ a8 , a1 + a2 + · · · + a8 = 8,

then
(a12 + 1)(a22 + 1) · · · (a82 + 1) ≥ (a1 + 1)(a2 + 1) · · · (a8 + 1).

2.18. If a, b, c, d are real numbers so that


−1
≤ a ≤ b ≤ 1 ≤ c ≤ d, a + b + c + d = 4,
2
then
1 1 1 1 1 1 1 1
 ‹  ‹
7 2 + 2 + 2 + 2 +3 + + + ≥ 40.
a b c d a b c d
HCF Method for Ordered Variables 159

2.19. Let a, b, c, d be real numbers. Prove that


(a) if −1 ≤ a ≤ b ≤ c ≤ 1 ≤ d, then
1 1 1 1 1 1 1 1
 ‹
3 2 + 2 + 2 + 2 ≥8+ + + + ;
a b c d a b c d
(b) if −1 ≤ a ≤ b ≤ 1 ≤ c ≤ d, then
1 1 1 1 1 1 1 1
 ‹
2 2 + 2 + 2 + 2 ≥4+ + + + .
a b c d a b c d

2.20. If a, b, c, d are positive real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, abcd = 1,

then
1 1 1 1
 ‹
a + b + c + d − 4 ≥ 18 a + b + c + d − − − −
2 2 2 2
.
a b c d

2.21. If a, b, c, d are positive real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, abcd = 1,

then
p p p p
a2 − a + 1 + b2 − b + 1 + c2 − c + 1 + d 2 − d + 1 ≥ a + b + c + d.

2.22. If a, b, c, d are positive real numbers so that

a ≤ b ≤ c ≤ 1 ≤ d, abcd = 1,

then
1 1 1 1 2
+ + + ≥ .
a3 + 3a + 2 b3 + 3b + 2 c 3 + 3c + 2 d 3 + 3d + 2 3

2.23. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,

then
1 1 1
+ + ··· + ≥ a1 + a2 + · · · + an .
a1 a2 an
160 Vasile Cîrtoaje

2.24. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If k ≥ 1, then
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2 1 + kan 1+k

2.25. If a1 , a2 , . . . , a9 are positive real numbers so that

a1 ≤ · · · ≤ a8 ≤ 1 ≤ a9 , a1 a2 · · · a9 = 1,

then
1 1 1
+ + · · · + ≥ 1.
(a1 + 2)2 (a2 + 2)2 (a9 + 2)2

2.26. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that
2pq
p+q ≥1+ ,
p + 4q
then
1 1 1 n
+ + ··· + ≥ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q

2.27. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If m ≥ 1 and 0 < k ≤ m, then


1 1 1 n
+ + · · · + ≥ .
(a1 + k)m (a2 + k)m (an + k)m (1 + k)m

2.28. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1,

then
1 1 1 n
+p + ··· + p ≥ .
1 + 3a1 1 + 3a2 1 + 3an
p
2
HCF Method for Ordered Variables 161

2.29. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.
1
If 0 < m < 1 and 0 < k ≤ , then
21/m −1
1 1 1 n
+ + · · · + ≥ .
(a1 + k)m (a2 + k)m (an + k)m (1 + k)m

2.30. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that

a1 ≥ a2 ≥ a3 ≥ 1 ≥ a4 ≥ · · · ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + ··· + ≥ .
3a1 + 1 3a2 + 1 3an + 1 4

2.31. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that

a1 ≥ a2 ≥ a3 ≥ 1 ≥ a4 ≥ · · · ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + ··· + ≥ .
(a1 + 1)2 (a2 + 1)2 (an + 1)2 4

2.32. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + · · · + ≤ .
(a1 + 3)2 (a2 + 3)2 (an + 3)2 16

2.33. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that p + q ≤ 1, then
1 1 1 n
+ + ··· + ≤ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q
162 Vasile Cîrtoaje

2.34. Let a1 , a2 , . . . , an be positive real numbers so that


a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.
1
If m > 1 and k ≥ , then
21/m −1
1 1 1 n
+ + ··· + ≤ .
(a1 + k) m (a2 + k) m (an + k) m (1 + k)m

2.35. If a1 , a2 , . . . , an are positive real numbers so that


a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,
then
1 1 1 n
+p + ··· + p ≤p .
1 + 2a1 1 + 2a2 1 + 2an
p
3

2.36. Let a1 , a2 , . . . , an be positive real numbers so that


a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.
If 0 < m < 1 and k ≥ m, then
1 1 1 n
+ + · · · + ≤ .
(a1 + k)m (a2 + k)m (an + k)m (1 + k)m

2.37. If a1 , a2 , . . . , an (n ≥ 3) are positive real numbers so that


a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , a1 a2 · · · an = 1,
then
1 1 1 n
+ + ··· + ≤ .
(a1 + 5)2 (a2 + 5)2 (an + 5)2 36

2.38. If a1 , a2 , . . . , an are nonnegative real numbers so that


a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a12 + a22 + · · · + an2 = n,
then
1 1 1 n
+ + ··· + ≤ .
3 − a1 3 − a2 3 − an 2

2.39. Let a1 , a2 , . . . , an be nonnegative real numbers so that


a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 + a2 + · · · + an = n.
Prove that
n − a1 3  n − a2 3  n − a 3
• ˜
n
a13 + a23 + ··· + an3 − n ≥ (n − 1) 2
+ + ··· + −n .
n−1 n−1 n−1
HCF Method for Ordered Variables 163

2.3 Solutions

P 2.1. If a, b, c, d are real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, a + b + c + d = 4,

then

(3a2 − 2)(a − 1)2 + (3b2 − 2)(b − 1)2 + (3c 2 − 2)(c − 1)2 + (3d 2 − 2)(d − 1)2 ≥ 0.

(Vasile C., 2007)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
f (u) = (3u2 − 2)(u − 1)2 , u ∈ I = R.
From
f 00 (u) = 2(18u2 − 18u + 1),
it follows that f 00 (u) > 0 for u ≥ 1, hence f is convex on I≥s . Therefore, we may
apply the RHCF-OV Theorem for n = 4 and m = 2. Thus, it suffices to show that
f (x) + 2 f ( y) ≥ 3 f (1) for all real x, y so that x + 2 y = 3. Using Note 1, we only
need to show that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

We have

g(u) = 3(u3 + u2 + u + 1) − 6(u2 + u + 1) + u + 1 = 3u3 − 3u2 − 2u − 2,

h(x, y) = 3(x 2 + x y + y 2 ) − 3(x + y) − 2 = (3 y − 4)2 ≥ 0.


From x + 2 y = 3 and h(x, y) = 0, we get x = 1/3, y = 4/3. Therefore, in
accordance with Note 4, the equality holds for a = b = c = d = 1, and also for

1 4
a= , b = 1, c=d= .
3 3

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , a2n be real numbers so that

a1 ≤ · · · ≤ an ≤ 1 ≤ an+1 ≤ · · · ≤ a2n , a1 + a2 + · · · + a2n = 2n.


164 Vasile Cîrtoaje

n
If k = , then
n2 − n + 1
(a12 − k)(a1 − 1)2 + (a22 − k)(a2 − 1)2 + · · · + (a2n
2
− k)(a2n − 1)2 ≥ 0,

with equality for a1 = a2 = · · · = a2n = 1, and also for

1 n2
a1 = 2 , a2 = · · · = an = 1, an+1 = · · · = an = 2 .
n −n+1 n −n+1

P 2.2. If a, b, c, d are nonnegative real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, a + b + c + d = 4,

then
1 1 1 1 4
+ 3 + 3 + 3 ≤ .
2a3 + 5 2b + 5 2c + 5 2d + 5 7
(Vasile C., 2009)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
−1
f (u) = , u ≥ 0.
2u3 + 5
From
12u(5 − 4u3 )
f 00 (u) = ,
(2u3 + 5)3
it follows that f 00 (u) ≥ 0 for u ∈ [0, 1], hence f is convex on [0, s]. Therefore, we
may apply the LHCF-OV Theorem for n = 4 and m = 2. Using Note 1, we only need
to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + 2 y = 3. We have

f (u) − f (1) 2(u2 + u + 1)


g(u) = = ,
u−1 7(2u3 + 5)

g(x) − g( y) 2E
h(x, y) = = ,
x−y 7(2x 3 + 5)(2 y 3 + 5)
where

E = −2x 2 y 2 − 2x y(x + y) − 2(x 2 + x y + y 2 ) + 5(x + y) + 5.

Since
E = (1 − 2 y)2 (2 + 3 y − 2 y 2 ) = (1 − 2 y)2 (2 + x y) ≥ 0,
HCF Method for Ordered Variables 165

the proof is completed. From x + 2 y = 3 and h(x, y) = 0, we get x = 2, y = 1/2.


Therefore, in accordance with Note 4, the equality holds for a = b = c = d = 1,
and also for
1
a = 2, b = 1, c = d = .
2

Remark. Similarly, we can prove the following generalization.


• If a1 , a2 , . . . , a2n are nonnegative real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n.

then
1 1 1 2n2
+ + ··· + ≥ 2 ,
a13 + n + 1n a23 + n + 1n 3
a2n + n + 1n n +n+1
with equality for a1 = a2 = · · · = a2n = 1, and also for

1
a1 = n, a2 = · · · = an = 1, an+1 = · · · = a2n = .
n

P 2.3. If

−2n − 1
≤ a1 ≤ · · · ≤ an ≤ 1 ≤ an+1 ≤ · · · ≤ a2n , a1 + a2 + · · · + a2n = 2n,
n−1
then
a13 + a23 + · · · + a2n
3
≥ 2n.

(Vasile C., 2007)

Solution. Write the inequality as


a1 + a2 + · · · + a2n
f (a1 ) + f (a2 ) + · · · + f (a2n ) ≥ 2n f (s), s= = 1,
2n
where
−2n − 1
f (u) = u3 , u≥ .
n−1
From f 00 (u) = 6u, it follows that f (u) is convex for u ≥ s. Therefore, we may apply
the RHCF-OV Theorem for 2n numbers and m = n. By Note 1, it suffices to show
−2n − 1
that h(x, y) ≥ 0 for all x, y ≥ so that x + n y = 1 + n. We have
n−1
f (u) − f (1)
g(u) = = u2 + u + 1,
u−1
166 Vasile Cîrtoaje

g(x) − g( y) (n − 1)x + 2n + 1
h(x, y) = = x + y +1= ≥ 0.
x−y n−1
From x + n y = 1 + n and h(x, y) = 0, we get

−2n − 1 n+2
x= , y= .
n−1 n−1
In accordance with Note 4, the equality holds for a1 = a2 = · · · = a2n = 1, and also
for
−2n − 1 n+2
a1 = , a2 = · · · = an = 1, an+1 = · · · = a2n = .
n−1 n−1

P 2.4. Let a1 , a2 , . . . , an (n ≥ 3) be real numbers so that a1 + a2 + · · · + an = n. Prove


that
(a) if −3 ≤ a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then

a13 + a23 + · · · + an3 ≥ a12 + a22 + · · · + an2 ;

n−1
(b) if − ≤ a1 ≤ a2 ≤ 1 ≤ · · · ≤ an , then
n−3

a13 + a23 + · · · + an3 + n ≥ 2(a12 + a22 + · · · + an2 ).

(Vasile C., 2007)

Solution. (a) Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = u3 − u2 , u ≥ −3.
For u ≥ 1, we have
f 00 (u) = 6u − 2 > 0,
hence f (u) is convex for u ≥ s. Thus, we may apply the RHCF-OV Theorem for
m = n − 2. According to this theorem, it suffices to show that

f (x) + 2 f ( y) ≥ 3 f (1)

for −3 ≤ x ≤ y satisfying x + 2 y = 3. Using Note 1, we only need to show that


h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
HCF Method for Ordered Variables 167

We have
g(u) = u2 ,
x +3
h(x, y) = x + y = ≥ 0.
2
From x + 2 y = 3 and h(x, y) = 0, we get x = −3 and y = 3. Therefore, in
accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = −3, a2 = · · · = an−2 = 1, an−1 = an = 3.

(b) Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
n−1
f (u) = u3 − 2u2 , u≥− .
n−3
For u ≥ 1, we have
f 00 (u) = 6u − 4 > 0,
hence f (u) is convex for u ≥ s. Thus, we may apply the RHCF-OV Theorem for
m = 2. According to this theorem, it suffices to show that

f (x) + (n − 2) f ( y) ≥ (n − 1) f (1)

n−1
for − ≤ x ≤ y satisfying x + (n − 2) y = n − 1. Using Note 1, we only need to
n−3
show that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

We have
g(u) = u2 − u − 1,
(n − 3)x + n − 1
h(x, y) = x + y − 1 = ≥ 0.
n−1
n−1 n−1
From x + (n − 2) y = n − 1 and h(x, y) = 0, we get x = − and y = .
n−3 n−3
Therefore, in accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1.
If n ≥ 4, then the equality holds also for

n−1 n−1
a1 = − , a2 = 1, a3 = · · · = a n = .
n−3 n−3
168 Vasile Cîrtoaje

P 2.5. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n


and let m ∈ {1, 2, . . . , n − 1}. Prove that
(a) if a1 ≤ a2 ≤ · · · ≤ am ≤ 1, then

(n − m)(a13 + a23 + · · · + an3 − n) ≥ (2n − 2m + 1)(a12 + a22 + · · · + an2 − n);

(b) if a1 ≥ a2 ≥ · · · ≥ am ≥ 1, then

a13 + a23 + · · · + an3 − n ≤ (n − m + 2)(a12 + a22 + · · · + an2 − n).

(Vasile C., 2007)

Solution. (a) Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = (n − m)u3 − (2n − 2m + 1)u2 , u ∈ I = [0, n].
For u ≥ 1, we have

f 00 (u) = 6(n − m)u − 2(2n − 2m + 1)


≥ 6(n − m) − 2(2n − 2m + 1) = 2(n − m − 1) ≥ 0,

hence f is convex on I≥s . Thus, by the RHCF-OV Theorem and Note 1, we only need
to show that h(x, y) ≥ 0 for all nonnegative numbers x, y so that x + (n − m) y =
n − m + 1. We have
f (u) − f (1)
g(u) = = (n − m)(u2 + u + 1) − (2n − 2m + 1)(u + 1)
u−1
= (n − m)u2 − (n − m + 1)u − n + m − 1,

g(x) − g( y)
h(x, y) = = (n − m)(x + y) − n + m − 1 = (n − m − 1)x ≥ 0.
x−y
From x +(n−m) y = 1+n−m and h(x, y) = 0, we get x = 0, y = (n−m+1)/(n−m).
Therefore, in accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1,
and also for
1
a1 = 0, a2 = · · · = am = 1, am+1 = · · · = an = 1 + .
n−m
(b) Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = (n − m + 2)u2 − u3 , u ∈ I = [0, n].
HCF Method for Ordered Variables 169

For u ≤ 1, we have

f 00 (u) = 2(n − m + 2 − 3u) ≥ 2(n − m + 2 − 3) = 2(n − m − 1) ≥ 0,

hence f is convex on I≤s . By the LHCF-OV Theorem and Note 1, it suffices to show
that h(x, y) ≥ 0 for all x, y ≥ 0 so that x + (n − m) y = 1 + n − m. We have

f (u) − f (1)
g(u) = = (n − m + 2)(u + 1) − (u2 + u + 1)
u−1
= −u2 + (n − m + 1)u + n − m + 1,

g(x) − g( y)
h(x, y) = = −(x + y) + n − m + 1 = (n − m − 1) y ≥ 0.
x−y
From x + (n − m) y = 1 + n − m and h(x, y) = 0, we get x = n − m + 1, y = 0.
Therefore, the equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = n − m + 1, a2 = · · · = am = 1, am+1 = · · · = an = 0.

Remark 1. For m = 1, we get the following results:


• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then

(n − 1)(a13 + a23 + · · · + an3 − n) ≥ (2n − 1)(a12 + a22 + · · · + an2 − n),

with equality for a1 = a2 = · · · = an = 1, and also for


n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n, then

a13 + a23 + · · · + an3 − n ≤ (n + 1)(a12 + a22 + · · · + an2 − n),

with equality for a1 = a2 = · · · = an = 1, and also for

a1 = n, a2 = a3 = · · · = an = 0

(or any cyclic permutation).

Remark 2. For m = n − 1, we get the following statements:


• If a1 , a2 , . . . , an are nonnegative real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 + a2 + · · · + an = n,

then
a13 + a23 + · · · + an3 + 2n ≥ 3(a12 + a22 + · · · + an2 ),
170 Vasile Cîrtoaje

with equality for a1 = a2 = · · · = an = 1, and also for

a1 = 0, a2 = · · · = an−1 = 1, an = 2.

• If a1 , a2 , . . . , an are nonnegative real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 + a2 + · · · + an = n,

then
a13 + a23 + · · · + an3 + 2n ≤ 3(a12 + a22 + · · · + an2 ),
with equality for a1 = a2 = · · · = an = 1, and also for

a1 = 2, a2 = · · · = an−1 = 1, an = 0.

Remark 3. Replacing n with 2n and choosing then m = n, we get the following


results:
• If a1 , a2 , . . . , a2n are nonnegative real numbers so that

a1 ≤ · · · ≤ an ≤ 1 ≤ an+1 ≤ · · · ≤ a2n , a1 + a2 + · · · + a2n = 2n,

then
n(a13 + a23 + · · · + a2n
3
− 2n) ≥ (2n + 1)(a12 + a22 + · · · + a2n
2
− 2n),
with equality for a1 = a2 = · · · = a2n = 1, and also for
1
a1 = 0, a2 = · · · = an = 1, an+1 = · · · = a2n = 1 + .
n

• If a1 , a2 , . . . , a2n are nonnegative real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n,

then
a13 + a23 + · · · + a2n
3
− 2n ≤ (n + 2)(a12 + a22 + · · · + a2n
2
− 2n),
with equality for a1 = a2 = · · · = a2n = 1, and also for

a1 = n + 1, a2 = · · · = an = 1, an+1 = · · · = a2n = 0.

P 2.6. Let a1 , a2 , . . . , an (n ≥ 3) be real numbers so that a1 + a2 + · · · + an = n. Prove


that
(a) if a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , then

a14 + a24 + · · · + an4 − n ≥ 6(a12 + a22 + · · · + an2 − n);


HCF Method for Ordered Variables 171

(b) if a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then


14 2
a14 + a24 + · · · + an4 − n ≥ (a1 + a22 + · · · + an2 − n);
3
(c) if a1 ≤ a2 ≤ 1 ≤ a3 ≤ · · · ≤ an , then

2(n2 − 3n + 3) 2
a14 + a24 + ··· + an4 −n≥ (a1 + a22 + · · · + an2 − n).
n − 5n + 7
2

(Vasile C., 2009)

Solution. Consider the inequality

a14 + a24 + · · · + an4 − n ≥ k(a12 + a22 + · · · + an2 − n), k ≤ 6,

and write it as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = u4 − ku2 , u ∈ R.
From f 00 (u) = 2(6u2 − k), it follows that f is convex for u ≥ 1. Therefore, we may
apply the RHCF-OV Theorem for m = n − 1, m = n − 2 and m = 2, respectively. By
Note 1, it suffices to show that h(x, y) ≥ 0 for all real x, y so that x + (n − m) y =
1 + n − m. We have
f (u) − f (1)
g(u) = = u3 + u2 + u + 1 − k(u + 1),
u−1
g(x) − g( y)
h(x, y) = = x 2 + x y + y 2 + x + y + 1 − k.
x−y

(a) We need to show that h(x, y) ≥ 0 for k = 6, m = n − 1, x + y = 2. Indeed,


we have
1
h(x, y) = 1 − x y = (x − y)2 ≥ 0.
4
From x + y = 2 and h(x, y) = 0, we get x = y = 1. Therefore, in accordance with
Note 4, the equality holds for a1 = a2 = · · · = an = 1.
(b) For k = 14/3, m = n − 2 and x + 2 y = 3, we have
1
h(x, y) = (3 y − 5)2 ≥ 0.
3
From x + 2 y = 3 and h(x, y) = 0, we get x = −1/3 and y = 5/3. Therefore, the
equality holds for a1 = a2 = · · · = an = 1, and also for
−1 5
a1 = , a2 = · · · = an−2 = 1, an−1 = an = .
3 3
172 Vasile Cîrtoaje

2(n2 − 3n + 3)
(c) We have k = , m = 2 and x +(n−2) y = n−1, which involve
n2 − 5n + 7
[(n2 − 5n + 7) y − n2 + 3n − 1]2
h(x, y) = ≥ 0.
n2 − 5n + 7
From x + (n − 2) y = n − 1 and h(x, y) = 0, we get

−n2 + 5n − 5 n2 − 3n + 1
x= , y= .
n2 − 5n + 7 n2 − 5n + 7
Therefore, the equality holds for a1 = a2 = · · · = an = 1, and also for

−n2 + 5n − 5 n2 − 3n + 1
a1 = , a2 = 1, a3 = · · · = a n = .
n2 − 5n + 7 n2 − 5n + 7

P 2.7. Let a, b, c, d, e be nonnegative real numbers so that a + b + c + d + e = 5. Prove


that
(a) if a ≥ b ≥ 1 ≥ c ≥ d ≥ e, then

21(a2 + b2 + c 2 + d 2 + e2 ) ≥ a4 + b4 + c 4 + d 4 + e4 + 100;

(b) if a ≥ b ≥ c ≥ 1 ≥ d ≥ e, then

13(a2 + b2 + c 2 + d 2 + e2 ) ≥ a4 + b4 + c 4 + d 4 + e4 + 60.

(Vasile C., 2009)

Solution. Consider the inequality

k(a2 + b2 + c 2 + d 2 + e2 − 5) ≥ a4 + b4 + c 4 + d 4 + e4 − 5, k ≥ 6,

and write it as
a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
f (u) = ku2 − u4 , u ≥ 0.
From f 00 (u) = 2(k − 6u2 ), it follows that f is convex on [0, 1]. Therefore, we may
apply the LHCF-OV Theorem for m = 2 and m = 3, respectively. By Note 1, it
suffices to show that h(x, y) ≥ 0 for all x, y ≥ 0 so that x + (5 − m) y = 6 − m. We
have
f (u) − f (1)
g(u) = = k(u + 1) − (u3 + u2 + u + 1),
u−1
HCF Method for Ordered Variables 173

g(x) − g( y)
h(x, y) = = k − (x 2 + x y + y 2 + x + y + 1).
x−y
(a) We need to show that h(x, y) ≥ 0 for k = 21, n = 5, m = 2 and x +3 y = 4;
indeed, we have

h(x, y) = 21 − (x 2 + x y + y 2 + x + y + 1) = y(22 − 7 y) = y(10 + 3x + 2 y) ≥ 0.

From x +3 y = 4 and h(x, y) = 0, we get x = 4 and y = 0. Therefore, in accordance


with Note 4, the equality holds for a = b = c = d = e = 1, and also for

a = 4, b = 1, c = d = e = 0.

(b) We have k = 13, n = 5, m = 3 and x + 2 y = 3, which involve

h(x, y) = 13 − (x 2 + x y + y 2 + x + y + 1) = y(10 − 3 y) = y(4 + 2x + y) ≥ 0.

From x + 2 y = 3 and h(x, y) = 0, we get x = 3 and y = 0. Therefore, the equality


holds for a = b = c = d = e = 1, and also for

a = 3, b = c = 1, d = e = 0.

P 2.8. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 + a2 +· · ·+ an = n.


Prove that
(a) if a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , then

7(a13 + a23 + · · · + an3 ) ≥ 3(a14 + a24 + · · · + an4 ) + 4n;

(b) if a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , then

13(a13 + a23 + · · · + an3 ) ≥ 4(a14 + a24 + · · · + an4 ) + 9n.

(Vasile C., 2009)

Solution. Consider the inequality

k(a13 + a23 + · · · + an3 − n) ≥ a14 + a24 + · · · + an4 − n, k ≥ 2,

and write it as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = ku3 − u4 , u ≥ 0.
174 Vasile Cîrtoaje

From f 00 (u) = 6u(k − 2u2 ), it follows that f is convex on [0, 1]. Therefore, we may
apply the LHCF-OV Theorem for m = n − 1 and m = n − 2, respectively. By Note 1,
it suffices to show that h(x, y) ≥ 0 for x ≥ y ≥ 0 so that x + m y = 1 + m. We have
f (u) − f (1)
g(u) = = k(u2 + u + 1) − (u3 + u2 + u + 1),
u−1
g(x) − g( y)
h(x, y) = = −(x 2 + x y + y 2 ) + (k − 1)(x + y + 1).
x−y

(a) We need to show that h(x, y) ≥ 0 for k = 7/3, m = n − 1, x + y = 2.


Indeed,
h(x, y) = x y ≥ 0.
From x > y, x + y = 2 and h(x, y) = 0, we get x = 2 and y = 0. Therefore, in
accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = 2, a2 = · · · = an−1 = 1, an = 0.

(b) We have k = 13/4, m = n − 2, x + 2 y = 3, which involve

h(x, y) = 3 y(9 − 4 y) = 3 y(3 + 2x) ≥ 0.

From x + 2 y = 3 and h(x, y) = 0, we get x = 3 and y = 0. Therefore, the equality


holds for a1 = a2 = · · · = an = 1, and also for

a1 = 3, a2 = · · · = an−2 = 1, an−1 = an = 0.

P 2.9. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n and

a1 ≥ · · · ≥ am ≥ 1 ≥ am+1 ≥ · · · ≥ an , m ∈ {1, 2, . . . , n − 1},

then
1 1 1
 ‹
(n − m + 1) 2
+ + ··· + − n ≥ 4(n − m)(a12 + a22 + · · · + an2 − n).
a1 a2 an

(Vasile C., 2007)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
(n − m + 1)2
f (u) = − 4(n − m)u2 , u > 0.
u
HCF Method for Ordered Variables 175

For u ∈ (0, 1], we have


2(n − m + 1)2
f 00 (u) = − 8(n − m)
u3
≥ 2(n − m + 1)2 − 8(n − m) = 2(n − m − 1)2 ≥ 0.
Since f is convex on (0, s], we may apply the LHCF-OV Theorem. By Note 1, it
suffices to show that h(x, y) ≥ 0 for all x, y > 0 so that x + (n − m) y = 1 + n − m.
We have
f (u) − f (1) −(n − m + 1)2
g(u) = = − 4(n − m)(u + 1),
u−1 u
(n − m + 1)2 [n − m + 1 − 2(n − m) y]2
h(x, y) = − 4(n − m) = ≥ 0.
xy xy
From x + (n − m) y = 1 + n − m and h(x, y) = 0, we get
n−m+1 n−m+1
x= , y= .
2 2(n − m)
Therefore, in accordance with Note 4, the equality holds for a1 = a2 = · · · = an = 1,
and also for
n−m+1 n−m+1
a1 = , a2 = a3 = · · · = am = 1, am+1 = · · · = an = .
2 2(n − m)
Remark 1. For m = n − 1, we get the following elegant statement:
• If a1 , a2 , . . . , an are positive real numbers so that
a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 + a2 + · · · + an = n,
then
1 1 1
+ + ··· + ≥ a12 + a22 + · · · + an2 ,
a1 a2 an
with equality for a1 = a2 = · · · = an = 1

Remark 2. Replacing n with 2n and choosing then m = n, we get the following


statement:
• If a1 , a2 , . . . , a2n are positive real numbers so that
a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n,
then
1 1 1
 ‹
(n + 1) 2
+ + ··· + − 2n ≥ 4n(a12 + a22 + · · · + a2n
2
− 2n),
a1 a2 a2n
with equality for a1 = a2 = · · · = a2n = 1, and also for
n+1 n+1
a1 = , a2 = a3 = · · · = an = 1, an+1 = · · · = a2n = .
2 2n
176 Vasile Cîrtoaje

1 1 1
P 2.10. If a1 , a2 , . . . , an are positive real numbers so that + +···+ = n and
a1 a2 an
a1 ≤ · · · ≤ am ≤ 1 ≤ am+1 ≤ · · · ≤ an , m ∈ {1, 2, . . . , n − 1},

then
 p 
n−m
a12 + a22 + ··· + an2 −n≥2 1+ (a1 + a2 + · · · + an − n).
n−m+1

(Vasile C., 2007)

Solution. Replacing each ai by 1/ai , we need to prove that

a1 ≥ · · · ≥ am ≥ 1 ≥ am+1 ≥ · · · ≥ an , a1 + a2 + · · · + a n = n

involves
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where p
1 2k m−n
f (u) = 2 − , k =1+ , u > 0.
u u n−m+1
For u ∈ (0, 1], we have
p
6 − 4ku 6 − 4k 2( n − m − 1)2
f (u) =
00
≥ = ≥ 0.
u4 u4 (n − m + 1)u4
Thus, f is convex on (0, 1]. By the LHCF-OV Theorem and Note 1, it suffices to
show that h(x, y) ≥ 0 for x, y > 0 so that x + (n − m) y = 1 + n − m, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
We have
−1 2k − 1
g(u) = +
u2 u
and
1 1 1
 ‹
h(x, y) = + + 1 − 2k .
xy x y
We only need to show that
p
1 1 2 n−m
+ ≥1+ .
x y n−m+1
Indeed, using the Cauchy-Schwarz inequality, we get
p p p
1 1 (1 + n − m)2 (1 + n − m)2 2 n−m
+ ≥ = =1+ .
x y x + (n − m) y n−m+1 n−m+1
HCF Method for Ordered Variables 177

From x + (n − m) y = 1 + n − m and h(x, y) = 0, we get


n−m+1 n−m+1
x= p , y= p .
1+ n−m n−m+ n−m
By Note 4, we have
f (a1 ) + f (a2 ) + · · · + f (an ) = n f (1)
for a1 = a2 = · · · = an = 1, and also for
n−m+1 n−m+1
a1 = p , a2 = a3 = · · · = am = 1, am+1 = · · · = an = p .
1+ n−m n−m+ n−m
Therefore, the original inequality becomes an equality for a1 = a2 = · · · = an = 1,
and also for
p p
1+ n−m n−m+ n−m
a1 = , a2 = a3 = · · · = am = 1, am+1 = · · · = an = .
n−m+1 n−m+1
Remark. Replacing n with 2n and choosing then m = n, we get the statement
below.
• If a1 , a2 , . . . , a2n are positive real numbers so that
1 1 1
a1 ≤ · · · ≤ an ≤ 1 ≤ an+1 ≤ · · · ≤ a2n , + + ··· + = 2n,
a1 a2 a2n
then
 p ‹
n
a12 + a22 + ··· + 2
a2n − 2n ≥ 2 1 + (a1 + a2 + · · · + a2n − 2n).
n+1
with equality for a1 = a2 = · · · = a2n = 1, and also for
p p
1+ n n+ n
a1 = , a2 = a3 = · · · = an = 1, an+1 = · · · = a2n = .
n+1 n+1

P 2.11. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 +a2 +· · ·+an = n.


Prove that
(a) if a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , then
1 1 1 n
+ + · · · + ≥ ;
a12 + 2 a22 + 2 an2 + 2 3
(b) if a1 ≤ · · · ≤ an−2 ≤ 1 ≤ an−1 ≤ an , then
1 1 1 n
+ 2 + ··· + 2 ≥ .
2a12 + 3 2a2 + 3 2an + 3 5

(Vasile C., 2007)


178 Vasile Cîrtoaje

Solution. Consider the inequality


1 1 1 n
+ 2 + ··· + 2 ≥ , k ∈ [0, 3];
a12 + k a2 + k an + k 1 + k

and write it as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
and
1
f (u) = , u ≥ 0.
u2 +k
For u ≥ 1, we have
2(3u2 − k) 2(3 − k)
f (u) =
00
≥ 2 ≥ 0,
(u + k)
2 3 (u + k)3
hence f (u) is convex for u ≥ s. Therefore, we may apply the RHCF-OV Theorem for
m = n−1 and m = n−2, respectively. By Note 1, it suffices to show that h(x, y) ≥ 0
for all x, y ≥ 0 so that x + (n − m) y = 1 + n − m. Since
f (u) − f (1) −u − 1
g(u) = = ,
u−1 (1 + k)(u2 + k)
g(x) − g( y) xy+x + y−k
h(x, y) = = ,
x−y (1 + k)(x 2 + k)( y 2 + k)
we only need to show that
x y + x + y − k ≥ 0.

(a) We need to show that x y + x + y − k ≥ 0 for k = 2, m = n − 1, x + y = 2;


indeed, we have
x y + x + y − k = x y ≥ 0.
From x < y, x + y = 2 and x y + x + y − k = 0, we get x = 0 and y = 2. Therefore,
by Note 4, the equality holds for a1 = a2 = · · · = an = 1, and also for

a1 = 0, a2 = · · · = an−1 = 1, an = 2.

(b) We have k = 3/2, m = n − 2, x + 2 y = 3, hence


x(4 − x) x(1 + 2 y)
xy+x + y−k= = ≥ 0.
2 2
From x + 2 y = 3 and x y + x + y − k = 0, we get x = 0 and y = 3/2. Therefore,
the equality holds for a1 = a2 = · · · = an = 1, and also for
3
a1 = 0, a2 = · · · = an−2 = 1, an−1 = an = .
2
HCF Method for Ordered Variables 179

P 2.12. If a1 , a2 , . . . , a2n are nonnegative real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n,

then
1 1 1 2n
+ 2 + ··· + 2 ≤ .
na12 + n + n + 1 na2 + n + n + 1
2 2 na2n + n + n + 1 (n + 1)2
2

(Vasile C., 2007)

Solution. Write the inequality as


a1 + a2 + · · · + a2n
f (a1 ) + f (a2 ) + · · · + f (a2n ) ≥ 2n f (s), s= = 1,
2n
where
−1
f (u) = , u ≥ 0.
nu2 + n2 + n + 1
For u ∈ [0, 1], we have

2nu(n2 + n + 1 − 3nu2 ) 2nu(n2 + n + 1 − 3n)


f 00 (u) = ≥ ≥ 0,
(nu2 + n2 + n + 1)3 (nu2 + n2 + n + 1)3

hence f is convex on [0, s]. Therefore, we may apply the LHCF-OV Theorem for 2n
numbers and m = n. By Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ≥ 0
so that x + n y = 1 + n. We have

f (u) − f (1) n(u + 1)


g(u) = = ,
u−1 (n + 1) (nu2 + n2 + n + 1)
2

g(x) − g( y)
h(x, y) =
x−y
n(n2 + n + 1 − nx − n y − nx y)
=
(n + 1)2 (nx 2 + n2 + n + 1)(n y 2 + n2 + n + 1)
n(n y − 1)2
= ≥ 0.
(n + 1)2 (nx 2 + n2 + n + 1)(n y 2 + n2 + n + 1)

From x + n y = 1 + n and h(x, y) = 0, we get x = n and y = 1/n. Therefore, the


equality holds for a1 = a2 = · · · = a2n = 1, and also for

a1 = n, a2 = · · · = an = 1, an+1 = · · · = an = f r ac1n.
180 Vasile Cîrtoaje

P 2.13. If a, b, c, d, e, f are nonnegative real numbers so that


a ≥ b ≥ c ≥ 1 ≥ d ≥ e ≥ f, a + b + c + d + e + f = 6,
then
3a + 4 3b + 4 3c + 4 3d + 4 3e + 4 3f + 4
+ + + + + ≤ 6.
3a2 + 4 3b2 + 4 3c 2 + 4 3d 2 + 4 3e2 + 4 3 f 2 + 4
(Vasile C., 2009)
Solution. Write the inequality as
a+b+c+d+e+ f
f (a) + f (b) + f (c) + f (d) + f (e) + f ( f ) ≥ 6 f (s), s= = 1,
6
where
−3u − 4
f (u) = , u ≥ 0.
3u2 + 4
For u ∈ [0, 1], we have
6(16 − 9u3 ) + 216u(1 − u)
f (u) =
00
> 0,
(3u2 + 4)3
hence f is convex on [0, s]. Therefore, we may apply the LHCF-OV Theorem for
n = 6 and m = 3. By Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ≥ 0 so
that x + 3 y = 4. We have
f (u) − f (1) 3u
g(u) = = 2 ,
u−1 3u + 4

g(x) − g( y) 3(4 − 3x y)
h(x, y) = =
x−y (3x + 4)(3 y 2 + 4)
2

3(x − 2)2
= ≥ 0.
(3x 2 + 4)(3 y 2 + 4)
From x + 3 y = 4 and h(x, y) = 0, we get x = 2 and y = 2/3. Therefore, in
accordance with Note 4, the equality holds for a = b = c = d = e = f = 1, and
also for
2
a = 2, b = c = 1, d = e = f = .
3

P 2.14. If a, b, c, d, e, f are nonnegative real numbers so that


a ≥ b ≥ 1 ≥ c ≥ d ≥ e ≥ f, a + b + c + d + e + f = 6,
then
a2 − 1 b2 − 1 c2 − 1 d2 − 1 e2 − 1 f2−1
+ + + + + ≥ 0.
(2a + 7)2 (2b + 7)2 (2c + 7)2 (2d + 7)2 (2e + 7)2 (2 f + 7)2
(Vasile C., 2009)
HCF Method for Ordered Variables 181

Solution. Write the inequality as

a+b+c+d+e+ f
f (a) + f (b) + f (c) + f (d) + f (e) + f ( f ) ≥ 6 f (s), s= = 1,
6

where
u2 − 1
f (u) = , u ≥ 0.
(2u + 7)2
For u ∈ [0, 1], we have
2(37 − 28u)
f 00 (u) = > 0,
(2u + 7)4
hence f is convex on [0, s]. Therefore, we may apply the LHCF-OV Theorem for
n = 6 and m = 2. By Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ≥ 0 so
that x + 4 y = 5. We have

f (u) − f (1) u+1


g(u) = = ,
u−1 (2u + 7)2

g(x) − g( y) 21 − 4x − 4 y − 4x y
h(x, y) = =
x−y (2x + 7)2 (2 y + 7)2
(x − 4)2
= ≥ 0.
(2x + 7)2 (2 y + 7)2

From x + 4 y = 5 and h(x, y) = 0, we get x = 4 and y = 1/4. Therefore, the


equality holds only for a = b = c = d = e = f = 1, and also for

1
a = 4, b = 1, c=d=e= f = .
4

P 2.15. If a, b, c, d, e, f are nonnegative real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d ≤ e ≤ f, a + b + c + d + e + f = 6,

then

a2 − 1 b2 − 1 c2 − 1 d2 − 1 e2 − 1 f2−1
+ + + + + ≤ 0.
(2a + 5)2 (2b + 5)2 (2c + 5)2 (2d + 5)2 (2e + 5)2 (2 f + 5)2

(Vasile C., 2009)


182 Vasile Cîrtoaje

Solution. Write the inequality as

a+b+c+d+e+ f
f (a) + f (b) + f (c) + f (d) + f (e) + f ( f ) ≥ 6 f (s), s= = 1,
6
where
1 − u2
f (u) = , u ≥ 0.
(2u + 5)2
For u ≥ 1, we have
2(20u − 13)
f 00 (u) = > 0,
(2u + 5)4
hence f (u) is convex for u ≥ s. Therefore, we may apply the RHCF-OV Theorem
for n = 6 and m = 2. By Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ≥ 0
so that x + 4 y = 5. We have

f (u) − f (1) −u − 1
g(u) = = ,
u−1 (2u + 5)2

g(x) − g( y)
h(x, y) =
x−y
4x y + 4x + 4 y − 5
=
(2x + 5)2 (2 y + 5)2
4x y + 3x
= ≥ 0.
(2x + 5)2 (2 y + 5)2

From x + 4 y = 5 and h(x, y) = 0, we get x = 0 and y = 5/4. Therefore, in


accordance with Note 4, the equality holds only for a = b = c = d = e = f = 1,
and also for
5
a = 0, b = 1, c = d = e = f = .
4

P 2.16. If a, b, c are nonnegative real numbers so that

a ≤ b ≤ 1 ≤ c, a + b + c = 3,

then v v v
t 2a t 2b t 2c
+ + ≥ 3.
b+c c+a a+b

(Vasile C., 2008)


HCF Method for Ordered Variables 183

Solution. Write the inequality as

a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where s
u
f (u) = , u ∈ [0, 3).
3−u
From
3(4u − 3)
f 00 (u) = ,
4u3/2 (3 − u)5/2
it follows that f (u) is convex for u ≥ s. Therefore, we may apply the RHCF-OV
Theorem for n = 3 and m = 2. So, it suffices to show that

f (x) + f ( y) ≥ 2 f (1)

for x + y = 2, 0 ≤ x ≤ 1 ≤ y. This inequality is true if g(x) ≥ 0, where

g(x) = f (x) + f ( y) − 2 f (1), y = 2 − x, x ∈ [0, 1].

Since y 0 = −1, we have


 
3 1 1
g (x) = f (x) − f ( y) =
0 0 0
p −p .
2 x(3 − x)3 y(3 − y)3

The derivative f 0 (x) has the same sign as h(x), where

h(x) = y(3 − y)3 − x(3 − x)3 = (2 − x)(1 + x)3 − x(3 − x)3


= 2(1 − 11x + 15x 2 − 5x 3 ) = 2(1 − x)(1 − 10x + 5x 2 ).

Let
2
x1 = 1 − p .
5
Since h(x 1 ) = 0, h(x) > 0 for x ∈ [0, x 1 ) and h(x) < 0 for x ∈ (x 1 , 1), it follows
that g is increasing on [0, x 1 ] and decreasing on [x 1 , 1]. From

g(0) = f (0) + f (2) − 2 f (1) = 0,

g(1) = f (1) + f (1) − 2 f (1) = 0,


it follows that g(x) ≥ 0 for x ∈ [0, 1].
The equality holds for a = b = c = 1, and also for a = 0, b = 1 and c = 2.
184 Vasile Cîrtoaje

P 2.17. If a1 , a2 , . . . , a8 are nonnegative real numbers so that

a1 ≥ a2 ≥ a3 ≥ a4 ≥ 1 ≥ a5 ≥ a6 ≥ a7 ≥ a8 , a1 + a2 + · · · + a8 = 8,

then
(a12 + 1)(a22 + 1) · · · (a82 + 1) ≥ (a1 + 1)(a2 + 1) · · · (a8 + 1).

(Vasile C., 2008)

Solution. Write the inequality as

a1 + a2 + · · · + a8
f (a1 ) + f (a2 ) + · · · + f (a8 ) ≥ 8 f (s), s= = 1,
8

where
f (u) = ln(u2 + 1) − ln(u + 1), u ≥ 0.
For u ∈ [0, 1], we have

2(1 − u2 ) 1 (u2 − u4 ) + 4u(1 − u2 ) + u2 + 3


f 00 (u) = + = > 0.
(u2 + 1)2 (u + 1)2 (u2 + 1)2 (u + 1)2

Therefore, f is convex on [0, s]. According to the LHCF-OV Theorem applied for
n = 8 and m = 4, it suffices to show that f (x) + 4 f ( y) ≥ 5 f (1) for x, y ≥ 0 so that
x + 4 y = 5. Using Note 2, we only need to show that H(x, y) ≥ 0 for x, y ≥ 0 so
that x + 4 y = 5, where

f 0 (x) − f 0 ( y) 2(1 − x y) 1
H(x, y) = = 2 + .
x−y (x + 1)( y + 1) (x + 1)( y + 1)
2

The inequality H(x, y) ≥ 0 is equivalent to

2(1 − x y)(x + 1)( y + 1) + (x 2 + 1)( y 2 + 1) ≥ 0.

Since 2(x 2 + 1) ≥ (x + 1)2 and 2( y 2 + 1) ≥ ( y + 1)2 , it suffices to prove that

8(1 − x y) + (x + 1)( y + 1) ≥ 0.

Indeed,

8(1 − x y) + (x + 1)( y + 1) = 28x 2 − 38x + 14 = 28(x − 19/28)2 + 31/28 > 0.

The proof is completed. The equality holds for a1 = a2 = · · · = a8 .


HCF Method for Ordered Variables 185

P 2.18. If a, b, c, d are real numbers so that

−1
≤ a ≤ b ≤ 1 ≤ c ≤ d, a + b + c + d = 4,
2

then
1 1 1 1 1 1 1 1
 ‹  ‹
7 2 + 2 + 2 + 2 +3 + + + ≥ 40.
a b c d a b c d

(Vasile C., 2011)

Solution. We have
1 1
d =4−a− b−c ≤4+ + − 1 = 4.
2 2

Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4

where
7 3 −1
•
˜
f (u) = 2 + , u∈I= , 4 \ {0}.
u u 2
7 3
Clearly, f (u) is convex for u ≥ 1 (because 2
and are convex). According to Note
u u
3, we may apply the RHCF-OV Theorem for n = 4 and m = 2. By Note 1, we only
need to show that h(x, y) ≥ 0 for x, y ∈ I so that x + 2 y = 3, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

We have
7 10
g(u) = − 2
− ,
u u
7(x + y) + 10x y (2x + 1)(−5x + 21)
h(x, y) = 2 2
= ≥ 0.
x y 2x 2 y 2
From x + 2 y = 3 and h(x, y) = 0, we get x = −1/2, y = 7/3. Therefore, in
accordance with Note 4, the equality holds for a = b = c = d = 1, and also for

−1 7
a= , b = 1, c=d= .
2 4
186 Vasile Cîrtoaje

P 2.19. Let a, b, c, d be real numbers. Prove that


(a) if −1 ≤ a ≤ b ≤ c ≤ 1 ≤ d, then

1 1 1 1 1 1 1 1
 ‹
3 2 + 2 + 2 + 2 ≥8+ + + + ;
a b c d a b c d
(b) if −1 ≤ a ≤ b ≤ 1 ≤ c ≤ d, then

1 1 1 1 1 1 1 1
 ‹
2 2 + 2 + 2 + 2 ≥4+ + + + .
a b c d a b c d

(Vasile C., 2011)

Solution. (a) We have

d = 4 − a − b − c ≤ 4 + 1 + 1 + 1 = 7.

Write the desired inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
3 1
f (u) = − , u ∈ I = [−1, 7] \ {0}.
u2 u
From
2(9 − u)
f 00 (u) = > 0,
u4
it follows that f is convex on I≥s . According to Note 3, we may apply the RHCF-OV
Theorem for n = 4 and m = 3. By Note 1, it suffices to show that h(x, y) ≥ 0 for
all x, y ∈ I so that x + y = 2. We have

f (u) − f (1) 2 3
g(u) = = − − 2,
u−1 u u

g(x) − g( y) 3(x + y) + 2x y
h(x, y) = =
x−y x2 y2
2(x + 1)(3 − x) 2(x + 1)( y + 1)
= = ≥ 0.
x2 y2 x2 y2

From x < y, x + y = 2 and h(x, y) = 0, we get x = −1 and y = 3. Therefore, in


accordance with Note 4, the equality holds for a = b = c = d = 1, and also for

a = −1, b = c = 1, d = 3.

(b) We have
d = 4 − a − b − c ≤ 4 + 1 + 1 − 1 = 5.
HCF Method for Ordered Variables 187

Write the desired inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
2 1
f (u) = − , u ∈ I = [−1, 5] \ {0}.
u2 u
From
2(6 − u)
f 00 (u) = > 0,
u4
it follows that f is convex on I≥s . According to Note 3, we may apply the RHCF-OV
Theorem for n = 4 and m = 2. By Note 1, it suffices to show that h(x, y) ≥ 0 for
all x, y ∈ I so that x + 2 y = 3. We have

f (u) − f (1) 1 2
g(u) = = − − 2,
u−1 u u

g(x) − g( y) 2(x + y) + x y
h(x, y) = =
x−y x2 y2
(x + 1)(6 − x)
= ≥ 0.
2x 2 y 2

From x +2 y = 3 and h(x, y) = 0, we get x = −1 and y = 2. Therefore, the equality


holds for a = b = c = d = 1, and also for

a = −1, b = 1, c = d = 2.

P 2.20. If a, b, c, d are positive real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, abcd = 1,

then
1 1 1 1
 ‹
a + b + c + d − 4 ≥ 18 a + b + c + d − − − −
2 2 2 2
.
a b c d
(Vasile C., 2008)

Solution. Using the substitution

a = ex , b = ey, c = ez , d = ew,

we need to show that

f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s),


188 Vasile Cîrtoaje

where
x + y +z+w
x ≥ y ≥ 0 ≥ z ≥ w, s= = 0,
4
f (u) = e2u − 1 − 18(eu − e−u ), u ∈ R.
For u ≤ 0, we have
f 00 (u) = 4e2u + 18(e−u − eu ) > 0,
hence f is convex on (−∞, s]. By the LHCF-OV Theorem applied for n = 4 and
m = 2, it suffices to show that f (x) + 2 f ( y) ≥ 3 f (0) for all real x, y so that
x + 2 y = 0; that is, to show that
1 2
 ‹
a + 2b − 3 − 18 a + 2b − −
2 2
≥0
a b
for all a, b > 0 so that ab2 = 1. This inequality is equivalent to

(b2 − 1)2 (2b2 + 1) 18(b − 1)3 (b + 1)


+ ≥ 0,
b4 b2
(b − 1)2 (2b − 1)2 (b + 1)(5b + 1)
≥ 0.
b4
The proof is completed. The equality holds for a = b = c = d = 1, and also for

a = 4, b = 1, c = d = 1/2.

P 2.21. If a, b, c, d are positive real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, abcd = 1,

then
p p p p
a2 − a + 1 + b2 − b + 1 + c2 − c + 1 + d 2 − d + 1 ≥ a + b + c + d.

(Vasile C., 2008)

Solution. Using the substitution

a = ex , b = ey, c = ez , d = ew,

we need to show that

f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s),

where
x + y +z+w
x ≤ y ≤ 0 ≤ z ≤ w, s= = 0,
4
HCF Method for Ordered Variables 189

p
f (u) = e2u − eu + 1 − eu , u ∈ R.
We claim that f is convex for u ≥ 0. Since

4e3u − 6e2u + 9eu − 2


e−u f 00 (u) = − 1,
4(e2u − eu + 1)3/2

we need to show that


4t 3 − 6t 2 + 9t − 2 ≥ 0
and
(4t 3 − 6t 2 + 9t − 2)2 ≥ 16(t 2 − t + 1)3 ,
where t = eu ≥ 1. Indeed, we have

4t 3 − 6t 2 + 9t − 2 ≥ 4t 3 − 6t 2 + 7t > 4t 3 − 6t 2 + 2t = 2t(t − 1)(2t − 1) ≥ 0

and

(4t 3 − 6t 2 + 9t − 2)2 − 16(t 2 − t + 1)3 = 12t 3 (t − 1) + 9t 2 + 12(t − 1) > 0.

By the RHCF-OV Theorem applied for n = 4 and m = 2, it suffices to show that


f (x) + 2 f ( y) ≥ 3 f (0) for all real x, y so that x + 2 y = 0; that is, to show that
p p
a2 − a + 1 + 2 b2 − b + 1 ≥ a + 2b

for all a, b > 0 so that ab2 = 1. This inequality is equivalent to


p
b4 − b2 + 1 p 1
2
+ 2 b2 − b + 1 ≥ 2 + 2b,
b b
p
b4 − b2 + 1 − 1 p
+ 2( b2 − b + 1 − 1) ≥ 0,
b2
b2 − 1 2(1 − b)
p +p ≥ 0.
b4 − b2 + 1 + 1 b2 − b + 1 + b
Since
b2 − 1 b2 − 1
p ≥ ,
b4 − b2 + 1 + 1 b2 + 1
it suffices to show that

b2 − 1 2(1 − b)
+p ≥ 0,
b +1
2
b2 − b + 1 + b
which is equivalent to

b+1 2
• ˜
(b − 1) 2 −p ≥ 0,
b +1 b2 − b + 1 + b
190 Vasile Cîrtoaje
” p —
(b − 1) (b + 1) b2 − b + 1 − b2 + b − 2 ≥ 0,
(b − 1)2 (3b2 − 2b + 3)
p ≥ 0.
(b + 1) b2 − b + 1 + b2 − b + 2
The last inequality is clearly true. The equality holds for a = b = c = d = 1.

P 2.22. If a, b, c, d are positive real numbers so that

a ≤ b ≤ c ≤ 1 ≤ d, abcd = 1,

then
1 1 1 1 2
+ 3 + 3 + 3 ≥ .
a3 + 3a + 2 b + 3b + 2 c + 3c + 2 d + 3d + 2 3
(Vasile C., 2007)
Solution. Using the substitution

a = ex , b = ey, c = ez , d = ew,

we need to show that

f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s),

where
x + y +z+w
x ≤ y ≤ z ≤ 0 ≤ w, s= = 0,
4
1
f (u) = , u ∈ R.
+ 3eu + 2
e3u
We claim that f is convex for u ≥ 0. Indeed, denoting t = eu , t ≥ 1, we have
3t(3t 5 + 2t 3 − 6t 2 + 3t − 2)
f 00 (u) =
(t 3 + 3t + 2)3
3t(t − 1)(3t 4 + 3t 3 + 5t 2 − t + 2)
= ≥ 0.
(t 3 + 3t + 2)3
By the RHCF-OV Theorem applied for n = 4 and m = 3, it suffices to show that
f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that is, to show that
1 1 1
+ 3 ≥
a3 + 3a + 2 b + 3b + 2 3
for all a, b > 0 so that ab = 1. This inequality is equivalent to

(a − 1)4 (a2 + a + 1) ≥ 0.

The equality holds for a = b = c = d = 1.


HCF Method for Ordered Variables 191

P 2.23. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,

then
1 1 1
+ + ··· + ≥ a1 + a2 + · · · + an .
a1 a2 an
(Vasile C., 2007)
Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
f (u) = e−u − eu , u ∈ R.
For u ≤ 0, we have
f 00 (u) = e−u − eu ≥ 0,
therefore f (u) is convex for u ≤ s. By the LHCF-OV Theorem applied for m = n − 1,
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that
is, to show that
1 1
−a+ −b≥0
a b
for all a, b > 0 so that ab = 1. This is true since
1 1 1 1
− a + − b = − a + a − = 0.
a b a a
The equality holds for

a1 ≥ 1, a2 = · · · = an−1 = 1, an = 1/a1 .

P 2.24. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If k ≥ 1, then
1 1 1 n
+ + ··· + ≥ .
1 + ka1 1 + ka2 1 + kan 1+k
(Vasile C., 2007)
192 Vasile Cîrtoaje

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1
f (u) = , u ∈ R.
1 + keu
For u ≥ 0, we have
keu (keu − 1)
f (u) =
00
≥ 0,
(1 + keu )3
therefore f (u) is convex for u ≥ s. By the RHCF-OV Theorem applied for m = n−1,
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that
is, to show that
1 1 2
+ ≥
1 + ka 1 + kb 1 + k
for all a, b > 0 so that ab = 1. This is true since

1 1 2 k(k − 1)(a − 1)2


+ − = ≥ 0.
1 + ka 1 + kb 1 + k (1 + ka)(a + k)

The equality holds for a1 = a2 = · · · = an = 1. If k = 1, then the equality holds for

a1 ≤ 1, a2 = · · · = an−1 = 1, an = 1/a1 .

P 2.25. If a1 , a2 , . . . , a9 are positive real numbers so that

a1 ≤ · · · ≤ a8 ≤ 1 ≤ a9 , a1 a2 · · · a9 = 1,

then
1 1 1
+ + ··· + ≥ 1.
(a1 + 2)2 (a2 + 2)2 (a9 + 2)2

(Vasile C., 2007)


HCF Method for Ordered Variables 193

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , 9,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x 9 ) ≥ 9 f (s),

where
x1 + x2 + · · · + x9
x1 ≤ · · · ≤ x8 ≤ 0 ≤ x9, s= = 0,
9
1
f (u) = , u ∈ R.
(eu + 2)2
For u ∈ [0, ∞), we have
4eu (eu − 1)
f 00 (u) = ≥ 0,
(eu + 2)4
hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case n = 9
and m = 8), it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that
x + y = 0; that is, to show that
1 1 2
+ ≥
(a + 2)2 (b + 2)2 9
for all a, b > 0 so that ab = 1. Write this inequality as
b2 1 2
+ ≥ ,
(2b + 1)2 (b + 2)2 9
which is equivalent to the obvious inequality

(b − 1)4 ≥ 0.

The equality holds for a1 = a2 = · · · = a9 = 1.

P 2.26. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that
2pq
p+q ≥1+ ,
p + 4q
then
1 1 1 n
+ + ··· + ≥ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan
2 1+p+q

(Vasile C., 2007)


194 Vasile Cîrtoaje

Solution. Using the substitution


ai = e x i , i = 1, 2, . . . , n,
we can write the inequality as
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),
where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1
f (u) = , u ∈ R.
1+ peu + qe2u
We have
eu f1 (u)
f 00 (u) = ,
(1 + peu + qe2u )3
where
f1 (u) = 4q2 e3u + 3pqe2u + (p2 − 4q)eu − p.
2pq
The hypothesis p + q ≥ 1 + is equivalent to
p + 4q
p2 + 3pq + 4q2 ≥ p + 4q.
For u ∈ [0, ∞), we have
f1 (u) ≥ 4q2 eu + 3pqeu + (p2 − 4q)eu − p ≥ p(eu − 1) ≥ 0,
hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = n−1),
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that
is, to show that
1 1 2
+ ≥
1 + pa + qa2 1 + pb + qb2 1+p+q
for all a, b > 0 so that ab = 1. Write this inequality as
1 a2 2
+ ≥
1 + pa + qa 2 a + pa + q 1 + p + q
2

which is equivalent to
(a − 1)2 h(a) ≥ 0,
where
h(a) = q(p + q − 1)(a2 + 1) + (p2 + pq + 2q2 − p − 2q)a
≥ 2q(p + q − 1)a + (p2 + pq + 2q2 − p − 2q)a
= (p2 + 3pq + 4q2 − p − 4q)a ≥ 0.
The equality holds for a1 = a2 = · · · = an = 1.
Remark. For p = 1, q = 1/4 and n = 9, we get the preceding P 2.25.
HCF Method for Ordered Variables 195

P 2.27. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If m ≥ 1 and 0 < k ≤ m, then

1 1 1 n
+ + ··· + ≥ .
(a1 + k) m (a2 + k) m (an + k) m (1 + k)m

(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1
f (u) = , u ∈ R.
(eu + k)m
For u ∈ [0, ∞), we have

meu (meu − k)
f (u) =
00
≥ 0,
(eu + k)m+2

hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = n−1),
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x ≤ y and
x + y = 0; that is, to show that

1 1 2
+ ≥
(a + k) m (b + k) m (1 + k)m

for all a, b > 0 so that a ∈ (0, 1] and ab = 1. Write this inequality as g(a) ≥ 0,
where
1 am 2
g(a) = + − ,
(a + k) m (ka + 1) m (1 + k)m
with
g 0 (a) a m−1 (a + k)m+1 − (ka + 1)m+1
= .
m (a + k)m+1 (ka + 1)m+1
If g 0 (a) ≤ 0 for a ∈ (0, 1], then g is decreasing, hence g(a) ≥ g(1) = 0. Thus, it
suffices to show that
ka + 1 m+1
 ‹
m−1
a ≤ .
a+k
196 Vasile Cîrtoaje

Since
ka + 1 ma + 1 (m − k)(1 − a2 )
− = ≥ 0,
a+k a+m (a + k)(a + m)
we only need to show that

ma + 1
 ‹m+1
m−1
a ≤ ,
a+m
which is equivalent to h(a) ≤ 0 for a ∈ (0, 1], where

h(a) = (m − 1) ln a + (m + 1) ln(a + m) − (m + 1) ln(ma + 1),

with
m − 1 m + 1 m(m + 1) m(m − 1)(a − 1)2
h (a) =
0
+ − = .
a a+m ma + 1 a(a + m)(ma + 1)
Since h0 (a) ≥ 0, h(a) is increasing for a ∈ (0, 1], therefore h(a) ≤ h(1) = 0. The
equality holds for a1 = a2 = · · · = an = 1.
Remark. For k = m = 2 and n = 9, we get the inequality in P 2.25.

P 2.28. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1,

then
1 1 1 n
+p + ··· + p ≥ .
1 + 3a1 1 + 3a2 1 + 3an
p
2

(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1
f (u) = p , u ∈ R.
1 + 3eu
For u ≥ 0, we have
3eu (3eu − 2)
f 00 (u) = > 0,
4(1 + 3eu )5/2
HCF Method for Ordered Variables 197

hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = n−1),
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that
is, to show that
1 1
p +p ≥1
1 + 3a 1 + 3b
for all a, b > 0 so that ab = 1. Write this inequality as
s
1 a
p + ≥ 1.
1 + 3a a+3
1
Substituting p = t, 0 < t < 1, the inequality becomes
1 + 3a
v
t 1 − t2
≥ 1 − t.
8t 2 + 1
By squaring, we get
t(1 − t)(2t − 1)2 ≥ 0,
which is true. The equality holds for a1 = a2 = · · · = an = 1.

P 2.29. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.
1
If 0 < m < 1 and 0 < k ≤ , then
21/m −1
1 1 1 n
+ + ··· + ≥ .
(a1 + k) m (a2 + k) m (an + k) m (1 + k)m

(Vasile C., 2007)

Solution. By Bernoulli’s inequality, we have


1
21/m > 1 + ,
m
hence
1
k≤ < m < 1.
21/m − 1
Using the substitution
ai = e x i , i = 1, 2, . . . , n,
we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),
198 Vasile Cîrtoaje

where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1
f (u) = , u ∈ R.
(eu + k)m
For u ∈ [0, ∞), we have

meu (meu − k)
f 00 (u) = ≥ 0,
(eu + k)m+2

hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = n−1),
it suffices to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that
is, to show that
1 1 2
+ ≥
(a + k)m (b + k)m (1 + k)m
for all a, b > 0 so that ab = 1. Write this inequality as g(a) ≥ 0 for a ≥ 1, where

1 am 2
g(a) = + − .
(a + k) m (ka + 1) m (1 + k)m

The derivative
g 0 (a) a m−1 (a + k)m+1 − (ka + 1)m+1
=
m (a + k)m+1 (ka + 1)m+1
has the same sign as the function

h(a) = (m − 1) ln a + (m + 1) ln(a + k) − (m + 1) ln(ka + 1).

We have

m−1 1 kh1 (a)


 ‹
k
h (a) =
0
+ (m + 1) − = ,
a a + k ka + 1 a(a + k)(ka + 1)

where  m
h1 (a) = (m − 1)(a2 + 1) − 2 k − a.
k
The discriminant D of the quadratic function h1 (a) is
 2
m 2
‹
D  m
= k− − (m − 1)2 = (1 − k2 ) − 1 .
4 k k2

Since D > 0, the roots a1 and a2 of h1 (a) are real and unequal. If a1 < a2 , then
h1 (a) ≥ 0 for a ∈ [a1 , a2 ] and h1 (a) ≤ 0 for a ∈ (−∞, a1 ] ∪ [a2 , ∞). Since

2(k + 1)(m − k)
h1 (1) = > 0,
k
HCF Method for Ordered Variables 199

it follows that a1 < 1 < a2 , therefore h1 (a) and h0 (a) are positive for a ∈ [1, a2 )
and negative for a ∈ (a2 , ∞), h is increasing on [1, a2 ] and decreasing on [a2 , ∞).
From h(1) = 0 and
lim h(a) = −∞,
a→∞

it follows that there is a3 > a2 so that h(a) and g 0 (a) are positive for a ∈ (1, a3 ) and
negative for a ∈ (a3 , ∞). As a result, g is increasing on [1, a3 ] and decreasing on
[a3 , ∞). Since g(1) = 0 and

1 2
lim g(a) = − ≥ 0,
a→∞ k m (1 + k)m

it follows that g(a) ≥ 0 for a ≥ 1. This completes the proof. The equality holds for
a1 = a2 = · · · = an = 1.
1 1
Remark. For k = and m = , we get the preceding P 2.28.
3 2

P 2.30. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that

a1 ≥ a2 ≥ a3 ≥ 1 ≥ a4 ≥ · · · ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + ··· + ≥ .
3a1 + 1 3a2 + 1 3an + 1 4
(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x1 ≥ x2 ≥ x3 ≥ 0 ≥ x4 ≥ · · · ≥ x n, s= = 0,
n
1
f (u) = , u ∈ R.
3eu+1
For u ∈ [0, ∞), we have

3eu (3eu − 1)
f 00 (u) = > 0,
(3eu + 1)3
200 Vasile Cîrtoaje

hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = n−3),
it suffices to show that f (x) + 3 f ( y) ≥ 4 f (0) for all real x, y so that x + 3 y = 0;
that is, to show that
1 3
+ ≥1
3a + 1 3b + 1
for all a, b > 0 so that ab3 = 1. The inequality is equivalent to

b3 3
+ ≥ 1,
b + 3 3b + 1
3

(b − 1)2 (b + 2) ≥ 0.
The equality holds for a1 = a2 = · · · = an = 1.

P 2.31. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers so that

a1 ≥ a2 ≥ a3 ≥ 1 ≥ a4 ≥ · · · ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + ··· + ≥ .
(a1 + 1)2 (a2 + 1)2 (an + 1)2 4
(Vasile C., 2007)
Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x1 ≥ x2 ≥ x3 ≥ 0 ≥ x4 ≥ · · · ≥ x n, s= = 0,
n
1
f (u) = , u ∈ R.
(eu + 1)2
For u ∈ [0, ∞), we have
2eu (2eu − 1)
f 00 (u) = > 0,
(eu + 1)4
hence f is convex on [s, ∞). According to the RHCF-OV Theorem (case m = 3), it
suffices to show that f (x) + 3 f ( y) ≥ 4 f (0) for all real x, y so that x + 3 y = 0; that
is, to show that
1 3
+ ≥1
(a + 1)2 (b + 1)2
HCF Method for Ordered Variables 201

for all a, b > 0 so that ab3 = 1. The inequality is equivalent to

b6 3
+ ≥ 1.
(b + 1)
3 2 (b + 1)2
Using the Cauchy-Schwarz inequality, it suffices to show that
(b3 + 3)2
≥ 1,
(b3 + 1)2 + 3(b + 1)2
which is equivalent to the obvious inequality

(b − 1)2 (4b + 5) ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1.

P 2.32. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+ + · · · + ≤ .
(a1 + 3)2 (a2 + 3)2 (an + 3)2 16
(Vasile C., 2007)
Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
−1
f (u) = , u ∈ R.
(eu + 3)2
For u ∈ (−∞, 0], we have
2eu (3 − 2eu )
f 00 (u) = > 0,
(eu + 3)4
hence f is convex on (−∞, s]. According to the LHCF-OV Theorem (case m =
n−1), it suffices to show that f (x)+ f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0;
that is, to show that
1 1 1
+ ≤
(a + 3)2 (b + 3)2 8
202 Vasile Cîrtoaje

for all a, b > 0 so that ab = 1. Write this inequality as

b2 1 1
+ ≤ ,
(3b + 1)2 (b + 3)2 8

which is equivalent to the obvious inequality

(b2 − 1)2 + 12b(b − 1)2 ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1.

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,
p
If k ≥ 1 + 2, then

1 1 1 n
+ + · · · + ≤ ,
(a1 + k)2 (a2 + k)2 (an + k)2 (1 + k)2

with equality for a1 = a2 = · · · = an = 1.

P 2.33. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that p + q ≤ 1, then

1 1 1 n
+ + ··· + ≤ .
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2 1+p+q

(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
HCF Method for Ordered Variables 203

−1
f (u) = , u ∈ R.
1 + peu + qe2u
For u ≤ 0, we have

eu [−4q2 e3u − 3pqe2u + (4q − p2 )eu + p]


f 00 (u) =
(1 + peu + qe2u )3
e [−4q − 3pq + (4q − p2 ) + p]
2u 2

(1 + peu + qe2u )3
e2u [(p + 4q)(1 − p − q) + 2pq]
= ≥ 0,
(1 + peu + qe2u )3

therefore f (u) is convex for u ≤ s. According to the LHCF-OV Theorem (case m =


n−1), it suffices to show that f (x)+ f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0;
that is, to show that

1 1 2
+ ≤
1 + pa + qa 2 1 + pb + qb 2 1+p+q

for all a, b > 0 so that ab = 1. Write this inequality as

(a − 1)2 [q(1 − p − q)a2 + (p + 2q − p2 − pq − 2q2 )a + q(1 − p − q)] ≥ 0,

which is true because

p + 2q − p2 − pq − 2q2 ≥ (p + 2q)(p + q) − p2 − pq − 2q2 = 2pq ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1.

P 2.34. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.

1
If m > 1 and k ≥ , then
21/m −1
1 1 1 n
+ + · · · + ≤ .
(a1 + k)m (a2 + k)m (an + k)m (1 + k)m

(Vasile C., 2007)

Solution. By Bernoulli’s inequality, we have

1
21/m < 1 + ,
m
204 Vasile Cîrtoaje

hence
1
k≥ > m > 1.
21/m − 1
Using the substitution
ai = e x i , i = 1, 2, . . . , n,
we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
−1
f (u) = , u ∈ R.
(eu + k)m
For u ≤ 0, we have
meu (k − meu )
f 00 (u) = ≥ 0,
(eu + k)m+2
hence f is convex u ≤ s. By the LHCF-OV Theorem (case m = n − 1), it suffices to
show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that is, to show
that
1 1 2
+ ≤
(a + k) m (b + k) m (1 + k)m
for all a, b > 0 so that ab = 1. Write this inequality as g(a) ≤ 0 for a ≥ 1, where

1 am 2
g(a) = + − .
(a + k)m (ka + 1)m (1 + k)m

The derivative
g 0 (a) a m−1 (a + k)m+1 − (ka + 1)m+1
=
m (a + k)m+1 (ka + 1)m+1
has the same sign as the function

h(a) = (m − 1) ln a + (m + 1) ln(a + k) − (m + 1) ln(ka + 1).

We have
m−1 1 kh1 (a)
 ‹
k
h (a) =
0
+ (m + 1) − = ,
a a + k ka + 1 a(a + k)(ka + 1)

where  m
h1 (a) = (m − 1)(a2 + 1) − 2 k − a.
k
The discriminant D of the quadratic function h1 (a) is

m 2 m2
 ‹
D 
= k− − (m − 1) = (k − 1) 1 − 2 .
2 2
4 k k
HCF Method for Ordered Variables 205

Since D > 0, the roots a1 and a2 of h1 (a) are real and unequal. If a1 < a2 , then
h1 (a) ≤ 0 for a ∈ [a1 , a2 ] and h1 (a) ≥ 0 for a ∈ (−∞, a1 ] ∪ [a2 , ∞). Since

2(k + 1)(m − k)
h1 (1) = < 0,
k
it follows that a1 < 1 < a2 , therefore h1 (a) and h0 (a) are negative for a ∈ [1, a2 )
and positive for a ∈ (a2 , ∞), h(a) is decreasing for a ∈ [1, a2 ] and increasing for
a ∈ [a2 , ∞). From h(1) = 0 and

lim h(a) = ∞,
a→∞

it follows that there is a3 > a2 so that h(a) and g 0 (a) are negative for a ∈ (1, a3 )
and positive for a ∈ (a3 , ∞). As a result, g is decreasing on [1, a3 ] and increasing
on [a3 , ∞). Since g(1) = 0 and

1 2
lim g(a) = − ≤ 0,
a→∞ k m (1 + k)m

it follows that g(a) ≤ 0 for a ≥ 1. This completes the proof. The equality holds for
a1 = a2 = · · · = an = 1.

P 2.35. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1,

then
1 1 1 n
+p + ··· + p ≤p .
1 + 2a1 1 + 2a2 1 + 2an
p
3
(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
−1
f (u) = p , u ∈ R.
1 + 2eu
206 Vasile Cîrtoaje

For u ≤ 0, we have
eu (1 − eu )
f 00 (u) = > 0,
(1 + 2eu )5/2
hence f is convex on (−∞, s]. According to the LHCF-OV Theorem (case m =
n−1), it suffices to show that f (x)+ f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0;
that is, to show that v v
t 3 t 3
+ ≤2
1 + 2a 1 + 2b
for all a, b > 0 so that ab = 1. By the Cauchy-Schwarz inequality, we get
v v v
3 t 3 3 3
t t ‹ ‹
+ ≤ +1 1+ = 2.
1 + 2a 1 + 2b 1 + 2a 1 + 2b

The equality holds for a1 = a2 = · · · = an = 1.

P 2.36. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a1 a2 · · · an = 1.

If 0 < m < 1 and k ≥ m, then

1 1 1 n
+ + · · · + ≤ .
(a1 + k)m (a2 + k)m (an + k)m (1 + k)m

(Vasile C., 2007)

Solution. Using the substitution

ai = e x i , i = 1, 2, . . . , n,

we can write the inequality as

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−1 ≥ 0 ≥ x n , s= = 0,
n
−1
f (u) = , u ∈ R.
(eu + k)m
For u ≤ 0, we have
meu (k − meu )
f 00 (u) = ≥ 0,
(eu + k)m+2
HCF Method for Ordered Variables 207

hence f is convex on (−∞, s]. According to the LHCF-OV Theorem (case m =


n−1), it suffices to show that f (x)+ f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0;
that is, to show that
1 1 2
+ ≤
(a + k) m (b + k) m (1 + k)m
for all a, b > 0 so that ab = 1. Write this inequality as g(a) ≤ 0 for a ≥ 1, where
1 am 2
g(a) = + − ,
(a + k) m (ka + 1) m (1 + k)m
with
g 0 (a) a m−1 (a + k)m+1 − (ka + 1)m+1
= .
m (a + k)m+1 (ka + 1)m+1
If g 0 (a) ≤ 0 for a ≥ 1, then g is decreasing, hence g(a) ≤ g(1) = 0. Thus, it suffices
to show that
ka + 1 m+1
 ‹
m−1
a ≤ .
a+k
Since
ka + 1 ma + 1 (k − m)(a2 − 1)
− = ≥ 0,
a+k a+m (a + k)(a + m)
we only need to show that
ma + 1 m+1
 ‹
m−1
a ≤ ,
a+m
which is equivalent to h(a) ≤ 0 for a ≥ 1, where
h(a) = (m − 1) ln a + (m + 1) ln(a + m) − (m + 1) ln(ma + 1),
m − 1 m + 1 m(m + 1) m(m − 1)(a − 1)2
h (a) =
0
+ − = .
a a+m ma + 1 a(a + m)(ma + 1)
Since h0 (a) ≤ 0, h(a) is decreasing for a ≥ 1, hence
h(a) ≤ h(1) = 0.
This completes the proof. The equality holds for a1 = a2 = · · · = an = 1.
1 1
Remark. For k = and m = , we get the preceding P 2.35.
2 2

P 2.37. If a1 , a2 , . . . , an (n ≥ 3)are positive real numbers so that


a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , a1 a2 · · · an = 1,
then
1 1 1 n
+ + · · · + ≤ .
(a1 + 5)2 (a2 + 5)2 (an + 5)2 36
(Vasile C., 2007)
208 Vasile Cîrtoaje

Solution. Using the substitution


ai = e x i , i = 1, 2, . . . , n,
we can write the inequality as
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),
where
x1 + x2 + · · · + x n
x 1 ≥ · · · ≥ x n−2 ≥ 0 ≥ x n−1 ≥ x n , s= = 0,
n
−1
f (u) = , u ∈ R.
(eu + 5)2
For u ∈ (−∞, 0], we have
2eu (5 − 2eu )
f 00 (u) = > 0,
(eu + 5)4
hence f is convex on (−∞, s]. According to the LHCF-OV Theorem (case m = n −
2), it suffices to show that f (x)+2 f ( y) ≥ 3 f (0) for all real x, y so that x +2 y = 0;
that is, to show that
1 2 1
+ ≤
(a + 5)2 (b + 5) 2 12
for all a, b > 0 so that ab2 = 1. Since
1 b4 b4 b2
= ≤ = ,
(a + 5)2 (5b2 + 1)2 (4b2 + 2b)2 4(2b + 1)2
it suffices to show that
b2 2 1
+ ≤ ,
4(2b + 1)2 (b + 5)2 12
which is equivalent to the obvious inequality
(b − 1)2 (b2 + 16b + 1) ≥ 0.
The equality holds for a1 = a2 = · · · = an = 1.

Remark. Similarly, we can prove the following refinement:


• Let a1 , a2 , . . . , an be positive real numbers so that
a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , a1 a2 · · · an = 1.
p
If k ≥ 2 + 6, then
1 1 1 n
+ + ··· + ≤ ,
(a1 + k)2 (a2 + k)2 (an + k)2 (1 + k)2
with equality for a1 = a2 = · · · = an = 1.
HCF Method for Ordered Variables 209

P 2.38. If a1 , a2 , . . . , an are nonnegative real numbers so that

a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , a12 + a22 + · · · + an2 = n,

then
1 1 1 n
+ + ··· + ≤ .
3 − a1 3 − a2 3 − an 2
(Vasile C., 2007)

Solution. From

n = a12 + (a22 + · · · + an−1


2
) + an2 ≥ a12 + (n − 2) + 0,

we get p
a1 ≤ 2.
p p p
Replacing a1 , a2 , . . . , an by a1 , a2 , . . . , an , we have to prove that

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s),

where
a1 + a2 + · · · + an
2 ≥ a1 ≥ · · · ≥ an−1 ≥ 1 ≥ an , s= = 1,
n
1
f (u) = p , u ∈ [0, 2].
u−3
For u ∈ [0, 1], we have
p
3(1 − u)
f (u) = p
00
p ≥ 0.
4u u(3 − u)3

Therefore, f is convex on [0, s]. According to the LHCF-OV Theorem and Note 1
(case m = n − 1), it suffices to show that h(x, y) ≥ 0 for x, y ≥ 0 so that x + y = 2.
Since
f (u) − f (1) −1
g(u) = = p p
u−1 2(3 − u)(1 + u)
and
p p
g(x) − g( y) 2− x − y
h(x, y) = = p p p p p p ,
x−y 2( x + y)(1 + x)(1 + y)(3 − x)(3 − y)

we need to show that p p


x+ y ≤ 2.
Indeed, we have
p p Æ
x+ y≤ 2(x + y) = 2.
This completes the proof. The equality holds for a1 = a2 = · · · = an = 1.
210 Vasile Cîrtoaje

P 2.39. Let a1 , a2 , . . . , an be nonnegative real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 + a2 + · · · + an = n.

Prove that
n − a1 3  n − a2 3  n − a 3
• ˜
n
a13 + a23 + ··· + an3 − n ≥ (n − 1) 2
+ + ··· + −n .
n−1 n−1 n−1

(Vasile C., 2010)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where  n − u 3
f (u) = u3 − (n − 1)2 , u ≥ 0.
n−1
For u ≥ 1, we have
6n(u − 1)
f 00 (u) =≥ 0.
n−1
Therefore, f (u) is convex for u ≥ s. Thus, by the RHCF-OV Theorem (case m =
n − 1), it suffices to show that f (x) + f ( y) ≥ 2 f (1) for x, y ≥ 0 so that x + y = 2.
We have
n − x 3  n − y 3
• ˜
f (x) + f ( y) − 2 f (1) = x + y − 2 − (n − 1)
3 3 2
+ −2
n−1 n−1
(n − x)(n − y)
• ˜
= 6(1 − x y) − 6(n − 1) 1 −
2
= 0.
(n − 1)2

This completes the proof. The equality holds for

a1 ≤ 1, a2 = · · · = an−1 = 1, a n = 2 − a1 .
Chapter 3

Partially Convex Function Method

3.1 Theoretical Basis


The following statement is known as the Right Partially Convex Function Theorem
(RPCF-Theorem).
Right Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval I and convex on [s, s0 ], where s, s0 ∈ I, s < s0 . In
addition, f is decreasing on I≤s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x ≤ s ≤ y and x + (n − 1) y = ns.

Proof. For
a1 = x, a2 = a3 = · · · = an = y,
the inequality
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ f (s)
becomes
f (x) + (n − 1) f ( y) ≥ n f (s);
therefore, the necessity is obvious.
The proof of sufficiency is based on Lemma below. According to this lemma, it
suffices to consider that a1 , a2 , . . . , an ∈ J, where

J = I≤s0 .

211
212 Vasile Cîrtoaje

Because f (u) is convex on J≥s , the desired inequality follows from the RHCF The-
orem (see Chapter 1) applied to the interval J.
Lemma. Let f be a real function defined on an interval I. In addition, f is decreasing
on I≤s0 , and f (u) ≥ f (s0 ) for u ∈ I, where s, s0 ∈ I, s < s0 . If the inequality

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s)

holds for all a1 , a2 , . . . , an ∈ I≤s0 so that a1 + a2 + · · · + an = ns, then it holds for all
a1 , a2 , . . . , an ∈ I so that a1 + a2 + · · · + an = ns.
Proof. For i = 1, 2, . . . , n, define the numbers
¨
ai , ai ≤ s0
bi =
s0 , ai > s0 .

Clearly, bi ∈ I≤s0 and bi ≤ ai . Since f (u) ≥ f (s0 ) for u ∈ I≥s0 , it follows that
f (bi ) ≤ f (ai ) for i = 1, 2, . . . , n. Therefore,

b1 + b2 + · · · + bn ≤ a1 + a2 + · · · + an = ns

and
f (b1 ) + f (b2 ) + · · · + f (bn ) ≤ f (a1 ) + f (a2 ) + · · · + f (an ).
Thus, it suffices to show that

f (b1 ) + f (b2 ) + · · · + f (bn ) ≥ n f (s)

for all b1 , b2 , . . . , bn ∈ I≤s0 so that b1 + b2 + · · · + bn ≤ ns. By hypothesis, this


inequality is true for b1 , b2 , . . . , bn ∈ I≤s0 and b1 + b2 + · · · + bn = ns. Since f (u)
is decreasing on I≤s0 , the more we have f (b1 ) + f (b2 ) + · · · + f (bn ) ≥ n f (s) for
b1 , b2 , . . . , bn ∈ I≤s0 and b1 + b2 + · · · + bn ≤ ns.
Similarly, we can prove the Left Partially Convex Function Theorem (LPCF-Theorem).
Left Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval I and convex on [s0 , s], where s0 , s ∈ I, s0 < s. In
addition, f is increasing on I≥s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x ≥ s ≥ y and x + (n − 1) y = ns.
Partially Convex Function Method 213

From the RPCF-Theorem and the LPCF-Theorem, we find the PCF-Theorem (Par-
tially Convex Function Theorem).
Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real function
defined on an interval I and convex on [s0 , s] or [s, s0 ], where s0 , s ∈ I. In addition, f
is decreasing on I≤s0 and increasing on I≥s0 . The inequality
a + a + ··· + a 
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
holds for all a1 , a2 , . . . , an ∈ I satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x + (n − 1) y = ns.

Note 1. Let us denote


f (u) − f (s) g(x) − g( y)
g(u) = , h(x, y) = .
u−s x−y

As shown in Note 1 from Chapter 1, we may replace the hypothesis condition in


the RPCF-Theorem and the LPCF-Theorem), namely

f (x) + (n − 1) f ( y) ≥ n f (s),

by the condition
h(x, y) ≥ 0 for all x, y ∈ I so that x + (n − 1) y = ns.

Note 2. Assume that f is differentiable on I, and let

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y

As shown in Note 2 from Chapter 1, the inequalities in the RPCF-Theorem and the
LPCF-Theorem hold true by replacing the hypothesis

f (x) + (n − 1) f ( y) ≥ n f (s)

with the more restrictive condition


H(x, y) ≥ 0 for all x, y ∈ I so that x + (n − 1) y = ns.

Note 3. The desired inequalities in the RPCF-Theorem and the LPCF-Theorem be-
come equalities for
a1 = a2 = · · · = an = s.
214 Vasile Cîrtoaje

In addition, if there exist x, y ∈ I so that

x + (n − 1) y = ns, f (x) + (n − 1) f ( y) = n f (s), x 6= y,

then the equality holds also for

a1 = x, a2 = · · · = an = y

(or any cyclic permutation). Notice that these equality conditions are equivalent to

x + (n − 1) y = ns, h(x, y) = 0

(x < y for the RPCF-Theorem, and x > y for the LPCF-Theorem).

Note 4. From the proof of the RPCF-Theorem, it follows that this theorem is also
valid in the case when f is defined on I \ {u0 }, where u0 ∈ I>s0 . Similarly, the LPCF-
Theorem is also valid in the case when f is defined on I \ {u0 }, where u0 ∈ I<s0 .

Note 5. The RPCF-Theorem holds true by replacing the condition


f is decreasing on I≤s0
with
ns − (n − 1)s0 ≤ inf I.
More precisely, the following theorem holds:
Theorem 1. Let f be a function defined on a real interval I, convex on [s, s0 ] and
satisfying
min f (u) = f (s0 ),
u∈I≥s

where
s, s0 ∈ I, s < s0 , ns − (n − 1)s0 ≤ inf I.
If
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x ≤ s ≤ y and x + (n − 1) y = ns, then
x + x + ··· + x 
1 2 n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f
n
for all x 1 , x 2 , . . . , x n ∈ I satisfying x 1 + x 2 + · · · + x n = ns .
In order to prove Theorem 1, we define the function

f (u), u ≤ s0 , u ∈ I
¨
f0 (u) =
f (s0 ), u ≥ s0 , u ∈ I,

which is convex on I≥s . Taking into account that f0 (s) = f (s) and f0 (u) ≤ f (u) for
all u ∈ I, it suffices to prove that

f0 (x 1 ) + f0 (x 2 ) + · · · + f0 (x n ) ≥ n f0 (s)
Partially Convex Function Method 215

for all x 1 , x 2 , . . . , x n ∈ I satisfying x 1 + x 2 + · · · + x n = ns. According to the HCF-


Theorem and Note 5 from Chapter 1, we only need to show that

f0 (x) + (n − 1) f0 ( y) ≥ n f0 (s)

for all x, y ∈ I so that x ≤ s ≤ y and x + (n − 1) y = ns. Since

ns − x ns − (n − 1)s0 − x ns − (n − 1)s0 − inf I


y − s0 = − s0 = ≤ ≤ 0,
n−1 n−1 n−1
the inequality f0 (x) + (n − 1) f0 ( y) ≥ n f0 (s) turns into f (x) + (n − 1) f ( y) ≥ n f (s),
which holds (by hypothesis) for all x, y ∈ I so that x ≤ s ≤ y and x +(n−1) y = ns.

Similarly, the LPCF-Theorem holds true by replacing the condition


f is increasing on I≥s0
with

ns − (n − 1)s0 ≥ sup I.
More precisely, the following theorem holds:
Theorem 2. Let f be a function defined on a real interval I, convex on [s0 , s] and
satisfying
min f (u) = f (s0 ),
u∈I≤s

where
s, s0 ∈ I, s > s0 , ns − (n − 1)s0 ≥ sup I.
If
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I so that x ≥ s ≥ y and x + (n − 1) y = ns, then
x + x + ··· + x 
1 2 n
f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f
n
for all x 1 , x 2 , . . . , x n ∈ I satisfying x 1 + x 2 + · · · + x n = ns.
The proof of Theorem 2 is similar to the proof of Theorem 1.

Note 6. From the proof of Theorem 1, it follows that Theorem 1 is also valid in
the case in which f is defined on I \ {u0 }, where u0 is an interior point of I so that
u0 ∈/ [s, s0 ]. Similarly, Theorem 2 is also valid in the case in which f is defined on
/ [s0 , s].
I \ {u0 }, where u0 is an interior point of I so that u0 ∈

Note 7. In the same manner, we can extend weighted Jensen’s inequality to right
and left partially convex functions establishing the WRPCF-Theorem, the WLPCF-
Theorem and the WPCF-Theorem (Vasile Cîrtoaje, 2014).
216 Vasile Cîrtoaje

WRPCF-Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1, p = min{p1 , p2 , . . . , pn },

and let f be a real function defined on an interval I and convex on [s, s0 ], where
s, s0 ∈ I, s < s0 . In addition, f is decreasing on I≤s0 and f (u) ≥ f (s0 ) for u ∈ I. The
inequality

p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) ≥ f (p1 a1 + p2 a2 + · · · + pn an )

holds for all a1 , a2 , . . . , an ∈ I satisfying

p1 a1 + p2 a2 + · · · + pn an = s,

if and only if
p f (x) + (1 − p) f ( y) ≥ f (s)
for all x, y ∈ I so that x ≤ s ≤ y and px + (1 − p) y = s.

WLPCF-Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1, p = min{p1 , p2 , . . . , pn },

and let f be a real function defined on an interval I and convex on [s0 , s], where
s0 , s ∈ I, s0 < s. In addition, f is increasing on I≥s0 and f (u) ≥ f (s0 ) for u ∈ I. The
inequality

p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) ≥ f (p1 a1 + p2 a2 + · · · + pn an )

holds for all a1 , a2 , . . . , an ∈ I satisfying

p1 a1 + p2 a2 + · · · + pn an = s,

if and only if
p f (x) + (1 − p) f ( y) ≥ f (s)
for all x, y ∈ I so that x ≥ s ≥ y and px + (1 − p) y = s.
Partially Convex Function Method 217

3.2 Applications

3.1. If a, b, c are real numbers so that a + b + c = 3, then

16a − 5 16b − 5 16c − 5


+ + ≤ 1.
32a + 1 32b + 1 32c 2 + 1
2 2

3.2. If a, b, c, d are real numbers so that a + b + c + d = 4, then

18a − 5 18b − 5 18c − 5 18d − 5


+ + + ≤ 4.
12a + 1 12b + 1 12c + 1 12d 2 + 1
2 2 2

3.3. If a, b, c, d, e, f are real numbers so that a + b + c + d + e + f = 6, then

5a − 1 5b − 1 5c − 1 5d − 1 5e − 1 5f − 1
+ + + + + ≤ 4.
5a2 + 1 5b2 + 1 5c 2 + 1 5d 2 + 1 5e2 + 1 5 f 2 + 1

3.4. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that a1 + a2 + · · · + an = n, then

n(n + 1) − 2a1 n(n + 1) − 2a2 n(n + 1) − 2an


+ + · · · + ≤ n.
n2 + (n − 2)a12 n2 + (n − 2)a22 n2 + (n − 2)an2

3.5. If a, b, c, d are real numbers so that a + b + c + d = 4, then

a(a − 1) b(b − 1) c(c − 1) d(d − 1)


+ + 2 + ≥ 0.
3a2 + 4 3b2 + 4 3c + 4 3d 2 + 4

3.6. If a, b, c are real numbers so that a + b + c = 3, then

1 1 1 3
+ 2 + 2 ≤ .
9a2 − 10a + 9 9b − 10b + 9 9c − 10c + 9 8

3.7. If a, b, c, d are real numbers so that a + b + c + d = 4, then

1 1 1 1 4
+ 2 + 2 + 2 ≤ .
4a2 − 5a + 4 4b − 5b + 4 4c − 5c + 4 4d − 5d + 4 3
218 Vasile Cîrtoaje

3.8. Let a1 , a2 , . . . , an 6= −k be real numbers so that a1 + a2 + · · · + an = n, where


n
k≥ p .
2 n−1
Then,
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + ··· + ≥ 0.
(a1 + k) 2 (a2 + k)2 (an + k)2

3.9. Let a1 , a2 , . . . , an 6= −k be real numbers so that a1 + a2 + · · · + an = n. If


n
k ≥1+ p ,
n−1
then
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (an + k)2

3.10. Let a1 , a2 , a3 , a4 , a5 be real numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5. If

1 25
• ˜
k∈ , ,
6 14
then X 1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4

3.11. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5.


If k ∈ [k1 , k2 ], where
p
29 − 761 25
k1 = ≈ 0.1414, k2 = ≈ 1.7857,
10 14
then X 1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4

3.12. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If k > 1, then
1 1 1
+ + ··· + ≤ 1.
a1k + a2 + · · · + an a1 + a2k + · · · + an a1 + a2 + · · · + ank
Partially Convex Function Method 219

3.13. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5.


If
4 61
• ˜
k∈ , ,
9 5
then X a1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4

3.14. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If k > 1, then
a1 a2 an
+ + ··· + ≤ 1.
a1k + a2 + · · · + an a1 + a2k + · · · + an a1 + a2 + · · · + ank

3.15. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


1
If k ≥ 1 − , then
n
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
ka12 + a2 + · · · + an a1 + ka2 + · · · + an
2
a1 + a2 + · · · + kan2

3.16. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


1
If k ≥ 1 − , then
n
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 − a1 + ka1 1 − a2 + ka2
2 2
1 − an + kan2

3.17. Let a1 , a2 , . . . , an be positive real numbers so that a1 + a2 + · · · + an = n. If


n
0<k≤ , then
n−1
k/a k/a
a1 1 + a2 2 + · · · + ank/an ≤ n.

3.18. If a, b, c, d, e are nonzero real numbers so that a + b + c + d + e = 5, then

5 2 5 2 5 2 5 2 5 2
 ‹  ‹  ‹  ‹  ‹
7− + 7− + 7− + 7− + 7− ≥ 20.
a b c d e
220 Vasile Cîrtoaje

3.19. If If a1 , a2 , . . . , a7 are real numbers so that a1 + a2 + · · · + a7 = 7, then

(a12 + 2)(a22 + 2) · · · (a72 + 2) ≥ 37 .

n2
3.20. Let a1 , a2 , . . . , an be real numbers so that a1 +a2 +· · ·+an = n. If k ≥ ,
4(n − 1)
then
(a12 + k)(a22 + k) · · · (an2 + k) ≥ (1 + k)n .

3.21. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 10,


then
(a12 − a1 + 1)(a22 − a2 + 1) · · · (an2 − an + 1) ≥ 1.

3.22. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 26,


then
(a12 − a1 + 2)(a22 − a2 + 2) · · · (an2 − an + 2) ≥ 2n .

3.23. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

(1 − a + a4 )(1 − b + b4 )(1 − c + c 4 ) ≥ 1.

3.24. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(1 − a + a3 )(1 − b + b3 )(1 − c + c 3 )(1 − d + d 3 ) ≥ 1.

3.25. If a, b, c, d, e are nonzero real numbers so that a + b + c + d + e = 5, then

1 1 1 1 1 1 1 1 1 1
 ‹  ‹
5 2 + 2 + 2 + 2 + 2 + 45 ≥ 14 + + + + .
a b c d e a b c d e

3.26. If a, b, c are positive real numbers so that abc = 1, then

7 − 6a 7 − 6b 7 − 6c
+ + ≥ 1.
2 + a2 2 + b2 2 + c 2
Partially Convex Function Method 221

3.27. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 1
+ + ≤ .
a + 5bc b + 5ca c + 5ab 2

3.28. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 3
+ + ≤ .
4 − 3a + 4a 2 4 − 3b + 4b 2 4 − 3c + 4c 2 5

3.29. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 3
+ + ≤ .
(3a + 1)(3a2 − 5a + 3) (3b + 1)(3b2 − 5b + 3) (3c + 1)(3c 2 − 5c + 3) 4

3.30. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If


p, q ≥ 0 so that p + 4q ≥ n − 1, then

1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

3.31. If a, b, c are positive real numbers so that abc = 1, then

1−a 1− b 1−c
+ + ≥ 0.
17 + 4a + 6a 2 17 + 4b + 6b 2 17 + 4c + 6c 2

3.32. If a1 , a2 , . . . , a8 are positive real numbers so that a1 a2 · · · a8 = 1, then

1 − a1 1 − a2 1 − a8
+ + · · · + ≥ 0.
(1 + a1 )2 (1 + a2 )2 (1 + a8 )2

−13 13
• ˜
3.33. Let a, b, c be positive real numbers so that abc = 1. If k ∈ p , p ,
3 3 3 3
then
a+k b+k c+k 3(1 + k)
+ + ≤ .
a2 + 1 b2 + 1 c 2 + 1 2
222 Vasile Cîrtoaje

p
3.34. If a, b, c are positive real numbers and 0 < k ≤ 2 + 2 2, then

a3 b3 c3 a+b+c
+ + ≥ .
ka + bc kb + ca kc + ab
2 2 2 k+1

3.35. If a, b, c, d, e are positive real numbers so that abcde = 1, then


1 1 1 1 1 1
 ‹  ‹
2 + + ··· + ≥3 + + ··· + .
a+1 b+1 e+1 a+2 b+2 e+2

3.36. If a1 , a2 , . . . , a14 are positive real numbers so that a1 a2 · · · a14 = 1, then


1 1 1 1 1 1
 ‹  ‹
3 + + ··· + ≥2 + + ··· + .
2a1 + 1 2a2 + 1 2a14 + 1 a1 + 1 a2 + 1 a14 + 1

3.37. Let a1 , a2 , . . . , a8 be positive real numbers so that a1 a2 · · · a8 = 1. If k > 1,


then
1 1 1 1 1 1
 ‹  ‹
(k + 1) + + ··· + ≥2 + + ··· + .
ka1 + 1 ka2 + 1 ka8 + 1 a1 + 1 a2 + 1 a8 + 1

3.38. If a1 , a2 , . . . , a9 are positive real numbers so that a1 a2 · · · a9 = 1, then


1 1 1 1 1 1
+ + ··· + ≥ + + ··· + .
2a1 + 1 2a2 + 1 2a9 + 1 a1 + 2 a2 + 2 a9 + 2

3.39. If a1 , a2 , . . . , an are real numbers so that

a1 , a2 , . . . , an ≤ π, a1 + a2 + · · · + an = π,

then
π
cos a1 + cos a2 + · · · + cos an ≤ n cos .
n

3.40. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that


−1
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
a12 a22 an2
+ + ··· + ≤ n.
a12 − a1 + 1 a22 − a2 + 1 an2 − an + 1
Partially Convex Function Method 223

3.41. If a1 , a2 , . . . , an (n ≥ 3) are nonzero real numbers so that


−n
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
1 1 1 1 1 1
2
+ 2 + ··· + 2 ≥ + + ··· + .
a1 a2 an a1 a2 an

3.42. If a1 , a2 , . . . , an ≥ −1 so that a1 + a2 + · · · + an = n, then


 
1 1 1 1 1 1
 ‹
(n + 1) 2 + 2 + · · · + 2 ≥ 2n + (n − 1) + + ··· + .
a1 a2 an a1 a2 an

3.43. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that

−(3n − 2)
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
1 − a1 1 − a2 1 − an
+ + · · · + ≥ 0.
(1 + a1 )2 (1 + a2 )2 (1 + an )2

3.44. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


2
If n ≥ 3 and k ≥ 2 − , then
n
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
(1 − ka1 )2 (1 − ka2 )2 (1 − kan )2
224 Vasile Cîrtoaje
Partially Convex Function Method 225

3.3 Solutions

P 3.1. If a, b, c are real numbers so that a + b + c = 3, then

16a − 5 16b − 5 16c − 5


+ + ≤ 1.
32a + 1 32b + 1 32c 2 + 1
2 2

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
5 − 16u
f (u) = , u ∈ R.
32u2 + 1
From
16(32u2 − 20u − 1)
f 0 (u) = ,
(32u2 + 1)2
it follows that f is increasing on
 p 
5 − 33
−∞, ∪ [s0 , ∞)
16

and decreasing on  p 
5 − 33
, s0 ,
16
where p
5 + 33
s0 = ≈ 0.6715.
16
Also, from
lim f (u) = 0
u→−∞

and
f (s0 ) < 0,
it follows that f (u) ≥ f (s0 ) for u ∈ R. In addition, for u ∈ [s0 , 1], we have

1 00 −512u3 + 480u2 + 48u − 5


f (u) =
64 (32u2 + 1)3
512u2 (1 − u) + 32u(1 − u) + (16u − 5)
= > 0,
(32u2 + 1)3
226 Vasile Cîrtoaje

hence f is convex on [s0 , s]. According to the LPCF-Theorem, we only need to show
that f (x) + 2 f ( y) ≥ 3 f (1) for all real x, y so that x + 2 y = 3. Using Note 1, it
suffices to prove that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
Indeed, we have
32(2u − 1)
g(u) = ,
3(32u2 + 1)
64(1 + 16x + 16 y − 32x y) 64(4x − 5)2
h(x, y) = = ≥ 0.
3(32x 2 + 1)(32 y 2 + 1) 3(32x 2 + 1)(32 y 2 + 1)
Thus, the proof is completed. From x + 2 y = 3 and h(x, y) = 0, we get
5 7
x= , y= .
4 8
Therefore, in accordance with Note 3, the equality holds for a = b = c = 1, and
also for
5 7
a= , b=c=
4 8
(or any cyclic permutation).

P 3.2. If a, b, c, d are real numbers so that a + b + c + d = 4, then

18a − 5 18b − 5 18c − 5 18d − 5


+ + + ≤ 4.
12a + 1 12b + 1 12c + 1 12d 2 + 1
2 2 2

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
5 − 18u
f (u) = , u ∈ R.
12u2 + 1
From
6(36u2 − 20u − 3)
f 0 (u) = ,
(12u2 + 1)2
it follows that f is increasing on
 p 
5 − 52
−∞, ∪ [s0 , ∞)
18
Partially Convex Function Method 227

and decreasing on
p p
5 + 52
 
5 − 52
, s0 , s0 = ≈ 0.678.
18 18

Also, from
lim f (u) = 0
u→−∞

and
f (s0 ) < 0,
it follows that f (u) ≥ f (s0 ) for u ∈ R. In addition, for u ∈ [s0 , 1], we have

1 00 −216u3 + 180u2 + 54u − 5


f (u) =
24 (12u2 + 1)3
216u (1 − u) + 36u(1 − u) + (18u − 5)
2
= > 0,
(32u2 + 1)3

hence f is convex on [s0 , s]. According to the LPCF-Theorem and Note 1, we only
need to show that h(x, y) ≥ 0 for x, y ∈ R so that x + 3 y = 4. We have

f (u) − f (1) 6(2u − 1)


g(u) = = ,
u−1 12u2 + 1

g(x) − g( y) 12(1 + 6x + 6 y − 12x y) 12(2x − 3)2


h(x, y) = = = ≥ 0.
x−y (12x 2 + 1)(12 y 2 + 1) (12x 2 + 1)(12 y 2 + 1)
Thus, the proof is completed. From x + 3 y = 4 and h(x, y) = 0, we get x = 3/2
and y = 5/6. Therefore, in accordance with Note 3, the equality holds for a = b =
c = d = 1, and also for
3 5
a= , b=c=d=
2 6
(or any cyclic permutation).

P 3.3. If a, b, c, d, e, f are real numbers so that a + b + c + d + e + f = 6, then

5a − 1 5b − 1 5c − 1 5d − 1 5e − 1 5f − 1
+ + + + + ≤ 4.
5a2 + 1 5b2 + 1 5c 2 + 1 5d 2 + 1 5e2 + 1 5 f 2 + 1

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c+d+e+ f
f (a) + f (b) + f (c) + f (d) + f (e) + f ( f ) ≥ 4 f (s), s= = 1,
6
228 Vasile Cîrtoaje

where
1 − 5u
f (u) = , u ∈ R.
5u2 + 1
From
5(5u2 − 2u − 1)
f (u) =
0
,
(5u2 + 1)2
it follows that f is increasing on
 p 
1− 6
−∞, ∪ [s0 , ∞)
5

and decreasing on
p p
1+ 6
 
1− 6
, s0 , s0 = ≈ 0.69.
5 5

Also, from
lim f (u) = 0
u→−∞

and
f (s0 ) < 0,
it follows that f (u) ≥ f (s0 ) for u ∈ R. In addition, for u ∈ [s0 , 1], we have

1 00 −216u3 + 180u2 + 54u − 5


f (u) =
24 (12u2 + 1)3
216u (1 − u) + 36u(1 − u) + (18u − 5)
2
= > 0,
(32u2 + 1)3

hence f is convex on [s0 , s]. According to the LPCF-Theorem and Note 1, we only
need to show that h(x, y) ≥ 0 for x, y ∈ R so that x + 5 y = 6. We have

f (u) − f (1) 5(2u − 1)


g(u) = = ,
u−1 3(5u2 + 1)

g(x) − g( y) 5(2 + 5x + 5 y − 10x y) 10(x − 2)2


h(x, y) = = = ≥ 0.
x−y 3(5x 2 + 1)(5 y 2 + 1) 3(5x 2 + 1)(5 y 2 + 1)
In accordance with Note 3, the equality holds for a = b = c = d = e = f = 1, and
also for
4
a = 2, b=c=d=e= f =
5
(or any cyclic permutation).
Partially Convex Function Method 229

P 3.4. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that a1 + a2 + · · · + an = n, then


n(n + 1) − 2a1 n(n + 1) − 2a2 n(n + 1) − 2an
+ + · · · + ≤ n.
n2 + (n − 2)a1
2
n2 + (n − 2)a2
2
n2 + (n − 2)an2

(Vasile C., 2008)


n(n + 1)
Solution. The desired inequality is true for a1 > since
2
n(n + 1) − 2a1
<0
n2 + (n − 2)a12
and
n(n + 1) − 2ai n
< , i = 2, 3, . . . , n.
n2 + (n − 2)ai2 n−1
The last inequalities are equivalent to

n(n − 2)ai2 + 2(n − 1)ai + n > 0,

which are true because

n(n − 2)ai2 + 2(n − 1)ai + n ≥ (n − 1)ai2 + 2(n − 1)ai + n > (n − 1)(ai + 1)2 ≥ 0.

Consider further that


n(n + 1)
a1 , a2 , . . . , a n ≤ ,
2
and rewrite the desired inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
2u − n(n + 1) n(n + 1)
 ˜
f (u) = , u ∈ I = −∞, .
(n − 2)u2 + n2 2
We have
f 0 (u) n2 + n(n + 1)u − u2
=
2(n − 2) [(n − 2)u2 + n2 ]2
and
f 00 (u) f1 (u)
= ,
2(n − 2) [(n − 2)u2 + n2 ]3
where

f1 (u) = 2(n − 2)u3 − 3n(n + 1)(n − 2)u2 − 2n2 (2n − 3)u + n3 (n + 1).

˜ of f , it follows that f is decreasing on (−∞, s0 ] and increasing


0
From
• the expression
n(n + 1)
on s0 , , where
2
n€ p Š
s0 = n + 1 − n2 + 2n + 5 ∈ (−1, 0);
2
230 Vasile Cîrtoaje

therefore,
min f (u) = f (s0 ).
u∈I

On the other hand, for −1 ≤ u ≤ 1, we have

f1 (u) > −2(n − 2) − 3n(n + 1)(n − 2) − 2n2 (2n − 3) + n3 (n + 1)


= n2 (n − 3)2 + 4(n + 1) > 0,

hence f 00 (u) > 0. Since [s0 , s] ⊂ [−1, 1], f is convex on [s0 , s]. By the LPCF-
Theorem and Note 1, we only need to show that h(x, y) ≥ 0 for x, y ∈ R and
x + (n − 1) y = n, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

Indeed, we have
(n − 2)u + n
g(u) =
(n − 2)u2 + n2
and
h(x, y) n2 − n(x + y) − (n − 2)x y
=
n−2 [(n − 2)x 2 + n2 ][(n − 2) y 2 + n2 ]
(n − 1)(n − 2) y 2
= ≥ 0.
[(n − 2)x 2 + n2 ][(n − 2) y 2 + n2 ]

The proof is completed. By Note 3, the equality holds for a1 = a2 = · · · = an = 1,


and also for
a1 = n, a2 = · · · = an = 0
(or any cyclic permutation).

P 3.5. If a, b, c, d are real numbers so that a + b + c + d = 4, then

a(a − 1) b(b − 1) c(c − 1) d(d − 1)


+ + 2 + ≥ 0.
3a2 + 4 3b2 + 4 3c + 4 3d 2 + 4

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
u2 − u
f (u) = , u ∈ R.
3u2 + 4
Partially Convex Function Method 231

From
3u2 + 8u − 4
f 0 (u) = ,
(3u2 + 4)2
 p 
−4 − 2 7
it follows that f is increasing on −∞, ∪ [s0 , ∞) and decreasing on
3
 p 
−4 − 2 7
, s0 , where
3
p
−4 + 2 7
s0 = ≈ 0.43.
3
Since
1
lim f (u) =
u→−∞ 3
and f (s0 ) < 0, it follows that

min f (u) = f (s0 ).


u∈R

For u ∈ [0, 1], we have


1 00 −9u3 − 36u2 + 36u + 14
f (u) =
2 (3u2 + 4)3
9u2 (1 − u) + 45u(1 − u) + (16 − 9u)
= > 0.
(3u2 + 4)3
Therefore, f is convex on [0, 1], hence on [s0 , s]. According to the LPCF-Theorem
and Note 1, we only need to show that h(x, y) ≥ 0 for x, y ∈ R so that x + 3 y = 4.
We have
f (u) − f (1) u
g(u) = = 2 ,
u−1 3u + 4

g(x) − g( y) 4 − 3x y
h(x, y) = =
x−y (3x + 4)(3 y 2 + 4)
2

(x − 2)2
= ≥ 0.
(3x 2 + 4)(3 y 2 + 4)
The proof is completed. From x + 3 y = 4 and h(x, y) = 0, we get x = 2 and
y = 2/3. By Note 3, the equality holds for a = b = c = d = 1, and also for
2
a = 2, b=c=d=
3
(or any cyclic permutation).
Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an are real numbers so that a1 + a2 + · · · + an = n, then
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + ··· + ≥ 0,
4(n − 1)a1 + n
2 2 4(n − 1)a2 + n
2 2 4(n − 1)an2 + n2
232 Vasile Cîrtoaje

with equality for a1 = a2 = · · · = an = 1, and also for


n n
a1 = , a2 = a3 = · · · = an =
2 2(n − 1)
(or any cyclic permutation).

P 3.6. If a, b, c are real numbers so that a + b + c = 3, then


1 1 1 3
+ + ≤ .
9a2 − 10a + 9 9b2 − 10b + 9 9c 2 − 10c + 9 8
(Vasile C., 2015)
Solution. Write the inequality as
a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
−1
f (u) = , u ∈ R.
9u2 − 10u + 9
From
2(9u − 5)
f 0 (u) = ,
(9u2 − 10u + 9)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞) and , where
5
s0 = < 1 = s.
9
For u ∈ [s0 , s] = [5/9, 1], we have
2(−243u2 + 270u − 19) 2(−243u2 + 270u − 27)
f 00 (u) = >
(9u2 − 10u + 9)3 (9u2 − 10u + 9)3
54(−9u2 + 10u − 1) 54(1 − u)(9u − 1)
= = ≥ 0.
(9u2 − 10u + 9)3 (9u2 − 10u + 9)3
Therefore, f is convex on [s0 , s]. According to the LPCF-Theorem and Note 1, we
only need to show that h(x, y) ≥ 0 for x, y ∈ R so that x + 2 y = 3. We have
f (u) − f (1) 9u − 1)
g(u) = = ,
u−1 8(9u2 − 10u + 9)

g(x) − g( y) 9(x + y) − 81x y + 71


h(x, y) = =
x−y 8(9x 2 − 10x + 9)(9 y 2 − 10 y + 9)
2(9 y − 7)2
= ≥ 0.
8(9x 2 − 10x + 9)(9 y 2 − 10 y + 9)
Partially Convex Function Method 233

The proof is completed. From x + 2 y = 3 and h(x, y) = 0, we get


13 7
x= , y= .
9 9
Thus, the equality holds for a = b = c = d = 1, and also for
13 7
a= , b=c=
9 9
(or any cyclic permutation).

P 3.7. If a, b, c, d are real numbers so that a + b + c + d = 4, then


1 1 1 1 4
+ 2 + 2 + 2 ≤ .
4a2 − 5a + 4 4b − 5b + 4 4c − 5c + 4 4d − 5d + 4 3

(Vasile C., 2015)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
−1
f (u) = , u ∈ R.
4u2 − 5u + 4
From
2(8u − 5)
f 0 (u) = ,
(4u2 − 5u + 4)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
5
s0 = < 1 = s.
8
For u ∈ [s0 , s] = [5/8, 1], we have

4(−48u2 + 60u − 9) 4(−48u2 + 60u − 12)


f (u) =
00
>
(4u2 − 5u + 4)3 (4u2 − 5u + 4)3
48(−4u2 + 5u − 1) 48(1 − u)(4u − 1)
= = ≥ 0.
(4u2 − 5u + 4)3 (4u2 − 5u + 4)3

Therefore, f is convex on [s0 , s]. According to the LPCF-Theorem and Note 1, we


only need to show that h(x, y) ≥ 0 for x, y ∈ R so that x + 3 y = 4. We have

f (u) − f (1) 4u − 1)
g(u) = = ,
u−1 3(4u2 − 5u + 4)
234 Vasile Cîrtoaje

g(x) − g( y) 4(x + y) − 16x y + 11


h(x, y) = =
x−y 3(4x 2 − 5x + 4)(4 y 2 − 5 y + 4)
(4 y − 3)2
= ≥ 0.
(4x 2 − 5x + 4)(4 y 2 − 5 y + 4)

From x + 3 y = 4 and h(x, y) = 0, we get

7 3
x= , y= .
4 4

In accord with Note 3, the equality holds for a = b = c = 1, and also for

7 3
a= , b=c=d=
4 4

(or any cyclic permutation).


Remark. Similarly, we can prove the following generalization:
• Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an = n. If

2(n − 1)
k =1− ,
n2

then
1 1 1 n
+ 2 + ··· + 2 ≥ ,
a12 − 2ka1 + 1 a2 − 2ka2 + 1 an − 2kan + 1 2(1 − k)
with equality for a1 = a2 = · · · = an = 1, and also for

3n2 − 6n + 4 n2 − 2n + 4
a1 = , a2 = a3 = · · · = an =
n2 n2

(or any cyclic permutation).

P 3.8. Let a1 , a2 , . . . , an 6= −k be real numbers so that a1 + a2 + · · · + an = n, where

n
k≥ p .
2 n−1

Then,
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + ··· + ≥ 0.
(a1 + k) 2 (a2 + k)2 (an + k)2

(Vasile C., 2008)


Partially Convex Function Method 235

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
u(u − 1)
f (u) = , u ∈ I = R \ {−k}.
(u + k)2
From
(2k + 1)u − k
f 0 (u) = ,
(u + k)3
it follows that f is increasing on (−∞, −k) ∪ [s0 , ∞) and decreasing on (−k, s0 ],
where
k
s0 = < 1 = s.
2k + 1
Since
lim f (u) = 1
u→−∞

and f (s0 ) < 0, we have


min f (u) = f (s0 ).
u∈I
From
1 00 k(k + 2) − (2k + 1)u
f (u) = ,
2 (u + k)4
k(k + 2)
• ˜
it follows that f is convex on 0, , hence on [s0 , 1]. According to the LPCF-
2k + 1
Theorem, Note 4 and Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ∈ I
which satisfy x + (n − 1) y = n, where
g(x) − g( y) f (u) − f (1)
h(x, y) = , g(u) = .
x−y u−1
Indeed, we have
u
g(u) =
(u + k)2
and
2
n
k2 − x y 4(n−1) − x y
h(x, y) = ≥
(x + k)2 ( y + k)2 (x + k)2 ( y + k)2
[2(n − 1) y − n] 2
= ≥ 0.
4(n − 1)(x + k)2 ( y + k)2
n
The equality holds for a1 = a2 = · · · = an = 1. If k = p , then the equality
2 n−1
holds also for
n n
a1 = , a2 = · · · = an =
2 2(n − 1)
(or any cyclic permutation).
236 Vasile Cîrtoaje

P 3.9. Let a1 , a2 , . . . , an 6= −k be real numbers so that a1 + a2 + · · · + an = n. If


n
k ≥1+ p ,
n−1
then
a12 − 1 a22 − 1 an2 − 1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (an + k)2
(Vasile C., 2008)

Solution. Write the inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
u2 − 1
f (u) = , u ∈ I = R \ {−k}.
(u + k)2
From
2(ku + 1)
f 0 (u) = ,
(u + k)3
it follows that f is increasing on (−∞, −k) ∪ [s0 , ∞) and decreasing on (−k, s0 ],
where
−1
s0 = < 0 = s, s0 > −1.
k
Since
lim f (u) = 1
u→−∞

and f (s0 ) < 0, we have


min f (u) = f (s0 ).
u∈I

For u ∈ [−1, 1], we have

2(k2 − 3 − 2ku) 2(k2 − 3 − 2k) 2(k + 1)(k − 3)


f 00 (u) = ≥ = ≥ 0,
(u + k)4 (u + k)4 (u + k4

hence f is convex on [s0 , 1]. According to the LPCF-Theorem, Note 4 and Note 1,
it suffices to show that h(x, y) ≥ 0 for x, y ∈ I which satisfy x + (n − 1) y = n. We
have
f (u) − f (1) u+1
g(u) = = ,
u−1 (u + k)2
g(x) − g( y) (k − 1)2 − 1 − x − y − x y
h(x, y) = = ≥ 0,
x−y (x + k)2 ( y + k)2
since
n2 [(n − 1) y − 1]2
(k − 1)2 − 1 − x − y − x y ≥ −1− x − y − xy = ≥ 0.
n−1 n−1
Partially Convex Function Method 237

n
The equality holds for a1 = a2 = · · · = an = 1. If k = 1 + p , then the equality
n−1
holds also for
1
a1 = n − 1, a2 = · · · = an =
n−1
(or any cyclic permutation).

P 3.10. Let a1 , a2 , a3 , a4 , a5 be real numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5. If

1 25
• ˜
k∈ , ,
6 14
then X 1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4
(Vasile C., 2006)

Solution. We see that


1 2 3 (a1 − 3)2
kai2 − ai + (a1 + a2 + a3 + a4 + a5 ) > ai − ai + = ≥0
6 2 6
for all i ∈ {1, 2, . . . , n}. Since each term of the left hand side of the inequality
decreases by increasing any number ai , it suffices to consider the case

a1 + a2 + a3 + a4 + a5 = 5,

when the desired inequality can be written as


a1 + a2 + a3 + a4 + a5
f (a1 ) + f (a2 ) + f (a3 ) + f (a4 ) + f (a5 ) ≥ 5 f (s), s= = 1,
5
where
−1
f (u) = , u ∈ R.
ku2 −u+5
From
2ku − 1
f 0 (u) = ,
(ku2 − u + 5)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
1
s0 = .
2k
We have
2g(u)
f 00 (u) = , g(u) = −3k2 u2 + 3ku + 5k − 1.
(ku2 − u + 5)3
238 Vasile Cîrtoaje

For
1 25
≤k≤ ,
2 14
we have
1
s0 = ≤ 1 = s,
2k
and for u ∈ [s0 , s], that is
1
≤ u ≤ 1,
2k
we have
(1 − u)(2ku − 1) ≥ 0,
−2ku2 ≥ (2k + 1)u + 1,
−2k2 u2 ≥ k(2k + 1)u + k,
therefore
3 −3k(2k − 1)u + 13k − 2
g(u) ≥ [k(2k + 1)u + k] + 3ku + 5k − 1 =
2 2
−3k(2k − 1) + 13k − 2
≥ = −3k2 + 8k − 1 = 3k(2 − k) + (2k − 1) > 0.
2
Consequently, f is convex on [s0 , s].
For
1 1
≤k≤ ,
6 2
we have
1
s0 = ≥ 1 = s,
2k
and for u ∈ [s, s0 ], that is
1
1≤u≤ ,
2k
we have

g(u) = −3k2 u2 + 3ku + 5k − 1 ≥ 3ku(1 − k) + 5k − 1


≥ 3k(1 − k) + 5k − 1 = −3k2 + 8k − 1
> −6k2 + 7k − 1 = (1 − k)(6k − 1) ≥ 0.

Consequently, f is convex on [s, s0 ].


In both cases, by the PCF-Theorem, it suffices to show that

1 4 5
+ ≤
kx 2 − x +5 ky − y +5 k +4
2

for
x + 4 y = 5, x, y ∈ R.
Partially Convex Function Method 239

Write this inequality as follows:

1 1 1 1
• ˜
− +4 − ≥ 0,
k + 4 kx 2 − x + 5 k + 4 k y2 − y + 5

(x − 1)(kx + k − 1) 4( y − 1)(k y + k − 1)
+ ≥ 0.
kx 2 − x + 5 k y2 − y + 5
Since
4( y − 1) = 1 − x,
the inequality is equivalent to

kx + k − 1 ky + k −1
 ‹
(x − 1) − ≥ 0,
kx 2 − x + 5 k y 2 − y + 5

5(x − 1)2 h(x, y)


≥ 0,
4(kx 2 − x + 5)(k y 2 − y + 5)
where

h(x, y) = −k2 x y − k(k − 1)(x + y) + 6k − 1


= 4k2 y 2 − k(2k + 3) y − 5k2 + 11k − 1
2k + 3 2 (25 − 14k)(6k − 1)
 ‹
= 2k y − + ≥ 0.
4 16

1
The equality holds for a1 = a2 = a3 = a4 = a5 = 1. If k = , then the equality
6
holds also for
5
a1 = −5, a2 = a3 = a4 = a5 =
2
25
(or any cyclic permutation). If k = , then the equality holds also for
14
79 23
a1 = , a2 = a3 = a4 = a5 =
25 50
(or any cyclic permutation).

Remark. In the same manner, we can prove the following generalization:


• Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an ≤ n. If k ∈ [k1 , k2 ],
where p
(n − 1)( 53n2 − 54n + 101 − 5n + 11)
k1 = ,
2(7n2 + 14n − 5)
p
2n2 − 2n + 1 + (n − 1)(3n3 − 4n2 + 3n − 1)
k2 = ,
2(n2 − n + 1)
240 Vasile Cîrtoaje

then X 1 n
≤ ,
ka12 + a2 + · · · + an k+n−1
with equality for a1 = a2 = · · · = an = 1. If k = k1 , then the equality holds also for
2n
a1 = −n, a2 = · · · = a n =
n−1
(or any cyclic permutation). If k = k2 , then the equality holds also for
(2k − 1)(n − 1) + 1 2k + n − 2
a1 = , a2 = · · · = an =
2k 2k(n − 1)
(or any cyclic permutation).

P 3.11. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5.


If k ∈ [k1 , k2 ], where
p
29 − 761 25
k1 = ≈ 0.1414, k2 = ≈ 1.7857,
10 14
then X 1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4
(Vasile C., 2006)
Solution. Since all terms of the left hand side of the inequality decrease by increas-
ing any number ai , it suffices to consider the case
a1 + a2 + a3 + a4 + a5 = 5.
The proof is similar to the one of the preceding P 3.10. Having in view P 3.10, it
suffices to consider the case
1
• ˜
k ∈ k1 , ,
6
when
1
s0 = > 1 = s.
2k
For u ∈ [s, s0 ], that is
1
1≤u≤ ,
2k
f is convex because
g(u) = −3k2 u2 + 3ku + 5k − 1 ≥ 3ku(1 − k) + 5k − 1
≥ 3k(1 − k) + 5k − 1 = −3k2 + 8k − 1
15 (2 − k)(15k − 2)
> − k2 + 87k − 1 = > 0.
4 4
Partially Convex Function Method 241

Thus, by the RPCF-Theorem, it suffices to show that


1 4 5
+ ≤
kx 2 − x + 5 k y 2 − y + 5 k + 4
for
5
x + 4 y = 5, 0≤ x ≤1≤ y ≤ .
4
As shown at P 3.10, this inequality is true if h(x, y) ≥ 0, where

h(x, y) = −k2 x y − k(k − 1)(x + y) + 6k − 1.

We have

h(x, y) = 4k2 y 2 − k(2k + 3) y − 5k2 + 11k − 1


= (5 − 4 y)(A − k2 y) + B = x(A − k2 y) + B,

where
3k(1 − k) −5k2 + 29k − 4
A= , B= .
4 4
Since B ≥ 0, it suffices to show that A − k2 y ≥ 0. Indeed, we have

3k(1 − k) 5k2 k(3 − 8k)


A − k2 y ≥ − = > 0.
4 4 4
The equality holds for a1 = a2 = a3 = a4 = a5 = 1. If k = k1 , then the equality
holds also for
5
a1 = 0, a2 = a3 = a4 = a5 =
4
(or any cyclic permutation). If k = k2 , then the equality holds also for
79 23
a1 = , a2 = a3 = a4 = a5 =
25 50
(or any cyclic permutatio

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.
If k ∈ [k1 , k2 ], where
p
n2 + n − 1 − n4 + 2n3 − 5n2 + 2n + 1
k1 = ,
2n
p
2n2 − 2n + 1 + (n − 1)(3n3 − 4n2 + 3n − 1)
k2 = ,
2(n2 − n + 1)
then X 1 n
≤ ,
ka12 + a2 + · · · + an k+n−1
242 Vasile Cîrtoaje

with equality for a1 = a2 = · · · = an = 1. If k = k1 , then the equality holds also for


n
a1 = 0, a2 = · · · = a n =
n−1
(or any cyclic permutation). If k = k2 , then the equality holds also for

(2k − 1)(n − 1) + 1 2k + n − 2
a1 = , a2 = · · · = an =
2k 2k(n − 1)

(or any cyclic permutation).

P 3.12. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If k > 1, then

1 1 1
+ + ··· + ≤ 1.
a1k + a2 + · · · + an a1 + a2k + · · · + an a1 + a2 + · · · + ank

(Vasile C., 2006)

Solution. It suffices to consider the case a1 + a2 + · · · + an = n, when the desired


inequality can be written as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−1
f (u) = , u ∈ [0, n].
uk − u + n
From
kuk−1 − 1
f 0 (u) = ,
(uk − u + n)2
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , n], where
1
s0 = k 1−k < 1 = s.

We will show that f is convex on [s0 , 1]. For u ∈ [s0 , 1], we have

−k(k + 1)u2k−2 + k(k + 3)uk−1 + nk(k − 1)uk−2 − 2 g(u)


f 00 (u) = > k ,
(u − u + n)
k 3 (u − u + n)3

where
g(u) = −k(k + 1)u2k−2 + k(k + 3)uk−1 − 2.
Denoting
t = kuk−1 , 1 ≤ t ≤ k,
Partially Convex Function Method 243

we get

kg(u) = −(k + 1)t 2 + k(k + 3)t − 2k


= (k + 1)(t − 1)(k − t) + (k − 1)(t + k) > 0.

By the LPCF-Theorem, it suffices to show that

1 n−1
+ k ≤1
xk −x +n y − y+n

for x ≥ 1 ≥ y ≥ 0 and x +(n−1) y = n. Since this inequality is trivial for x = y = 1,


assume next that x > 1 > y ≥ 0, and write the desired inequality as follows:

yk − y + n
xk − x + n ≥ ,
yk − y + 1

(n − 1)( y − y k )
xk − x ≥ ,
yk − y + 1
xk − x y − yk
≥ .
x −1 (1 − y)( y k − y + 1)
xk − x
Let h(x) = , x > 1. By the weighted AM-GM inequality, we have
x −1

(k − 1)x k + 1 − kx k−1
h0 (x) = > 0.
(x − 1)2

Therefore, h is increasing. Since

x − 1 = (n − 1)(1 − y) ≥ 1 − y, x ≥ 2 − y > 1,

we get
(2 − y)k + y − 2
h(x) ≥ h(2 − y) = .
1− y
Thus, it suffices to show that

y − yk
(2 − y) + y − 2 ≥ k
k
,
y − y +1

which is equivalent to

1
(2 − y)k + y − 1 ≥ .
yk − y +1

Using the substitution


t = 1 − y, 0 < t ≤ 1,
244 Vasile Cîrtoaje

the inequality becomes


1
(1 + t)k − t ≥ ,
(1 − t)k + t
(1 − t 2 )k + t(1 + t)k ≥ 1 + t 2 + t(1 − t)k .
By Bernoulli’s inequality,

(1 − t 2 )k + t(1 + t)k ≥ 1 − kt 2 + t(1 + kt) = 1 + t.

So, we only need to show that

1 + t ≥ 1 + t 2 + t(1 − t)k ,

which is equivalent to the obvious inequality


 
t(1 − t) 1 − (1 − t)k−1 ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1.


Remark. Using this result, we can formulate the following statement:
• Let x 1 , x 2 , . . . , x n be nonnegative real numbers so that x 1 + x 2 + · · · + x n ≥ n. If
k > 1, then

x 1k − x 1 x 2k − x 2 x nk − x n
+ + ··· + ≥ 0.
x 1k + x 2 + · · · + x n x 1 + x 2k + · · · + x n x 1 + x 2 + · · · + x nk

This inequality is equivalent to


1 1 1 n
+ +· · ·+ ≤ .
x 1k + x2 + · · · + x n x1 + x 2k + · · · + xn x1 + x2 + · · · + x n
k x1 + x2 + · · · + x n

Using the substitutions


x1 + x2 + · · · + x n
s= , s ≥ 1,
n
and
xi
ai =
, i = 1, 2, . . . , n,
s
which yields a1 + a2 + · · · + an = n, the desired inequality becomes
X 1
≤ 1.
s k−1 a1k + a2 + · · · + a n

Since s k−1 ≥ 1, it suffices to show that


X 1
≤ 1,
a1k + a2 + · · · + an
Partially Convex Function Method 245

which follows immediately from the inequality in P 3.12.


Since x 1 x 2 · · · x n ≥ 1 involves x 1 + x 2 + · · · + x n ≥ n, the inequality is also true
under the more restrictive condition x 1 x 2 · · · x n ≥ 1. For n = 3 and k = 5/2, we get
the inequality from IMO-2005:
• If x, y, z are nonnegative real numbers so that x yz ≥ 1, then

x5 − x2 y5 − y2 z5 − z2
+ + ≥ 0.
x 5 + y 2 + z2 x 2 + y 5 + z2 x 2 + y 2 + z5

P 3.13. Let a1 , a2 , . . . , a5 be nonnegative numbers so that a1 + a2 + a3 + a4 + a5 ≥ 5.


If
4 61
• ˜
k∈ , ,
9 5
then X a1 5
≤ .
ka12 + a2 + a3 + a4 + a5 k+4

(Vasile C., 2006)

Solution. Using the substitution


a1 a a3 a4 a5
x1 = , x2 = 2 , x3 = , x4 = , x5 = ,
s s s s s
where
a1 + a2 + a3 + a4 + a5
s= ≥ 1,
5
we need to show that x 1 + x 2 + x 3 + x 4 + x 5 = 5 involves

x1 x5 5
+ ··· + ≤ .
ksx 12 + x2 + x3 + x4 + x5 x 1 + x 2 + x 3 + x 4 + ksx 5
2
k+4

Since s ≥ 1, it suffices to prove the inequality for s = 1; that is, to show that

a1 a a5 5
+ 2 2 + ··· + 2 ≤
ka12 − a1 + 5 ka2 − a1 + 5 ka5 − an + 5 k + 4

for
a1 + a2 + a3 + a4 + a5 = 5.
Write the desired inequality as

f (a1 ) + f (a2 ) + f (a3 ) + f (a4 ) + f (a5 ) ≥ 5 f (s),


246 Vasile Cîrtoaje

where
a1 + a2 + a3 + a4 + a5
s= =1
5
and
−u
f (u) = , u ∈ [0, 5].
ku2 − u + 5
From
ku2 − 5
f 0 (u) = ,
(ku2 − u + 5)2
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , 5], where
v
t5
s0 = .
k

We have
2g(u)
f 00 (u) = , g(u) = −k2 u3 + 15ku − 5, g 0 (u) = 3k(5 − ku2 ).
(u2 − u + 5)3
4
Case 1: ≤ k ≤ 5. We have
9
v
t5
s0 = ≥ 1 = s.
k

For u ∈ [1, s0 ], the derivative g 0 is nonnegative, g is increasing, hence

4 86k − 25
 ‹
g(u) ≥ g(1) = −k + 15k − 5 = k −
2
(5 − k) + > 0.
9 9

Consequently, f 00 (u) > 0 for u ∈ [1, s0 ], hence f is convex on [s, s0 ].


61
Case 2: 5 ≤ k ≤ . We have
5
v
t5
s0 = < 1 = s.
k

For u ∈ [s0 , 1], we have g 0 (u) ≤ 0, g(u) is decreasing, hence

g(u) ≥ g(1) = −k2 + 15k − 5 = (k − 1)(13 − k) + k + 8 > 0.

Consequently, f 00 (u) > 0 for u ∈ [s0 , 1], hence f is convex on [s0 , s].

In both cases, by the PCF-Theorem, it suffices to show that

x 4y 5
+ ≤
kx 2 − x +5 ky − y +5 k +4
2
Partially Convex Function Method 247

for
x + 4 y = 5, x, y ≥ 0.
Write this inequality as follows:

1 1
• ˜
x y
− +4 − ≥ 0,
k + 4 kx 2 − x + 5 k + 4 k y2 − y + 5

(x − 1)(kx − 5) 4( y − 1)(k y − 5)
+ ≥ 0.
kx 2 − x + 5 k y2 − y + 5
Since
4( y − 1) = 1 − x,
the inequality is equivalent to

kx − 5 ky −5
 ‹
(x − 1) − ≥ 0,
kx 2 − x + 5 k y 2 − y + 5

(x − 1)2 h(x, y)
≥ 0,
(kx 2 − x + 5)(k y 2 − y + 5)
where

h(x, y) = −k2 x y + 5k(x + y) + 5k − 5


= 4k2 y 2 − 5k(k + 3) y + 5(6k − 1).

4 61 4
• ˜ • ˜
We need to show that h(x, y) ≥ 0 for k ∈ , . For k ∈ , 1 , we have
9 5 9

15k 5(9k − 4)
 ‹
h(x, y) = (5 − 4 y) −k y +
2
+
4 4
kx(15 − 4k y) 5(9k − 4)
= +
4 4
kx(kx + 15 − 5k) 5(9k − 4)
= + ≥ 0,
4 4
61
• ˜
while for k ∈ 1, , we have
5

5k + 15 2 (61 − 5k)(k − 1)
 ‹
h(x, y) = 2k y − + ≥ 0.
4 16

4
The equality holds for a1 = a2 = a3 = a4 = a5 = 1. If k = , then the equality
9
holds also for
5
a1 = 0, a2 = a3 = a4 = a5 =
4
248 Vasile Cîrtoaje

61
(or any cyclic permutation). If k = , then the equality holds also for
5
115 95
a1 = , a2 = a3 = a4 = a5 =
61 122
(or any cyclic permutation).

Remark. In the same manner, we can prove the following generalization:


• Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an ≤ n. If k ∈ [k1 , k2 ],
where
n−1
k1 = ,
2n − 1
p
n2 + 2n − 2 + 2 (n − 1)(2n2 − 1)
k2 = ,
n
then X a1 n
≤ ,
ka1 + a2 + · · · + an
2
k+n−1
with equality for a1 = a2 = · · · = an = 1. If k = k1 , then the equality holds also for
n
a1 = 0, a2 = a3 = a4 = a5 =
n−1
(or any cyclic permutation). If k = k2 , then the equality holds also for

n(k − n + 2) n(k + n − 2)
a1 = , a2 = · · · = a n =
2k 2k(n − 1)

(or any cyclic permutation).

P 3.14. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≥ n.


If k > 1, then
a1 a2 an
+ + ··· + ≤ 1.
a1k + a2 + · · · + an a1 + a2k + · · · + an a1 + a2 + · · · + ank

(Vasile C., 2006)

Solution. Using the substitution


a1 a an
x1 = , x2 = 2 , . . . , x n = ,
s s s
where
a1 + a2 + · · · + an
s= ≥ 1,
n
Partially Convex Function Method 249

we need to show that x 1 + x 2 + · · · + x n = n involves


x1 xn
+ ··· + ≤ 1.
s k−1 x 1k + x 2 + · · · + x n x 1 + x 2 + · · · + s k−1 x nk

Since s k−1 ≥ 1, it suffices to prove the inequality for s = 1; that is, to show that
a1 a2 an
+ + ··· + ≤1
a1k − a1 + n a2k − a2 + n ank − an + n

for
a1 + a2 + · · · + an = n.
Case 1: 1 < k ≤ n + 1. By Bernoulli’s inequality, we have

a1k ≥ 1 + k(a1 − 1), a1k − a1 + n ≥ (k − 1)a1 + n − k + 1.

Thus, it suffices to show that


X a1
≤ 1.
(k − 1)a1 + n − k + 1

This is an equality for k = n − 1. If 1 < k < n + 1, then the inequality is equivalent


to X 1
≥ 1,
(k − 1)a1 + n − k + 1
which follows from the the AM-HM inequality
X 1 n2
≥ .
(k − 1)a1 + n − k + 1 [(k − 1)a1 + n − k + 1]
P

Case 2: k > n + 1. Write the desired inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
−u
f (u) = , u ∈ [0, n].
uk −u+n
We have
(k − 1)uk − n
f 0 (u) =
(uk − u + n)2
and
f1 (u)
f 00 (u) = ,
(uk − u + n)3
where

f1 (u) = k(k − 1)uk−1 (uk − u + n) − 2(kuk−1 − 1)[(k − 1)uk − n].


250 Vasile Cîrtoaje

From the expression of f 0 , it follows that f is decreasing on [0, s0 ] and increasing


on [s0 , n], where
 n 1/k
s0 = < 1 = s.
k−1
For u ∈ [s0 , 1], we have

(k − 1)uk − n ≥ (k − 1)s0k − n = 0,

hence

f1 (u) ≥ k(k − 1)uk−1 (uk − u + n) − 2kuk−1 [(k − 1)uk − n]


= kuk−1 [−(k − 1)(uk + u) + n(k + 1)]
≥ kuk−1 [−2(k − 1) + 2(k + 1)] = 4kuk−1 > 0.

Since f 00 (u) > 0, it follows that f is convex on [s0 , s]. By the LPCF-Theorem, we
need to show that
f (x) + (n − 1) f ( y) ≥ n f (1)
for
x ≥ 1 ≥ y ≥ 0, x + (n − 1) y = n.
Consider the nontrivial case where x > 1 > y ≥ 0 and write the required inequality
as follows:
x (n − 1) y
+ ≤ 1,
xk − x + n yk − y + n
x( y k − y + n)
xk − x + n ≥ ,
yk − ny + n
(n − 1) y( y − y k )
xk − x ≥ .
yk − ny + n
Since y < 1 and y k − n y + n > y k − y + 1, it suffices to show that
(n − 1)( y − y k )
xk − x ≥ ,
yk − y + 1
which has been proved at P 3.12.
The equality holds for a1 = a2 = · · · = an = 1.

P 3.15. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


1
If k ≥ 1 − , then
n
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
ka12 + a2 + · · · + an a1 + ka2 + · · · + an
2
a1 + a2 + · · · + kan2

(Vasile C., 2006)


Partially Convex Function Method 251

Solution. Let
a1 + a2 + · · · + an
s= , s ≤ 1.
n
We have three cases to consider.
1
Case 1: s ≤ . The inequality is trivial because
n
ai ≤ a1 + a2 + · · · + an = ns ≤ 1

for i = 1, 2, . . . , n.
1
Case 2: < s < 1. Without loss of generality, assume that
n
a1 ≤ · · · ≤ a j < 1 ≤ a j+1 · · · ≤ an , j ∈ {1, 2, . . . , n}.

Clearly, there are b1 , b2 , . . . , bn so that b1 + b2 + · · · + bn = n and

a1 ≤ b1 ≤ 1, . . . , a j ≤ b j ≤ 1, b j+1 = a j+1 , . . . , bn = an .

Write the desired inequality as

f (a1 ) + f (a2 ) + · · · + f (an ) ≥ 0,

where
1−u
f (u) = , u ∈ [0, ns].
ku2 − u + ns
For u ∈ [0, 1], we have

k[(1 − u)2 − 1] + (1 − ns)


f 0 (u) = < 0,
(ku2 − u + ns)2

hence f is strictly decreasing on [0, 1] and

f (b1 ) ≤ f (a1 ), . . . , f (b j ) ≤ f (a j ), f (b j+1 ) = f (a j+1 ), . . . , f (bn ) = f (an ).

Since
f (b1 ) + f (b2 ) + · · · + f (bn ) ≤ f (a1 ) + f (a2 ) + · · · + f (an ),
it suffices to show that f (b1 ) + f (b2 ) + · · · + f (bn ) ≥ 0 for b1 + b2 + · · · + bn = n.
This inequality is proved at Case 3.
Case 3: s = 1. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1−u
f (u) = , u ∈ [0, n].
ku2 −u+n
252 Vasile Cîrtoaje

From
k[(u − 1)2 − 1] − (n − 1)
f 0 (u) = ,
(ku2 − u + n)2
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , n], where
v
t n−1
s0 = 1 + 1 + > 1 = s, s0 < n.
k
We will show that f is convex on [1, s0 ]. We have

2g(u)
f 00 (u) = ,
(ku2− u + n)3
where

g(u) = −k2 u3 + 3k2 u2 + 3k(n − 1)u − kn − n + 1, g 0 (u) = 3k(−ku2 + 2ku + n − 1).

For u ∈ [1, s0 ], we have g 0 (u) ≥ 0, g is increasing, therefore

g(u) ≥ g(1) = 2k2 + (2n − 3)k − n + 1


2(n − 1)2 (2n − 3)(n − 1)
≥ + −n+1
n2 n
(n2 − 1)(n − 2)
= ≥ 0,
n2
f 00 (u) ≥ 0, f (u) is convex for u ∈ [s, s0 ]. By the RPCF-Theorem, it suffices to show
that
1− x (n − 1)(1 − y)
+ ≥0
kx 2 − x + n k y2 − y + n
for 0 ≤ x ≤ 1 ≤ y and x + (n − 1) y = n. Since (n − 1)(1 − y) = x − 1, we have

1− x (n − 1)(1 − y) 1 1
 ‹
+ = (x − 1) − 2 +
kx 2 − x + n k y2 − y + n kx − x + n k y 2 − y + n
(x − 1)(x − y)(kx + k y − 1)
=
(kx 2 − x + n)(k y 2 − y + n)
n(x − 1)2 (kx + k y − 1)
= ≥ 0,
(n − 1)(kx 2 − x + n)(k y 2 − y + n)
because
n−1 (n − 2)x
k(x + y) − 1 ≥ (x + y) − 1 = ≥ 0.
n n
1
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1. If k = 1 − ,
n
then the equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
Partially Convex Function Method 253

(or any cyclic permutation).


Remark. For k = 1, we get the following statement:
• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an ≤ n, then
1 − a1 1 − a2 1 − an
+ + · · · + ≥ 0.
a12 + a2 + · · · + an a1 + a22 + · · · + an a1 + a2 + · · · + an2

with equality for a1 = a2 = · · · = an = 1.

P 3.16. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an ≤ n.


1
If k ≥ 1 − , then
n
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 − a1 + ka1 1 − a2 + ka2
2 2
1 − an + kan2

(Vasile C., 2006)


Solution. The proof is similar to the one of the preceding P 3.15. For the case 3,
we need to show that
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1−u
f (u) = , u ∈ [0, n].
1 − u + ku2
From
ku(u − 2)
f 0 (u) = ,
(1 − u + ku2 )2
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , n], where

s0 = 2 > s.

We will show that f is convex on [1, s0 ]. For u ∈ [1, s0 ], we have


2kg(u)
f 00 (u) = , g(u) = −ku3 + 3ku2 − 1.
(1 − u + ku2 )3
Since
g 0 (u) = 3ku(2 − u) ≥ 0,
g is increasing, g(u) ≥ g(1) = 2k − 1 ≥ 0, hence f 00 (u) ≥ 0 for u ∈ [1, s0 ]. By the
RPCF-Theorem, it suffices to show that
1− x (n − 1)(1 − y)
+ ≥0
1 − x + kx 2 1 − y + k y2
254 Vasile Cîrtoaje

for 0 ≤ x ≤ 1 ≤ y and x + (n − 1) y = n. Since (n − 1)(1 − y) = x − 1, we have

1− x (n − 1)(1 − y) 1 1
 ‹
+ = (1 − x) −
1 − x + kx 2 1 − y + k y2 1 − x + kx 2 1 − y + k y 2
(1 − x)( y − x)(kx + k y − 1)
=
(1 − x + kx 2 )(1 − y + k y 2 )
n(x − 1)2 (kx + k y − 1)
= .
(n − 1)(1 − x + kx 2 )(1 − y + k y 2 )

Since
n−1 (n − 2)x
k(x + y) − 1 ≥ (x + y) − 1 = ≥ 0,
n n
1
the conclusion follows. The equality holds for a1 = a2 = · · · = an = 1. If k = 1 − ,
n
then the equality holds also for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).

P 3.17. Let a1 , a2 , . . . , an be positive real numbers so that a1 + a2 + · · · + an = n. If


n
0<k≤ , then
n−1
k/a k/a
a1 1 + a2 2 + · · · + ank/an ≤ n.

(Vasile C., 2006)

Solution. According to the power mean inequality, we have

p/a1 p/a2
Œ1/p q/a1 q/a2
Œ1/q
+ a2 + · · · + anp/an + a2 + · · · + anq/an
‚ ‚
a1 a1

n n

for all p ≥ q > 0. Thus, it suffices to prove the desired inequality for
n
k= , 1 < k ≤ 2.
n−1
Rewrite the desired inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = −uk/u , u ∈ I = (0, n).
Partially Convex Function Method 255

We have
k
f 0 (u) = ku u −2 (ln u − 1),
k
f 00 (u) = ku u −4 [u + (1 − ln u)(2u − k + k ln u)].
For n = 2, when k = 2 and I = (0, 2), f is convex on [1, 2) because

1 − ln u > 0, 2u − k + k ln u = 2u − 2 + 2 ln u ≥ 2u − 2 ≥ 0.

Therefore, we may apply the RHCF-Theorem. Consider now that n ≥ 3. From the
expression of f 0 , it follows that f is decreasing on (0, s0 ] and increasing on [s0 , n),
where
s0 = e > 1 = s.
In addition, we claim that f is convex on [1, s0 ]. Indeed, since

1 − ln u ≥ 0, 2u − k + k ln u ≥ 2 − k > 0,

we have f 00 > 0 for u ∈ [1, s0 ]. Therefore, by the RHCF-Theorem (for n = 2) and


the RPCF-Theorem (for n ≥ 3), we only need to show that

x k/x + (n − 1) y k/ y ≤ n

for
0 < x ≤ 1 ≤ y, x + (n − 1) y = n.
We have
k
≥ k > 1.
x
Also, from
k n n k n 2
= > = 1, = ≤ ≤ 2,
y (n − 1) y x + (n − 1) y y (n − 1) y y
we get
k
0<
− 1 ≤ 1.
y
Therefore, by Bernoulli’s inequality, we have
1
x k/x + (n − 1) y k/ y − n =  + (n − 1) y · y
k/ y−1
−n
1 k/x
x
1
•  ‹ ˜
k
≤ 1
 + (n − 1) y 1 + − 1 ( y − 1) − n
1+ k
x x −1 y
x2
= − (k − 1)x 2 − (2 − k)x
x 2 − kx + k
−x(x − 1)2 [(k − 1)x + k(2 − k)]
= ≤ 0.
x 2 − kx + k
The proof is completed. The equality holds for a1 = a2 = · · · = an = 1.
256 Vasile Cîrtoaje

P 3.18. If a, b, c, d, e are nonzero real numbers so that a + b + c + d + e = 5, then

5 2 5 2 5 2 5 2 5 2
 ‹  ‹  ‹  ‹  ‹
7− + 7− + 7− + 7− + 7− ≥ 20.
a b c d e

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
5 2
 ‹
f (u) = 7 − , u ∈ I = R \ {0}.
u
From
10(7u − 5)
f 0 (u) = ,
u3
it follows that f is increasing on (−∞, 0)∪[s0 , ∞) and decreasing on (0, s0 ], where
5
s0 = < 1 = s.
7
Since
lim f (u) = 49
u→−∞

and f (s0 ) = 0, we have


min f (u) = f (s0 ).
u∈I

Also, f is convex on [s0 , s] = [5/7, 1] because

10(15 − 14u)
f 00 (u) = > 0.
u4
According to the LPCF-Theorem and Note 4, we only need to show that

f (x) + 4 f ( y) ≥ 5 f (1)

for all nonzero real x, y so that x + 4 y = 5. Using Note 1, it suffices to prove that
h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
We have
9 5
 ‹
g(u) = 5 − ,
u u2
5(5x + 5 y − 9x y) 5(6 y − 5)2
h(x, y) = = ≥ 0.
x2 y2 x2 y2
Partially Convex Function Method 257

In accordance with Note 3, the equality holds for a = b = c = d = e = 1, and also


for
5 5
a= , b=c=d=e=
3 6
(or any cyclic permutation).
Remark. Similarly, we can prove the following generalization:
• Let a1 , a2 , . . . , an be nonzero real numbers so that a1 + a2 + · · · + an = n. If
n
k= p , then
n+ n−1

k 2 k 2 k 2
 ‹  ‹  ‹
1− + 1− + ··· + 1 − ≥ n(1 − k)2 ,
a1 a2 an

with equality for a1 = a2 = · · · = an = 1, and also for


n n
a1 = p , a2 = a3 = · · · = an = p
1+ n−1 n−1+ n−1

(or any cyclic permutation).

P 3.19. If a1 , a2 , . . . , a7 are real numbers so that a1 + a2 + · · · + a7 = 7, then

(a12 + 2)(a22 + 2) · · · (a72 + 2) ≥ 37 .

(Vasile C., 2007)

Solution. Write the inequality as

a1 + a2 + · · · + a7
f (a1 ) + f (a2 ) + · · · + f (a7 ) ≥ 7 f (s), s= = 1,
7
where
f (u) = ln(u2 + 2), u ∈ R.
From
2u
f 0 (u) = ,
u2 + 2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞], where

s0 = 0.

From
2(2 − u2 )
f 00 (u) = ,
(u2 + 2)2
258 Vasile Cîrtoaje

it follows that f 00 (u) > 0 for u ∈ [0, 1], therefore f is convex on [s0 , s]. By the
LPCF-Theorem, it suffices to prove that

f (x) + 6 f ( y) ≥ 7 f (1)

for x, y ∈ R so that x + 6 y = 7. The inequality can be written as g( y) ≥ 0, where

g( y) = ln [(7 − 6 y)2 + 2] + 6 ln ( y 2 + 2) − 7 ln 3, y ∈ R.

From
4(6 y − 7) 12 y
g 0 ( y) = +
12 y 2 − 28 y + 17 y 2 + 2
28(6 y 3 − 13 y 2 + 9 y − 2)
=
(12 y 2 − 28 y + 17)( y 2 + 2)
28(2 y − 1)(3 y − 2)( y − 1)
= ,
(12 y 2 − 28 y + 17)( y 2 + 2)

1 2 1 2
 ˜ • ˜ • ˜
it follows that g is decreasing on −∞, ∪ , 1 and increasing on , ∪
2 3 2 3
[1, ∞); therefore,
g ≥ min{g(1/2), g(1)}.
Since g(1) = 0, we only need to show that g(1/2) ≥ 0; that is, to show that x = 4
and y = 1/2 involve
(x 2 + 2)( y 2 + 2)6 ≥ 37 .
Indeed, we have

37 139 · 37
 ‹
(x + 2)( y + 2) − 3 = 3
2 2 6 7 7
− 1 = > 0.
211 211

The equality holds for a1 = a2 = · · · = a7 = 1.

n2
P 3.20. Let a1 , a2 , . . . , an be real numbers so that a1 +a2 +· · ·+an = n. If k ≥ ,
4(n − 1)
then
(a12 + k)(a22 + k) · · · (an2 + k) ≥ (1 + k)n .

(Vasile C., 2007)

Solution. Write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
Partially Convex Function Method 259

where
f (u) = ln(u2 + k), u ∈ R.
From
2u
f 0 (u) =
,
+k u2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞], where

s0 = 0.

From
2(k − u2 )
f 00 (u) = ,
(u2 + k)2
it follows that f 00 (u) ≥ 0 for u ∈ [0, 1], therefore f is convex on [s0 , s]. By the
LPCF-Theorem and Note 2, it suffices to prove that H(x, y) ≥ 0 for x, y ∈ R so that
x + (n − 1) y = n, where
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
We have
1 k−xy
H(x, y) = 2
2 (x + k)( y 2 + k)
n2 − 4(n − 1)x y
≥ ,
4(n − 1)(x 2 + k)( y 2 + k)
[x + (n − 1) y]2 − 4(n − 1)x y
=
4(n − 1)(x 2 + k)( y 2 + k)
[x − (n − 1) y)]2
= ≥ 0.
4(n − 1)(x 2 + k)( y 2 + k)
The equality holds for a1 = a2 = · · · = an = 1.

P 3.21. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 10,


then
(a12 − a1 + 1)(a22 − a2 + 1) · · · (an2 − an + 1) ≥ 1.

(Vasile C., 2007)


Solution. Write the inequality as
a1 + a2 + · · · + a n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= =1
n
where
f (u) = ln(u2 − u + 1), u ∈ R.
260 Vasile Cîrtoaje

From
2u − 1
f 0 (u) = ,
u2 − u + 1
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
1
s0 = < 1 = s.
2
In addition, from
1 + 2u(1 − u)
f 00 (u) = ,
(u2 − u + 1)2
it follows that f 00 (u) > 0 for u ∈ [s0 , 1], hence f is convex on [s0 , s]. According to
LPCF-Theorem, we only need to show that

f (x) + (n − 1) f ( y) ≥ n f (1)

for all real x, y so that x + (n − 1) y = n. Write this inequality as g(x) ≥ 0, where


n− x
g(x) = ln(x 2 − x + 1) + (n − 1) ln( y 2 − y + 1), y= .
n−1
−1
Since y 0 (x) = , we have
n−1
2x − 1 2y − 1 2x − 1 2y − 1
g 0 (x) = + (n − 1) y 0 2 = 2 − 2
x2 − x +1 y − y +1 x − x +1 y − y +1

(x − y)(1 + x + y − 2x y) (x − 1)[2x 2 − (n + 2)x + 2n − 1]


= = .
(x 2 − x + 1)( y 2 − y + 1) (n − 1)2 (x 2 − x + 1)( y 2 − y + 1)
Because 2x 2 − (n + 2)x + 2n − 1 > 0 for n ≤ 10, we have g 0 (x) ≤ 0 for x ∈ (−∞, 1]
and g 0 (x) ≥ 0 for x ∈ [1, ∞). Therefore, since g(x) is decreasing on (−∞, 1] and
increasing on [1, ∞), we have

g(x) ≥ g(1) = 0.

The equality occurs for a1 = a2 = · · · = an = 1.


Remark 1. The inequality holds also for n = 11, n = 12 and n = 13, when the
equation
2x 2 − (n + 2)x + 2n − 1 = 0
has two positive roots, namely
p p
n + 2 − n2 − 12(n − 1) n+2+ n2 − 12(n − 1)
x1 = , x2 = ,
4 4
satisfying 1 < x 1 < x 2 . Thus, g(x) is decreasing on (−∞, 1] ∪ [x 1 , x 2 ] and increas-
ing on [1, x 1 ] ∪ [x 2 , ∞). Therefore, it suffices to show that

min{g(1), g(x 2 )} ≥ 0.
Partially Convex Function Method 261

We have g(1) = 0. For n = 13, we have


13 − x 2 2
x 2 = 5, y2 = = ,
12 3
hence
7 713
g(x 2 ) = ln(x 22 − x 2 + 1) + (n − 1) ln( y22 − y2 + 1) = ln 21 + 12 · ln = ln 23 > 0.
9 3
For n = 14, the inequality does not hold.
Remark 2. By replacing a1 , a2 , . . . , an respectively with 1 − a1 , 1 − a2 , . . . , 1 − an , we
get the following statement:
• Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = 0. If n ≤ 13, then

(1 − a1 + a12 )(1 − a2 + a22 ) · · · (1 − an + an2 ) ≥ 1,

with equality for a1 = a2 = · · · = an = 0.

P 3.22. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 26,


then
(a12 − a1 + 2)(a22 − a2 + 2) · · · (an2 − an + 2) ≥ 2n .
(Vasile C., 2007)
Solution. Write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = ln(u2 − u + 2), u ∈ R.
From
2u − 1
f 0 (u) =
,
−u+2 u2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
1
s0 = < 1 = s.
2
In addition, from
3 + 2u(1 − u)
f 00 (u) = ,
(u2 − u + 2)2
it follows that f 00 (u) > 0 for u ∈ [s0 , 1], hence f is convex on [s0 , s]. According to
LPCF-Theorem, we only need to show that

f (x) + (n − 1) f ( y) ≥ n f (1)
262 Vasile Cîrtoaje

for all real x, y so that x + (n − 1) y = n. Write this inequality as g(x) ≥ 0, where


n− x
g(x) = ln(x 2 − x + 2) + (n − 1) ln( y 2 − y + 2), y= .
n−1
−1
Since y 0 (x) = , we have
n−1
2x − 1 0 2y − 1 2x − 1 2y − 1
g 0 (x) = + (n − 1) y = −
x2 − x + 2 y2 − y + 2 x2 − x + 2 y2 − y + 2

(x − y)(3 + x + y − 2x y) (x − 1)[2x 2 − (n + 2)x + 4n − 3]


= = .
(x 2 − x + 2)( y 2 − y + 2) (n − 1)2 (x 2 − x + 1)( y 2 − y + 1)
Because 2x 2 − (n + 2)x + 4n − 3 > 0 for n ≤ 26, we have g 0 (x) ≤ 0 for x ∈ (−∞, 1]
and g 0 (x) ≥ 0 for x ∈ [1, ∞). Therefore, since g(x) is decreasing on (−∞, 1] and
increasing on [1, ∞), we have

g(x) ≥ g(1) = 0.

The equality occurs for a1 = a2 = · · · = an = 1.


Remark 1. The inequality holds also for 27 ≤ n ≤ 38, when the equation

2x 2 − (n + 2)x + 4n − 3 = 0

has two positive roots, namely


p p
n + 2 − n2 − 28(n − 1) n+2+ n2 − 28(n − 1)
x1 = , x2 = ,
4 4
satisfying 1 < x 1 < x 2 . Thus, g(x) is decreasing on (−∞, 1] ∪ [x 1 , x 2 ] and increas-
ing on [1, x 1 ] ∪ [x 2 , ∞). Therefore, it suffices to show that

min{g(1), g(x 2 )} ≥ 0.

We have g(1) = 0 and g(x 2 ) > 0 for 27 ≤ n ≤ 38. For n = 39, the inequality does
not hold.
Remark 2. By replacing a1 , a2 , . . . , an respectively with 1 − a1 , 1 − a2 , . . . , 1 − an , we
get the following statement:
• Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = 0. If n ≤ 38, then

(2 − a1 + a12 )(2 − a2 + a22 ) · · · (2 − an + an2 ) ≥ 2n ,

with equality for a1 = a2 = · · · = an = 0.


Partially Convex Function Method 263

P 3.23. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

(1 − a + a4 )(1 − b + b4 )(1 − c + c 4 ) ≥ 1.

Solution. Write the inequality as

a+b+c
f (a) + f (b) + f (c) ≥ 3 f (s), s= = 1,
3
where
f (u) = ln(1 − u + u4 ), u ∈ [0, 3].
From
4u3 − 1
f 0 (u) = ,
1 − u + u4
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , 3], where
1
s0 = p
3
< 1 = s.
4
Also, f is convex on [s0 , 1] because

−4u6 − 4u3 + 12u2 − 1 −4u2 − 4u2 + 12u2 − 1 4u2 − 1


f 00 (u) = ≥ = > 0.
(1 − u + u4 )2 (1 − u + u4 )2 (1 − u + u4 )2
According to the LPCF-Theorem, we only need to show that

f (x) + 2 f ( y) ≥ 3 f (1)

for all x, y ≥ 0 so that x+2 y = 3. Using Note 2, it suffices to prove that H(x, y) ≥ 0,
where
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
We have
(x + y)(x − y)2 − 1 + 4(x 2 + y 2 + x y) − 2x y(x + y) − 4x 3 y 3
H(x, y) =
(1 − x + x 4 )(1 − y + y 4 )
−1 + 4(x 2 + y 2 + x y) − 2x y(x + y) − 4x 3 y 3

(1 − x + x 4 )(1 − y + y 4 )
h(x, y)
= ,
(1 − x + x 4 )(1 − y + y 4 )
where
h(x, y) = −1 + 2(x + y)[2(x + y) − x y] − 4x y − 4x 3 y 3 .
From 3 = x + 2 y ≥ 2 2x y and (1 − x)(1 − y) ≤ 0, we get
p

9
xy ≤ , x + y ≥ 1 + x y.
8
264 Vasile Cîrtoaje

Therefore,

h(x, y) ≥ −1 + 2(1 + x y)[2(1 + x y) − x y] − 4x y − 4x 3 y 3


= 3 + 2x y + 2x 2 y 2 − 4x 3 y 3 ≥ 3 + 2x y + 2x 2 y 2 − 5x 2 y 2
= 3 + 2x y − 3x 2 y 2 ≥ 3 + 2x y − 4x y = 3 − 2x y > 0.

The proof is completed. The equality holds for a = b = c = 1.

P 3.24. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(1 − a + a3 )(1 − b + b3 )(1 − c + c 3 )(1 − d + d 3 ) ≥ 1.

(Vasile C., 2012)

Solution. Write the inequality as


a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
f (u) = ln(1 − u + u3 ), u ∈ [0, 4].
From
3u2 − 1
f 0 (u) = ,
1 − u + u3
it follows that f is decreasing on [0, s0 ] and increasing on [s0 , 4], where
1
s0 = p < 1 = s.
3
In addition, f is convex on [s0 , 1] because

−3u4 + 6u − 1 −3u + 6u − 1 3u − 1
f 00 (u) = ≥ = > 0.
(1 − u + u )
3 2 (1 − u + u )
3 2 (1 − u + u3 )2
According to the LPCF-Theorem, we only need to show that

f (x) + 3 f ( y) ≥ 4 f (1)

for all x, y ≥ 0 so that x+3 y = 4. Using Note 2, it suffices to prove that H(x, y) ≥ 0,
where
f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
We have
(x − y)2 + 3(x + y) − 1 − 3x 2 y 2 3(x + y) − 1 − 3x 2 y 2
H(x, y) = ≥ .
(1 − x + x 3 )(1 − y + y 3 ) (1 − x + x 3 )(1 − y + y 3 )
Partially Convex Function Method 265

From 4 = x + 3 y ≥ 2 3x y and (1 − x)(1 − y) ≤ 0, we get


p

4
xy ≤ , x + y ≥ 1 + x y.
3
Therefore,

3(x + y) − 1 − 3x 2 y 2 ≥ 3(1 + x y) − 1 − 3x 2 y 2
≥ 3(1 + x y) − 1 − 4x y = 2 − x y > 0,

hence H(x, y) > 0. The equality holds for a = b = c = d = 1.

P 3.25. If a, b, c, d, e are nonzero real numbers so that a + b + c + d + e = 5, then

1 1 1 1 1 1 1 1 1 1
 ‹  ‹
5 2 + 2 + 2 + 2 + 2 + 45 ≥ 14 + + + + .
a b c d e a b c d e

(Vasile C., 2013)

Solution. Write the desired inequality as

a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
5 14
f (u) = 2
− + 9, u ∈ I = R \ {0}.
u u
From
2(7u − 5)
f 0 (u) = ,
u3
it follows that f is increasing on (−∞, 0)∪[s0 , ∞) and decreasing on (0, s0 ], where

5
s0 = < 1 = s.
7
Since
lim f (u) = 9
u→−∞

and f (s0 ) < f (1) = 0, we have

min f (u) = f (s0 ).


u∈I

From
2(15 − 14u)
f 00 (u) = ,
u4
266 Vasile Cîrtoaje

it follows that f is convex on [s0 , 1]. By the LPCF-Theorem, Note 4 and Note 1, it
suffices to show that h(x, y) ≥ 0 for all x, y ∈ I which satisfy x + 4 y = 5, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

Indeed, we have
9 5
g(u) = − 2,
u u
5x + 5 y − 9x y (6 y − 5)2
h(x, y) = = ≥ 0.
x2 y2 x2 y2
In accordance with Note 3, the equality holds for a = b = c = d = e = 1, and also
for
5 5
a= , b=c=d=e=
3 6
(or any cyclic permutation).

P 3.26. If a, b, c are positive real numbers so that abc = 1, then

7 − 6a 7 − 6b 7 − 6c
+ + ≥ 1.
2 + a2 2 + b2 2 + c 2

(Vasile C., 2008)

Solution. Using the substitution

a = e x , b = e y , c = ez ,

we need to show that


f (x) + f ( y) + f (z) ≥ 3 f (s),
where
x + y +z
s= =0
3
and
7 − 6eu
f (u) = , u ∈ R.
2 + e2u
From
2(3eu + 2)(eu − 3)
f 0 (u) = ,
(2 + e2u )2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where

s0 = ln 3 > s.
Partially Convex Function Method 267

We have
2t · h(t)
f 00 (u) = , h(t) = −3t 4 + 14t 3 + 36t 2 − 28t − 12, t = eu .
(2 + t 2 )3

We will show that h(t) > 0 for t ∈ [1, 3], hence f is convex on [0, s0 ]. We have

h(t) = 3(t 2 − 1)(9 − t 2 ) + 14t 3 + 6t 2 − 28t + 15


≥ 14t 3 + 6t 2 − 28t + 15
= 14t 2 (t − 1) + 14(t − 1)2 + 6t 2 + 1 > 0.

By the RPCF-Theorem, we only need to prove that

f (x) + 2 f ( y) ≥ 3 f (0)

for all real x, y so that x + 2 y = 0. That is, to show that the original inequality
holds for b = c and a = 1/c 2 . Write this inequality as

c 2 (7c 2 − 6) 2(7 − 6c)


+ ≥ 1,
2c 4 + 1 2 + c2

(c − 1)2 (c − 2)2 (5c 2 + 6c + 3) ≥ 0.


By Note 3, the equality holds for a = b = c = 1, and also for

1
a= , b=c=2
4
(or any cyclic permutation).

P 3.27. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 1
+ + ≤ .
a + 5bc b + 5ca c + 5ab 2

(Vasile C., 2008)

Solution. Write the inequality as

a b c 1
+ 2 + 2 ≤ .
a2 +5 b +5 c +5 2
Using the substitution
a = e x , b = e y , c = ez ,
we need to show that
f (x) + f ( y) + f (z) ≥ 3 f (s),
268 Vasile Cîrtoaje

where
x + y +z
s= =0
3
and
−eu
f (u) = , u ∈ R.
e2u + 5
From
eu (e2u − 5)
f 0 (u) = ,
(e2u + 5)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where

ln 5
s0 = > 0 = s.
2
Also, from
eu (−e4u + 30e2u − 25)
f 00 (u) = ,
(e2u + 5)3
p
it follows that f is convex on [s, s0 ], because u ∈ [0, s0 ] involves eu ∈ [1, 5 ] and
e2u ∈ [1, 5], hence

−e4u + 30e2u − 25 = e2u (5 − e2u ) + 25(e2u − 1) > 0.

By the RPCF-Theorem, we only need to prove the original inequality for b = c and
a = 1/c 2 . Write this inequality as

c2 2c 1
+ 2 ≤ ,
5c + 1 c + 5 2
4

(c − 1)2 (5c 4 − 10c 3 − 2c 2 + 6c + 5) ≥ 0,


(c − 1)2 [5(c − 1)4 + 2c(5c 2 − 16c + 13)] ≥ 0.
The equality holds for a = b = c = 1.

P 3.28. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 3
+ + ≤ .
4 − 3a + 4a 2 4 − 3b + 4b 2 4 − 3c + 4c 2 5

(Vasile Cirtoaje, 2008)

Solution. Let
a = ex , b = ey, c = ez .
We need to show that
f (x) + f ( y) + f (z) ≥ 3 f (s),
Partially Convex Function Method 269

where
x + y +z
s= =0
3
and
−1
f (u) = , u ∈ R.
4 − 3eu + 4e2u
From
eu (8eu − 3)
f 0 (u) = ,
(4 − 3eu + 4e2u )2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
3
s0 = ln < 0 = s.
8
We claim that f is convex on [s0 , 0]. Since

eu (−64e3u + 36e2u + 55eu − 12)


f 00 (u) = ,
(4 − 3eu + 4e2u )3
we need to show that
−64t 3 + 36t 2 + 55t − 12 ≥ 0,
where
3
˜ •
t=e ∈ u
,1 .
8
Indeed, we have

−64t 3 + 36t 2 + 55t − 12 > −72t 3 + 36t 2 + 48t − 12


= 12(1 − t)(6t 2 + 3t − 1) ≥ 0.

By the LPCF-Theorem, we only need to prove the original inequality for b = c and
a = 1/c 2 . Write this inequality as follows:

c4 2 3
+ ≤ ,
4c − 3c + 4 4 − 3c + 4c
4 2 2 5

28c 6 − 21c 5 − 48c 4 + 27c 3 + 42c 2 − 36c + 8 ≥ 0,


(c − 1)2 (28c 4 + 35c 3 − 6c 2 − 20c + 8) ≥ 0.
It suffices to show that

7(4c 4 + 5c 3 − c 2 − 3c + 1) ≥ 0.

Indeed,
4c 4 + 5c 3 − c 2 − 3c + 1 = c 2 (2c − 1)2 + 9c 3 − 2c 2 − 3c + 1
and
9c 3 − 2c 2 − 3c + 1 = c(3c − 1)2 + (2c − 1)2 > 0.
270 Vasile Cîrtoaje

The equality holds for a = b = c = 1.


Remark. Since
1 1 1
≥ = ,
4 − 3a + 4a2 4 − 3a + 4a2 + (1 − a)2 5(1 − a + a2 )

we get the following known inequality


1 1 1
+ + ≤ 3.
1−a+a 2 1− b+ b 2 1 − c + c2

P 3.29. If a, b, c are positive real numbers so that abc = 1, then


1 1 1 3
+ + ≤ .
(3a + 1)(3a − 5a + 3) (3b + 1)(3b − 5b + 3) (3c + 1)(3c − 5c + 3) 4
2 2 2

Solution. Let
a = ex , b = ey, c = ez .
We need to show that
f (x) + f ( y) + f (z) ≥ 3 f (s),
where
x + y +z
s= =0
3
and
−1
f (u) = , u ∈ R.
(3eu + 1)(3e2u − 5eu + 3)
From
(3eu − 2)(9eu − 2)
f 0 (u) = ,
(3eu + 1)2 (3e2u − 5eu + 3)2
it follows that f is increasing on (−∞, s1 ] ∪ [s0 , ∞) and decreasing on [s1 , s0 ],
where
s1 = ln 2 − ln 9, s0 = ln 2 − ln 3, s1 < s0 < 0 = s.
Since
−1
lim f (u) = f (s0 ) = ,
u→−∞ 3
we get
min f (u) = f (s0 ).
u∈R

We claim that f is convex on [s0 , 0]. We have

t · h(t)
f 00 (u) = ,
(3t + 1)3 (3t 2 − 5t + 3)3
Partially Convex Function Method 271

where
2
•˜
t=e ∈u
,1 , h(t) = −729t 5 + 1188t 4 − 648t 3 + 387t 2 − 160t + 12.
3

Since the polynomial h(t) has the real roots

t 1 ≈ 0.0933, t 2 ≈ 0.5072, t 3 ≈ 1.11008,

it follows that h(t) > 0 for t ∈ [2/3, 1] ⊂ [t 2 , t 3 ], hence f is convex on [s0 , 0]. By
the LPCF-Theorem, we only need to prove the original inequality for b = c ≤ 1 and
a = 1/c 2 . Write this inequality as follows:

c6 2 3
+ ≤ .
(c 2 + 3)(3c 4 − 5c 2 + 3) (3c + 1)(3c 2 − 5c + 3) 4

Since
c 2 + 3 ≥ 2(c + 1)
and
3c 4 − 5c 2 + 3 ≥ c(3c 2 − 5c + 3),
it suffices to prove that

c5 2 3
+ ≤ .
2(c + 1)(3c 2 − 5c + 3) (3c + 1)(3c 2 − 5c + 3) 4

This is equivalent to the obvious inequality

(1 − c)2 (1 + 15c + 5c 2 − 14c 3 − 6c 4 ) ≥ 0.

The equality holds for a = b = c = 1.

P 3.30. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers so that a1 a2 · · · an = 1. If


p, q ≥ 0 so that p + 4q ≥ n − 1, then

1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

(Vasile C., 2008)

Solution. For q = 0, we get a known inequality (see Remark 2 from the proof
of P 1.63). Consider further that q > 0. Using the substitutions ai = e x i for i =
1, 2, . . . , n, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),
272 Vasile Cîrtoaje

where
x1 + x2 + · · · + x n
s= =0
n
and
1 − eu
f (u) = , u ∈ R.
1 + peu + qe2u
From
eu (qe2u − 2qeu − p − 1)
f 0 (t) = ,
(1 + peu + qe2u )2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
v
t p+1
s0 = ln r0 > 0 = s, r0 = 1 + 1 + .
q

Also, we have
t · h(t)
f 00 (u) = ,
(1 + pt + qt 2 )3
where

h(t) = −q2 t 4 + q(p + 4q)t 3 + 3q(p + 2)t 2 + (p − 4q + p2 )t − p − 1, t = eu .

We will show that h(t) ≥ 0 for t ∈ [1, r0 ], hence f is convex on [0, s0 ]. We have

h0 (t) = −4q2 t 3 + 3q(p + 4q)t 2 + 6q(p + 2)t + p − 4q + p2 ,

h00 (t) = 6q[−2qt 2 + (p + 4q)t + p + 2].


Since

h00 (t) = 6q[2(−qt 2 + 2qt + p + 1) + p(t − 1)] ≥ 12q(−qt 2 + 2qt + p + 1) ≥ 0,

h0 (t) is increasing,

h0 (t) ≥ h0 (1) = p2 + 9pq + 8q2 + p + 8q > 0,

h is increasing, hence

h(t) ≥ h(1) = p2 + 4pq + 3q2 + 2q − 1 = (p + 2q)2 − (q − 1)2


= (p + q + 1)(p + 3q − 1).

Since
p + 4q p + 2q
p + 3q − 1 ≥ p + 3q − = > 0,
n−1 2
f 00 (u) > 0 for u ∈ [0, s0 ], therefore f is convex on [s, s0 ]. By the RPCF-Theorem, we
only need to prove the original inequality for

a2 = · · · = an := t, a1 = 1/t n−1 , t ≥ 1.
Partially Convex Function Method 273

Write this inequality as

t n−1 (t n−1 − 1) (n − 1)(1 − t)


+ ≥ 0,
t 2n−2 + pt n−1 + q 1 + pt + qt 2
or
pA + qB ≥ C,
where

A = t n−1 (t n − nt + n − 1),
B = t 2n − t n+1 − (n − 1)(t − 1),
C = t n−1 [(n − 1)t n + 1 − nt n−1 ].

Since p + 4q ≥ n − 1 and C ≥ 0 (by the AM-GM inequality applied to n positive


numbers), it suffices to show that

(p + 4q)C
pA + qB ≥ ,
n−1
which is equivalent to

p[(n − 1)A − C] + q[(n − 1)B − 4C] ≥ 0.

This is true if
(n − 1)A − C ≥ 0
and
(n − 1)B − 4C ≥ 0
for t ≥ 1. By the AM-GM inequality, we have

(n − 1)A − C = nt n−1 [t n−1 + n − 2 − (n − 1)t] ≥ 0.

For n = 3, we have

B = (t − 1)2 (t 4 + 2t 3 + 2t 2 + 2t + 2),

C = t 2 (t − 1)2 (2t + 1),

B − 2C = (t − 1)2 (t 4 − 2t 3 + 2t + 2)
= (t − 1)2 [(t − 1)2 (t 2 − 1) + 3] ≥ 0.

Consider further that


n ≥ 4.
Since
t − 1 ≤ t n−1 (t − 1),
274 Vasile Cîrtoaje

we have

B ≥ t 2n − t n+1 − (n − 1)t n−1 (t − 1)


= t n−1 [t n+1 − t 2 − (n − 1)t + n − 1].

Thus, the inequality (n − 1)B − 4C ≥ 0 is true if

(n − 1)[t n+1 − t 2 − (n − 1)t + n − 1] − 4(n − 1)t n − 4 − 4nt n−1 ≥ 0,

which is equivalent to g(t) ≥ 0, where

g(t) = (n − 1)t n+1 − 4(n − 1)t n + 4nt n−1 − (n − 1)t 2 − (n − 1)2 t + n2 − 2n − 3.

We have

g 0 (t) = (n − 1)g1 (t), g1 (t) = (n + 1)t n − 4nt n−1 + 4nt n−2 − 2t − n + 1,

g10 (t) = n(n + 1)t n−1 − 4n(n − 1)t n−2 + 4n(n − 2)t n−3 − 2.
Since Æ
n(n + 1)t n−1 + 4n(n − 2)t n−3 ≥ 4n (n + 1)(n − 2)t n−2 ,
we get
”Æ —
g10 (t) ≥ 4n (n + 1)(n − 2) − n + 1 t n−2 − 2
”Æ —
≥ 4n (n + 1)(n − 2) − n + 1 − 2
4n(n − 3)
=p −2
(n + 1)(n − 2) + n − 1
4n(n − 3)
> − 2 = 2(n − 4) ≥ 0.
(n + 1) + n − 1

Therefore, g1 (t) is increasing for t ≥ 1, g1 (t) ≥ g1 (1) = 0, g(t) is increasing for


t ≥ 1, hence
g(t) ≥ g(1) = 0.
The equality holds for a1 = a2 = · · · = an = 1.
Remark. For p = 0 and q = 1, we get the inequality (Vasile C., 2006)

1−a 1− b 1−c 1−d 1−e


+ + + + ≥ 0,
1 + a2 1 + b2 1 + c 2 1 + d 2 1 + e2
where a, b, c, d, e are positive real numbers so that abcde = 1. Replacing a, b, c, d, e
by 1/a, 1/b, 1/c, 1/d, 1/e, we get

1+a 1+ b 1+c 1+d 1+e


+ + + + ≤ 5,
1+a 2 1+ b 2 1+c 2 1+d 2 1 + e2
Partially Convex Function Method 275

where a, b, c, d, e are positive real numbers so that abcde = 1.


Notice that the inequality

1 − a1 1 − a2 1 − a3 1 − a4 1 − a5 1 − a6
+ + + + + ≥0
1 + a12 1 + a22 1 + a32 1 + a42 1 + a52 1 + a62

is not true for all positive numbers a1 , a2 , a3 , a4 , a5 , a6 satisfying a1 a2 a3 a4 a5 a6 = 1.


Indeed, for a2 = a3 = a4 = a5 = a6 = 2, the inequality becomes

1 − a1
− 1 ≥ 0,
1 + a12

which is false for a1 > 0.

P 3.31. If a, b, c are positive real numbers so that abc = 1, then

1−a 1− b 1−c
+ + ≥ 0.
17 + 4a + 6a2 17 + 4b + 6b2 17 + 4c + 6c 2

(Vasile C., 2008)

Solution. Using the substitution

a = e x , b = e y , c = ez ,

we need to show that


f (x) + g( y) + g(z) ≥ 3 f (s),
where
x + y +z
s= =0
3
and
1 − eu
f (u) = , u ∈ R,
1 + peu + qe2u
with
4 6
p= , q= .
17 17
As we have shown in the proof of the preceding P 3.30, f is decreasing on (−∞, s0 ]
and increasing on [s0 , ∞), where
v v
t p+1 t9
s0 = ln r0 > 0 = s, r0 = 1 + 1 + =1+ .
q 2
276 Vasile Cîrtoaje

5
In addition, since p + 3q − 1 = > 0 (see the proof of P 3.30), f is convex on
17
[0, s0 ]. By the RPCF-Theorem, we only need to prove the original inequality for
b = c ≥ 1 and a = 1/c 2 . Write this inequality as follows:

c 2 (c 2 − 1) 2(1 − c)
+ ≥ 0,
c 4 + pc 2 + q 1 + pc + qc 2
pA + qB ≥ C,
where
A = c 2 (c − 1)2 (c + 2),
B = (c − 1)2 (c 4 + 2c 3 + 2c 2 + 2c + 2),
C = c 2 (c − 1)2 (2c + 1).
Indeed, we have

3(c − 1)2 (c − 2)2 (2c 2 + 2c + 1)


pA + qB − C = ≥ 0.
17
In accordance with Note 3, the equality holds for a = b = c = 1, and also for
1
a= , b=c=2
4
(or any cyclic permutation).

P 3.32. If a1 , a2 , . . . , a8 are positive real numbers so that a1 a2 · · · a8 = 1, then


1 − a1 1 − a2 1 − a8
+ + · · · + ≥ 0.
(1 + a1 )2 (1 + a2 )2 (1 + a8 )2

(Vasile C., 2006)

Solution. Using the substitutions ai = e x i for i = 1, 2, . . . , 8, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x 8 ) ≥ 8 f (s),

where
x1 + x2 + · · · + x8
s= =0
8
and
1 − eu
f (u) = , u ∈ R.
(1 + eu )2
From
eu (eu − 3)
f 0 (t) = ,
(1 + eu )3
Partially Convex Function Method 277

it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where

s0 = ln 3 > 1 = s.

We have
eu (8eu − e2u − 3)
f 00 (u) = .
(1 + eu )4
For u ∈ [0, ln 3], that is eu ∈ [1, 3], we have

8eu − e2u − 3 > 8eu − 3eu − 7 = (eu − 1)(7 − eu ) ≥ 0;

therefore, f is convex on [s, s0 ]. By the RPCF-Theorem, we only need to prove the


original inequality for a2 = · · · = a8 := t and a1 = 1/t 7 , where t ≥ 1. For the
nontrivial case t > 1, write this inequality as follows:
t 7 (t 7 − 1) 7(t − 1)
≥ .
(t 7 + 1)2 (t + 1)2
t 7 (t 7 − 1)(t + 1)2
≥ 7,
(t − 1)(t 7 + 1)2
t 7 (t 6 + t 5 + t 4 + t 3 + t 2 + t + 1)
≥ 7.
(t 6 − t 5 + t 4 − t 3 + t 2 − t + 1)2
Since

t 6 − t 5 + t 4 − t 3 + t 2 − t + 1 = t 4 (t 2 − t + 1) − (t − 1)(t 2 + 1) < t 4 (t 2 − t + 1),

it suffices to show that


t6 + t5 + t4 + t3 + t2 + t + 1
≥ 7,
t(t 2 − t + 1)2
which is equivalent to the obvious inequality

(t − 1)6 ≥ 0.

Thus, the proof is completed. The equality holds for a1 = a2 = · · · = a8 = 1.


Remark. The inequality
1 − a1 1 − a2 1 − a9
+ + · · · + ≥0
(1 + a1 )2 (1 + a2 )2 (1 + a9 )2
is not true for all positive numbers a1 , a2 , . . . , a9 satisfying a1 a2 · · · a9 = 1. Indeed,
for a2 = a3 = · · · = a9 = 3, the inequality becomes
1 − a1
− 1 ≥ 0,
(1 + a1 )2
which is false for a1 > 0.
278 Vasile Cîrtoaje

−13 13
˜ •
P 3.33. Let a, b, c be positive real numbers so that abc = 1. If k ∈ p , p ,
3 3 3 3
then
a+k b+k c+k 3(1 + k)
+ 2 + 2 ≤ .
a +1 b +1 c +1
2 2
(Vasile C., 2012)

Solution. The inequality is equivalent to


1 3 1
X ‹ X ‹
a
k − ≤ − ,
a2 + 1 2 2 a2 + 1
X (a − 1)2 X
2
‹
≥k −3 . (*)
a2 + 1 a2 + 1
13
Thus, it suffices to prove it for |k| = p . On the other hand, replacing a, b, c by
3 3
1/a, 1/b, 1/c, the inequality becomes
X (a − 1)2  X 2 ‹
≥ k 3− . (**)
a2 + 1 a2 + 1
Based on (∗) and (∗∗), we only need to prove the desired inequality for
13
k= p .
3 3
Using the substitution
a = ex , b = ey, c = ez ,
we need to show that
f (x) + g( y) + g(z) ≥ 3 f (s),
where
x + y +z
s= =0
3
and
−eu − k
f (u) = , u ∈ R.
e2u + 1
From
e2u + 2keu − 1
f 0 (t) = ,
(e2u + 1)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
p 1
s0 = ln r0 < 0 = s, r0 = −k + k2 + 1 = p .
3 3
Also, we have
t · h(t)
f 00 (u) = ,
(1 + t 2 )3
Partially Convex Function Method 279

where
h(t) = −t 4 − 4kt 3 + 6t 2 + 4kt − 1, t = eu .
We will show that h(t) > 0 for t ∈ [r0 , 1], hence f is convex on [s0 , s]. Indeed,
since
52t 52
4kt = p ≥ > 1,
3 3 27
we have

h(t) = −t 4 + 6t 2 − 1 + 4kt(1 − t 2 ) ≥ −t 4 + 6t 2 − 1 + (1 − t 2 ) = t 2 (5 − t 2 ) > 0.

By the LPCF-Theorem, we only need to prove the original inequality for b = c := t


and a = 1/t 2 , where t > 0. Write this inequality as

t 2 (kt 2 + 1) 2(t + k) 3(1 + k)


+ 2 ≤ ,
t4 + 1 t +1 2

3t 6 − 4t 5 + t 4 + t 2 − 4t + 3 − k(1 − t 2 )3 ≥ 0,
(t − 1)2 [(3 + k)t 4 + 2(1 + k)t 3 + 2t 2 + 2(1 − k)t + 3 − k] ≥ 0,
p 2  p p p 
(t − 1)2 t − 2 + 3 (27 + 13 3)t 2 + 24(2 + 3)t + 33 + 17 3 ≥ 0.
13
The equality holds for a = b = c = 1. If k = p , then the equality holds also for
3 3
p p
a = 7 + 4 3, b = c =2− 3
−13
(or any cyclic permutation). If k = p , then the equality holds also for
3 3
p p
a = 7 − 4 3, b = c =2+ 3

(or any cyclic permutation).

p
P 3.34. If a, b, c are positive real numbers and 0 < k ≤ 2 + 2 2, then

a3 b3 c3 a+b+c
+ 2 + 2 ≥ .
ka + bc kb + ca kc + ab
2 k+1

(Vasile C., 2011)

Solution. Due to homogeneity, we may assume that abc = 1. On this hypothesis,


we write the inequality as follows:

a4 b4 b4 a b c
+ + ≥ + + ,
ka + 1 kb + 1 kb + 1 k + 1 k + 1 k + 1
3 3 3
280 Vasile Cîrtoaje

a4 − a b4 − b c4 − c
+ + ≥ 0.
ka3 + 1 kb3 + 1 kc 3 + 1
Using the substitution
a = ex , b = ey, c = ez ,
we need to show that
f (x) + g( y) + g(z) ≥ 3 f (s),
where
x + y +z
s= =0
3
and
e4u − eu
f (u) = , u ∈ R.
ke3u + 1
From
ke6u + 2(k + 2)e3u − 1
f 0 (t) = ,
(ke3u + 1)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
v
p
3 −k − 2 + (k + 1)(k + 4)
u
t
s0 = ln r0 < 0, r0 = ∈ (0, 1).
k
Also, we have
t · h(t)
f 00 (u) = ,
(kt 3 + 1)3
where
h(t) = k2 t 9 − k(4k + 1)t 6 + (13k + 16)t 3 − 1, t = eu .
If h(t) > p
0 for t ∈ [r0 , 1], then f is convex on [s0 , 0]. We will prove this only for
k = 2 + 2 2, when r0 ≈ 0.415 and h(t) ≥ 0 for t ∈ [t 1 , t 2 ], where t 1 ≈ 0.2345 and
t 2 ≈ 1.02. Since [r0 , 1] ⊂ [t 1 , t 2 ], the conclusion follows. By the LPCF-Theorem,
we only need to prove the original inequality for b = c. Due to homogeneity, we
may consider that b = c = 1. Thus, we need to show that
a3 2 a+2
+ ≥ ,
ka + 1 a + k
2 k+1
which is equivalent to the obvious inequality
(a − 1)2 [a2 − (k − 2)a + 2] ≥ 0.
p
For k = 2 + 2 2, this inequality has the form
p
(a − 1)2 (a − 2)2 ≥ 0.
p
The equality holds for a = b = c. If k = 2 + 2 2, then the equality holds also for
a
p =b=c
2
(or any cyclic permutation).
Partially Convex Function Method 281

P 3.35. If a, b, c, d, e are positive real numbers so that abcde = 1, then

1 1 1 1 1 1
 ‹  ‹
2 + + ··· + ≥3 + + ··· + .
a+1 b+1 e+1 a+2 b+2 e+2

(Vasile C., 2012)

Solution. Write the inequality as

1−a 1− b 1−c 1−d 1−e


+ + + + ≥ 0.
(a + 1)(a + 2) (b + 1)(b + 2) (c + 1)(c + 2) (d + 1)(d + 2) (e + 1)(e + 2)

Using the substitution

a = ex , b = ey, c = ez , d = et , e = ew,

we need to show that

f (x) + f ( y) + f (z) + f (t) + f (w) ≥ 5 f (s),

where
x + y +z+ t +w
s= =0
5
and
1 − eu
f (u) = u , u ∈ R.
(e + 1)(eu + 2)
From
eu (e2u − 2eu − 5)
f 0 (u) = ,
(eu + 1)2 (eu + 2)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
p
s0 = ln(1 + 6) < 2, s < s0 .

Also, we have
t · h(t)
f 00 (u) = , t = eu ,
(t + 1)3 (t + 2)3
where
h(t) = −t 4 + 7t 3 + 21t 2 + 7t − 10.
We will show that h(t) > 0 for t ∈ [1, 2], hence f is convex on [0, s0 ]. We have

h(t) ≥ −2t 3 + 7t 3 + 21t 2 + 7t − 10 = 5t 3 + 21t 2 + 7t − 10 > 0.

By the RPCF-Theorem, we only need to prove the original inequality for

b = c = d = e := t, a = 1/t 4 , t ≥ 1.
282 Vasile Cîrtoaje

Write this inequality as

t 4 (t 4 − 1) 4(t − 1)
≥ ,
(t 4 + 1)(2t 4 + 1) (t + 1)(t + 2)

which is true if

t 4 (t + 1)(t + 2)(t 3 + t 2 + t + 1) ≥ 4(t 4 + 1)(2t 4 + 1).

Since
(t 4 + 1)(2t 4 + 1) = 2t 8 + 3t 4 + 1 ≤ 2t 4 (t 4 + 2),
it suffices to show that

(t + 1)(t + 2)(t 3 + t 2 + t + 1) ≥ 8(t 4 + 2).

This inequality is equivalent to

t 5 − 4t 4 + 6t 3 + 6t 2 + 5t − 14 ≥ 0,

t(t − 1)4 + 10(t 2 − 1) + 4(t − 1) ≥ 0.


The equality holds for a = b = c = d = e = 1.

P 3.36. If a1 , a2 , . . . , a14 are positive real numbers so that a1 a2 · · · a14 = 1, then

1 1 1 1 1 1
 ‹  ‹
3 + + ··· + ≥2 + + ··· + .
2a1 + 1 2a2 + 1 2a14 + 1 a1 + 1 a2 + 1 a14 + 1

(Vasile C., 2012)

Solution. Write the inequality as

1 − a1 1 − a2 1 − a14
+ + ··· + ≥ 0.
(a1 + 1)(2a1 + 1) (a2 + 1)(2a2 + 1) (a14 + 1)(2a14 + 1)

Using the substitutions ai = e x i for i = 1, 2, . . . , 14, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x 14 ) ≥ 14 f (s),

where
x 1 + x 2 + · · · + x 14
s= =0
14
and
1 − eu
f (u) = , u ∈ R.
(eu + 1)(2eu + 1)
Partially Convex Function Method 283

From
2eu (e2u − 2eu − 2)
f 0 (u) = ,
(eu + 1)2 (2eu + 1)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
p
s0 = ln(1 + 3) < 2, s < s0 .

Also, we have
2t · h(t)
f 00 (u) = , t = eu ,
(t + 1)3 (2t + 1)3
where
h(t) = −2t 4 + 11t 3 + 15t 2 + 2t − 2.
We will show that h(t) > 0 for t ∈ [1, 2], hence f is convex on [0, s0 ]. We have

h(t) ≥ −4t 3 + 11t 3 + 15t 2 + 2t − 2 = 7t 3 + 15t 2 + 2t − 2 > 0.

By the RPCF-Theorem, we only need to prove the original inequality for

a2 = a3 = · · · = a14 := t, a1 = 1/t 13 , t ≥ 1.

Write this inequality as

t 13 (t 13 − 1) 13(t − 1)
≥ .
(t + 1)(t + 2) (t + 1)(2t + 1)
13 13

Since
(t 13 + 1)(t 13 + 2) = t 26 + 3t 13 + 2 ≤ t 13 (t 13 + 5),
it suffices to show that
t 13 − 1 13(t − 1)
≥ ,
t 13 + 5 (t + 1)(2t + 1)
which is equivalent to

t 13 (t 2 − 5t + 7) − t 2 − 34t + 32 ≥ 0.

Substituting
t = 1 + x, x ≥ 0,
the inequality becomes

(1 + x)13 (x 2 − 3x + 3) − x 2 − 36x − 3 ≥ 0.

Since
(1 + x)13 ≥ 1 + 13x + 78x 2 ,
it suffices to show that

(78x 2 + 13x + 1)(x 2 − 3x + 3) − x 2 − 36x − 3 ≥ 0.


284 Vasile Cîrtoaje

This inequality, equivalent to

x 2 (78x 2 − 221x + 196) ≥ 0,

is true since

78x 2 − 221x + 196 ≥ 64x 2 − 224x + 196 = 4(4x − 7)2 ≥ 0.

The equality holds for a1 = a2 = · · · = a14 = 1.

P 3.37. Let a1 , a2 , . . . , a8 be positive real numbers so that a1 a2 · · · a8 = 1. If k > 1,


then
1 1 1 1 1 1
 ‹  ‹
(k + 1) + + ··· + ≥2 + + ··· + .
ka1 + 1 ka2 + 1 ka8 + 1 a1 + 1 a2 + 1 a8 + 1

(Vasile C., 2012)

Solution. Write the inequality as

1 − a1 1 − a2 1 − a8
+ + ··· + ≥ 0.
(a1 + 1)(ka1 + 1) (a2 + 1)(ka2 + 1) (a8 + 1)(ka8 + 1)

Using the substitutions ai = e x i for i = 1, 2, . . . , 8, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x 8 ) ≥ 8 f (s),

where
x1 + x2 + · · · + x8
s= =0
8
and
1 − eu
f (u) = , u ∈ R.
(eu + 1)(keu + 1)
From
eu (ke2u − 2keu − k − 2)
f 0 (u) = ,
(eu + 1)2 (keu + 1)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
‚ v Œ
t 2
s0 = ln 1 + 2+ < 2, s < s0 .
k

Also, we have
t · h(t)
f 00 (u) = , t = eu ,
(t + 1)3 (kt + 1)3
Partially Convex Function Method 285

where

h(t) = −k2 t 4 + k(5k + 1)t 3 + 3k(k + 3)t 2 + (k2 − k + 2)t − k − 2.

We will show that h(t) > 0 for t ∈ [1, 2], hence f is convex on [0, s0 ]. We have

h(t) > −2k2 t 3 + k(5k + 1)t 3 + 3k(k + 3)t 2 + (k2 − k + 2)t − k − 2


= k(3k + 1)t 3 + 3k(k + 3)t 2 + (k2 − k + 2)t − k − 2
> 3k(k + 3) + (k2 − k + 2) − k − 2 > 0.

By the RPCF-Theorem, we only need to prove the original inequality for

a2 = a3 = · · · = a8 := t, a1 = 1/t 7 , t ≥ 1.

Write this inequality as

t 7 (t 7 − 1) 7(t − 1)
≥ .
(t + 1)(t + k) (t + 1)(kt + 1)
7 7

Since
(t 7 + 1)(t 7 + k) = t 14 + (k + 1)t 7 + k ≤ t 7 (t 7 + 2k + 1),
it suffices to show that
t7 − 1 7(t − 1)
≥ ,
t 7 + 2k + 1 (t + 1)(kt + 1)

which is equivalent to
k(t − 1)P(t) + Q(t) ≥ 0,
where
P(t) = t(t + 1)(t 6 + t 5 + t 4 + t 3 + t 2 + t + 1) − 14,
Q(t) = (t + 1)(t 7 − 1) − 7(t − 1)(t 7 + 1).
Since (t − 1)P(t) ≥ 0 for t ≥ 1, it suffices to consider the case k = 1. So, we need
to show that
t 7 − 1 7(t − 1)
≥ ,
t7 + 3 (t + 1)2
which is equivalent to

t 7 (t 2 − 5t + 8) − t 2 − 23t + 20 ≥ 0.

Substituting
t = 1 + x, x ≥ 0,
the inequality becomes

(1 + x)7 (x 2 − 3x + 4) − x 2 − 25x − 4 ≥ 0.
286 Vasile Cîrtoaje

Since
(1 + x)7 ≥ 1 + 7x + 21x 2 ,
it suffices to show that

(21x 2 + 7x + 1)(x 2 − 3x + 4) − x 2 − 25x − 4 ≥ 0.

This inequality, equivalent to

x 2 (21x 2 − 56x + 63) ≥ 0.

is true since

21x 2 − 56x + 63 > 16x 2 − 56x + 49 = (4x − 7)2 ≥ 0.

The equality holds for a1 = a2 = · · · = a8 = 1.

P 3.38. If a1 , a2 , . . . , a9 are positive real numbers so that a1 a2 · · · a9 = 1, then

1 1 1 1 1 1
+ + ··· + ≥ + + ··· + .
2a1 + 1 2a2 + 1 2a9 + 1 a1 + 2 a2 + 2 a9 + 2

(Vasile C., 2012)

Solution. Write the inequality as

1 − a1 1 − a2 1 − a9
+ + ··· + ≥ 0.
(2a1 + 1)(a1 + 2) (2a2 + 1)(a2 + 2) (2a9 + 1)(a9 + 2)

Using the substitutions ai = e x i for i = 1, 2, . . . , 9, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x 9 ) ≥ 9 f (s),

where
x1 + x2 + · · · + x9
s= =0
9
and
1 − eu
f (u) = , u ∈ R.
(2eu + 1)(eu + 2)
From
eu (2e2u − 4eu − 7)
f 0 (u) = ,
(2eu + 1)2 (eu + 2)2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where
 p 
3 2
s0 = ln 1 + < 2, s < s0 .
2
Partially Convex Function Method 287

Also, we have
t · h(t)
f 00 (u) = , t = eu ,
(2t + 1)3 (t + 2)3
where
h(t) = −4t 4 + 26t 3 + 54t 2 + 19t − 14.
We will show that h(t) > 0 for t ∈ [1, 2], hence f is convex on [0, s0 ]. We have

h(t) ≥ −8t 3 + 26t 3 + 54t 2 + 19t − 14 = 18t 3 + 54t 2 + 19t − 14 > 0.

By the RPCF-Theorem, we only need to prove the original inequality for

a2 = a3 = · · · = a9 := t, a1 = 1/t 8 , t ≥ 1.

Write this inequality as


t 8 (t 8 − 1) 8(t − 1)
≥ .
(t 8 + 2)(2t 8 + 1) (2t + 1)(t + 2)
Since
(t 8 + 2)(2t 8 + 1) = 2t 16 + 5t 8 + 2 ≤ t 8 (2t 8 + 7),
it suffices to show that
t8 − 1 8(t − 1)
≥ ,
2t + 7 (2t + 1)(t + 2)
8

which is equivalent to

t 8 (2t 2 − 11t + 18) − 2t 2 − 61t + 54 ≥ 0.

Substituting
t = 1 + x, x ≥ 0,
the inequality becomes

(1 + x)8 (2x 2 − 7x + 9) − 2x 2 − 65x − 9 ≥ 0.

Since
(1 + x)8 ≥ 1 + 8x + 28x 2 ,
it suffices to show that

(28x 2 + 8x + 1)(2x 2 − 7x + 9) − 2x 2 − 65x − 9 ≥ 0.

This inequality, equivalent to

x 2 (56x 2 − 180x + 196) ≥ 0.

is true since

56x 2 − 180x + 196 ≥ 49x 2 − 196x + 196 = 49(x − 2)2 ≥ 0.

The equality holds for a1 = a2 = · · · = a9 = 1.


288 Vasile Cîrtoaje

P 3.39. If a1 , a2 , . . . , an are real numbers so that


a1 , a2 , . . . , an ≤ π, a1 + a2 + · · · + an = π,
then
π
cos a1 + cos a2 + · · · + cos an ≤ n cos .
n
(Vasile C., 2000
Solution. Write the inequality as
a1 + a2 + · · · + an π
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = ,
n n
where
f (u) = − cos u, u ∈ I = [−(n − 2)π, π].
Let
s0 = 0 < s.
We see that f is increasing on [s0 , π] = I≥s0 and f (u) ≥ f (s0 ) = −1 for u ∈ I. In
addition, f is convex on [s0 , s]. Thus, by the LPCF-Theorem, we only need to prove
that g(x) ≤ 0, where
g(x) = cos x + (n − 1) cos y − n cos s, x + (n − 1) y = π, π ≥ x ≥ s ≥ y ≥ 0.
−1
Since y 0 = , we get
n−1
x−y x+y
g 0 (x) = − sin x + sin y = −2 sin cos .
2 2
We have g 0 (x) ≤ 0 because
x+y x + (n − 1) y π
0< ≤ =
2 2 2
and
x−y π
0≤ < .
2 2
From g ≤ 0, it follows that g is decreasing, hence g(x) ≤ g(s) = 0.
0
π
The equality holds for a1 = a2 = · · · = an = . If n = 2, then the inequality is an
n
identity.
Remark. In the same manner, we can prove the following generalization:
• If a1 , a2 , . . . , an are real numbers so that
a1 + a2 + · · · + an π
a1 , a2 , . . . , an ≤ π, = s, 0<s≤ ,
n 4
then
cos a1 + cos a2 + · · · + cos an ≤ n cos s,
with equality for a1 = a2 = · · · = an = s.
Partially Convex Function Method 289

P 3.40. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that

−1
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
a12 a22 an2
+ + ··· + ≤ n.
a12 − a1 + 1 a22 − a2 + 1 an2 − an + 1

(Vasile Cirtoaje, 2012)

Solution. Write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1−u −1 n2 − n − 1
• ˜
f (u) = 2 , u∈I= , .
u −u+1 n−2 n−2
Let s0 = 2. We have s < s0 and

min f (u) = f (s0 )


u∈I

because
1−u 1 (u − 2)2
f (u) − f (2) = + = ≥ 0.
u2 − u + 1 3 3(u2 − u + 1)
From
u(u − 2)
f 0 (u) = ,
(u2− u + 1)2
2(3u2 − u3 − 1) 2u2 (2 − u) + 2(u2 − 1)
f 00 (u) = = ,
(u2 − u + 1)3 (u2 − u + 1)3
it follows that f is convex on [1, s0 ]. However, we can’t apply the RPCF-Theorem
in its original form because f is not decreasing on I≤s0 . According to Theorem 1,
we may replace this condition with ns − (n − 1)s0 ≤ inf I. Indeed, we have

−1
ns − (n − 1)s0 = n − 2(n − 1) = −n + 2 ≤ = inf I.
n−2
So, it suffices to show that f (x) + (n − 1) f ( y) ≥ n f (1) for all x, y ∈ I so that

x + (n − 1) y = n.

According to Note 1, we only need to show that h(x, y) ≥ 0, where

f (u) − f (1) g(x) − g( y)


g(u) = , h(x, y) = .
u−1 x−y
290 Vasile Cîrtoaje

We have
−1
g(u) = ,
u2 −u+1
x + y −1 (n − 2)x + 1
h(x, y) = = ≥ 0.
(x 2 − x + 1)( y 2 − y + 1) (n − 1)(x 2 − x + 1)( y 2 − y + 1)
The equality holds for a1 = a2 = · · · = an = 1, and also for

−1 n−1
a1 = , a2 = a3 = · · · = a n =
n−2 n−2
(or any cyclic permutation).

P 3.41. If a1 , a2 , . . . , an (n ≥ 3) are nonzero real numbers so that

−n
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
1 1 1 1 1 1
2
+ 2 + ··· + 2 ≥ + + ··· + .
a1 a2 an a1 a2 an

(Vasile Cirtoaje, 2012)

Solution. According to P 2.25-(a) in Volume 1, the inequality is true for n = 3.


Assume further that n ≥ 4 and write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1 1 −n n(2n − 3)
• ˜
f (u) = 2 − , u∈I= , \ {0}.
u u n−2 n−2
Let
s0 = 2, s < s0 .
From
1 1 1 (u − 2)2
f (u) − f (2) = − + = ≥ 0,
u2 u 4 4u2
it follows that
min f (u) = f (s0 ),
u∈I

while from
u−2 2(3 − u)
f 0 (u) = , f 00 (u) = ,
u3 u4
Partially Convex Function Method 291

it follows that f is convex on [s, s0 ]. However, we can’t apply the RPCF-Theorem


because f is not decreasing on I≤s0 . According to Theorem 1 and Note 6, we may
replace this condition with ns − (n − 1)s0 ≤ inf I. For n ≥ 4, we have

−n
ns − (n − 1)s0 = n − 2(n − 1) = −n + 2 ≤ = inf I.
n−2

So, according to Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ∈ I so that
x + (n − 1) y = n. We have

f (u) − f (1) −1
g(u) = = 2,
u−1 u

g(x) − g( y) x+y (n − 2)x + n


h(x, y) = = 2 2 = ≥ 0.
x−y x y (n − 1)x 2 y 2
The proof is completed. By Note 3, the equality holds for a1 = a2 = · · · = an = 1,
and also for
−n n
a1 = , a2 = a3 = · · · = a n =
n−2 n−2
(or any cyclic permutation).
Remark. Similarly, we can prove the following generalization:
−n
• Let a1 , a2 , . . . , an ≥ so that a1 + a2 + · · · + an = n. If n ≥ 3 and k ≥ 0, then
n−2

1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0,
k + a1 k + a2
2 2
k + an2

with equality for a1 = a2 = · · · = an = 1, and also for

−n n
a1 = , a2 = a3 = · · · = a n =
n−2 n−2

(or any cyclic permutation).

P 3.42. If a1 , a2 , . . . , an ≥ −1 so that a1 + a2 + · · · + an = n, then


 
1 1 1 1 1 1
 ‹
(n + 1) 2 + 2 + · · · + 2 ≥ 2n + (n − 1) + + ··· + .
a1 a2 an a1 a2 an

(Vasile C., 2013)


292 Vasile Cîrtoaje

Solution. Write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
n+1 n−1
f (u) = − , u ∈ I = [−1, 2n − 1] \ {0}.
u2 u
Let
2(n + 1)
s0 = ∈ I, s < s0 .
n−1
Since
[(n − 1)u − 2(n + 1)]2
f (u) − f (s0 ) = ≥ 0,
4(n + 1)u2
we have
min f (u) = f (s0 ).
u∈I

From
(n − 1)u − 2(n + 1) 6(n + 1) − 2(n − 1)u
f 0 (u) = , f 00 (u) = ,
u3 u4
it follows that f is convex on [1, s0 ]. Since f is not decreasing on I≤s0 , according
to Theorem 1 and Note 6, we may replace this condition in RPCF-Theorem with
ns − (n − 1)s0 ≤ inf I. We have

ns − (n − 1)s0 = n − 2(n + 1) = −n − 2 < −1 = inf I.

According to Note 1, we only need to show that h(x, y) ≥ 0 for −1 ≤ x ≤ 1 ≤ y


and x + (n − 1) y = n. We have

f (u) − f (1) 2 n+1


g(u) = =− − 2
u−1 u u
and

g(x) − g( y) 2x y + (n + 1)(x + y) (x + 1)(n2 + n − 2x)


h(x, y) = = = ≥ 0.
x−y x2 y2 (n − 1)x 2 y 2

According to Note 4, the equality holds for a1 = a2 = · · · = an = 1, and also for

n+1
a1 = −1, a2 = · · · = an =
n−1
(or any cyclic permutation).
Partially Convex Function Method 293

P 3.43. If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that

−(3n − 2)
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
then
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
(1 + a1 )2 (1 + a2 )2 (1 + an )2
(Vasile C., 2014)

Solution. According to P 2.25-(b) in Volume 1, the inequality is true for n = 3.


Assume further that n ≥ 4 and write the inequality as
a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1−u −(3n − 2) 4n2 − 7n + 2
• ˜
f (u) = , u∈I= , \ {−1}.
(1 + u)2 n−2 n−2
Let
s0 = 3, s < s0 .
From
1−u 1 (u − 3)2
f (u) − f (3) = + = ≥ 0,
(1 + u)2 8 8(u + 1)2
it follows that
min f (u) = f (s0 ).
u∈I

From
u−3 2(5 − u)
f 0 (u) = , f 00 (u) = ,
(u + 1)3 (u + 1)4
it follows that f is convex on [1, s0 ]. We can’t apply the RPCF-Theorem in its original
form because f is not decreasing on I≤s0 . However, according to Theorem 1 and
Note 6, we may replace this condition with ns − (n − 1)s0 ≤ inf I. Indeed, for n ≥ 4,
we have
−(3n − 2)
ns − (n − 1)s0 = n − 3(n − 1) = −2n + 3 ≤ = inf I.
n−2
According to Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ∈ I so that
x ≤ 1 ≤ y and x + (n − 1) y = n. We have

f (u) − f (1) −1
g(u) = = ,
u−1 (u + 1)2

g(x) − g( y) x + y +2 (n − 2)x + 3n − 2
h(x, y) = = = ≥ 0.
x−y (x + 1) ( y + 1)
2 2 (n − 1)(x + 1)2 ( y + 1)2
294 Vasile Cîrtoaje

In accordance with Note 3, the equality holds for a1 = a2 = · · · = an = 1, and also


for
−(3n − 2) n+2
a1 = , a2 = a3 = · · · = an =
n−2 n−2
(or any cyclic permutation).

P 3.44. Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n.


2
If n ≥ 3 and k ≥ 2 − , then
n
1 − a1 1 − a2 1 − an
+ + · · · + ≥ 0.
(1 − ka1 )2 (1 − ka2 )2 (1 − kan )2

(Vasile C., 2012)

Solution. According to P 3.99 in Volume 1, the inequality is true for n = 3. Assume


further that n ≥ 4 and write the inequality as

a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
1−u
f (u) = , u ∈ I = [0, n] \ {1/k}.
(1 − ku)2
Let
s0 = 2 − 1/k, 1 = s < s0 .
Since

1−u 1 (ku − 2k + 1)2


f (u) − f (s0 ) = + = ≥ 0,
(1 − ku)2 4k(k − 1) 4k(k − 1)(1 − ku)2

we have
min f (u) = f (s0 ).
u∈I

From
ku − 2k + 1 2k(−ku + 3k − 2)
f 0 (u) = , f 00 (u) = ,
(ku − 1)3 (1 − ku)4
it follows that f is convex on [1, s0 ]. We can’t apply the RPCF-Theorem because f
is not decreasing on I≤s0 . According to Theorem 1 and Note 6, we may replace this
condition with ns − (n − 1)s0 ≤ inf I. Indeed, we have

3n − 4 4−n
ns − (n − 1)s0 ≤ n − (n − 1) · = ≤ 0 = inf I.
2(n − 1) 2
Partially Convex Function Method 295

So, it suffices to show that f (x) + (n − 1) f ( y) ≥ n f (1) for all x, y ∈ I so that


x ≤ 1 ≤ y and x + (n − 1) y = n. According to Note 1, we only need to show that
h(x, y) ≥ 0, where

f (u) − f (1) g(x) − g( y)


g(u) = , h(x, y) = .
u−1 x−y

Since
−1 k[k(x + y) − 2]
g(u) = , h(x, y) = ,
(1 − ku)2 (1 − kx)2 (1 − k y)2
we need to show that k(x + y) − 2 ≥ 0. Indeed, we have

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


≥ −1 = − = ≥ 0.
2 n n n n
2
The equality holds for a1 = a2 = · · · = an = 1. If k = 2 − , then the equality also
n
holds for
n
a1 = 0, a2 = a3 = · · · = an =
n−1
(or any cyclic permutation).
296 Vasile Cîrtoaje
Chapter 4

Partially Convex Function Method


for Ordered Variables

4.1 Theoretical Basis


The following statement is known as Right Partially Convex Function Theorem for
Ordered Variables (RPCF-OV Theorem).
RPCF-OV Theorem (Vasile Cirtoaje, 2014). Let f be a real function defined on an
interval I and convex on [s, s0 ], where s, s0 ∈ I, s < s0 . In addition, f is decreasing on
I≤s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≤ a2 ≤ · · · ≤ am ≤ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so that x ≤ s ≤ y and x + (n − m) y = (1 + n − m)s.

Proof. For
a1 = x, a2 = · · · = am = s, am+1 = · · · = an = y,
the inequality
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s)
becomes
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s);

297
298 Vasile Cîrtoaje

therefore, the necessity is obvious. By Lemma from Chapter 3, to prove the suffi-
ciency, it suffices to consider that a1 , a2 , . . . , an ∈ J, where

J = I≤s0 .

Because f is convex on J≥s , the desired inequality follows from HCF-OV Theorem
applied to the interval J.
Similarly, we can prove Left Partially Convex Function Theorem for Ordered Vari-
ables (LPCF-OV Theorem).
LPCF-OV Theorem. Let f be a real function defined on an interval I and convex on
[s0 , s], where s0 , s ∈ I, s0 < s. In addition, f is increasing on I≥s0 and f (u) ≥ f (s0 )
for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≥ a2 ≥ · · · ≥ am ≥ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so that x ≥ s ≥ y and x + (n − m) y = (1 + n − m)s.
The RPCF-OV Theorem and the LPCF-OV Theorems are respectively generaliza-
tions of the RPCF Theorem and LPCF Theorem, because the last theorems can be
obtained from the first theorems for m = 1.

Note 1. Let us denote


f (u) − f (s) g(x) − g( y)
g(u) = , h(x, y) = .
u−s x−y

We may replace the hypothesis condition in the RPCF-OV Theorem and the LPCF-OV
Theorem, namely
f (x) + m f ( y) ≥ (1 + m) f (s),
by the condition
h(x, y) ≥ 0 for all x, y ∈ I so that x + m y = (1 + m)s.

Note 2. Assume that f is differentiable on I, and let

f 0 (x) − f 0 ( y)
H(x, y) = .
x−y
PCF Method for Ordered Variables 299

The desired inequality of Jensen’s type in the RPCF-OV Theorem and the LPCF-OV
Theorem holds true by replacing the hypothesis

f (x) + m f ( y) ≥ (1 + m) f (s)

with the more restrictive condition


H(x, y) ≥ 0 for all x, y ∈ I so that x + m y = (1 + m)s.

Note 3. The desired inequalities in the RPCF-OV Theorem and the LPCF-OV Theo-
rem become equalities for

a1 = a2 = · · · = an = s.

In addition, if there exist x, y ∈ I so that

x + (n − m) y = (1 + n − m)s, f (x) + (n − m) f ( y) = (1 + n − m) f (s), x 6= y,

then the equality holds also for

a1 = x, a2 = · · · = am = s, am+1 = · · · = an = y

(or any cyclic permutation). Notice that these equality conditions are equivalent to

x + (n − m) y = (1 + n − m)s, h(x, y) = 0

(x < y for RHCF-OV Theorem, and x > y for LHCF-OV Theorem).

Note 4. The RPCF-OV Theorem is also valid in the case where f is defined on
I \ {u0 }, where u0 is an interior point of I so that u0 > s0. Similarly, LPCF Theorem
is also valid in the case in which f is defined on I \ {u0 }, where u0 is an interior
point of I so that u0 < s0.

Note 5. The RPCF-Theorem holds true by replacing the condition


f is decreasing on I≤s0
with
ns − (n − 1)s0 ≤ inf I.
More precisely, the following theorem holds:
Theorem 1. Let f be a function defined on a real interval I, convex on [s, s0 ] and
satisfying
min f (u) = f (s0 ),
u∈I≥s

where
s, s0 ∈ I, s < s0 , (1 + n − m)s − (n − m)s0 ≤ inf I.
The inequality
a + a + ··· + a 
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
300 Vasile Cîrtoaje

holds for all a1 , a2 , . . . , an ∈ I satisfying

a1 + a2 + · · · + an = ns

and
a1 ≤ a2 ≤ · · · ≤ am ≤ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so that x ≤ s ≤ y and x + (n − m) y = (1 + n − m)s.
The proof of this theorem is similar to the one of Theorem 1 from chapter 3.

Similarly, the LPCF-Theorem holds true by replacing the condition


f is increasing on I≥s0
with

ns − (n − 1)s0 ≥ sup I.
More precisely, the following theorem holds:
Theorem 2. Let f be a function defined on a real interval I, convex on [s0 , s] and
satisfying
min f (u) = f (s0 ),
u∈I≤s

where
s, s0 ∈ I, s > s0 , (1 + n − m)s − (n − m)s0 ≥ sup 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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≥ a2 ≥ · · · ≥ am ≥ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I so that x ≥ s ≥ y and x + (n − m) y = (1 + n − m)s.

Note 6. Theorem 1 is also valid in the case in which f is defined on I \ {u0 }, where
/ [s, s0 ]. Similarly, Theorem 2 is also valid in
u0 is an interior point of I so that u0 ∈
the case in which f is defined on I \ {u0 }, where u0 is an interior point of I so that
/ [s0 , s].
u0 ∈
PCF Method for Ordered Variables 301

Note 7. We can extend weighted Jensen’s inequality to right and left partially con-
vex functions with ordered variables establishing the WRPCF-OV Theorem and the
WLPCF-OV Theorem (Vasile Cirtoaje, 2014).
WRPCF-OV Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1,

and let f be a real function defined on an interval I and convex on [s, s0 ], where
s, s0 ∈ int(I), s < s0 . In addition, f is decreasing on I≤s0 and f (u) ≥ f (s0 ) for u ∈ I.
The inequality

p1 f (x 1 ) + p2 f (x 2 ) + · · · + pn f (x n ) ≥ f (p1 x 1 + p2 x 2 + · · · + pn x n )

holds for all x 1 , x 2 , . . . , x n ∈ I so that p1 x 1 + p2 x 2 + · · · + pn x n = s and

x1 ≤ x2 ≤ · · · ≤ x n, x m ≤ s, m ∈ {1, 2, . . . , n − 1},

if and only if
f (x) + k f ( y) ≥ (1 + k) f (s)
for all x, y ∈ I satisfying

x ≤ s ≤ y, x + k y = (1 + k)s,

where
pm+1 + pm+2 + · · · + pn
k= .
p1

WLPCF-OV Theorem. Let p1 , p2 , . . . , pn be positive real numbers so that

p1 + p2 + · · · + pn = 1,

and let f be a real function defined on an interval I and convex on [s0 , s], where
s0 , s ∈ I, s0 < s. In addition, f is increasing on I≥s0 and f (u) ≥ f (s0 ) for u ∈ I. The
inequality

p1 f (x 1 ) + p2 f (x 2 ) + · · · + pn f (x n ) ≥ f (p1 x 1 + p2 x 2 + · · · + pn x n )

holds for all x 1 , x 2 , . . . , x n ∈ I so that p1 x 1 + p2 x 2 + · · · + pn x n = s and

x1 ≥ x2 ≥ · · · ≥ x n, x m ≥ s, m ∈ {1, 2, . . . , n − 1},

if and only if
f (x) + k f ( y) ≥ (1 + k) f (s)
for all x, y ∈ I satisfying

x ≥ s ≥ y, x + k y = (1 + k)s,
302 Vasile Cîrtoaje

where
pm+1 + pm+2 + · · · + pn
k= .
p1

For the most commonly used case

1
p1 = p2 = · · · = pn = ,
n
the WRPCF-OV Theorem and the WLPCF-OV Theorem yield the RPCF-OV Theorem
and the LPCF-OV Theorem, respectively.
PCF Method for Ordered Variables 303

4.2 Applications

4.1. If a, b, c, d are real numbers so that

a ≤ 1 ≤ b ≤ c ≤ d, a + b + c + d = 4,

then
a b c d
+ 2 + 2 + 2 ≤ 1.
3a2 + 1 3b + 1 3c + 1 3d + 1

4.2. If a, b, c, d are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, a + b + c + d = 4,

then
16a − 5 16b − 5 16c − 5 16d − 5 4
+ + + ≤ .
32a + 1 32b + 1 32c + 1 32d + 1 3
2 2 2 2

4.3. If a, b, c, d, e are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d ≥ e, a + b + c + d + e = 5,

then
18a − 5 18b − 5 18c − 5 18d − 5 18e − 5
+ + + + ≤ 5.
12a2 + 1 12b2 + 1 12c 2 + 1 12d 2 + 1 12e2 + 1

4.4. If a, b, c, d, e are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d ≥ e, a + b + c + d + e = 5,

then
a(a − 1) b(b − 1) c(c − 1) d(d − 1) e(e − 1)
+ + 2 + + 2 ≥ 0.
3a2 + 4 3b2 + 4 3c + 4 3d 2 + 4 3e + 4

4.5. Let a1 , a2 , . . . , a2n 6= −k be real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n.

n+1
If k ≥ p , then
2 n

a1 (a1 − 1) a2 (a2 − 1) a2n (a2n − 1)


+ + ··· + ≥ 0.
(a1 + k) 2 (a2 + k)2 (a2n + k)2
304 Vasile Cîrtoaje

4.6. Let a1 , a2 , . . . , a2n 6= −k be real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n.

n+1
If k ≥ 1 + p , then
n

a12 − 1 a22 − 1 2
a2n −1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (a2n + k)2

4.7. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ 1 ≥ a2 ≥ · · · ≥ an , a1 + a2 + · · · + an = n,

then
3/a1 3/a2
a1 + a2 + · · · + an3/an ≤ n.

4.8. If a1 , a2 , . . . , a11 are real numbers so that

a1 ≥ a2 ≥ 1 ≥ a3 ≥ · · · ≥ a11 , a1 + a2 + · · · + a11 = 11,

then
(1 − a1 + a12 )(1 − a2 + a22 ) · · · (1 − a11 + a11
2
) ≥ 1.

4.9. If a1 , a2 , . . . , a8 are nonzero real numbers so that

a1 ≥ a2 ≥ a3 ≥ a4 ≥ 1 ≥ a5 ≥ a6 ≥ a7 ≥ a8 , a1 + a2 + · · · + a8 = 8,

then  
1 1 1 1 1 1
 ‹
5 2 + 2 + · · · + 2 + 72 ≥ 14 + + ··· + .
a1 a2 a8 a1 a2 a8

4.10. If a, b, c, d are positive real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, abcd = 1,

then
7 − 6a 7 − 6b 7 − 6c 7 − 6d 4
+ + + ≥ .
2+a 2 2+ b 2 2+c 2 2+d 2 3
PCF Method for Ordered Variables 305

4.11. If a, b, c are positive real numbers so that

a ≤ b ≤ 1 ≤ c, abc = 1,

then
7 − 4a 7 − 4b 7 − 4c
+ + ≥ 3.
2 + a2 2 + b2 2 + c 2

4.12. If a, b, c are positive real numbers so that

a ≥ 1 ≥ b ≥ c, abc = 1,

then
23 − 8a 23 − 8b 23 − 8c
+ + ≥ 9.
3 + 2a2 3 + 2b2 3 + 2c 2

4.13. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that p + 3q ≥ 1, then

1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

4.14. If a, b, c, d, e are real numbers so that

−2 ≤ a ≤ b ≤ 1 ≤ c ≤ d ≤ e, a + b + c + d + e = 5,

then
1 1 1 1 1 1 1 1 1 1
2
+ 2+ 2+ 2+ 2≥ + + + + .
a b c d e a b c d e
306 Vasile Cîrtoaje
PCF Method for Ordered Variables 307

4.3 Solutions

P 4.1. If a, b, c, d are real numbers so that

a ≤ 1 ≤ b ≤ c ≤ d, a + b + c + d = 4,

then
a b c d
+ + + ≤ 1.
3a2 + 1 3b2 + 1 3c 2 + 1 3d 2 + 1

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
−u
f (u) = , u ∈ R.
3u2 + 1
From
3u2 − 1
f 0 (u) = ,
(3u2 + 1)2
it follows that pf is increasing on (−∞, −s0 ] ∪ [s0 , ∞) and decreasing on [−s0 , s0 ],
where s0 = 1/ 3. Since
lim f (u) = 0
u→−∞

and f (s0 ) < 0, it follows that

min f (u) = f (s0 ).


u∈R

From
18u(1 − u2 )
f 00 (u) = ,
(3u2 + 1)3
it follows that f is convex on [0, 1], hence on [s0 , 1]. Therefore, we may apply the
LPCF-OV Theorem for n = 4 and m = 1. We only need to show that f (x) + f ( y) ≥
2 f (1) for all real x, y so that x + y = 2. Using Note 1, it suffices to prove that
h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

Indeed, we have
3u − 1
g(u) = ,
4(3u2 + 1)
3(1 + x + y − 3x y) 9(1 − x y)
h(x, y) = = ≥ 0,
4(3x + 1)(3 y + 1) 4(3x + 1)(3 y 2 + 1)
2 2 2
308 Vasile Cîrtoaje

since
4(1 − x y) = (x + y)2 − 4x y = (x − y)2 ≥ 0.
Thus, the proof is completed. The equality holds for a = b = c = d = 1.
Remark. Similarly, we can prove the following generalization:
• If a1 , a2 , . . . , an are real numbers so that

a1 ≤ 1 ≤ a2 ≤ · · · ≤ an , a1 + a2 + · · · + an = n,

then
a1 a an n
+ 22 + ··· + 2 ≤ ,
3a12+ 1 3a2 + 1 3an + 1 4
with equality for a1 = a2 = · · · = an = 1.

P 4.2. If a, b, c, d are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d, a + b + c + d = 4,

then
16a − 5 16b − 5 16c − 5 16d − 5 4
+ + + ≤ .
32a + 1 32b + 1 32c + 1 32d + 1 3
2 2 2 2

(Vasile C., 2012)

Solution. Write the inequality as

a+b+c+d
f (a) + f (b) + f (c) + f (d) ≥ 4 f (s), s= = 1,
4
where
5 − 16u
f (u) = , u ∈ R.
32u2 + 1
As shown in the proof of P 3.1, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈R

where p
5 + 33
s0 = ≈ 0.6715.
16
Therefore, we may apply the LPCF-OV Theorem for n = 4 and m = 2. We only need
to show that f (x) + 2 f ( y) ≥ 3 f (1) for all real x, y so that x + 2 y = 3. Using Note
1, it suffices to prove that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1
PCF Method for Ordered Variables 309

Indeed, we have
32(2u − 1)
g(u) = ,
3(32u2 + 1)
64(1 + 16x + 16 y − 32x y) 64(4x − 5)2
h(x, y) = = ≥ 0.
3(32x 2 + 1)(32 y 2 + 1) 3(32x 2 + 1)(32 y 2 + 1)
From x + 2 y = 3 and h(x, y) = 0, we get x = 5/4 and y = 7/8. Therefore, in
accordance with Note 3, the equality holds for a = b = c = d = 1, and also for
5 7
a= , b = 1, c=d= .
4 8

Remark. Similarly, we can prove the following generalization:


• If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that

a1 ≥ · · · ≥ an−2 ≥ 1 ≥ an−1 ≥ an , a1 + a2 + · · · + an = n,

then
16a1 − 5 16a2 − 5 16an − 5 n
+ + · · · + ≤ ,
32a12 + 1 32a22 + 1 32an2 + 1 3
with equality for a1 = a2 = · · · = an = 1, and also for
5 7
a1 = , a2 = · · · = an−2 = 1, an−1 = an = .
4 8

P 4.3. If a, b, c, d, e are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d ≥ e, a + b + c + d + e = 5,

then
18a − 5 18b − 5 18c − 5 18d − 5 18e − 5
+ + + + ≤ 5.
12a + 1 12b + 1 12c + 1 12d + 1 12e2 + 1
2 2 2 2

(Vasile C., 2012)

Solution. Write the inequality as


a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
5 − 18u
f (u) = , u ∈ R.
12u2 + 1
As shown in the proof of P 3.2, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈R
310 Vasile Cîrtoaje

where p
5 + 52
s0 = ≈ 0.678.
18
Therefore, applying the LPCF-OV Theorem for n = 5 and m = 3, we only need to
show that f (x) + 3 f ( y) ≥ 4 f (1) for all real x, y so that x + 3 y = 4. Using Note 1,
it suffices to prove that h(x, y) ≥ 0, where

g(x) − g( y) f (u) − f (1)


h(x, y) = , g(u) = .
x−y u−1

Indeed, we have
6(2u − 1)
g(u) = ,
12u2 + 1
12(1 + 6x + 6 y − 12x y) 12(2x − 3)2
h(x, y) = = ≥ 0.
(12x 2 + 1)(12 y 2 + 1) (12x 2 + 1)(12 y 2 + 1)
From x + 3 y = 4 and h(x, y) = 0, we get x = 3/2 and y = 5/6. Therefore, in
accordance with Note 3, the equality holds for a = b = c = d = e = 1, and also for

3 5
a= , b = 1, c=d=e= .
2 6

Remark. Similarly, we can prove the following generalization:


• If a1 , a2 , . . . , an (n ≥ 4) are real numbers so that

a1 ≥ · · · ≥ an−3 ≥ 1 ≥ an−2 ≥ an−1 ≥ an , a1 + a2 + · · · + an = n,

then
18a1 − 5 18a2 − 5 18an − 5
+ + ··· + ≤ n,
12a1 + 1 12a2 + 1
2 2
12an2 + 1
with equality for a1 = a2 = · · · = an = 1, and also for

3 5
a1 = , a2 = · · · = an−3 = 1, an−2 = an−1 = an = .
2 6

P 4.4. If a, b, c, d, e are real numbers so that

a ≥ b ≥ 1 ≥ c ≥ d ≥ e, a + b + c + d + e = 5,

then
a(a − 1) b(b − 1) c(c − 1) d(d − 1) e(e − 1)
+ + 2 + + 2 ≥ 0.
3a2 + 4 3b2 + 4 3c + 4 3d 2 + 4 3e + 4
(Vasile C., 2012)
PCF Method for Ordered Variables 311

Solution. Write the inequality as

a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
u2 − u
f (u) = , u ∈ R.
3u2 + 4
As shown in the proof of P 3.5, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈R

where p
−4 + 2 7
s0 = ≈ 0.43.
3
Therefore, we may apply the LPCF-OV Theorem for n = 5 and m = 2. We only need
to show that f (x) + 3 f ( y) ≥ 4 f (1) for all real x, y so that x + 3 y = 4. Using Note
1, it suffices to prove that h(x, y) ≥ 0. Indeed, we have

f (u) − f (1) u
g(u) = = 2 ,
u−1 3u + 4
g(x) − g( y) 4 − 3x y (x − 2)2
h(x, y) = = = ≥ 0.
x−y (3x 2 + 4)(3 y 2 + 4) (12x 2 + 1)(12 y 2 + 1)
From x + 3 y = 4 and h(x, y) = 0, we get x = 2 and y = 2/3. Therefore, in
accordance with Note 3, the equality holds for

a = b = c = d = e = 1,

and also for


2
a = 2, b = 1, c = d = e = .
3
Remark. Similarly, we can prove the following generalizations:
• If a1 , a2 , . . . , an (n ≥ 4) are real numbers so that

a1 ≥ · · · ≥ an−3 ≥ 1 ≥ an−2 ≥ an−1 ≥ an , a1 + a2 + · · · + an = n,

then
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + ··· + ≥ 0,
3a1 + 4
2
3a2 + 4
2
3an2 + 4
with equality for a1 = a2 = · · · = an = 1, and also for

2
a1 = 2, a2 = · · · = an−3 = 1, an−2 = an−1 = an = .
3
312 Vasile Cîrtoaje

• If a1 , a2 , . . . , an (n ≥ 3) are real numbers so that

a1 ≥ a2 ≥ 1 ≥ a3 ≥ · · · ≥ an , a1 + a2 + · · · + an = n,

then
a1 (a1 − 1) a2 (a2 − 1) an (an − 1)
+ + · · · + ≥ 0,
4(n − 2)a12 + (n − 1)2 4(n − 2)a22 + (n − 1)2 4(n − 2)an2 + (n − 1)2

with equality for a1 = a2 = · · · = an = 1, and also for


n−1 n−1
a1 = , a2 = 1, a3 = · · · = a n = .
2 2(n − 2)

P 4.5. Let a1 , a2 , . . . , a2n 6= −k be real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n.


n+1
If k ≥ p , then
2 n
a1 (a1 − 1) a2 (a2 − 1) a2n (a2n − 1)
+ + · · · + ≥ 0.
(a1 + k)2 (a2 + k)2 (a2n + k)2

(Vasile C., 2012)


Solution. Write the inequality as
a1 + a2 + · · · + a2n
f (a1 ) + f (a2 ) + · · · + f (a2n ) ≥ 2n f (s), s= = 1,
2n
where
u(u − 1)
f (u) = , u ∈ I = R \ {−k}.
(u + k)2
As shown in the proof of P 3.8, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈I

where
k
s0 = < 1.
2k + 1
Having in view Note 4, we may apply the LPCF-OV Theorem for 2n real numbers
and m = n. We only need to show that f (x) + n f ( y) ≥ (n + 1) f (1) for x, y ∈ I so
that x + n y = n + 1. Using Note 1, it suffices to prove that h(x, y) ≥ 0. We have
f (u) − f (1) u
g(u) = = ,
u−1 (u + k)2
PCF Method for Ordered Variables 313

g(x) − g( y) k2 − x y
h(x, y) = = ≥ 0,
x−y (x + k)2 ( y + k)2
because
(n + 1)2 (x + n y)2 (x − n y)2
k2 − x y ≥ −xy = −xy = ≥ 0.
4n 4n 4n
n+1
The equality holds for a1 = a2 = · · · = an = 1. If k = p , then the equality holds
2 n
also for
n+1 n+1
a1 = , a2 = · · · = an = 1, an+1 = · · · = a2n = .
2 2n

P 4.6. Let a1 , a2 , . . . , a2n 6= −k be real numbers so that

a1 ≥ · · · ≥ an ≥ 1 ≥ an+1 ≥ · · · ≥ a2n , a1 + a2 + · · · + a2n = 2n.


n+1
If k ≥ 1 + p , then
n
a12 − 1 a22 − 1 2
a2n −1
+ + ··· + ≥ 0.
(a1 + k)2 (a2 + k)2 (a2n + k)2

(Vasile C., 2012)


Solution. Write the inequality as
a1 + a2 + · · · + a2n
f (a1 ) + f (a2 ) + · · · + f (a2n ) ≥ 2n f (s), s= = 1,
2n
where
u2 − 1
f (u) = , u ∈ I = R \ {−k}.
(u + k)2
As shown in the proof of P 3.9, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈I

where
−1
s0 =
∈ (−1, 0).
k
According to Note 4, we may apply the LPCF-OV Theorem for 2n real numbers and
m = n. Thus, we only need to show that f (x) + n f ( y) ≥ (n + 1) f (1) for x, y ∈ I
so that x + n y = n + 1. Using Note 1, it suffices to prove that h(x, y) ≥ 0. We have
f (u) − f (1) u+1
g(u) = = ,
u−1 (u + k)2
314 Vasile Cîrtoaje

g(x) − g( y) (k − 1)2 − 1 − x − y − x y
h(x, y) = = ≥ 0,
x−y (x + k)2 ( y + k)2
because
(n + 1)2 (n y − 1)2
(k − 1)2 − 1 − x − y − x y ≥ −1− x − y − xy = ≥ 0.
n n
n+1
The equality holds for a1 = a2 = · · · = an = 1. If k = 1 + p , then the equality
n
holds also for
1
a1 = n, a2 = · · · = an = 1, an+1 = · · · = a2n = .
n

P 4.7. If a1 , a2 , . . . , an are positive real numbers so that

a1 ≥ 1 ≥ a2 ≥ · · · ≥ an , a1 + a2 + · · · + an = n,

then
3/a1 3/a2
a1 + a2 + · · · + an3/an ≤ n.

(Vasile C., 2012)

Solution. Rewrite the desired inequality as


a1 + a2 + · · · + an
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f (s), s= = 1,
n
where
f (u) = −u3/u , u ∈ I = (0, n).
We have
3
f 0 (u) = 3u u −2 (ln u − 1),
3
f 00 (u) = 3u u −4 g(t), g(t) = u + (1 − ln u)(2u − 3 + 3 ln u).
From the expression of f 0 , it follows that f is decreasing on (0, s0 ] and increasing
on [s0 , n), where
s0 = e.
In addition, we claim that f 00 (u) ≥ for u ∈ [1, e]. If u ∈ [3/2, e], then

g(t) > (1 − ln u)(2u − 3) ≥ 0.

Also,for u ∈ [1, 3/2], we have


3
 ‹
g(t) = 3(u − 1) + (6 − 2u − 3 ln u) ln u ≥ (6 − 2u − 3 ln u) ln u ≥ 3 1 − ln ln u > 0.
2
PCF Method for Ordered Variables 315

Since f is convex on [1, s0 ], we may apply the RPCF-OV Theorem for m = n − 1.


We only need to show that f (x) + f ( y) ≥ 2 f (1) for all x, y > 0 so that x + y = 2.
The inequality f (x) + f ( y) ≥ 2 f (1) is equivalent to

x 3/x + y 3/ y ≤ 2,

which is just the inequality in P 3.32 from Volume 2. The equality holds for

a1 = a2 = · · · = an = 1.

P 4.8. If a1 , a2 , . . . , a11 are real numbers so that

a1 ≥ a2 ≥ 1 ≥ a3 ≥ · · · ≥ a11 , a1 + a2 + · · · + a11 = 11,

then
(1 − a1 + a12 )(1 − a2 + a22 ) · · · (1 − a11 + a11
2
) ≥ 1.

(Vasile C., 2012)

Solution. Rewrite the desired inequality as


a1 + a2 + · · · + a11
f (a1 ) + f (a2 ) + · · · + f (a11 ) ≥ 11 f (s), s= = 1,
11
where
f (u) = ln(1 − u + u2 ), u ∈ R.
From
2u − 1
f 0 (u) = ,
1 − u + u2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where

s0 = 1/2.

Also, from
1 + 2u(1 − u)
f 00 (u) = ,
(1 − u + u2 )2
it follows that f 00 (u) > 0 for u ∈ [s0 , 1], hence f is convex on [s0 , 1]. Therefore,
applying the LPCF-OV Theorem for n = 11 and m = 2, we only need to show that
f (x) + 9 f ( y) ≥ 9 f (1) for all real x, y so that x + 9 y = 10. Using Note 2, it suffices
to prove that H(x, y) ≥ 0, where

f 0 (x) − f 0 ( y) 1 + x + y − 2x y
H(x, y) = = .
x−y (1 − x + x 2 )(1 − y + y 2 )
316 Vasile Cîrtoaje

Since
1 + x + y − 2x y = 18 y 2 − 8 y + 1 = 2 y 2 + (4 y − 1)2 > 0,
the conclusion follows. The equality holds for a1 = a2 = · · · = a11 = 1.
Remark. By replacing a1 , a2 , . . . , a11 respectively with 1 − a1 , 1 − a2 , . . . , 1 − a11 , we
get the following statement.
• If a1 , a2 , . . . , a11 are real numbers so that

a1 ≤ a2 ≤ 0 ≤ a3 ≤ · · · ≤ a11 , a1 + a2 + · · · + a11 = 0,

then
(1 − a1 + a12 )(1 − a2 + a22 ) · · · (1 − a11 + a11
2
) ≥ 1,
with equality for a1 = a2 = · · · = an = 0.

P 4.9. If a1 , a2 , . . . , a8 are nonzero real numbers so that

a1 ≥ a2 ≥ a3 ≥ a4 ≥ 1 ≥ a5 ≥ a6 ≥ a7 ≥ a8 , a1 + a2 + · · · + a8 = 8,

then  
1 1 1 1 1 1
 ‹
5 2 + 2 + · · · + 2 + 72 ≥ 14 + + ··· + .
a1 a2 a8 a1 a2 a8
(Vasile C., 2012)

Solution. Write the desired inequality as


a1 + a2 + · · · + a8
f (a1 ) + f (a2 ) + · · · + f (a8 ) ≥ 8 f (s), s = = 1,
8
where
5 14
f (u) =
− + 9, u ∈ I = R \ {0}.
u2 u
As shown in the proof of P 3.25, f is convex on [s0 , 1], increasing for u ≥ s0 and

min f (u) = f (s0 ),


u∈I

where
5
s0 = .
7
Taking into account Note 4, we may apply the LPCF-OV Theorem for n = 8 and
m = 4. We only need to show that f (x) + 4 f ( y) ≥ 5 f (1) for x, y ∈ I so that
x + 4 y = 5. Using Note 1, it suffices to prove that h(x, y) ≥ 0. Indeed, we have

f (u) − f (1) 9 5
g(u) = = − 2,
u−1 u u
PCF Method for Ordered Variables 317

g(x) − g( y) 5(x + y) − 9x y
h(x, y) = =
x−y x2 y2
(x + 4 y)(x + y) − 9x y (x − 2 y)2
= = ≥ 0.
x2 y2 x2 y2
In accordance with Note 3, the equality holds for a1 = a2 = · · · = a8 = 1, and also
for
5 5
a1 = , a2 = a3 = a4 = 1, a5 = a6 = a7 = a8 = .
3 6

P 4.10. If a, b, c, d are positive real numbers so that

a ≤ b ≤ 1 ≤ c ≤ d, abcd = 1,

then
7 − 6a 7 − 6b 7 − 6c 7 − 6d 4
+ + + ≥ .
2+a 2 2+ b 2 2+c 2 2+d 2 3
(Vasile C., 2012)

Solution. Using the substitution

a = e x , b = e y , c = ez , d = e w ,

we need to show that

f (x) + f ( y) + f (z) + f (w) ≥ 4 f (s),

where
x + y +z+w
x ≤ y ≤ 0 ≤ z ≤ w, s= = 0,
4
7 − 6eu
f (u) = , u ∈ R.
2 + e2u
As shown in the proof of P 3.26, f is convex on [0, s0 ], is decreasing on (−∞, s0 ]
and increasing on [s0 , ∞), where

s0 = ln 3.

Therefore, we may apply the RPCF-OV Theorem for n = 4 and m = 2. We only


need to show that f (x) + 2 f ( y) ≥ 3 f (0) for all real x, y so that x + 2 y = 0; that
is, to prove that
7 − 6a 2(7 − 6d)
+ ≥1
2 + a2 2 + d2
for a, d > 0 so that ad 2 = 1. This is equivalent to

(d − 1)2 (d − 2)2 (5d 2 + 6d + 3) ≥ 0,


318 Vasile Cîrtoaje

which is clearly true. In accordance with Note 3, the equality holds for a = b =
c = d = 1, and also for

1
a= , b = 1, c = d = 2.
4

P 4.11. If a, b, c are positive real numbers so that

a ≤ b ≤ 1 ≤ c, abc = 1,

then
7 − 4a 7 − 4b 7 − 4c
+ + ≥ 3.
2 + a2 2 + b2 2 + c 2
(Vasile C., 2012)

Solution. Using the substitution

a = e x , b = e y , c = ez ,

we need to show that


f (x) + f ( y) + f (z) ≥ 3 f (s),
where
x + y +z
x ≤ y ≤ 0 ≤ z, s= = 0,
3
7 − 4eu
f (u) = , u ∈ R.
2 + e2u
From
2eu (2eu + 1)(eu − 4)
f (u) =
0
,
(2 + e2u )2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where

s0 = ln 4.

Also, we have
4t · h(t)
f 00 (u) = , t = eu ,
(2 + t 2 )3
where
h(t) = −t 4 + 7t 3 + 12t 2 − 14t − 4.
We will show that h(t) ≥ 0 for t ∈ [1, 4], hence f is convex on [0, s0 ]. Indeed,

h(t) = (t − 1)[t 2 (−t + 6) + 18t + 4] ≥ 0.


PCF Method for Ordered Variables 319

Therefore, we may apply the RPCF-OV Theorem for n = 3 and m = 2. We only


need to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0. That is,
to prove that
7 − 4a 7 − 4b
+ ≥2
2 + a2 2 + b2
for all a, b > 0 so that ab = 1. This is equivalent to

(a − 1)4 ≥ 0.

The equality holds for a = b = c = 1.

P 4.12. If a, b, c are positive real numbers so that

a ≥ 1 ≥ b ≥ c, abc = 1,

then
23 − 8a 23 − 8b 23 − 8c
+ + ≥ 9.
3 + 2a2 3 + 2b2 3 + 2c 2
(Vasile C., 2012)

Solution. Using the substitution

a = e x , b = e y , c = ez ,

we need to show that


f (x) + f ( y) + f (z) ≥ 3 f (s),
where
x + y +z
x ≥ 1 ≥ y ≥ z, s= = 0,
3
23 − 8eu
f (u) = , u ∈ R.
3 + 2e2u
From
4eu (4eu + 1)(eu − 6)
f 0 (u) = ,
(3 + 2e2u )2
it follows that f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞), where s0 =
ln 6. Also, we have
8t · h(t)
f 00 (u) = , t = eu ,
(3 + 2t 2 )3
where
h(t) = −4t 4 + 46t 3 + 36t 2 − 69t − 9.
We will show that h(t) ≥ 0 for t ∈ [1, 6], hence f is convex on [0, s0 ]. Indeed,

h(t) = (t − 1)(2t + 3)[2t(−t + 12) + 3] ≥ 0.


320 Vasile Cîrtoaje

Therefore, we may apply the RPCF-OV Theorem for n = 3 and m = 2. We only


need to show that f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0. That is,
to prove that
23 − 8a 23 − 8b
+ ≥ 6.
3 + 2a2 3 + 2b2
for all a, b > 0 so that ab = 1. This is equivalent to

(a − 1)4 ≥ 0.

The equality holds for a = b = c = 1.

P 4.13. Let a1 , a2 , . . . , an be positive real numbers so that

a1 ≤ · · · ≤ an−1 ≤ 1 ≤ an , a1 a2 · · · an = 1.

If p, q ≥ 0 so that p + 3q ≥ 1, then
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 + pa1 + qa1 1 + pa2 + qa2
2 2
1 + pan + qan2

(Vasile C., 2012)


Solution. For q = 0, we need to show that p ≥ 1 involves
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
1 + pa1 1 + pa2 1 + pan
This is just the inequality from P 2.24. Consider next that q > 0. Using the substi-
tutions ai = e x i for i = 1, 2, . . . , n, we need to show that

f (x 1 ) + f (x 2 ) + · · · + f (x n ) ≥ n f (s),

where
x1 + x2 + · · · + x n
x 1 ≤ · · · ≤ x n−1 ≤ 0 ≤ x n , s= = 0,
n
1 − eu
f (u) = , u ∈ R.
1 + peu + qe2u
As shown in the proof of P 3.30, if p + 3q − 1 ≥ 0, then f is convex on [0, s0 ], where
v
t p+1
s0 = ln r0 > 0, r0 = 1 + 1 + .
q
In addition, f is decreasing on (−∞, s0 ] and increasing on [s0 , ∞). Therefore,
we may apply the RPCF-OV Theorem for m = n − 1. We only need to show that
f (x) + f ( y) ≥ 2 f (0) for all real x, y so that x + y = 0; that is, to prove that
1−a 1− b
+ ≥0
1 + pa + qa 2 1 + pb + qb2
PCF Method for Ordered Variables 321

for a, b > 0 so that ab = 1. This is equivalent to

(a − 1)2 [(p − 1)a + q(a2 + a + 1)] ≥ 0,

which is true because

(p − 1)a + q(a2 + a + 1) ≥ (p − 1)a + q(3a) = (p + 3q − 1)a ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1.

P 4.14. If a, b, c, d, e are real numbers so that

−2 ≤ a ≤ b ≤ 1 ≤ c ≤ d ≤ e, a + b + c + d + e = 5,

then
1 1 1 1 1 1 1 1 1 1
2
+ 2+ 2+ 2+ 2≥ + + + + .
a b c d e a b c d e

Solution. Write the inequality as

a+b+c+d+e
f (a) + f (b) + f (c) + f (d) + f (e) ≥ 5 f (s), s= = 1,
5
where
1 1
f (u) = 2
− , u ∈ I = [−2, 7] \ {0}.
u u
Let
s0 = 2, s < s0 .
From
1 1 1 (u − 2)2
f (u) − f (2) = − + = ≥ 0,
u2 u 4 4u2
it follows that
min f (u) = f (s0 ),
u∈I

while from
u−2 2(3 − u)
f 0 (u) =3
, f 00 (u) = ,
u u4
it follows that f is convex on [s, s0 ]. We can’t apply the the RPCF-OV Theorem
because f is not decreasing on I≤s0 . According to Theorem 1 (applied for n = 5 and
m = 2) and Note 6, we may replace this condition with (1+n−m)s−(n−m)s0 ≤ inf I.
Indeed, we have

(1 + n − m)s − (n − m)s0 = 4 − 6 = −2 = inf I.


322 Vasile Cîrtoaje

So, according to Note 1, it suffices to show that h(x, y) ≥ 0 for all x, y ∈ I so that
x + 3 y = 4. We have
f (u) − f (1) −1
g(u) = = 2,
u−1 u
g(x) − g( y) x+y 2(x + 2)
h(x, y) = = 2 2 = ≥ 0.
x−y x y 3x 2 y 2
The proof is completed. By Note 3, the equality holds for a = b = c = d = e = 1,
and also for
a = −2, b = 1, c = d = e = 2.
Chapter 5

EV Method for Nonnegative Variables

5.1 Theoretical Basis


The Equal Variables Method is an effective tool for solving some difficult symmetric
inequalities.
EV-Theorem (Vasile Cirtoaje, 2005). Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative
real numbers, and let
0 ≤ x1 ≤ x2 ≤ · · · ≤ x n
so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,

where k is a nonnegative real number (k 6= 1); k = 0 means x 1 x 2 · · · x n = a1 a2 · · · an .


Let f be a real-valued function, continuous on [0, ∞) and differentiable on (0, ∞),
so that the joined function € 1 Š
g(x) = f 0 x k−1

is strictly convex on (0, ∞). Then, the sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal only for


x 1 = x 2 = · · · = x n−1 ≤ x n ,
and minimal only for x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .
To prove the EV-Theorem, we need the EV-Lemma and the EV-Proposition be-
low.

EV-Lemma. Let a, b, c be fixed nonnegative real numbers, not all equal and, for k ≥ 0,
at most one of them equal to zero, and let x ≤ y ≤ z be nonnegative real numbers so
that
x + y + z = a + b + c, x k + y k + z k = ak + bk + c k ,

323
324 Vasile Cîrtoaje

where k is a real number (k 6= 1); for k = 0, the second equation is x yz = abc. Then,
the range of y is an interval [m, M ] with m < M ; in addition,
(1) y = m if and only if x = y < z;
(2) y = M if and only if 0 = x < y ≤ z or 0 < x ≤ y = z.

Proof. We show first, by the contradiction method, that x < z. Indeed, if x = z,


then
 x + y + z k
x = z ⇒ x = y = z ⇒ x k + y k + zk = 3
3
+ +
 ‹k
a b c
⇒ ak + bk + c k = 3 ⇒ a = b = c,
3
which is false. Notice that the last implication follows from Jensen’s inequalities

a+b+c k
 ‹
a +b +c ≥3
k k k
, k ∈ (−∞, 0) ∪ (1, ∞),
3

a+b+c k
 ‹
a +b +c ≤3
k k k
, k ∈ (0, 1),
3
a+b+c 3
 ‹
abc ≤ , k = 0,
3
where the equality holds if and only if a = b = c.
According to the relations

x + z = a + b + c − y, x k + z k = ak + bk + c k − y k ,

we may consider x and z as functions of y. From

x 0 + z 0 = −1, x k−1 x 0 + z k−1 z 0 = − y k−1 ,

we get
y k−1 − z k−1 y k−1 − x k−1
x0 = ≤ 0, z0 = ≤ 0. (*)
z k−1 − x k−1 x k−1 − z k−1
Let us define the nonnegative functions

f1 ( y) = y − x( y), f2 ( y) = z( y) − y. f3 ( y) = x( y).

Since

f10 ( y) = 1 − x 0 ( y) > 0, f20 ( y) = z 0 ( y) − 1 < 0, f30 ( y) = x 0 ( y) ≤ 0,

these functions are strictly increasing, decreasing and decreasing, respectively. Thus,
the inequality f1 ( y) ≥ 0 (with f1 increasing) involves y ≥ m, where m is a root of
the equation x( y) = y, and the inequality f2 ( y) ≥ 0 (with f2 decreasing) involves
involves y ≤ y2 , where y2 is a root of the equation z( y) = y. If x( y2 ) ≥ 0, then
EV Method for Nonnegative Variables 325

y2 is the maximal value of y. Otherwise, the maximal value of y is given by the


inequality f3 ( y) ≥ 0 (with f3 decreasing), which involves y ≤ y3 , where y3 is a
root of the equation x( y) = 0. Therefore, y ∈ [m, M ], with y = m for x = y, and
y = M for either y = z or x = 0.

EV-Proposition. Let a, b, c be fixed nonnegative real numbers, and let 0 ≤ x ≤ y ≤ z


so that
x + y + z = a + b + c, x k + y k + z k = ak + bk + c k ,
where k is a real number (k 6= 1); k = 0 means x yz = abc. Let f be a real-valued
function, continuous on [0, ∞) and differentiable on (0, ∞), so that the joined func-
tion € 1 Š
g(x) = f 0 x k−1

is strictly convex on (0, ∞). Then, the sum

S3 = f (x) + f ( y) + f (z)

is maximal only when 0 ≤ x = y ≤ z, and minimal only when x = 0 or 0 < x ≤


y = z.

Proof. If a = b = c, then

a+b+c
 ‹k
a +b +c =3
k k k
,
3

hence
 x + y + z k
x k + y k + zk = 3 ,
3
which involves x = y = z. If k > 0 and two of a, b, c are equal to zero, then

a k + b k + c k = (a + b + c)k ,

hence
x k + y k + z k = (x + y + z)k ,
which involves x = y = 0. In both cases, the extremum conditions in the statement
(x = y and either x = 0 or y = z) are satisfied. Consider further that a, b, c are
not all equal and at most one of them is equal to zero. As shown in the proof of the
EV-Lemma, we have x < z. According to the relations

x + z = a + b + c − y, x k + z k = ak + bk + c k − y k ,

we may consider x and z as functions of y. Thus, we have

S3 = f (x( y)) + f ( y) + f (z( y)) := F ( y).


326 Vasile Cîrtoaje

According to the EV-Lemma, it suffices to show that F is maximal for y = m and is


minimal for y = M . Using (*), we have

F 0 ( y) = x 0 f 0 (x) + f 0 ( y) + z 0 f 0 (z)
y k−1 − z k−1 y k−1 − x k−1
= g(x k−1
) + g( y k−1
) + g(z k−1 ),
z k−1 − x k−1 x k−1 − z k−1
which, for x < y < z, is equivalent to

F 0 ( y) g(x k−1 )
=
( y k−1 − x k−1 )( y k−1 − z k−1 ) (x k−1 − y k−1 )(x k−1 − z k−1 )

g( y k−1 ) g(z k−1 )


+ + .
( y k−1 − z k−1 )( y k−1 − x k−1 ) (z k−1 − x k−1 )(z k−1 − y k−1 )
Since g is strictly convex, the right hand side is positive. Moreover, since

( y k−1 − x k−1 )( y k−1 − z k−1 ) < 0,

we have F 0 ( y) < 0 for y ∈ (m, M ) (see the EV-Lemma), hence F is strictly decreas-
ing on [m, M ]. Therefore, F is maximal for y = m (when 0 ≤ x = y ≤ z) and is
minimal for y = M (when x = 0 or 0 < x ≤ y = z.

Proof of the EV-Theorem. Since X = {x 1 , x 2 , . . . , x n } is defined as a compact set


in R+n ,Sn attains its minimum and maximum. For n = 3, the EV-Theorem follows
immediately from the EV-Proposition. To prove the theorem for n ≥ 4, we use the
contradiction method.
(a) For the sake of contradiction, assume that Sn is maximal at (b1 , b2 , . . . , bn ),
where b1 ≤ b2 ≤ · · · ≤ bn and b1 < bn−1 . Let x 1 , x n−1 and x n be real numbers so
that x 1 ≤ x n−1 ≤ x n and

x 1 + x n−1 + x n = b1 + bn−1 + bn , x 1k + x n−1


k
+ x nk = b1k + bn−1
k
+ bnk .

According to the EV-Proposition, the sum f (x 1 ) + f (x n−1 ) + f (x n ) is maximal for


x 1 = x n−1 , when

f (x 1 ) + f (x n−1 ) + f (x n ) > f (b1 ) + f (bn−1 ) + f (bn ).

This result contradicts the assumption that Sn attains its maximum at (b1 , b2 , . . . , bn )
with b1 < bn−1 .
(b) Similarly, we can prove that Sn is minimal for n ≥ 4 when either x 1 = 0 or

0 < x1 ≤ x2 = · · · = x n.

Corollary 1. Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n
EV Method for Nonnegative Variables 327

so that
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + a n ,
x 12 + x 22 + · · · + x n2 = a12 + a22 + · · · + an2 .
Let f be a real-valued function, continuous on [0, ∞) and differentiable on (0, ∞),
so that the joined function
g(x) = f 0 (x)
is strictly convex on (0, ∞). The sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal only when


x 1 = x 2 = · · · = x n−1 ≤ x n ,
and is minimal only when either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .

Corollary 2. Let a1 , a2 , . . . , an (n ≥ 3) be fixed positive real numbers, and let

0 < x1 ≤ x2 ≤ · · · ≤ x n

so that
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + a n ,
1 1 1 1 1 1
+ + ··· + = + + ··· + .
x1 x2 xn a1 a2 an
Let f be a real-valued function, continuous and differentiable on (0, ∞), so that the
joined function
1
 ‹
g(x) = f p
0
x
is strictly convex on (0, ∞). The sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal only when


x 1 = x 2 = · · · = x n−1 ≤ x n ,
and is minimal only when

x1 ≤ x2 = x3 = · · · = x n.

Corollary 3. Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1 x 2 · · · x n = a1 a2 · · · an .
328 Vasile Cîrtoaje

Let f be a real-valued function, continuous on [0, ∞) and differentiable on (0, ∞),


so that the joined function
g(x) = f 0 (1/x)
is strictly convex on (0, ∞). The sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal only when


x 1 = x 2 = · · · = x n−1 ≤ x n ,
and is minimal only when either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .

Corollary 4. Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,

where k is a real number (k 6= 0, k 6= 1).


(1) For k < 0, the product Pn = x 1 x 2 · · · x n is maximal when

0 < x1 ≤ x2 = x3 = · · · = x n,

and is minimal only when

0 < x 1 = x 2 = · · · = x n−1 ≤ x n ;

(2) For k > 0, the product Pn = x 1 x 2 · · · x n is maximal when

x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is minimal only when either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .

Note 1. The EV-Theorem, Corollary 1 and Corollary 3 are also valid for the cases
when x 1 , x 2 , . . . , x n > 0, f is continuous and differentiable on (0, ∞), f (0+) =
±∞ and the sum Sn has a global maximum (minimum).

From the EV-Theorem and Note 1, we can obtain some interesting particular
results, which are useful in many applications.

Corollary 5. Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank .


EV Method for Nonnegative Variables 329

Let us denote
Sn = x 1m + x 2m + · · · + x nm .

Case 1 : k < 0.
(a) If m ∈ (k, 0) ∪ (1, ∞), then Sn is maximal only for

0 < x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is minimal only for

0 < x1 ≤ x2 = x3 = · · · = x n.

(b) If m ∈ (−∞, k) ∪ (0, 1), then Sn is minimal only for

0 < x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is maximal only for

0 < x1 ≤ x2 = x3 = · · · = x n.

Case 2 : 0 ≤ k < 1 (k = 0 means x 1 x 2 · · · x n = a1 a2 · · · an ).


(a) If m ∈ (0, k) ∪ (1, ∞), then Sn is maximal only for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is minimal only for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .


(b) If m ∈ (−∞, 0), then Sn is minimal only for

0 < x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is maximal (if it has a global maximum) only for

0 < x1 ≤ x2 = x3 = · · · = x n.

(c) If m ∈ (k, 1), then Sn is minimal only for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is maximal only for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .

Case 3 : k > 1.
(a) If m ∈ (0, 1) ∪ (k, ∞), then Sn is maximal only for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,
330 Vasile Cîrtoaje

and is minimal only for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .


(b) If m ∈ (−∞, 0), then Sn is minimal only for

0 < x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is maximal (if it has a global maximum) only for

0 < x1 ≤ x2 = x3 = · · · = x n.

(c) If m ∈ (1, k), then Sn is minimal only for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is maximal only for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .


Proof. We apply the EV-Theorem and Note 1 to the function

f (u) = m(m − 1)(m − k)um .

We have
f 0 (u) = m2 (m − 2)(m − k)um−1
and

m−1 m2 (m − 1)2 (m − k)2 1+m−2k


g(x) = m2 (m − 1)(m − k)x k−1 , g 00 (x) = x k−1 .
(k − 1)2

Since g 00 (x) > 0 for x > 0, g is strictly convex on (0, ∞).

Corollary 6. Let a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

so that
p p p p q q q q
x 1 + x 2 + · · · + x np = a1 + a2 + · · · + anp , x 1 + x 2 + · · · + x nq = a1 + a2 + · · · + anq ,

where
p, q ∈ {1, 2, 3}, p 6= q.
The symmetric sum X
Sn = x i1 x i2 x i3
1≤i1 <i2 <i3 ≤n

is maximal only for


0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,
and is minimal only for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n .
EV Method for Nonnegative Variables 331

Proof. Taking into account that


X €X Š3 €X Š €X Š X
6 x i1 x i2 x i3 = x1 −3 x1 x 12 +2 x 13 ,
1≤i1 <i2 <i3 ≤n

Corollary 6 is a consequence
P of Corollary 5. For p = 2 and q = 3, according
P to
this identity, the sum 1≤i1 <i2 <i3 ≤n x i1 x i2 x i3 is maximal/minimal when x 1 is max-
imal/minimal. Therefore, we need to show that if

x 12 + x 22 + · · · + x n2 = const ant, x 13 + x 23 + · · · + x n3 = const ant,


P
then the sum x 1 is maximal for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ,

and is minimal for either x 1 = 0 or 0 < x 1 ≤ x 2 = x 3 = · · · = x n . This follows by


replacing x 1 , x 2 , . . . , x n with x 12 , x 22 , . . . , x n2 in Corollary 5, case k = 3/2 and m = 1/2.

Note 2. The EV-Theorem and Corollaries 1-3 can be extended to the cases where:
(a) x 1 , x 2 , . . . , x n ≥ m ≥ 0, f is continuous on [m, ∞) and differentiable on
1
(m, ∞), and g(x) is strictly convex for x k−1 > m; so, the sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal for x 1 = x 2 = · · · = x n−1 ≤ x n , and is minimal for either x 1 = m or


m < x1 ≤ x2 = x3 = · · · = x n;
(b) 0 ≤ x 1 , x 2 , . . . , x n ≤ M , f is continuous on [0, M ] and differentiable on
1
(0, M ), and g(x) is strictly convex for x k−1 < M ; so, the sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximal for either x n = M or x 1 = x 2 = · · · = x n−1 ≤ x n , and is minimal


x1 ≤ x2 = x3 = · · · = x n;

Note 3. The EV-Theorem and Corollaries 1-3 can be extended to the cases where:
(a) x 1 , x 2 , . . . , x n > m ≥ 0, f is continuous and differentiable on (m, ∞),
1
f (m+) = ±∞, g(x) is strictly convex for x k−1 > m and the sum Sn has a global
maximum (minimum);
(b) 0 ≤ x 1 , x 2 , . . . , x n < M , f is continuous and differentiable on [0, M ),
1
f (M −) = ±∞, g(x) is strictly convex for x k−1 < M and the sum Sn has a global
maximum (minimum).
332 Vasile Cîrtoaje
EV Method for Nonnegative Variables 333

5.2 Applications

5.1. If a, b, c, d are nonnegative real numbers so that

a + b + c + d = a3 + b3 + c 3 + d 3 = 2,

then
7
≤ a2 + b2 + c 2 + d 2 ≤ 2.
4

5.2. If a1 , a2 , . . . , a9 are nonnegative real numbers so that

a1 + a2 + · · · + a9 = a12 + a22 + · · · + a92 = 3,

then
14
3 ≤ a13 + a23 + · · · + a93 ≤ .
3

5.3. If a, b, c, d are nonnegative real numbers so that

27
a + b + c + d = a2 + b2 + c 2 + d 2 = ,
7
then
5427 1377
≤ a3 + b3 + c 3 + d 3 ≤ .
1372 343

5.4. If a, b, c are positive real numbers so that abc = 1, then


Æ
a5 + b5 + c 5 ≥ 3(a7 + b7 + c 7 ).

5.5. If a, b, c, d are positive real numbers so that abcd = 1, then


Æ
a3 + b3 + c 3 + d 3 ≥ 4(a4 + b4 + c 4 + d 4 ).

5.6. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

bcd cda dab abc 4


+ + + ≤ .
11a + 16 11b + 16 11c + 16 11d + 16 27
334 Vasile Cîrtoaje

5.7. If a, b, c are real numbers, then

bc ca ab 3
+ 2 + 2 ≤ .
3a2 +b +c
2 2 3b + c + a
2 2 3c + a + b
2 2 5

5.8. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


bc ca ab 9
(a) + 2 + 2 ≤ ;
a2 +2 b +2 c +2 8
p
bc ca ab 11 33 − 45
(b) + + ≤ ;
a2 + 3 b2 + 3 c 2 + 3 24
bc ca ab 3
(c) + 2 + 2 ≤ .
a2 +4 b +4 c +4 5

5.9. If a, b, c, d are nonnegative real numbers so that

(3a + 1)(3b + 1)(3c + 1)(3d + 1) = 64,

then
abc + bcd + cda + dab ≤ 1.

5.10. If a1 , a2 , . . . , an and p, q are nonnegative real numbers so that

a1 + a2 + · · · + an = p + q, a13 + a23 + · · · + an3 = p3 + q3 ,

then
a12 + a22 + · · · + an2 ≤ p2 + q2 .

5.11. If a, b, c are nonnegative real numbers, then


p p p
a a2 + 4b2 + 4c 2 + b b2 + 4c 2 + 4a2 + c c 2 + 4a2 + 4b2 ≥ (a + b + c)2 .

5.12. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then

1 1 1 3 a+b+c
+ + ≤ + .
a + b b + c c + a 2(a + b + c) 3
EV Method for Nonnegative Variables 335

5.13. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


1 1 1 3 a+b+c
+ + ≥ + .
a+b b+c c+a a+b+c 6

5.14. Let a, b, c be nonnegative real numbers, no two of which are zero. If

a2 + b2 + c 2 = 3,

then
1 1 1 a+b+c 11
+ + + ≥ .
a+b b+c c+a 9 2(a + b + c)

5.15. Let a, b, c be nonnegative real numbers, no two of which are zero. If

a + b + c = 4,

then
1 1 1 15
+ + ≥ .
a + b b + c c + a 8 + ab + bc + ca

5.16. If a, b, c are nonnegative real numbers, no two of which are zero, then
1 1 1 1 2
+ + ≥ +p .
a+b b+c c+a a+b+c ab + bc + ca

5.17. If a, b, c are nonnegative real numbers, no two of which are zero, then
p p
1 1 1 3− 3 2+ 3
+ + ≥ + p .
a+b b+c c+a a + b + c 2 ab + bc + ca

5.18. Let a, b, c be nonnegative real numbers, no two of which are zero, so that

ab + bc + ca = 3.

If p
9+5 3
0≤k≤ ≈ 2.943,
6
then
2 2 2 9(1 + k)
+ + ≥ .
a+b b+c c+a a + b + c + 3k
336 Vasile Cîrtoaje

5.19. If a, b, c are nonnegative real numbers, no two of which are zero, then
1 1 1 20
+ + ≥ p .
a+b b+c c+a a + b + c + 6 ab + bc + ca

5.20. If a, b, c are positive real numbers so that

7(a2 + b2 + c 2 ) = 11(ab + bc + ca),

then
51 a b c
≤ + + ≤ 2.
28 b+c c+a a+b

5.21. If a1 , a2 , . . . , an are nonnegative real numbers so that

a12 + a22 + · · · + an2  a + a + · · · + a 2


1 2 n
= ,
n+3 n+1
then
(n + 1)(2n − 1) 1 1 1 3n2 (n + 1)
 ‹
≤ (a1 + a2 + · · · + an ) + + ··· + ≤ .
2 a1 a2 an 2(n + 2)

5.22. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then

176
abc + bcd + cda + dab ≤ 1 + abcd.
81

5.23. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then


3
a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + abcd ≤ 1.
4

5.24. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then


4
a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + (abcd)3/2 ≤ 1.
3

5.25. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + 2(abcd)3/2 ≤ 6.
EV Method for Nonnegative Variables 337

5.26. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

11(ab + bc + ca) + 4(a2 b2 + b2 c 2 + c 2 a2 ) ≤ 45.

5.27. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

a2 b2 + b2 c 2 + c 2 a2 + a3 b3 + b3 c 3 + c 3 a3 ≥ 6abc.

5.28. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


€p p p Š
2(a2 + b2 + c 2 ) + 5 a + b + c ≥ 21.

5.29. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


v v v
t 1 + 2a t 1 + 2b t 1 + 2c
+ + ≥ 3.
3 3 3

5.30. Let a, b, c be nonnegative real numbers, no two of which are zero. If

0 ≤ k ≤ 15,

then
1 1 1 k 9+k
+ + + ≥ .
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 4(ab + bc + ca)

5.31. If a, b, c are nonnegative real numbers, no two of which are zero, then
1 1 1 24 8
+ + + ≥ .
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 ab + bc + ca

5.32. If a, b, c are nonnegative real numbers, no two of which are zero, so that

k(a2 + b2 + c 2 ) + (2k + 3)(ab + bc + ca) = 9(k + 1), 0 ≤ k ≤ 6,

then
1 1 1 9k 3
+ + + ≥ + k.
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 4
338 Vasile Cîrtoaje

5.33. If a, b, c are nonnegative real numbers, no two of which are zero, then
2 2 2 8 1
(a) + 2 + 2 ≥ 2 + ;
a2 +b 2 b +c 2 c +a 2 a +b +c
2 2 ab + bc + ca
2 2 2 7 6
(b) + 2 + 2 ≥ 2 + ;
a2 +b 2 b +c 2 c +a 2 a +b +c
2 2 (a + b + c)2
2 2 2 45
(c) + 2 + 2 ≥ .
a2 +b 2 b +c 2 c +a 2 4(a + b + c ) + ab + bc + ca
2 2 2

5.34. If a, b, c are nonnegative real numbers, no two of which are zero, then

1 1 1 3 4
+ 2 + 2 + 2 ≥ .
a2 +b 2 b +c 2 c +a 2 a +b +c
2 2 ab + bc + ca

5.35. If a, b, c are nonnegative real numbers, no two of which are zero, then
3 3 3 5 4
(a) + + ≥ + ;
a2 + ab + b2 b2 + bc + c 2 c 2 + ca + a2 ab + bc + ca a2 + b2 + c 2
3 3 3 1 24
(b) + 2 + 2 ≥ + ;
a2 + ab + b 2 b + bc + c 2 c + ca + a 2 ab + bc + ca (a + b + c)2
1 1 1 21
(c) + + ≥ .
a2 + ab + b2 b2 + bc + c 2 c 2 + ca + a2 2(a2 + b2 + c 2 ) + 5(ab + bc + ca)

5.36. Let f be a real-valued function, continuous on [0, ∞) and differentiable on


(0, ∞), so that f 000 (u) ≥ 0 for u ∈ (0, ∞). If a, b, c ≥ 0, then

f (a2 + 2bc) + f (b2 + 2ca) + f (c 2 + 2ab) ≤ f (a2 + b2 + c 2 ) + 2 f (ab + bc + ca).

5.37. If a, b, c are the lengths of the side of a triangle, then

1 1 1 85
+ + ≤ .
(a + b)2 (b + c)2 (c + a)2 36(ab + bc + ca)

5.38. If a, b, c are the lengths of the side of a triangle so that a + b + c = 3, then

1 1 1 3(a2 + b2 + c 2 )
+ + ≤ .
(a + b)2 (b + c)2 (c + a)2 4(ab + bc + ca)
EV Method for Nonnegative Variables 339

2
5.39. Let a, b, c ≥ so that a + b + c = 3. Then,
5
1 1 1 3
+ + ≤ .
3 + 2(a + b ) 3 + 2b + c ) 3 + 2(c + a ) 7
2 2 2 2 2 2

5.40. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


2 2 2 99
+ + ≤ .
2 + a2 + b2 2 + b2 + c 2 2 + c 2 + a2 63 + a2 + b2 + c 2

5.41. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


1 1 1 18
+ + ≤ .
3+a + b
2 2 3+ b +c
2 2 3+c +a
2 2 27 + a + b2 + c 2
2

5.42. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


5 5 5 27
+ + ≥ .
3+a + b
2 2 3+ b +c
2 2 3+c +a
2 2 6 + a + b2 + c 2
2

5.43. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then


X 3 296
≤ .
3 + 2(a2 + b2 + c 2 ) 218 + a2 + b2 + c 2 + d 2

5.44. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


4 4 4 21
+ + ≥ .
2+a + b
2 2 2+ b +c
2 2 2+c +a
2 2 4 + a + b2 + c 2
2

5.45. If a, b, c are nonnegative real numbers so that a2 + b2 + c 2 = 3, then


1 1 1 1
+ + ≤ .
10 − (a + b)2 10 − (b + c)2 10 − (c + a)2 2

5.46. If a, b, c are nonnegative real numbers, no two of which are zero, so that
a4 + b4 + c 4 = 3, then
1 1 1 3
+ + ≥ .
a5 + b5 b5 + c 5 c 5 + a5 2
340 Vasile Cîrtoaje

5.47. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
v 
1
q q Æ t ‹
a1 + 1+ a2 + 1+· · ·+ an + 1 ≥ 2 1 −
2 2 2 (a12 + a22 + · · · + an2 ) + 2(n2 − n + 1).
n

5.48. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
Xq q
(3n − 4)a12 + n ≥ (3n − 4)(a12 + a22 + · · · + an2 ) + n(4n2 − 7n + 4).

5.49. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


v
p p p t8
a2 + 4 + b2 + 4 + c 2 + 4 ≤ (a2 + b2 + c 2 ) + 37.
3

5.50. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


p p p Æ
32a2 + 3 + 32b2 + 3 + 32c 2 + 3 ≤ 32(a2 + b2 + c 2 ) + 219.

5.51. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then


p
1 1 1 2n n − 1 p
+ + ··· + + 2 ≥ n + 2 n − 1.
a1 a2 an a1 + a22 + · · · + an2

5.52. If a, b, c ∈ [0, 1], then

(1 + 3a2 )(1 + 3b2 )(1 + 3c 2 ) ≥ (1 + ab + bc + ca)3 .

5.53. If a, b, c are nonnegative real numbers so that a + b + c = ab + bc + ca, then


1 1 1 1
+ + ≥ .
4 + 5a2 4 + 5b2 4 + 5c 2 3

5.54. If a, b, c, d are positive real numbers so that a + b + c + d = 4abcd, then


1 1 1 1
+ + + ≥ 1.
1 + 3a 1 + 3b 1 + 3c 1 + 3d
EV Method for Nonnegative Variables 341

5.55. If a1 , a2 , . . . , an are positive real numbers so that


1 1 1
a1 + a2 + · · · + an = + + ··· + ,
a1 a2 an

then
1 1 1
+ + ··· + ≥ 1.
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an

5.56. If a, b, c, d, e are nonnegative real numbers so that a4 + b4 + c 4 + d 4 + e4 = 5,


then
7(a2 + b2 + c 2 + d 2 + e2 ) ≥ (a + b + c + d + e)2 + 10.

5.57. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
n(n − 1) 
(a12 + a22 + · · · + an2 )2 − n2 ≥ 2 a14 + a24 + · · · + an4 − n .
n −n+1

5.58. If a1 , a2 , . . . , an are nonnegative real numbers so that a12 + a22 + · · · + an2 = n,


then v
1
t  ‹
a1 + a2 + · · · + an ≥ n2 − n + 1 + 1 −
3 3 3
(a16 + a26 + · · · + an6 ).
n

5.59. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 50
 ‹
4 + + + ≥ 27.
a b c a+b+c

5.60. If a, b, c are positive real numbers so that abc = 1, then

1 1 1
 ‹
a + b + c + 15 ≥ 6
3 3 3
+ + .
a b c

5.61. Let a1 , a2 , . . . , an be positive numbers so that a1 a2 · · · an = 1. If k ≥ n − 1,


then
1 1 1
 ‹
a1 + a2 + · · · + an + (2k − n)n ≥ (2k − n + 1)
k k k
+ + ··· + .
a1 a2 an
342 Vasile Cîrtoaje

5.62. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 +a2 +· · ·+an = n,


and let k be an integer satisfying 2 ≤ k ≤ n + 2. If
 n k−1
r= − 1,
n−1
then
a1k + a2k + · · · + ank − n ≥ nr(1 − a1 a2 · · · an ).

1 1 1
5.63. If a, b, c are positive real numbers so that + + = 3, then
a b c
4(a2 + b2 + c 2 ) + 9 ≥ 21abc.

1 1 1
5.64. If a1 , a2 , . . . , an are positive real numbers so that + + ··· + = n,
a1 a2 an
then,
a1 + a2 + · · · + an − n ≤ en−1 (a1 a2 · · · an − 1),
where ‹n−1
1

en−1 = 1 + .
n−1

5.65. If a1 , a2 , . . . , an are positive real numbers, then

a1n + a2n + · · · + ann 1 1 1


 ‹
+ n(n − 1) ≥ (a1 + a2 + · · · + an ) + + ··· + .
a1 a2 · · · an a1 a2 an

5.66. If a1 , a2 , . . . , an are nonnegative real numbers, then

(n−1)(a1n + a2n +· · ·+ ann )+ na1 a2 · · · an ≥ (a1 + a2 +· · ·+ an )(a1n−1 + a2n−1 +· · ·+ ann−1 ).

5.67. If a1 , a2 , . . . , an are nonnegative real numbers, then

(n − 1)(a1n+1 + a2n+1 + · · · + ann+1 ) ≥ (a1 + a2 + · · · + an )(a1n + a2n + · · · + ann − a1 a2 · · · an ).

5.68. If a1 , a2 , . . . , an are positive real numbers, then

1 1 1 1
 ‹
(a1 + a2 + · · · + an − n) + + ··· + − n + a1 a2 · · · an + ≥ 2.
a1 a2 an a1 a2 · · · an
EV Method for Nonnegative Variables 343

5.69. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1
−Ç < 1.
a1 + a2 + · · · + an − n
p
1 1 1
a1 + a2 + ··· + an −n

5.70. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

n2 (n − 2) 1 1 1
 ‹
a1n−1 + a2n−1 + ··· + ann−1 + ≥ (n − 1) + + ··· + .
a1 + a2 + · · · + an a1 a2 an

5.71. If a, b, c are nonnegative real numbers, then

abc − 1 2
(a + b + c − 3)2 ≥ (a + b2 + c 2 − 3).
abc + 1

5.72. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then


p1
(a1 a2 · · · an ) n−1 (a12 + a22 + · · · + an2 ) ≤ n.

5.73. If a1 , a2 , . . . , an are positive real numbers such that a1 + a2 + · · · + an = n − 1,


then v
t a1 + a22 + · · · + an2
v u 2
tn n−1
≥ 4 .
a1 a2 · · · an n(n − 1)

5.74. If a1 , a2 , . . . , an are positive real numbers so that a13 + a23 + · · · + an3 = n, then

a1 + a2 + · · · + an ≥ n n+1 a1 a2 · · · an .
p

5.75. Let a, b, c be nonnegative real numbers so that ab + bc + ca = 3. If

ln 4
k ≥2− ≈ 0.738,
ln 3
then
a k + b k + c k ≥ 3.
344 Vasile Cîrtoaje

5.76. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 9 − ln 8
k≥ ≈ 0.29,
ln 3 − ln 2
then
a k + b k + c k ≥ ab + bc + ca.

5.77. If a1 , a2 , . . . , an (n ≥ 4) are nonnegative numbers so that a1 +a2 +· · ·+an = n,


then
1 1 1
+ + ··· + ≤ 1.
n + 1 − a2 a3 · · · an n + 1 − a3 a4 · · · a1 n + 1 − a1 a2 · · · an−1

5.78. If a, b, c are nonnegative real numbers so that

a + b + c ≥ 2, ab + bc + ca ≥ 1,

then p p p
3
a+ b+
3 3
c ≥ 2.

5.79. If a, b, c, d are positive real numbers so that abcd = 1, then


p
(a + b + c + d)4 ≥ 36 3 (a2 + b2 + c 2 + d 2 ).

5.80. If a, b, c are nonnegative real numbers so that ab + bc + ca = 1, then


p p p
33a2 + 16 + 33b2 + 16 + 33c 2 + 16 ≤ 9(a + b + c).

5.81. If a, b, c are positive real numbers so that a + b + c = 3, then


3
a2 b2 + b2 c 2 + c 2 a2 ≤ p
3
.
abc

5.82. If a1 , a2 , . . . , an (n ≤ 81) are nonnegative real numbers so that

a12 + a22 + · · · + an2 = a15 + a25 + · · · + an5 ,

then
a16 + a26 + · · · + an6 ≤ n.
EV Method for Nonnegative Variables 345

5.83. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


p Æ
1 + 1 + a3 + b3 + c 3 ≥ 3(a2 + b2 + c 2 ).

5.84. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


v
p p p t 2
a + b + b + c + c + a ≤ 16 + (ab + bc + ca).
3

5.85. If a, b, c ∈ [0, 4] and ab + bc + ca = 4, then


p p p p
a + b + b + c + c + a ≤ 3 + 5.

5.86. If a, b, c are positive real numbers so that abc = 1, then


v
a+b+c t 3 2+ a + b + c
2 2 2
(a) ≥ ;
3 5
p
(b) a3 + b3 + c 3 ≥ 3(a4 + b4 + c 4 ).

5.87. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 18) ≤ 10(a3 + b3 + c 3 + d 3 − 4).

5.88. If a, b, c, d are nonnegative real numbers such that

a + b + c + d = 4,

then
(a4 + b4 + c 4 + d 4 )2 ≥ (a2 + b2 + c 2 + d 2 )(a5 + b5 + c 5 + d 5 ).

5.89. If a, b, c, d are nonnegative real numbers such that

a + b + c + d = 4,

then
13(a2 + b2 + c 2 + d 2 )2 ≥ 12(a4 + b4 + c 4 + d 4 ) + 160.
346 Vasile Cîrtoaje

5.90. If a1 , a2 , . . . , a8 are nonnegative real numbers, then

19(a12 + a22 + · · · + a82 )2 ≥ 12(a1 + a2 + · · · + a8 )(a13 + a23 + · · · + a83 ).

5.91. If a, b, c are nonnegative real numbers so that

5(a2 + b2 + c 2 ) = 17(ab + bc + ca),

then v v p
t3 s a 1+ 7
t b s
c
3 ≤ + + ≤ p .
5 b+c c+a a+b 2

5.92. If a, b, c are nonnegative real numbers so that

8(a2 + b2 + c 2 ) = 9(ab + bc + ca),

then
19 a b c 141
≤ + + ≤ .
12 b+c c+a a+b 88

5.93. If a, b, c ∈ (0, 2] such that a + b + c = 3, then


v v v
t 2(b + c) t 2(c + a) t 2(a + b) 9
−1+ −1+ −1≥ p .
a b c ab + bc + ca

5.94. Let a, b, c and x, y, z be nonnegative real numbers such that

x 3 + y 3 + z 3 = a3 + b3 + c 3 .

Then,
(a + b + c)(x + y + z) p
3
≥ 3.
ab + bc + ca + x y + yz + z x

5.95. If a, b, c, d are positive numbers such that


1 1 1 1
a+b+c+d = + + + ,
a b c d
then
ab + ac + ad + bc + bd + cd + 3abcd ≥ 9.
EV Method for Nonnegative Variables 347

5.96. If a1 , a2 , a3 , a4 , a5 are nonnegative real numbers, then


(a13 + a23 + a33 + a43 + a53 )2 1X
≥ ai a j .
a14 + a24 + a34 + a44 + a54 2 i< j

5.97. If a1 , a2 , . . . , an ≥ 0 such that

a1 + a2 + · · · + an = n,

then v
1 X
 ‹
p u
a1 + a2 + · · · + an ≤ 2n − 1 + 2 1 −
p p t ai a j .
n i< j

5.98. If a1 , a2 , . . . , an ≥ 0 such that


X
a1 + a2 + · · · + a n = ai a j > 0,
i< j

then
(n − 1)(n − 2) Xp
(a1 + a2 + · · · + an ) + ai a j ≥ n(n − 1).
2 i< j

5.99. Let

F (a1 , a2 , . . . , an ) = n(a12 + a22 + · · · + an2 ) − (a1 + a2 + · · · + an )2 ,

where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and

a12 (a22 + a32 + · · · + an2 ) ≥ n − 1.

Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an

5.100. Let

F (a1 , a2 , . . . , an ) = a1 + a2 + · · · + an − n n a1 a2 · · · an ,
p

where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and

a1 (a2 + a3 + · · · + an ) ≥ n − 1.

Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an
348 Vasile Cîrtoaje

5.101. Let
v
t a2 + a2 + · · · + a2
1 2 n a1 + a2 + · · · + an
F (a1 , a2 , . . . , an ) = − ,
n n
where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and

a1n−1 (a2 + a3 + · · · + an ) ≥ n − 1.

Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an

5.102. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers such that

a1 + a2 + · · · + an = n, an = max{a1 , a2 , . . . , an },

then
1 1 1
 ‹
n + + ··· + ≥ 4(a12 + a22 + · · · + an2 ) + n(n − 5).
a1 a2 an−1

5.103. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then

1 1 1
+ + ≤ 1.
a+ b+1 b+c+1 c+a+1
EV Method for Nonnegative Variables 349

5.3 Solutions

P 5.1. If a, b, c, d are nonnegative real numbers so that

a + b + c + d = a3 + b3 + c 3 + d 3 = 2,

then
7
≤ a2 + b2 + c 2 + d 2 ≤ 2.
4
(Vasile C., 2010)

Solution. The right inequality follows from the Cauchy-Schwarz inequality

(a2 + b2 + c 2 + d 2 )2 ≤ (a + b + c + d)(a3 + b3 + c 3 + d 3 ).

The equality holds for a = b = 0 and c = d = 1 (or any permutation).


To prove the left inequality, assume that a ≤ b ≤ c ≤ d, then apply Corollary 5
for k = 3 and m = 2:
• If a, b, c, d are nonnegative real numbers so that

a+b+c+d =2 , a3 + b3 + c 3 + d 3 = 2, a ≤ b ≤ c ≤ d,

then
S4 = a2 + b2 + c 2 + d 2
is minimal for a = b = c.
So, we only need to prove that the equations

3a + d = 3a3 + d 3 = 2, a, d ≥ 0,

imply
7
≤ 3a2 + d 2 .
4
Indeed, from 3a + d = 3a3 + d 3 = 2, we get a = 1/4 and d = 5/4, when

7
3a2 + d 2 = .
4
The left inequality is an equality for

1 5
a=b=c= , d=
4 4
(or any cyclic permutation).
350 Vasile Cîrtoaje

P 5.2. If a1 , a2 , . . . , a9 are nonnegative real numbers so that

a1 + a2 + · · · + a9 = a12 + a22 + · · · + a92 = 3,

then
14
3 ≤ a13 + a23 + · · · + a93 ≤ .
3
(Vasile C., 2010)

Solution. The left inequality follows from the Cauchy-Schwarz inequality

(a1 + a2 + · · · + a9 )(a13 + a23 + · · · + a93 ) ≥ (a12 + a22 + · · · + a92 )2 .

The equality holds for a1 = a2 = · · · = a6 = 0 and a7 = a8 = a9 = 1 (or any


permutation).
To prove the right inequality, assume that

a1 ≤ a2 ≤ · · · ≤ a9 ,

then apply Corollary 5 for k = 2 and m = 3:


• If a1 , a2 , . . . , a9 are nonnegative real numbers so that

a1 + a2 + · · · + a9 = 3 , a12 + a22 + · · · + a92 = 3, a1 ≤ a2 ≤ · · · ≤ a9 ,

then
S9 = a13 + a23 + · · · + a93
is maximal for a1 = a2 = · · · = a8 ≤ a9 .
Thus, we only need to prove that the equations

8a + b = 3, 8a2 + b2 = 3, a, b ≥ 0,

involve
14
8a3 + b3 ≤
.
3
Indeed, from the equations above, we get a = 1/6 and b = 5/3, when

1 125 14
8a3 + b3 = + = .
27 27 3
The equality holds for

1 5
a1 = a2 = · · · = a8 = , a9 =
6 3
(or any cyclic permutation).
EV Method for Nonnegative Variables 351

P 5.3. If a, b, c, d are nonnegative real numbers so that


27
a + b + c + d = a2 + b2 + c 2 + d 2 = ,
7
then
5427 1377
≤ a3 + b3 + c 3 + d 3 ≤ .
1372 343
(Vasile C., 2014)

Solution. Assume that a ≤ b ≤ c ≤ d.


(a) To prove the right inequality, we apply Corollary 5 for k = 2 and m = 3:
• If a, b, c, d are nonnegative real numbers so that
27 27
a+b+c+d = , a2 + b2 + c 2 + d 2 = , a ≤ b ≤ c ≤ d,
7 7
then
S4 = a3 + b3 + c 3 + d 3
is maximal for a = b = c ≤ d
Thus, we only need to prove that the equations
27 27
3a + d = , 3a2 + d 2 = , a, d ≥ 0,
7 7
involve
1377
3a3 + d 3 ≤
.
343
Indeed, from the equations above, we get a = 6/7 and d = 9/7, when
 ‹3  ‹3
6 9 1377
3a + d = 3
3 3
+ = .
7 7 343
The equality holds for
6 9
a=b=c= , d=
7 7
(or any cyclic permutation).

(b) To prove the left inequality, we apply Corollary 5 for k = 2 and m = 3:


• If a, b, c, d are nonnegative real numbers so that
27 27
a+b+c+d = , a2 + b2 + c 2 + d 2 = , a ≤ b ≤ c ≤ d,
7 7
then
S4 = a3 + b3 + c 3 + d 3
352 Vasile Cîrtoaje

is minimal for either a = 0 or a ≤ b = c = d.


The case a = 0 is not possible because from
27 27
b+c+d = , b2 + c 2 + d 2 = ,
7 7
we get
27 27
 ‹
3(b + c + d ) − (b + c + d) =
2 2 2
3− < 0,
2
7 7
which contradicts the known inequality

3(b2 + c 2 + d 2 ) ≥ b + c + d)2 .

For a ≤ b = c = d, we need to prove that the equations


27 27
a + 3d = , a2 + 3d 2 = , a, d ≥ 0,
7 7
involve
5427
a3 + 3d 3 ≥
.
1372
Indeed, from the equations above, we get a = 9/14 and d = 15/14, when
 ‹3  ‹3
9 15 5427
a + 3d =
3 3
+3 = .
14 14 1372
The equality holds for
9 15
a= , b=c=d=
14 14
(or any cyclic permutation).
Remark. In the same manner, we can prove the following generalization:
• Let k be a positive real number (k > 2), and let a1 , a2 , . . . , an be nonnegative real
numbers so that
(n − 1)3
a1 + a2 + · · · + an = a12 + a22 + · · · + an2 = .
n2 − 3n + 3
The sum
Sn = a1k + a2k + · · · + ank
is maximal for
(n − 1)(n − 2) (n − 1)2
a1 = · · · = an−1 = , an = ,
n2 − 3n + 3 n2 − 3n + 3
and is minimal for
(n − 1)2 (n − 2) (n − 1)(n2 − 2n + 2)
a1 = , a2 = · · · = an = .
n(n2 − 3n + 3) n(n2 − 3n + 3)
EV Method for Nonnegative Variables 353

P 5.4. If a, b, c are positive real numbers so that abc = 1, then


Æ
a5 + b5 + c 5 ≥ 3(a7 + b7 + c 7 ).

(Vasile C., 2014)

Solution. Substituting

a = x 1/5 , b = y 1/5 , c = z 1/5 ,

we need to show that x yz = 1 involves


Æ
x + y + z ≥ 3(x 7/5 + y 7/5 + z 7/5 ).

Assume that x ≤ y ≤ z, then apply Corollary 5 for k = 0 and m = 7/5:


• If x, y, z are positive real numbers so that

x + y + z = const ant , x yz = 1, x ≤ y ≤ z,

then
S3 = x 7/5 + y 7/5 + z 7/5
is maximal for x = y.
So, it suffices to prove the original inequality for a = b. Write this inequality in
the homogeneous form

(a5 + b5 + c 5 )2 ≥ 3abc(a7 + b7 + c 7 ).

We only need to prove this inequality for a = b = 1; that is, to show that f (c) ≥ 0,
where
f (c) = (c 5 + 2)2 − 3c(c 7 + 2), c > 0.
We have
f 0 (c) = 10c 4 (c 5 + 2) − 24c 7 − 6,
f 00 (c) = 2c 3 g(t), g(t) = 45c 5 − 84c 3 + 40.
By the AM-GM inequality, we get
Æ
5
g(t) = 15c 5 + 15c 5 + 15c 5 + 20 + 20 − 84c 3 ≥ 5 (15c 5 )3 · 202 − 84c 3
p5
p5

= 27 · 16 25 − 14 18 c 3 > 0,

hence f 00 (c) > 0, f 0 (c) is increasing. Since f 0 (0) = 1, it follows that f 0 (c) ≤ 0 for
c ≤ 1, f 0 (c) ≥ 0 for c ≥ 1, therefore f is decreasing on (0, 1] and increasing on
[1, ∞); consequently, f (c) ≥ f (1) = 0. The equality occurs for a = b = c = 1.
354 Vasile Cîrtoaje

P 5.5. If a, b, c, d are positive real numbers so that abcd = 1, then


Æ
a3 + b3 + c 3 + d 3 ≥ 4(a4 + b4 + c 4 + d 4 ).

(Vasile C., 2014)

Solution. Substituting

a = x 1/3 , b = y 1/3 , c = z 1/3 , d = t 1/3 ,

we need to show that x yz t = 1 involves


Æ
x + y + z + t ≥ 4(x 4/3 + y 4/3 + z 4/3 + t 4/3 ).

Apply Corollary 5, case k = 0 and m = 4/3:


• If x, y, z, t are positive real numbers so that

x + y + z + t = const ant , x yz t = 1, x ≤ y ≤ z ≤ t,

then
S4 = x 4/3 + y 4/3 + z 4/3 + t 4/3
is maximal for x = y = z.
Therefore, it suffices to prove the original inequality for a = b = c. Write the
original inequality in the homogeneous form
p
(a3 + b3 + c 3 + d 3 )2 ≥ 4 abcd (a4 + b4 + c 4 + d 4 ).

We only need to prove this inequality for a = b = c = 1; that is, to show that
p
(d 3 + 3)2 ≥ 4 d (d 4 + 3).
p
Putting u = d, we have
p
(d 3 + 3)2 − 4 d (d 4 + 3) = (u6 + 3)2 − 4u(u8 + 3)
= (u3 − 1)4 + 4(u + 2)(u − 1)2 ≥ 0.

The equality holds for a = b = c = d = 1.

P 5.6. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

bcd cda dab abc 4


+ + + ≤ .
11a + 16 11b + 16 11c + 16 11d + 16 27

(Vasile C., 2008)


EV Method for Nonnegative Variables 355

Solution. For a = 0, the inequality becomes


64
bcd ≤,
27
where b, c, d ≥ 0, b + c + d = 4. By the AM-GM inequality, we have
b+c+d 3
 ‹3
4 64
 ‹
bcd ≤ = = .
3 3 27
For a bcd 6= 0, we write the inequality in the form
4
f (a) + f (b) + f (c) + f (d) + ≥ 0,
(1 + k)abcd
where
−1 16
f (u) = , k= , u > 0.
u(u + k) 11
We have f (0+) = −∞ and
2u + k
f 0 (u) = ,
(u2 + ku)2
kx 4 + 2x 3
g(x) = f 0 (1/x) = ,
(kx + 1)2
2x(k3 x 3 + 4k2 x 2 + 6kx + 6)
g 00 (x) = .
(kx + 1)4
Since g 00 (x) > 0 for x > 0, g is strictly convex on (0, ∞). By Corollary 3 and Note
1, if
a + b + c + d = 4, abcd = const ant, 0 < a ≤ b ≤ c ≤ d,
then the sum
S4 = f (a) + f (b) + f (c) + f (d)
is minimal for b = c = d. Thus, we only need to prove that
b3 3ab2 4
+ ≤
11a + 16 11b + 16 27
for a + 3b = 4. The inequality is equivalent to
b3 3b2 (4 − 3b) 4
+ ≤ ,
3(20 − 11b) 11b + 16 21
(b − 1)2 (4 − 3b)(231b + 80) ≥ 0,
(b − 1)2 a(231b + 80) ≥ 0.
The equality holds for a = b = c = d = 1, and also for
4
a = 0, b=c=d=
3
(or any cyclic permutation).
356 Vasile Cîrtoaje

P 5.7. If a, b, c are real numbers, then


bc ca ab 3
+ 2 + 2 ≤ .
3a2 +b +c
2 2 3b + c + a
2 2 3c + a + b
2 2 5

(Vasile Cirtoaje and Pham Kim Hung, 2005)

Solution. For a = 0, the inequality is true because


bc 1 3
≤ < .
b2 + c 2 2 5
Consider further that a, b, c are different from zero. The inequality remains un-
changed by replacing a, b, c with −a, −b, −c, respectively. Thus, we only need to
consider the case a < 0, b, c > 0, and the case a, b, c > 0. In the first case, it suffices
to show that
bc 3
≤ .
3a + b + c
2 2 2 5
Indeed, we have
bc bc 1 3
< 2 ≤ < .
3a + b + c
2 2 2 b +c 2 2 5
p p p
Consider now the case a, b, c > 0. Replacing a, b, c with a, b, c, the inequality
becomes
1 1 1 3
p +p +p ≤ p .
a(3a + b + c) b(3b + c + a) c(3c + a + b) 5 abc

Due to homogeneity, we may consider that a + b + c = 2. So, we need to show that


6
f (a) + f (b + f (c) + p ≥ 0,
5 abc
where
−1
f (u) = p , u > 0.
u(u + 1)
We have f (0+) = −∞ and
3u + 1
f 0 (u) = p ,
2u u(u + 1)2
p
x 2 x(x + 3)
g(x) = f (1/x) =
0
,
2(x + 1)2
p
x(3x 3 + 11x 2 + 5x + 45)
g (x) =
00
.
8(x + 1)4
Since g 00 (x) > 0 for x > 0, g is strictly convex on (0, ∞). By Corollary 3 and Note
1, if
a + b + c = 2, abc = const ant, 0 < a ≤ b ≤ c,
EV Method for Nonnegative Variables 357

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for b = c. Thus, we only need to prove the original homogeneous in-
equality for b = c = 1; that is,
1 2a 3
+ 2 ≤ ,
3a2 +2 a +4 5
9a4 − 30a3 + 37a2 − 20a + 4 ≥ 0,
(a − 1)2 (3a − 2)2 ≥ 0.
The equality holds for a = b = c, and also for

3a = 2b = 2c

(or any cyclic permutation).

P 5.8. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


bc ca ab 9
(a) + + ≤ ;
a2 + 2 b2 + 2 c 2 + 2 8
p
bc ca ab 11 33 − 45
(b) + + ≤ ;
a2 + 3 b2 + 3 c 2 + 3 24
bc ca ab 3
(c) + + ≤ .
a2 + 4 b2 + 4 c 2 + 4 5
(Vasile C., 2008)

Solution. For the nontrivial case abc 6= 0, we can write the desired inequalities in
the form
m
f (a) + f (b) + f (c) + ≥ 0,
abc
where
−1
f (u) = , k ∈ {2, 3, 4}, u > 0.
u(u2 + k)
We have f (0+) = −∞ and

3u2 + k
f 0 (u) = ,
u2 (u2 + k)2

kx 6 + 3x 4
g(x) = f 0 (1/x) = ,
(kx 2 + 1)2
2x 2 (k3 x 6 + 4k2 x 4 − 3kx 2 + 18)
g 00 (x) = .
(kx 2 + 1)4
358 Vasile Cîrtoaje

Since
k3 x 6 + 4k2 x 4 − 3kx 2 + 18 > 4k2 x 4 − 3kx 2 + 18 > 0,
we have g 00 (x) > 0, hence g is strictly convex on (0, ∞). According to Corollary 3
and Note 1, if

a + b + c = 3, abc = const ant, 0 < a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for b = c. Thus, we only need to prove the original inequalities for b = c.

(a) We only need to prove the homogeneous inequality

bc ca ab 1
+ + ≤
9a2 + 2(a + b + c)2 9b2 + 2(a + b + c)2 9c 2 + 2(a + b + c)2 8

for b = c = 1, that is

1 2a 1
+ ≤ ,
11a2 + 8a + 8 2a2 + 8a + 17 8

2a a(11a + 8)
≤ ,
2a2 + 8a + 17 8(11a2 + 8a + 8)
a(22a3 − 72a2 + 123a + 8) ≥ 0.
Since

22a3 − 72a2 + 123a + 8 > 20a3 − 80a2 + 80a = 20a(a − 2)2 ≥ 0,

the conclusion follows. The equality holds for a = 0 and b = c = 3/2 (or any cyclic
permutation).

(b) Let
p p
11 33 − 45 33 − 5
m= ≈ 0.253, r= ≈ 0.186.
72 4
We only need to prove the homogeneous inequality

bc ca ab
+ 2 + 2 ≤m
3a2 + (a + b + c)2 3b + (a + b + c)2 3c + (a + b + c)2

for b = c = 1; that is, to show that f (a) ≤ m, where

1 2a
f (a) = + .
4(a2 + a + 1) a2 + 4a + 7
EV Method for Nonnegative Variables 359

We have

−8a6 − 18a5 + 15a4 + 28a3 + 18a2 − 42a + 7


f 0 (a) =
4(a2 + a + 1)2 (a2 + 4a + 7)2
(1 − a) (7 + 7a + 4a2 )(1 − 5a − 2a2 )
2
= .
4(a2 + a + 1)2 (a2 + 4a + 7)2

Since f 0 (a) ≥ 0 for a ∈ [0, r], and f 0 (a) ≤ 0 for a ∈ [r, ∞), f is increasing on
[0, r] and decreasing on [r, ∞); therefore,

f (a) ≥ f (r) = m.

The equality holds for


a/r = b = c
(or any cyclic permutation).

(c) We only need to prove the homogeneous inequality

bc ca ab 1
+ + ≤
9a2 + 4(a + b + c)2 9b2 + 4(a + b + c)2 9c 2 + 4(a + b + c)2 15

for b = c = 1, that is

1 2a 1
+ 2 ≤ ,
13a2 + 16a + 16 4a + 16a + 25 15

52a4 − 118a3 + 105a2 − 64a + 25 ≥ 0,

(a − 1)2 (52a2 − 14a + 25) ≥ 0.


Since
52a2 − 14a + 25 > 7a2 − 14a + 7 = 7(a − 1)2 ≥ 0,
the conclusion follows. The equality holds for a = b = c = 1.

P 5.9. If a, b, c, d are nonnegative real numbers so that

(3a + 1)(3b + 1)(3c + 1)(3d + 1) = 64,

then
abc + bcd + cda + dab ≤ 1.

(Vasile C., 2014)


360 Vasile Cîrtoaje

Solution. For d = 0, we need to show that

(3a + 1)(3b + 1)(3c + 1) = 64

involves abc ≤ 1. Indeed, by the AM-GM inequality, we have


€ p
4
Š€ p4
Š€ p4
Š Æ
4
64 = (3a + 1)(3b + 1)(3c + 1) ≥ 4 a3 4 b3 4 c 3 = 64 (abc)3 ,

hence abc ≤ 1. Consider further that a, b, c, d > 0 and use the contradiction
method. Assume that
abc + bcd + cda + dab > 1,
and prove that
(3a + 1)(3b + 1)(3c + 1) > 64.
It suffices to show that
abc + bcd + cda + dab ≥ 1
involves
(3a + 1)(3b + 1)(3c + 1) ≥ 64.
Replacing a, b, c, d by 1/a, 1/b, 1/c, 1/d, we need to show that

a + b + c + d = abcd

involves
3 3 3 3
 ‹ ‹ ‹ ‹
+1 +1 +1 + 1 ≥ 64,
a b c d
which is equivalent to

f (a) + f (b) + f (c) + f (d) ≤ −6 ln 2,

where
3
 ‹
f (u) = − ln +1 , u > 0.
u
We have f (0+) = −∞ and

3x 2 6
g(x) = f 0 (1/x) = , g 00 (x) = > 0,
3x + 1 (3x + 1)3

hence g is strictly convex on (0, ∞). By Corollary 3 and Note 1, if a, b, c, d are


positive real numbers so that

a + b + c + d = const ant , abcd = const ant , a ≤ b ≤ c ≤ d,

then
S4 = f (a) + f (b) + f (c) + f (d)
is maximal for a = b = c.
EV Method for Nonnegative Variables 361

Thus, we only need to prove that


‹3 
3 3
 ‹
+1 + 1 ≥ 64
a d

for 3a + d = a3 d, that is

3 a3 − 1
= , 1 < a ≤ d.
d a
Write this inequality as
(3 + a)3 (3 + d) ≥ 64a3 d,
(3 + a)4 (3 + d) ≥ 64a3 d(3 + a),
a−1 4
 ‹
4 1+ (3 + d) ≥ a3 d(3 + a).
4
By Bernoulli’s inequality, we have

a−1 4 a−1
 ‹
1+ ≥1+4· = a.
4 4

Thus, it suffices to show that

4(3 + d) ≥ a2 d(3 + a),

which is equivalent to
12
≥ a3 + 3a2 − 4,
d
4(a3 − 1)
≥ a3 + 3a2 − 4,
a
a4 − a3 − 4a + 4 ≤ 0,
(a − 1)(a3 − 4) ≤ 0.
This is true if a3 ≤ 4. Indeed, we have

3 3 3 a3 − 1 4 − a3
0≤ − = − = .
a d a a a
The proof is completed. The original inequality is an equality for

a = b = c = 1, d =0

(or any cyclic permutation).


362 Vasile Cîrtoaje

P 5.10. If a1 , a2 , . . . , an and p, q are nonnegative real numbers so that

a1 + a2 + · · · + an = p + q, a13 + a23 + · · · + an3 = p3 + q3 ,

then
a12 + a22 + · · · + an2 ≤ p2 + q2 .
(Vasile C., 2013)
Solution. For n = 2, the inequality is an equality. Consider now that n ≥ 3 and
a1 ≤ a2 ≤ · · · ≤ an . We will apply Corollary 5 for k = 3 and m = 2:
• If a1 , a2 , . . . , an are nonnegative real numbers so that a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = p + q, a13 + a23 + · · · + an3 = p3 + q3 ,

then
Sn = a12 + a22 + · · · + an2
is maximal for either a1 = 0 or a2 = a3 = · · · = an .
In the first case a1 = 0, the conclusion follows by induction method. In the
second case, for
a1 = a, a2 = a3 = · · · = an = b,
we need to show that
a2 + (n − 1)b2 ≤ p2 + q2
for
a + (n − 1)b = p + q, a3 + (n − 1)b3 = p3 + q3 .
Since
2(p3 + q3 )
3(p2 + q2 ) = (p + q)2 + ,
p+q
the inequality can be written as
2[a3 + (n − 1)b3 ]
3a2 + 3(n − 1)b2 ≤ [a + (n − 1)b]2 + ,
a + (n − 1)b
which is equivalent to

(n − 1)(n − 2)b2 [3a + (n − 3)b] ≥ 0.

The equality holds when n − 2 of a1 , a2 , . . . , an are equal to zero.

P 5.11. If a, b, c are nonnegative real numbers, then


p p p
a a2 + 4b2 + 4c 2 + b b2 + 4c 2 + 4a2 + c c 2 + 4a2 + 4b2 ≥ (a + b + c)2 .

(Vasile C., 2010)


EV Method for Nonnegative Variables 363

Solution. Due to homogeneity and symmetry, we may assume that


p
a2 + b2 + c 2 = 3, 0 ≤ a ≤ b ≤ c ≤ 3.

Under this assumption, we write the desired inequality as

1
f (a) + f (b) + f (c) + p (a + b + c)2 ≤ 0,
3
where p p
f (u) = −u 4 − u2 , 0≤u≤ 3.
We have
2(x 2 − 2)
g(x) = f 0 (x) = p ,
4 − x2
48
g 00 (x) = .
(4 − x 2 )5/2
p
Since g 00 (x) > 0 for x ∈ (0, 2), g is strictly convex on [0, 3]. According to Corol-
lary 1, if

a + b + c = const ant , a2 + b2 + c 2 = 3 , 0 ≤ a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
is maximal for a = b ≤ c. Thus, we only need to prove the original inequality for
a = b. Since the inequality is an identity for a = b = 0, we may consider a = b = 1
and c ≥ 1. We need to prove that
p p
2 4c 2 + 5 + c c 2 + 8 ≥ (c + 2)2 .

By squaring, the inequality becomes


Æ
c (4c 2 + 5)(c 2 + 8) ≥ 2c 3 + 8c − 1.

This is true if
c 2 (4c 2 + 5)(c 2 + 8) ≥ (2c 3 + 8c − 1)2 ,
which is equivalent to

5c 4 + 4c 3 − 24c 2 + 16c − 1 ≥ 0,

(c − 1)2 (5c 2 + 14c − 1) ≥ 0.


The equality holds for a = b = c, and also for a = b = 0 (or any cyclic permutation).
364 Vasile Cîrtoaje

P 5.12. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then

1 1 1 3 a+b+c
+ + ≤ + .
a + b b + c c + a 2(a + b + c) 3

(Vasile C., 2010)

Solution. Write the inequality in the homogeneous form

1 1 1 3 a+b+c
+ + ≤ + .
a + b b + c c + a 2(a + b + c) ab + bc + ca

Due to homogeneity and symmetry, we may assume that

a + b + c = 1, 0 ≤ a ≤ b ≤ c, ab + bc + ca > 0.

Under this assumption, we write the desired inequality as

3 1
f (a) + f (b) + f (c) ≤ + ,
2 ab + bc + ca
where
1
f (u) = , 0 ≤ u < 1.
1−u
We will apply Corollary 1 to the function f , which satisfies f (1−) = ∞ and

1
g(x) = f 0 (x) = ,
(1 − x)2

6
g 00 (x) = .
(1 − x)4
Since g 00 (x) > 0, g is strictly convex on [0, 1). According to Corollary 1 and Note
3, if
a+b+c =1 , ab + bc + ca = const ant , 0 ≤ a ≤ b ≤ c,
then the sum
S3 = f (a) + f (b) + f (c)
is maximal for a = b ≤ c. Thus, we only need to prove the homogeneous inequality
for a = b = 1 and c ≥ 1; that is,

4 3 2(c + 2)
1+ ≤ + ,
c+1 c+2 2c + 1
which reduces to
(c − 1)2 ≥ 0.
The original inequality is an equality for a = b = c = 1.
EV Method for Nonnegative Variables 365

P 5.13. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


1 1 1 3 a+b+c
+ + ≥ + .
a+b b+c c+a a+b+c 6
(Vasile C., 2010)

Solution. Proceeding in the same manner as in the proof of the preceding P 5.12,
we only need to prove the homogeneous inequality
1 1 1 3 a+b+c
+ + ≥ +
a+b b+c c+a a + b + c 2(ab + bc + ca)
for a = 0 and for a ≤ b = c = 1.
Case 1: a = 0. The homogeneous inequality reduces to
1 1 2 b+c
+ ≥ + ,
b c b+c 2bc
which is equivalent to
(b − c)2 ≥ 0.
Case 2: a ≤ b = c = 1. The homogeneous inequality becomes
1 2 3 a+2
+ ≥ + ,
2 a + 1 a + 2 2(2a + 1)
1 a+2 3 2
− ≥ − ,
2 2(2a + 1) a + 2 a + 1
a−1 a−1
≥ ,
2(2a + 1) (a + 1)(a + 2)
a(a − 1)2 ≥ 0.
The equality holds for a = b = c = 1, and also for
p
a = 0, b=c= 3

(or any cyclic permutation).

P 5.14. Let a, b, c be nonnegative real numbers, no two of which are zero. If

a2 + b2 + c 2 = 3,

then
1 1 1 a+b+c 11
+ + + ≥ .
a+b b+c c+a 9 2(a + b + c)
(Vasile C., 2010)
366 Vasile Cîrtoaje

Solution. Using the same method as in the proof of P 5.12, we only need to prove
the homogeneous inequality

1 1 1 a+b+c 11
+ + + ≥
a + b b + c c + a 3(a + b + c ) 2(a + b + c)
2 2 2

for a = 0 and for a ≤ b = c = 1.


Case 1: a = 0. The homogeneous inequality reduces to

1 1 1 b+c 11
+ + + ≥ ,
b c b + c 3(b + c ) 2(b + c)
2 2

b+c b+c 9
+ ≥ ,
bc 3(b + c ) 2(b + c)
2 2

1 1 9
• ˜
(b + c)2
+ ≥ .
bc 3(b2 + c 2 ) 2
Using the substitution
b2 + c 2
x= , x ≥ 2,
bc
the inequality becomes
1 9
 ‹
(x + 2) 1 + ≥ ,
3x 2
which is equivalent to
6x 2 − 13x + 4 ≥ 0,

x + 2(x − 2)(3x − 1) ≥ 0.

Case 2: a ≤ 1 = b = c. The homogeneous inequality becomes

1 2 a+2 11
+ + ≥ ,
2 a + 1 3(a + 2) 2(a + 2)
2

a+2 a2 − 4a − 1
+ ≥ 0,
3(a2 + 2) 2(a + 1)(a + 2)
3a4 − 10a3 + 13a2 − 8a + 2 ≥ 0,

(a − 1)2 (3a2 − 4a + 2) ≥ 0,

(a − 1)2 [a2 + 2(a − 1)2 ] ≥ 0.

The equality holds for a = b = c = 1.


EV Method for Nonnegative Variables 367

P 5.15. Let a, b, c be nonnegative real numbers, no two of which are zero. If

a + b + c = 4,

then
1 1 1 15
+ + ≥ .
a + b b + c c + a 8 + ab + bc + ca
(Vasile C., 2010)
Solution. Using the same method as in P 5.12, we only need to prove the homo-
geneous inequality
2 2 2 15(a + b + c)
+ + ≥
a + b b + c c + a (a + b + c)2 + 2(ab + bc + ca)
for a = 0 and for a ≤ b = c = 1.
Case 1: a = 0. The homogeneous inequality reduces to
2(b + c) 2 15(b + c)
+ ≥ ,
bc b+c (b + c)2 + 2bc
2(b + c)2 15(b + c)2
+2≥ .
bc (b + c)2 + 2bc
Using the substitution
(b + c)2
x= , x ≥ 4,
bc
the inequality becomes
15x
2x + 2 ≥ ,
x +2
which is equivalent to
2x 2 − 9x + 4 ≥ 0,
(x − 4)(2x − 1) ≥ 0.
Case 2: a ≤ 1, b = c = 1. The homogeneous inequality becomes
4 15(a + 2)
1+ ≥ ,
a + 1 (a + 2)2 + 2(2a + 1)
a+5 15(a + 2)
≥ 2 ,
a + 1 a + 8a + 6
a(a − 1)2 ≥ 0.
The equality holds for a = b = c = 4/3, and also for

a = 0, b=c=2

(or any cyclic permutation).


368 Vasile Cîrtoaje

P 5.16. If a, b, c are nonnegative real numbers, no two of which are zero, then
1 1 1 1 2
+ + ≥ +p .
a+b b+c c+a a+b+c ab + bc + ca
(Vasile C., 2010)
Solution. Using the same method as in P 5.12, we only need to prove the desired
homogeneous inequality for a = 0 and for 0 < a ≤ b = c = 1.
Case 1: a = 0. The inequality reduces to the obvious form
1 1 2
+ ≥p .
b c bc
Case 2: 0 < a ≤ 1 = b = c. The inequality becomes
1 2 1 2
+ ≥ +p ,
2 a+1 a+2 2a + 1
1 1 2 2
− ≥p − ,
2 a+2 2a + 1 a + 1
p
a 2(a + 1 − 2a + 1)
≥ p ,
2(a + 2) (a + 1) 2a + 1
a 2a2
≥ p p .
2(a + 2) (a + 1) 2a + 1 (a + 1 + 2a + 1)
Since
p p p p p
2a + 1 (a + 1 + 2a + 1) ≥ 2a + 1( 2a + 1 + 2a + 1) = 2(2a + 1),

it suffices to show that


a a2
≥ ,
2(a + 2) (a + 1)(2a + 1)
which is equivalent to
a(1 − a) ≥ 0.
The equality holds for
a = 0, b=c
(or any cyclic permutation).

P 5.17. If a, b, c are nonnegative real numbers, no two of which are zero, then
p p
1 1 1 3− 3 2+ 3
+ + ≥ + p .
a+b b+c c+a a + b + c 2 ab + bc + ca

(Vasile C., 2010)


EV Method for Nonnegative Variables 369

Solution. As shown in the proof of P 5.12, it suffices to consider the cases a = 0


and a ≤ b = c = 1.
Case 1: a = 0. The inequality reduces to
p p
1 1 2− 3 2+ 3
+ ≥ + p .
b c b+c 2 bc
It suffices to show that p p
1 1 2− 3 2+ 3
+ ≥ p + p ,
b c 2 bc 2 bc
which is equivalent to the obvious inequality
1 1 2
+ ≥p .
b c bc
Case 2: a ≤ 1 = b = c. The inequality reduces to
p p
1 2 3− 3 2+ 3
+ ≥ + p .
2 a+1 a+2 2 2a + 1
Using the substitution p
3
2a + 1 = 3x , 2
x≥ ,
3
the inequality becomes
p p
1 4 6−2 3 2+ 3
+ ≥ + p ,
2 3x 2 + 1 3(x 2 + 1) 2 3 x
1 4 2 1 1 2
+ 2 − 2 − ≥p −p ,
2 3x + 1 x + 1 2x 3x 3 (x 2 + 1)
3x 5 − 3x 4 − 4x 2 + 5x − 1 1 1 2
 ‹
≥p − ,
2x(x 2 + 1)(3x 2 + 1) 3 x x2 + 1
(x − 1)2 (3x 3 + 3x 2 + 3x − 1) (x − 1)2
≥ p .
2x(x 2 + 1)(3x 2 + 1) 3 x(x 2 + 1)
This is true if p
3x 3 + 3x 2 + 3x − 1 3
≥ ,
2(3x + 1)
2 3
which is equivalent to
p p
9x 3 + 3(3 − 2 3)x 2 + 9x − 3 − 2 3 ≥ 0,
p p p
(3x − 3 )[3x 2 + (3 − 3)x + 2 + 3] ≥ 0.
The equality holds for a = b = c, and also for
a = 0, b=c
(or any cyclic permutation).
370 Vasile Cîrtoaje

P 5.18. Let a, b, c be nonnegative real numbers, no two of which are zero, so that

ab + bc + ca = 3.

If p
9+5 3
0≤k≤ ≈ 2.943,
6
then
2 2 2 9(1 + k)
+ + ≥ .
a+b b+c c+a a + b + c + 3k
(Vasile Cirtoaje and Lorian Saceanu, 2014)

Solution. From
(a + b + c)2 ≥ 3(ab + bc + ca),
we get
a + b + c ≥ 3.
Let p
9+5 3
m= , m ≥ k.
6
We claim that
1+m 1+k
≥ .
a + b + c + 3m a + b + c + 3k
Indeed, this inequality is equivalent to the obvious inequality

(m − k)(a + b + c − 3) ≥ 0.

Thus, we only need to show that

2 2 2 9(1 + m)
+ + ≥ ,
a+b b+c c+a a + b + c + 3m
which can be rewritten in the homogeneous form

2 2 2 9(1 + m)
+ + ≥ .
a+b b+c c+a
p
a + b + c + m 3(ab + bc + ca)

As shown in the proof of P 5.12, it suffices to prove this homogeneous inequality


for a = 0 and for a ≤ b = c = 1.
Case 1: a = 0. The inequality reduces to

2 2 2 9(1 + m)
+ + ≥ p .
b c b+c b + c + m 3bc
Substituting
b+c
x= p , x ≥ 2,
bc
EV Method for Nonnegative Variables 371

the inequality becomes


2 9(1 + m)
2x + ≥ p ,
x x +m 3
p p
2x 3 + 2 3 mx 2 − (7 + 9m)x + 2 3 m ≥ 0,
p p
(x − 2)[2x 2 + 2( 3 m + 2)x − 3 m] ≥ 0.
Case 2: a ≤ 1 = b = c. The inequality has the form

4 9(1 + m)
1+ ≥ .
a + 1 a + 2 + m 3(2a + 1)
p

Using the substitution


p
3
2a + 1 = 3x ,
2
x≥ ,
3
the inequality becomes
3x 2 + 9 6(1 + m)
≥ 2 ,
3x + 1
2 x + 2mx + 1
x 4 + 2mx 3 − 2(3m + 1)x 2 + 6mx + 1 − 2m ≥ 0,

(x − 1)2 [x 2 + 2(m + 1)x + 1 − 2m] ≥ 0,


which is true since
p
1 2(m + 1) 3
x + 2(m + 1)x + 1 − 2m ≥ +
2
+ 1 − 2m
3 p 3 p
2[2 + 3 − (3 − 3)m]
= = 0.
3
p
9+5 3
The equality holds for a = b = c = 1. If k = , then the equality holds also
6
for p
a = 0, b=c= 3
(or any cyclic permutation).

P 5.19. If a, b, c are nonnegative real numbers, no two of which are zero, then

1 1 1 20
+ + ≥ p .
a+b b+c c+a a + b + c + 6 ab + bc + ca

(Vasile C., 2010)


372 Vasile Cîrtoaje

Solution. The proof is similar to the one of P 5.12. Finally, we only need to prove
the inequality for a = 0 and for a ≤ b = c = 1.
Case 1: a = 0. The inequality reduces to

1 1 1 20
+ + ≥ p .
b c b+c b + c + 6 bc

Substituting
b+c
x= p , x ≥ 2,
bc
the inequality becomes
1 20
x+ ≥ ,
x x +6
x 3 + 6x 2 − 19x + 6 ≥ 0,

(x − 2)(x 2 + 8x − 3) ≥ 0.

Case 2: a ≤ 1 = b = c. We need to show that

1 2 20
+ ≥ p .
2 a + 1 a + 2 + 6 2a + 1

Using the substitution


2a + 1 = x 2 , x ≥ 1,
the inequality becomes
x2 + 9 40
≥ ,
2(x 2 + 1) x 2 + 12x + 3

x 4 + 12x 3 − 68x 2 + 108x − 53 ≥ 0,

(x − 1)(x 3 + 13x 2 − 55x + 53) ≥ 0.


It is true since

x 3 + 13x 2 − 55x + 53 = (x − 1)3 + 16x 2 − 58x + 54


29 2 23
 ‹
= (x − 1) + 16 x −
3
+ > 0.
16 16

The equality holds for


a = 0, b=c
(or any cyclic permutation).
EV Method for Nonnegative Variables 373

P 5.20. If a, b, c are positive real numbers so that

7(a2 + b2 + c 2 ) = 11(ab + bc + ca),

then
51 a b c
≤ + + ≤ 2.
28 b+c c+a a+b
(Vasile C., 2008)

Solution. Due to homogeneity and symmetry, we may consider that

a + b + c = 1, 0 < a ≤ b ≤ c < 1.

Thus, we need to show that


11
a + b + c = 1, a2 + b2 + c 2 = , 0<a≤ b≤c<1
25
involves
51 a b c
≤ + + ≤ 2.
28 1 − a 1 − b 1 − c
We apply Corollary 1 to the function
u
f (u) = , 0 ≤ u < 1.
1−u
We have f (1−) = ∞ and
1 6
g(x) = f 0 (x) = , g 00 (x) = .
(1 − x)2 (1 − x)4

Since g 00 (x) > 0, g is strictly convex on [0, 1). According to Corollary 1 and Note
3, if
11
a+b+c =1 , a2 + b2 + c 2 = , 0 ≤ a ≤ b ≤ c < 1,
25
then the sum
S3 = f (a) + f (b) + f (c)
is maximal for a = b ≤ c, and is minimal for either a = 0 or 0 < a ≤ b = c. Note
that the case a = 0 is not possible because it involves 7(b2 + c 2 ) = 11bc, which is
false.
(1) To prove the right original inequality for a = b ≤ c, let us denote
c
t= , t ≥ 1.
a
The hypothesis 7(a2 + b2 + c 2 ) = 11(ab + bc + ca) involves t = 3, hence
a b c 2a c 2 t
+ + = + = + = 2.
b+c c+a a+b a + c 2a 1 + t 2
374 Vasile Cîrtoaje

c
The right inequality is an equality for a = b = (or any cyclic permutation).
3
(2) To prove the left original inequality for 0 < a ≤ b = c, let us denote
a
t= , 0 < t ≤ 1.
b
1
The hypothesis 7(a2 + b2 + c 2 ) = 11(ab + bc + ca) involves t = , hence
7
a b c a 2b t 2 51
+ + = + = + = .
b + c c + a a + b 2b a + b 2 t + 1 28
The left inequality is an equality for 7a = b = c (or any cyclic permutation).

P 5.21. If a1 , a2 , . . . , an are nonnegative real numbers so that

a12 + a22 + · · · + an2  a + a + · · · + a 2


1 2 n
= ,
n+3 n+1
then
(n + 1)(2n − 1) 1 1 1 3n2 (n + 1)
 ‹
≤ (a1 + a2 + · · · + an ) + + ··· + ≤ .
2 a1 a2 an 2(n + 2)

(Vasile C., 2008)

Solution. For n = 2, both inequalities are identities. For n ≥ 3, assume that

a1 ≤ a2 ≤ · · · ≤ an .

The case a1 = 0 is not possible because the hypothesis involves

a22 + · · · + an2 n+3 1


= < ,
(a2 + · · · + an )2 (n + 1)2 n−1

which contradicts the Cauchy-Schwarz inequality

a22 + · · · + an2 1
≥ .
(a2 + · · · + an )2 n−1

Due to homogeneity and symmetry, we may consider that

a1 + a2 + · · · + an = n + 1,

which implies
a12 + a22 + · · · + an2 = n + 3.
EV Method for Nonnegative Variables 375

Thus, we need to show that

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = n + 3, 0 < a1 ≤ a2 ≤ · · · ≤ a n

involves
2n − 1 1 1 1 3n2
≤ + + ··· + ≤ .
2 a1 a2 an 2(n + 2)
We apply Corollary 5 for k = 2 and m = −1:
• If a1 , a2 , . . . , an are positive real numbers so that 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = n + 3,

then
1 1 1
Sn = + + ··· +
a1 a2 an
is minimal for
0 < a1 = a2 = · · · = an−1 ≤ an ,
and is maximal for
a1 ≤ a2 = a3 = · · · = an .

(1) To prove the left original inequality, we only need to consider the case

a1 = a2 = · · · = an−1 ≤ an .

The hypothesis
a12 + a22 + · · · + an2  a + a + · · · + a 2
1 2 n
=
n+3 n+1
implies
(n − 1)a12 + an2
˜2
(n − 1)a1 + an
•
= ,
n+3 n+1
(2a1 − an )[2a1 − (n + 2)an ] = 0,
an
a1 = ,
2
hence
1 1 1 n−1 1
 ‹  ‹
(a1 + a2 + · · · + an ) + + ··· + = [(n − 1)a1 + an ] +
a1 a2 an a1 an
a1 a n
 ‹
= (n − 1) + 1 + (n − 1)
2
+
a n a1
(n + 1)(2n − 1)
= .
2
The equality holds for
an
a1 = a2 = · · · = an−1 =
2
376 Vasile Cîrtoaje

(or any cyclic permutation).

(2) To prove the right original inequality, we only need to consider the case

a1 ≤ a2 = a3 = · · · = an .

The hypothesis involves

(a1 − 2an )[(n + 2)a1 − 2an ] = 0,

2an
a1 = ,
n+2
hence
1 1 1 n−1 1
 ‹  ‹
(a1 + a2 + · · · + an ) + + ··· + = [(n − 1)a1 + an ] +
a1 a2 an a1 an
a1 a n
 ‹
= (n − 1) + 1 + (n − 1)
2
+
a n a1
3n2 (n + 1)
= .
2(n + 2)

The equality holds for


2an
a1 = a2 = · · · = an−1 =
n+2
(or any cyclic permutation).

P 5.22. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then

176
abc + bcd + cda + dab ≤ 1 + abcd.
81

(Vasile C., 2005)

Solution. Assume that


a ≤ b ≤ c ≤ d.
For a = 0, we need to show that b + c + d = 3 implies

bcd ≤ 1,

which follows immediately from the AM-GM inequality:


‹3
b+c+d

bcd ≤ = 1.
3
EV Method for Nonnegative Variables 377

For a > 0, rewrite the inequality in the form

1 1 1 1 176
 ‹
abcd + + + ≤1+ abcd
a b c d 81
and apply Corollary 5 for k = 0 and m = −1:
• If

a + b + c + d = 3, abcd = const ant, 0 < a ≤ b ≤ c ≤ d,

then
1 1 1 1
S4 = + + +
a b c d
is maximal for
a ≤ b = c = d.
Thus, we only need to prove the homogeneous inequality

27(a + b + c + d)(abc + bcd + cda + dab) ≤ (a + b + c + d)4 + 176abcd

for a ≤ b = c = d = 1. The inequality becomes

27(a + 3)(3a + 1) ≤ (a + 3)4 + 176a,

a4 + 12a3 − 27a2 + 14a ≥ 0,


a(a − 1)2 (a + 14) ≥ 0.
The equality holds for a = b = c = d = 3/4, and also for

a = 0, b=c=d =1

(or any cyclic permutation).

P 5.23. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then

3
a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + abcd ≤ 1.
4

(Gabriel Dospinescu and Vasile Cirtoaje, 2005)

Solution. Assume that


a ≤ b ≤ c ≤ d.
For a = 0, we need to show that

b2 c 2 d 2 ≤ 1,
378 Vasile Cîrtoaje

which follows immediately from the AM-GM inequality:


‹3
b+c+d

bcd ≤ = 1.
3
For a > 0, rewrite the inequality in the form

1 1 1 1 3
 ‹
2 2 2 2
a b c d + + + + abcd ≤ 1,
a2 b2 c 2 d 2 4
and apply Corollary 5 for k = 0 and m = −2:
• If

a + b + c + d = 3, abcd = const ant, 0 < a ≤ b ≤ c ≤ d,

then
1 1 1 1
S4 = 2
+ 2+ 2+ 2
a b c d
is maximal for a ≤ b = c = d.
Thus, we only need to prove the homogeneous inequality
‹6
a+b+c+d 1

≥ a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + abcd(a + b + c + d)2
3 12
for a ≤ b = c = d = 1; that is, to show that 0 < a ≤ 1 implies
 a 6 1
1+ ≥ 1 + 3a2 + a(a + 3)2 .
3 12
Since  a 3 a2 a3 a2
1+ =1+a+ + >1+a+ ,
3 3 27 3
it suffices to show that
‹2
a2 1

1+a+ ≥ 1 + 3a2 + a(a + 3)2 ,
3 12
which is equivalent to the obvious inequality

4a4 + 3a(1 − a)(15 − 7a) ≥ 0.

The equality holds for


a = 0, b=c=d =1
(or any cyclic permutation).
EV Method for Nonnegative Variables 379

P 5.24. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 3, then

4
a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + (abcd)3/2 ≤ 1.
3

(Vasile C., 2005)

Solution. The proof is similar to the one of the preceding P 5.23. We need to prove
that  a 6 4
1+ ≥ 1 + 3a2 + a3/2
3 3
for 0 ≤ a ≤ 1. Since
2a3/2 ≤ a2 + a,
it suffices to show that  a 6 2 11 2
1+ ≥1+ a+ a .
3 3 3
Since  a 3 a2 a3 a2
1+ =1+a+ + ≥1+a+
3 3 27 3
and
‹2
a2 5 2 1

1+a+ = 1 + 2a + a2 + a3 + a4
3 3 3 9
5 2 2 3
≥ 1 + 2a + a + a ,
3 3
it suffices to show that
5 2 2 11 2
1 + 2a + a2 + a3 ≥ 1 + a + a ,
3 3 3 3
which is equivalent to the obvious inequality

a(1 − a)(2 − a) ≥ 0.

The equality holds for


a = 0, b=c=d =1
(or any cyclic permutation).

P 5.25. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 + 2(abcd)3/2 ≤ 6.

(Vasile C., 2005)


380 Vasile Cîrtoaje

Solution. The proof is similar to the one of P 5.23. We need to prove that

a+3 6
 ‹
6 ≥ 1 + 3a2 + 2a3/2
4
for 0 ≤ a ≤ 1. Since
2a3/2 ≤ a2 + a,
it suffices to show that ‹6
a+3

6 ≥ 1 + a + 4a2 .
4
Using the substitution
1−a 1
x= , 0≤ x ≤ ,
4 4
the inequality becomes

3(1 − x)6 ≥ 3 − 18x + 32x 2 ,

x 2 (13 − 60x + 45x 2 − 18x 3 + 3x 4 ) ≥ 0.


It is true since

2(13 − 60x + 45x 2 − 18x 3 + 3x 4 ) > 25 − 120x + 90x 2 − 40x 3


= 5(1 − 4x)(5 − 4x + 2x 2 ) ≥ 0.

The equality holds for a = b = c = d = 1.

P 5.26. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

11(ab + bc + ca) + 4(a2 b2 + b2 c 2 + c 2 a2 ) ≤ 45.

(Vasile C., 2005)


Solution. Assume that a ≤ b ≤ c. For a = 0, we need to show that b + c = 3
involves
11bc + 4b2 c 2 ≤ 45.
We have ‹2
b+c 9

bc ≤ = ,
2 4
hence
99 81
11bc + 4b2 c 2 ≤ + = 45.
4 4
For a > 0, rewrite the desired inequality in the form
1 1 1 1 1 1
 ‹  ‹
11abc + + + 4a b c
2 2 2
+ + ≤ 45.
a b c a2 b2 c 2
EV Method for Nonnegative Variables 381

According to Corollary 5 (case k = 2 and m < 0), if

a + b + c = 3, abc = const ant, 0 < a ≤ b ≤ c,


1 1 1 1 1 1
then the sums + + and 2 + 2 + 2 are maximal for 0 < a ≤ b = c.
a b c a b c
Therefore, we only need to prove that a + 2b = 3 involves

11(2ab + b2 ) + 4(2a2 b2 + b4 ) ≤ 45,

which is equivalent to

15 − 22b − 13b2 + 32b3 − 12b4 ≥ 0,

(3 − 2b)(1 − b)2 (5 + 6b) ≥ 0,


a(1 − b)2 (5 + 6b) ≥ 0.
The equality holds for a = b = c = 1, and also for

3
a = 0, b=c=
2
(or any cyclic permutation).
Remark. In the same manner, we can prove the following statement:
• If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

abc + bcd + cda + dab + a2 b2 c 2 + b2 c 2 d 2 + c 2 d 2 a2 + d 2 a2 b2 ≤ 8,

with equality for a = b = c = d = 1.

P 5.27. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

a2 b2 + b2 c 2 + c 2 a2 + a3 b3 + b3 c 3 + c 3 a3 ≥ 6abc.

(Vasile C., 2005)

Solution. Assume that a ≤ b ≤ c. For a = 0, the inequality is trivial. For a > 0,


rewrite the desired inequality in the form

1 1 1 1 1 1
 ‹  ‹
abc 2 + 2 + 2 + a b c 2 2 2
+ + ≥ 6.
a b c a3 b3 c 3
According to Corollary 5 (case k = 0 and m < 0), if

a + b + c = 3, abc = const ant, 0 < a ≤ b ≤ c,


382 Vasile Cîrtoaje

1 1 1 1 1 1
then the sums 2
+ 2 + 2 and 3 + 3 + 3 are maximal for 0 < a ≤ b = c.
a b c a b c
Thus, we only need to prove that

2a2 b2 + b4 + 2a3 b3 + b6 ≥ 6ab2

for
a + 2b = 3, 1 ≤ b < 3/2.
The inequality is equivalent to

b3 (14 − 33b + 24b2 − 5b3 ) ≥ 0,

b3 (1 − b)2 (14 − 5b) ≥ 0.


The equality holds for a = b = c = 1, and also for

a = b = 0, c=3

(or any cyclic permutation).

P 5.28. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


€p p p Š
2(a2 + b2 + c 2 ) + 5 a + b + c ≥ 21.

(Vasile C., 2008)

Solution. Apply Corollary 5 for k = 2 and m = 1/2:


• If
a + b + c = 3, a2 + b2 + c 2 = const ant, 0 ≤ a ≤ b ≤ c,
then
p p p
S3 = a+ b+ c
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. We need to show that b + c = 3 involves
€p p Š
2(b2 + c 2 ) + 5 b + c ≥ 21,

which is equivalent to q p
5 3 + 2 bc ≥ 3 + 4bc.
Substituting
p b+c 3
x= bc, 0 ≤ x ≤ = ,
2 2
EV Method for Nonnegative Variables 383

the inequality becomes p


5 3 + 2x ≥ 3 + 4x 2 ,
25(3 + 2x) ≥ (3 + 4x 2 )2 .
This inequality is equivalent to f (x) ≥ 0, where

66
f (x) = + 50 − 24x − 16x 3 , 0 < x ≤ 3/2.
x
Since f is decreasing, we have

f (x) ≥ f (3/2) = 4 > 0.

Case 2: 0 < a ≤ b = c. We need to show that


€p p Š
2(a2 + 2b2 ) + 5 a + 2 b ≥ 21

for
3
a + 2b = 3, 1≤ b< .
2
Write the inequality as
p p
5 3 − 2b + 10 b ≥ 3 + 24b − 12b2 .

Substituting v
p t3
x= b, 1 ≤ x < ,
2
the inequality becomes
p
5 3 − 2x 2 ≥ 3 − 10x + 24x 2 − 12x 4 ,
€ p Š
12(x 2 − 1)2 ≥ 5 3 − 2x − 3 − 2x 2 ,

30(x − 1)2
12(x 2 − 1)2 ≥ p ,
3 − 2x + 3 − 2x 2
which is true if
5
2(x + 1)2 ≥ p .
3 − 2x + 3 − 2x 2
It suffices to show that
5
2(x + 1)2 ≥ ,
3 − 2x
which is equivalent to
1 + 8x − 2x 2 − 4x 3 ≥ 0,
7 4 − 3x
 ‹
x(5 − 4x) +x + ≥ 0.
4 4
384 Vasile Cîrtoaje

Since v
t3 5 4
x< < < ,
2 4 3
the conclusion follows.
The equality holds for a = b = c = 1.

P 5.29. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


v v v
t 1 + 2a t 1 + 2b t 1 + 2c
+ + ≥ 3.
3 3 3

(Vasile C., 2008)

Solution. Write the hypothesis ab + bc + ca = 3 as

(a + b + c)2 = 6 + a2 + b2 + c 2 ,

and apply Corollary 1 to


v
t 1 + 2u
f (u) = , u ≥ 0.
3
We have
1
g(x) = f 0 (x) = p ,
3(1 + 2x)
p
3
g 00 (x) = .
(1 + 2x)5/2
Since g 00 (x) > 0 for x ≥ 0, g is strictly convex on [0, ∞). According to Corollary
1, if

a + b + c = const ant, a2 + b2 + c 2 = const ant, 0 ≤ a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. We need to show that bc = 3 involves
p p p
1 + 2b + 1 + 2c ≥ 3 3 − 1.

By squaring, the inequality becomes


Æ p
b + c + 13 + 2(b + c) ≥ 13 − 3 3.
EV Method for Nonnegative Variables 385

p p
We have b + c ≥ 2 bc = 2 3, hence
Æ p Æ p p p
b+c+ 13 + 2(b + c) ≥ 2 3 + 13 + 4 3 = 4 3 + 1 > 13 − 3 3.

Case 2: 0 < a ≤ b = c. From ab + bc + ca = 3, it follows that

3 − b2 p
a= . 0< b< 3.
2b
Thus, the inequality can be written as
v
t 3 − b2 p p
1+ + 2 1 + 2b ≥ 3 3.
b
Substituting
v
t 1 + 2b
v p
1 1+2 3 5
t
t= , p <t< < ,
3 3 3 4
the inequality turns into
v
t 3 + 4t 2 − 3t 4
≥ 3 − 2t.
2(3t 2 − 1)

By squaring, we need to show that

7 − 8t − 14t 2 + 24t 3 − 9t 4 ≥ 0,

which is equivalent to
(1 − t)2 (7 + 6t − 9t 2 ) ≥ 0.
This is true since
‹2
15 7

7 + 6t − 9t = 8 − (3t − 1) > 8 −
2 2
−1 = > 0.
4 16

The equality holds for a = b = c = 1.

P 5.30. Let a, b, c be nonnegative real numbers, no two of which are zero. If

0 ≤ k ≤ 15,

then
1 1 1 k 9+k
+ + + ≥ .
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 4(ab + bc + ca)
(Vasile C., 2007)
386 Vasile Cîrtoaje

Solution. Due to homogeneity and symmetry, we may consider that

a + b + c = 1, 0 ≤ a ≤ b ≤ c.

On this assumption, the inequality becomes

1 1 1 9+k
+ + +k≥ .
(1 − a)2 (1 − b)2 (1 − c)2 2(1 − a2 − b2 − c 2 )

To prove it, we apply Corollary 1 to the function

1
f (u) = , 0 ≤ u < 1.
(1 − u)2

We have f (1−) = ∞ and

2 24
g(x) = f 0 (x) = , g 00 (x) = .
(1 − x)3 (1 − x)5

Since g 00 (x) > 0, g is strictly convex on [0, 1). According to Corollary 1 and Note
3, if
a + b + c = 1, a2 + b2 + c 2 = const ant, 0 ≤ a ≤ b ≤ c,
then the sum
S3 = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. For
b c
x= + , x ≥ 2,
c b
the original inequality becomes

1 1 1+k 9+k
+ 2+ ≥ ,
b 2 c (b + c)2 4bc

1+k 9+k
x+ ≥ ,
x +2 4
(x − 2)(4x + 7 − k) ≥ 0.
This is true since
4x + 7 − k ≥ 15 − k ≥ 0.
Case 2: 0 < a ≤ b = c. The original inequality becomes

2 1 k 9+k
+ 2+ ≥ ,
(a + b)2 4b (a + 2b)2 4b(2a + b)

a(a − b)2 ka(4b − a)


+ ≥ 0.
2b2 (a + b)2 (2a + b) 4b(a + 2b)2 (2a + b)
EV Method for Nonnegative Variables 387

The equality holds for


a = 0, b=c
(or any cyclic permutation). If k = 0 (Iran 1996 inequality), then the equality holds
also for a = b = c.

P 5.31. If a, b, c are nonnegative real numbers, no two of which are zero, then

1 1 1 24 8
+ + + ≥ .
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 ab + bc + ca

(Vasile C., 2007)

Solution. As shown in the proof of the preceding P 5.30, it suffices to prove the
inequality for a = 0, and for 0 < a ≤ b = c.
Case 1: a = 0. For
b c
x= + , x ≥ 2,
c b
the original inequality becomes

1 1 25 8
+ + ≥ ,
b2 c 2 (b + c)2 bc

25
x+ ≥ 8,
x +2
(x − 3)2 ≥ 0.
Case 2: 0 < a ≤ b = c. Due to homogeneity, we only need to prove the homoge-
neous inequality for 0 < a ≤ b = c = 1; that is,

2 1 24 8
+ + ≥ .
(a + 1)2 4 (a + 2)2 2a + 1

It suffices to show that


2 8 24
≥ − ,
(a + 1)2 2a + 1 (a + 2)2

which is equivalent to
1 4(1 − a)2
≥ ,
(1 + a)2 (2a + 1)(a + 2)2
a(2a2 + 9a + 12) ≥ 4a2 (a2 − 2).
This is true since
a(2a2 + 9a + 12) ≥ 0 ≥ 4a2 (a2 − 2).
388 Vasile Cîrtoaje

The equality holds for


b c
a = 0, + =3
c b
(or any cyclic permutation).
Remark. Actually, the following generalization holds:
• Let a, b, c be nonnegative real numbers, no two of which are zero. If k ≥ 15, then
p
1 1 1 k 2( k + 1 − 1)
+ + + ≥ ,
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 ab + bc + ca

with equality for


b c p
a = 0, + = k+1−2
c b
(or any cyclic permutation).

P 5.32. If a, b, c are nonnegative real numbers, no two of which are zero, so that

k(a2 + b2 + c 2 ) + (2k + 3)(ab + bc + ca) = 9(k + 1), 0 ≤ k ≤ 6,

then
1 1 1 9k 3
+ + + ≥ + k.
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 4
(Vasile C., 2007)

Solution. Write the inequality in the homogeneous form

4 4 4 36k 9(k + 1)(4k + 3)


+ + + ≥ .
(a + b)2 (b + c)2 (c + a)2 (a + b + c)2 k(a2 + b2 + c 2 ) + (2k + 3)(ab + bc + ca)

As shown in the proof of P 5.30, it suffices to prove this inequality for a = 0, and
for 0 < a ≤ b = c.
Case 1: a = 0. Let
b c
x=+ , x ≥ 2.
c b
The homogeneous inequality becomes

1 1 36k + 4 9(k + 1)(4k + 3)


 ‹
4 2+ 2 + ≥ ,
b c (b + c)2 k(b2 + c 2 ) + (2k + 3)bc

36k + 4 9(k + 1)(4k + 3)


4x + ≥ ,
x +2 kx + 2k + 3
4kx 3 + 4(4k + 3)x 2 − (43k + 3)x − 2(5k + 21) ≥ 0,
EV Method for Nonnegative Variables 389

(x − 2)[4kx 2 + 4(6k + 3)x + 5k + 21] ≥ 0.


Case 2: 0 < a ≤ b = c. We only need to prove the homogeneous inequality for
b = c = 1. The inequality becomes
8 36k 9(k + 1)(4k + 3)
+ 1 + ≥ ,
(a + 1)2 (a + 2)2 ka2 + (4k + 6)a + 4k + 3

ka6 + (10k + 6)a5 − (14k − 12)a4 − (10k + 18)a3 + (17k − 24)a2 + (24 − 4k)a ≥ 0,
a(a − 1)2 [ka3 + 6(2k + 1)a2 + 3(3k + 8)a + 4(6 − k)] ≥ 0.
Clearly, the last inequality is true for 0 ≤ k ≤ 6.
The equality holds for a = b = c, and also for

a = 0, b=c

(or any cyclic permutation).

P 5.33. If a, b, c are nonnegative real numbers, no two of which are zero, then
2 2 2 8 1
(a) + 2 + 2 ≥ 2 + ;
a2 +b 2 b +c 2 c +a 2 a +b +c
2 2 ab + bc + ca
2 2 2 7 6
(b) + + ≥ + ;
a2 + b2 b2 + c 2 c 2 + a2 a2 + b2 + c 2 (a + b + c)2
2 2 2 45
(c) + 2 + 2 ≥ .
a2 +b 2 b +c 2 c +a 2 4(a + b + c ) + ab + bc + ca
2 2 2

(Vasile C., 2007)


Solution. (a) Due to homogeneity and symmetry, we may consider that

a2 + b2 + c 2 = 1, 0 ≤ a ≤ b ≤ c.

On this assumption, the inequality can be written as


2 2 2 2
+ + ≥ 8 + .
1 − a2 1 − b2 1 − c 2 (a + b + c)2 − 1
To prove it, we apply Corollary 1 to the function
1
f (u) = , 0 ≤ u < 1.
1 − u2
We have f (1−) = ∞ and

2x 24x(1 + x 2 )
g(x) = f 0 (x) = , g 00 (x) = .
(1 − x 2 )2 (1 − x 2 )4
390 Vasile Cîrtoaje

Since g 00 (x) > 0 for x ∈ (0, 1), g is strictly convex on [0, 1). According to Corollary
1 and Note 3, if

a + b + c = const ant, a2 + b2 + c 2 = 1, 0 ≤ a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. For
b c
x= + , x ≥ 2,
c b
the original inequality becomes

2 2 6 1
+ 2≥ 2 + ,
b 2 c b +c 2 bc
6
2x ≥ + 1,
x
(x − 2)(2x + 3) ≥ 0.
Case 2: 0 < a ≤ b = c. Due to homogeneity, it suffices to prove the original
inequality for b = c = 1. Thus, we need to show that

4 8 1
1+ ≥ + ,
a2 + 1 a2 + 2 2a + 1
which is equivalent to
2a 4a2
≥ 2 ,
2a + 1 (a + 1)(a2 + 2)
a(a4 − a2 − 2a + 2) ≥ 0,
a(a − 1)2 (a2 + 2a + 2) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permu-
tation).
(b) The proof is similar to the one of the inequality in (a). For a = 0 and

b c
x= + , x ≥ 2,
c b
the original inequality becomes

2 2 5 6
+ 2≥ 2 + ,
b 2 c b +c 2 (b + c)2

5 6
2x ≥ + ,
x x +2
EV Method for Nonnegative Variables 391

(x − 2)(2x 2 + 8x + 5) ≥ 0.
For b = c = 1, the original inequality is

4 7 6
1+ ≥ 2 + ,
a2 + 1 a + 2 (a + 2)2

a(a5 + 4a4 − 2a3 − 15a + 12) ≥ 0,


a(a − 1)2 (a3 + 6a2 + 9a + 12) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permu-
tation).

(c) The proof is also similar to the one of the inequality in (a). For a = 0 and

b c
x= + , x ≥ 2,
c b
the original inequality becomes

1 1 2 45
 ‹
2 2+ 2 + 2 ≥ ,
b c b +c 2 4(b + c 2 ) + bc
2

2 45
2x + ≥ ,
x 4x + 1
(x − 2)(8x 2 + 18x − 1) ≥ 0.
For b = c = 1, the original inequality can be written as

4 45
1+ ≥ ,
a2 + 1 4a2 + 2a + 9

a(2a3 + a2 − 8a + 5) ≥ 0,
a(a − 1)2 (2a + 5) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permu-
tation).

P 5.34. If a, b, c are nonnegative real numbers, no two of which are zero, then

1 1 1 3 4
+ + + ≥ .
a2 + b2 b2 + c 2 c 2 + a2 a2 + b2 + c 2 ab + bc + ca

(Vasile C., 2007)


392 Vasile Cîrtoaje

Solution. As shown in the proof of the preceding P 5.33, it suffices to prove the
inequality for a = 0, and for 0 < a ≤ b = c.
Case 1: a = 0. For
b c
x= + , x ≥ 2,
c b
the original inequality becomes
1 1 4 4
+ 2+ 2 ≥ ,
b 2 c b +c 2 bc
4
x+ ≥ 4,
x
(x − 2)2 ≥ 0.
Case 2: 0 < a ≤ b = c. Due to homogeneity, it suffices to prove the original
inequality for 0 < a ≤ b = c = 1. Thus, we need to show that
1 2 3 4
+ 2 + 2 ≥ .
2 a + 1 a + 2 2a + 1
It suffices to show that
2 3 4 1
+ ≥ − ,
a + 1 a + 2 2a + 1 2
which is equivalent to
5a + 7 7 − 2a
≥ ,
a2 + 3a + 2 4a + 2
a(2a2 + 19a + 21) ≥ 0,

The equality holds for


a = 0, b=c
(or any cyclic permutation).
Remark. Actually, the following generalization holds:
• Let a, b, c be nonnegative real numbers, no two of which are zero.
(a) If −4 ≤ k ≤ 3, then
2 2 2 2k k+5
+ 2 + 2 + 2 ≥ ,
a2 +b 2 b +c 2 c +a 2 a +b +c
2 2 ab + bc + ca
with equality for
a = 0, b=c
(or any cyclic permutation).
(b) If k ≥ 3, then
p
1 1 1 k 2 k+1
+ + + ≥ ,
a2 + b2 b2 + c 2 c 2 + a2 a2 + b2 + c 2 ab + bc + ca
EV Method for Nonnegative Variables 393

with equality for


b c p
a = 0, + = k+1
c b
(or any cyclic permutation).

P 5.35. If a, b, c are nonnegative real numbers, no two of which are zero, then
3 3 3 5 4
(a) + 2 + 2 ≥ + 2 ;
a2 + ab + b 2 b + bc + c 2 c + ca + a 2 ab + bc + ca a + b2 + c 2
3 3 3 1 24
(b) + 2 + 2 ≥ + ;
a2 + ab + b 2 b + bc + c 2 c + ca + a 2 ab + bc + ca (a + b + c)2
1 1 1 21
(c) + 2 + 2 ≥ .
a2 + ab + b b + bc + c c + ca + a
2 2 2 2(a + b + c ) + 5(ab + bc + ca)
2 2 2

(Vasile C., 2007)


Solution. (a) Due to homogeneity and symmetry, we may consider that

a + b + c = 1, 0 ≤ a ≤ b ≤ c.

Let
1 + a2 + b2 + c 2
p= .
2
Since
1 1 1
= = ,
2(b2 + bc + c ) (a + b + c) + a + b + c − 2a(a + b + c) 2(p − a)
2 2 2 2 2

the inequality can be written as


3 3 3 5 4
+ + ≥ + .
p−a p−b p−c 1 − p 2p − 1
To prove it, we apply Corollary 1 to the function
3
f (u) = , 0 ≤ u < p.
p−u
We have f (p−) = ∞ and
3 18
g(x) = f 0 (x) = , g 00 (x) = .
(p − x)2 (p − x)4
Since g 00 (x) > 0, g is strictly convex on [0, p). According to Corollary 1 and Note
3, if

a + b + c = 1, a2 + b2 + c 2 = 2p − 1 = const ant, 0 ≤ a ≤ b ≤ c,
394 Vasile Cîrtoaje

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. For
b c
x= + , x ≥ 2,
c b
the original inequality becomes

1 1 3 5 4
 ‹
3 2+ 2 + 2 ≥ + 2 ,
b c b + bc + c 2 bc b + c 2

which is equivalent to
3 4
3x + ≥5+ ,
x +1 x
(x − 2)(3x 2 + 4x + 2) ≥ 0.
Case 2: 0 < a ≤ b = c. Due to homogeneity, it suffices to prove the original
inequality for b = c = 1. Thus, we need to show that

6 5 4
+1≥ + 2 ,
a2 +a+1 2a + 1 a + 2
which is equivalent to

a(a4 − a3 + 3a2 − 7a + 4) ≥ 0,

a(a − 1)2 (a2 + a + 4) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permuta-
tion).
(b) The proof is similar to the one of the inequality in (a). For a = 0, the
original inequality becomes

1 1 3 1 24
 ‹
3 2+ 2 + 2 ≥ + ,
b c b + bc + c 2 bc (b + c)2

which is equivalent to

3 24 b c
3x + ≥1+ , x= + ,
x +1 x +2 c b

(x − 2)(3x 2 + 14x + 10) ≥ 0.


For b = c = 1, the original inequality becomes

6 1 24
+1≥ + 2 ,
a2 +a+1 2a + 1 a + 2
EV Method for Nonnegative Variables 395

which is equivalent to

a(a4 + 5a3 − 9a2 − a + 4) ≥ 0,

a(a − 1)2 (a2 + 7a + 4) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permuta-
tion).
(c) The proof is similar to the one of the inequality in (a). For a = 0, the
original inequality becomes

1 1 1 21
+ 2+ 2 ≥ ,
b 2 c b + bc + c 2 2(b + c 2 ) + 5bc
2

which is equivalent to

1 21 b c
x+ ≥ , x= + ,
x + 1 2x + 5 c b

(x − 2)(2x 2 + 11x + 8) ≥ 0.
For b = c = 1, the original inequality becomes

2 1 21
+ ≥ 2 ,
a2 + a + 1 3 2a + 10a + 9
which is equivalent to
a(a3 + 6a2 − 15a + 8) ≥ 0,
a(a − 1)2 (a + 8) ≥ 0.

The equality holds for a = b = c, and also for a = 0, b = c (or any cyclic permuta-
tion).

P 5.36. Let f be a real-valued function, continuous on [0, ∞) and differentiable on


(0, ∞), so that f 000 (u) ≥ 0 for u ∈ (0, ∞). If a, b, c ≥ 0, then

f (a2 + 2bc) + f (b2 + 2ca) + f (c 2 + 2ab) ≤ f (a2 + b2 + c 2 ) + 2 f (ab + bc + ca).

Solution. Denoting

x = a2 + 2bc, y = b2 + 2ca, z = c 2 + 2ab,

the inequality becomes

f (x) + f ( y) + f (z) ≤ f (a2 + b2 + c 2 ) + 2 f (ab + bc + ca).


396 Vasile Cîrtoaje

Assume that

a + b + c = const ant, a2 + b2 + c 2 = const ant,

which involve

2(ab + bc + ca) = (a + b + c)2 − (a2 + b2 + c 2 ) = const ant.

We have
x + y + z = (a + b + c)2 = const ant,
x 2 + y 2 + z 2 = (a2 + b2 + c 2 )2 + 2(ab + bc + ca)2 = const ant.
According to the EV-Theorem (Corollary 1), since f 000 (u) ≥ 0 for u ∈ (0, ∞), the
sum f (x) + f ( y) + f (z) is maximal for x = y ≤ z, that is

a2 + 2bc = b2 + 2ca ≤ c 2 + 2ab.

From a2 + 2bc = b2 + 2ca, we get a = b or a + b = 2c. If a + b = 2c, the


inequality b2 + 2ca ≤ c 2 + 2ab is equivalent to (b − c)2 ≤ 0, which involves b = c.
Thus it suffices to prove the required inequality for two equal variables, when the
inequality is an identity.
The equality holds for a = b or b = c or c = a.
Remark 1. The inequality is also true for a real-valued function f , continuous
on (0, ∞) and differentiable on (0, ∞), so that f 000 (u) ≥ 0 for u ∈ (0, ∞) and
limu→0 f (u) = ±∞.
Remark 2. The following inequalities hold:

1 1 1 1 2
+ 2 + 2 ≥ 2 + ,
+ 2bc b + 2ca c + 2ab
a2 a +b +c
2 2 ab + bc + ca
p p p p p
a2 + 2bc + b2 + 2ca + c 2 + 2ab ≤ a2 + b2 + c 2 + 2 ab + bc + ca,
1 1 1 1 2
p +p +p ≥p +p ,
a2 + 2bc b2 + 2ca c2 + 2ab a2 + b2 + c2 ab + bc + ca

(a2 + 2bc)(b2 + 2ca)(c 2 + 2ab) ≤ (a2 + b2 + c 2 )(ab + bc + ca)2 .

P 5.37. If a, b, c are the lengths of the side of a triangle, then

1 1 1 85
+ + ≤ .
(a + b)2 (b + c)2 (c + a)2 36(ab + bc + ca)

(Vasile C., 2007)


EV Method for Nonnegative Variables 397

Solution. Use the substitution

a = y + z, b = z + x, c = x + y,

where x, y, z are nonnegative real numbers. Due to homogeneity and symmetry,


we may consider that

x + y + z = 2, 0 ≤ x ≤ y ≤ z.

We need to show that


1 1 1 85
+ + ≤ ,
(x + 2)2 ( y + 2)2 (z + 2)2 18(12 − x 2 − y 2 − z 2 )
which can be written as
85
f (x) + f ( y) + f (z) + ≥ 0,
18(12 − x 2 − y 2 − z 2 )
where
−1
f (u) = , u ≥ 0.
(u + 2)2
We have
2 24
g(x) = f 0 (x) = , g 00 (x) = .
(x + 2)3 (x + 2)5
Since g 00 (x) > 0 for x ≥ 0, g is strictly convex on [0, ∞). According to Corollary
1, if
x + y + z = 2, x 2 + y 2 + z 2 = const ant, 0 ≤ x ≤ y ≤ z,
then the sum
S3 = f (x) + f ( y) + f (z)
is minimal for either x = 0 or 0 < x ≤ y = z.
Case 1: x = 0. This implies a = b + c. Since

1 1 5(b2 + c 2 ) + 8bc
+ =
(a + b)2 (c + a)2 (2b2 + 2c 2 + 5bc)2
and
ab + bc + ca = a(b + c) + bc = (b + c)2 + bc = b2 + c 2 + 3bc,
we need to show that
5(b2 + c 2 ) + 8bc 1 85
+ ≤ .
(2b + 2c + 5bc)
2 2 2 (b + c)2 36(b + c 2 + 3bc)
2

For bc = 0, the inequality is true. For bc 6= 0, substituting

b c
t= + , t ≥ 2,
c b
398 Vasile Cîrtoaje

the inequality becomes

5t + 8 1 85
+ ≤ ,
(2t + 5)2 t + 2 36(t + 3)

5t + 8 49t + 62
≤ .
(2t + 5)2 36(t + 2)(t + 3)
It suffices to show that
5t + 8 48t + 64
≤ ,
(2t + 5)2 36(t + 2)(t + 3)

which is equivalent to
5t + 8 12t + 16
≤ ,
(2t + 5)2 9(t + 2)(t + 3)
3t 3 + 7t 2 − 10t − 32 ≥ 0,
(t − 2)(3t 2 + 13t + 16) ≥ 0.
Case 2: 0 < x ≤ y = z. This involves b = c. Since the original inequality is
homogeneous, we may consider b = c = 1 and 0 ≤ a ≤ b + c = 2. Thus, we only
need to show that
1 2 85
+ ≤ ,
4 (a + 1) 2 36(2a + 1)
which is equivalent to
(a − 2)(9a2 − 2a + 1) ≤ 0.

The equality holds for a degenerated isosceles triangle with a = b + c, b = c (or


any cyclic permutation).

P 5.38. If a, b, c are the lengths of the side of a triangle so that a + b + c = 3, then

1 1 1 3(a2 + b2 + c 2 )
+ + ≤ .
(a + b)2 (b + c)2 (c + a)2 4(ab + bc + ca)

(Vasile C., 2007)

Solution. Write the inequality in the homogeneous form

1 1 1 27(a2 + b2 + c 2 )
+ + ≤ .
(a + b)2 (b + c)2 (c + a)2 4(a + b + c)2 (ab + bc + ca)

As shown in the proof of the preceding P 5.37, it suffices to prove this inequality
for a = b + c and for b = c = 1.
EV Method for Nonnegative Variables 399

Case 1: a = b + c. Since
1 1 5(b2 + c 2 ) + 8bc
+ =
(a + b)2 (c + a)2 (2b2 + 2c 2 + 5bc)2
and
27(a2 + b2 + c 2 ) 27(b2 + c 2 + bc)
= ,
4(a + b + c)2 (ab + bc + ca) 8(b + c)2 (b2 + c 2 + 3bc)
we need to show that
5(b2 + c 2 ) + 8bc 1 27(b2 + c 2 + bc)
+ ≤ .
(2b2 + 2c 2 + 5bc)2 (b + c)2 8(b + c)2 (b2 + c 2 + 3bc)
For bc = 0, the inequality is true. For bc 6= 0, substituting
b c
t= + , t ≥ 2,
c b
the inequality becomes
5t + 8 1 27(t + 1)
+ ≤ ,
(2t + 5)2 t + 2 8(t + 2)(t + 3)
9t 2 + 38t + 41 27(t + 1)
≤ .
(2t + 5)2 8(t + 3)
It suffices to show that
9t 2 + 45t + 27 27(t + 1)
≤ ,
(2t + 5)2 8(t + 3)
which is equivalent to
t 2 + 5t + 3 3(t + 1)
≤ ,
(2t + 5)2 8(t + 3)
4t 3 + t(8t − 9) + 3 ≥ 0.

Case 2: b = c = 1, a ≤ b + c = 2. The homogeneous inequality becomes


2 1 27(a2 + 2)
+ ≤ .
(a + 1)2 4 4(2a + 1)(a + 2)2
Since
4(2a + 1)(a + 2) ≤ 9(a + 1)2 ,
it suffices to show that
2 1 3(a2 + 2)
+ ≤ ,
(a + 1)2 4 (a + 1)2 (a + 2)
which is equivalent to
(a − 6)(a − 1)2 ≤ 0.
The equality holds for a an equilateral triangle.
400 Vasile Cîrtoaje

2
P 5.39. Let a, b, c ≥ so that a + b + c = 3. Then,
5
1 1 1 3
+ + ≤ .
3 + 2(a + b ) 3 + 2(b + c ) 3 + 2(c + a ) 7
2 2 2 2 2 2

(Vasile C., 2006)


Solution. For a ≤ b ≤ c, we have
2 11
≤a≤b≤c≤ .
5 5
Indeed,
2 2 11
c =3−a− b ≤3− − = .
5 5 5
Using the substitution
3 3 1 9
m= + a2 + b2 + c 2 , m≥ + (a + b + c)2 = ,
2 2 3 2
we have to show that
6
f (a) + f (b) + f (c) ≤
7
for
3 2 11
a + b + c = 3, a2 + b2 + c 2 = m − , ≤a≤b≤c≤ ,
2 5 5
1 2 11
f (u) = , ≤ u ≤ .
m − u2 5 5
From
2x 24x(m + x 2 )
g(x) = f 0 (x) = , g 00 (x) = ,
(m − x 2 )2 (m − x 2 )4
it follows that g 00 (x) > 0, hence g is strictly convex. By Corollary 1 and Note 2, if
2 11
a + b + c = 3, a2 + b2 + c 2 = const ant, ≤a≤b≤c≤ ,
5 5
then the sum
S3 = f (a) + f (b) + f (c)
is maximal for either c = 11/5 or a = b ≤ c. The case c = 11/5 leads to a = b = 2/5,
when the inequality is an equality. In the second case, we need to prove that
1 2 3
+ ≤
3 + 4a2 3 + 2(a2 + c 2 ) 7
2
for 2a + c = 3, ≤ a ≤ c. Write the inequality as follows
5
1 2 3
+ ≤ ,
3 + 4a 2 21 − 24a + 10a 2 7
EV Method for Nonnegative Variables 401

1 49 − 72a + 30a2
≤ ,
3 + 4a2 7(21 − 24a + 10a2 )
a(a − 1)2 (5a − 2) ≥ 0.

The equality holds for a = b = c = 1, and also for

2 11
a=b= , c=
5 5
(or any cyclic permutation).

Remark In the same manner, we can prove the following statement:


• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If
n2 − 1
k≥ 2 , then
n −n−1
X 1 n
≤ ,
k + a2 + · · · + an
2 2 k+n−1

n2 − 1
with equality for a1 = a2 = · · · = an = 1. If k = 2 , then the equality holds
n −n−1
also for
1 n−1
a1 = · · · = an−1 = 2 , an = n − 2
n −n−1 n −n−1
(or any cyclic permutation).

P 5.40. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

2 2 2 99
+ + ≤ .
2 + a2 + b2 2 + b2 + c 2 2 + c 2 + a2 63 + a2 + b2 + c 2

(Vasile C., 2009)

Solution. The proof is similar to the one of P 5.39. Thus, we only need to prove
the inequality for 0 ≤ a = b ≤ c; that is, to show that 2a + c = 3 involves

1 4 99
+ ≤ .
1 + a2 2 + a2 + c 2 63 + 2a2 + c 2
Write this inequality as follows

1 4 33
+ 2 ≤ ,
a2 + 1 5a − 12a + 11 2(a − 2a + 12)
2

49a4 − 112a3 + 78a2 − 16a + 1 ≥ 0,


402 Vasile Cîrtoaje

(a − 1)2 (7a − 1)2 ≥ 0.

The equality holds for a = b = c = 1, and also for

1 19
a=b= , c=
7 7
(or any cyclic permutation).
Remark. In the same manner, we can prove the following generalization:
8
• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If ≤ k ≤ 3, then
5
k+2 k+2 k+2 9(3k2 + 11k + 10)
+ + ≤ ,
k + a2 + b2 k + b2 + c 2 k + c 2 + a2 9(k2 + 2k + 6) + (5k − 8)(a2 + b2 + c 2 )

with equality for a = b = c = 1, and also for

3−k 2k + 15
a=b= , c=
7 7
(or any cyclic permutation).

P 5.41. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

1 1 1 18
+ + ≤ .
3+a + b
2 2 3+ b +c
2 2 3+c +a
2 2 27 + a + b2 + c 2
2

(Vasile C., 2009)

Solution. The proof is similar to the one of P 5.39. Thus, we only need to prove
the inequality for 0 ≤ a = b ≤ c. Therefore, we only need to show that 2a + c = 3
involves
1 2 18
+ ≤ .
3 + 2a 2 3+a +c
2 2 27 + 2a2 + c 2
Write this inequality as follows

1 2 3
+ 2 ≤ 2 ,
2a2 + 3 5a − 12a + 12 a − 2a + 6
a2 (a − 1)2 ≥ 0.

The equality holds for a = b = c = 1, and also for

a = b = 0, c=3

(or any cyclic permutation).


EV Method for Nonnegative Variables 403

Remark. In the same manner, we can prove the following generalization:


• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If
n
k≥ , then
n−2
X 1 n2 (n + k)
≤ ,
k + a22 + · · · + an2 n(n2 + kn + k2 ) + (kn − n − k)(a12 + a22 + · · · + an2 )

with equality for a1 = a2 = · · · = an = 1, and also for

a1 = · · · = an−1 = 0, an = n

(or any cyclic permutation).

P 5.42. If a, b, c are nonnegative real numbers so that a + b + c = 3, then

5 5 5 27
+ + ≥ .
3+a + b
2 2 3+ b +c
2 2 3+c +a
2 2 6 + a + b2 + c 2
2

(Vasile C., 2014)

Solution. Using the substitution

m = 3 + a2 + b2 + c 2 ,

we have to show that


27
f (a) + f (b) + f (c) ≥
24 + m
for
a + b + c = 3, a2 + b2 + c 2 = m − 3, 0 ≤ a ≤ b ≤ c,
5 p
f (u) = , 0 ≤ u ≤ m − 3.
m − u2
From
10x 120x(m + x 2 )
g(x) = f 0 (x) = , g 00 (x) = ,
(m − x 2 )2 (m − x 2 )4
p
it follows that g 00 (x) ≥ 0 for 0 ≤ x ≤ m − 3, hence g is strictly convex. By
Corollary 1, if

a + b + c = 3, a2 + b2 + c 2 = const ant, 0 ≤ a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
404 Vasile Cîrtoaje

is minimal for either a = 0 or 0 < a ≤ b = c. Write the inequality in the homoge-


neous form
X 5 27
≥ .
(a + b + c) + 3(a + b ) 2(a + b + c) + 3(a2 + b2 + c 2 )
2 2 2 2

Case 1: a = 0. The homogeneous inequality becomes

5 5 5 27
+ + ≥ ,
(b + c)2 + 3b2 (b + c)2 + 3c 2 (b + c)2 + 3(b2 + c 2 ) 2(b + c)2 + 3(b2 + c 2 )

5[5(b2 + c 2 ) + 4bc] 5 27
+ ≥ .
4(b + c ) + 10bc(b + c ) + 13b c
2 2 2 2 2 2 2 4(b + c ) + 2bc
2 2 5(b + c 2 ) + 4bc
2

For the nontrivial case bc 6= 0, substituting

b c
+ = t, t ≥ 2,
c b
we may write the inequality as

5(5t + 4) 5 27
+ ≥ ,
4t 2 + 10t + 13 4t + 2 5t + 4
5(5t + 4) 83t + 34
≥ .
4t 2+ 10t + 13 2(2t + 1)(5t + 4)
Since
83t + 34 ≤ 90t + 20,
it suffices to show that
5t + 4 9t + 2
≥ ,
4t 2 + 10t + 13 (2t + 1)(5t + 4)

which is equivalent to
14t 3 + 7t 2 − 65t − 10 ≥ 0,
(t − 2)(14t 2 + 35t + 5) ≥ 0.
Case 2: 0 < a ≤ b = c. We only need to prove the homogeneous inequality for
b = c = 1; that is,
10 5 27
+ ≥ ,
(a + 2)2 + 3(a + 1) (a + 2) + 6 2(a + 2) + 3(a2 + 2)
2 2 2

10 5 27
+ 2 ≥ 2 ,
4a2 + 4a + 7 a + 4a + 10 5a + 8a + 14
a(a3 − 3a + 2) ≥ 0,
a(a − 1)2 (a + 2) ≥ 0.
EV Method for Nonnegative Variables 405

The equality holds for a = b = c = 1, and also for


3
a = 0, b=c=
2
(or any cyclic permutation).

Remark 1. Similarly, we can prove the following generalization:


• Let a, b, c be nonnegative real numbers so that a + b + c = 3. If k ≥ 0, then
1 1 1 9(4k + 15)
+ + ≥ .
k+a +b k+b +c k+c +a
2 2 2 2 2 2 3(4k + 15k + 9) + (8k + 21)(a2 + b2 + c 2 )
2

with equality for a = b = c = 1, and also for


3
a = 0, b=c=
2
(or any cyclic permutation).
For k = 0, we get the inequality in P 1.171 from Volume 2:
1 1 1 45
+ 2 + 2 ≥ .
a2 +b 2 b +c 2 c +a 2 (a + b + c) + 7(a2 + b2 + c 2 )
2

Remark 2. More general, the following statement holds:


• Let a1 , a2 , . . . , an be nonnegative real numbers so that a1 + a2 + · · · + an = n. If
k ≥ 0, then
X 1 p
≥ ,
k + a2 + · · · + an2
2
q + a1 + a22 + · · · + an2
2

where
n2 (n − 1)2 k + n3 (n2 − n − 1) n(n − 1)2 k2 + n2 (n2 − n − 1)k + n3
p= , q= ,
(n − 1)3 k + n(n3 − 2n2 − n + 1) (n − 1)3 k + n(n3 − 2n2 − n + 1)
with equality for a1 = a2 = · · · = an = 1, and also for
n
a1 = 0, a2 = · · · = a n =
n−1
(or any cyclic permutation).
For k = 0 and k = n, we get the inequalities
X 1 n2 (n2 − n − 1)
≥ ,
a22 + · · · + an2 n2 + (n3 − 2n2 − n + 1)(a12 + a22 + · · · + an2 )
X 2n − 1 n2 (2n − 3)
≥ .
n + a22 + · · · + an2 n(n − 1) + (n − 2)(a12 + a22 + · · · + an2 )
406 Vasile Cîrtoaje

P 5.43. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then


X 3 296
≤ .
3 + 2(a2 + b + c ) 218 + a + b2 + c 2 + d 2
2 2 2

(Vasile C., 2009)

Solution. The proof is similar to the one of P 5.39. Thus, we only need to prove
the inequality for 0 ≤ a = b = c ≤ d, that is to show that 3a + d = 4 involves

1 9 296
+ ≤ .
1 + 2a2 3 + 4a2 + 2d 2 218 + 3a2 + d 2
Write this inequality as follows

1 9 148
+ ≤ ,
1 + 2a 2 35 − 48a + 22a 2 3(39 − 4a + 2a2 )

(a − 1)2 (14a − 1)2 ≥ 0.


The equality holds for a = b = c = d = 1, and also for

1 53
a=b=c= , d=
14 14
(or any cyclic permutation).

P 5.44. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then

4 4 4 21
+ + ≥ .
2 + a2 + b2 2 + b2 + c 2 2 + c 2 + a2 4 + a2 + b2 + c 2

(Vasile C., 2014)

Solution. The proof is similar to the one of P 5.42. Thus, we only need to prove
the inequality for a = 0 and for 0 < a ≤ b = c.
Case 1: a = 0. We need to show that bc = 3 involves
1 1 1 21
+ + ≥ .
2+ b 2 2+c 2 2+ b +c
2 2 4(4 + b2 + c 2 )

Denote
x = b2 + c 2 , x ≥ 2bc = 6.
Since
1 1 4 + b2 + c 2 x +4
+ = = ,
2 + b2 2 + c 2 b2 c 2 + 2(b2 + c 2 ) + 4 2x + 13
EV Method for Nonnegative Variables 407

we only need to show that


x +4 1 21
+ ≥ .
2x + 13 x + 2 4(x + 4)
Since
x +4 1 x 2 + 8x + 21 7(2x + 3)
+ = ≥ ,
2x + 13 x + 2 (2x + 13)(x + 2) (2x + 13)(x + 2)
it suffices to show that
2x + 3 3
≥ .
(2x + 13)(x + 2) 4(x + 4)
This inequality reduces to
(x − 6)(2x + 5) ≥ 0.

Case 2: 0 < a ≤ b = c. Let


q = ab + bc + ca.
We only need to prove the homogeneous inequality
4 4 4 21
+ + ≥
2q + 3(a + b ) 2q + 3(b + c ) 2q + 3(c + a ) 4q + 3(a + b2 + c 2 )
2 2 2 2 2 2 2

for b = c = 1. Thus, we need to show that


8 4 21
+ ≥ ,
2(2a + 1) + 3(a + 1) 2(2a + 1) + 6 4(2a + 1) + 3(a2 + 2)
2

which is equivalent to
8 1 21
+ ≥ ,
3a2 + 4a + 5 a + 2 3a2 + 8a + 10
a2 + 4a + 7 7
≥ 2 ,
(3a + 4a + 5)(a + 2) 3a + 8a + 10
2

a(3a3 − a2 − 7a + 5) ≥ 0,
a(a − 1)2 (3a + 5) ≥ 0.
The equality holds for a = b = c = 1, and also for
p
a = 0, b=c= 3

(or any cyclic permutation).

Remark. In the same manner, we can prove the following generalization:


• Let a, b, c be nonnegative real numbers so that ab + bc + ca = 3. If k ≥ 0, then
1 1 1 9(k + 5)
+ + ≥ .
k + a2 + b2 k + b2 + c 2 k + c 2 + a2 3(k2 + 5k + 2) + 2(k + 4)(a2 + b2 + c 2 )
408 Vasile Cîrtoaje

with equality for a = b = c = 1, and also for


p
a = 0, b=c= 3

(or any cyclic permutation).


For k = 0, we get the inequality in P 1.171 from Volume 2:

1 1 1 45
+ 2 + 2 ≥ .
a2 +b 2 b +c 2 c +a 2 2(ab + bc + ca) + 8(a2 + b2 + c 2 )

P 5.45. If a, b, c are nonnegative real numbers so that a2 + b2 + c 2 = 3, then

1 1 1 1
+ + ≤ .
10 − (a + b)2 10 − (b + c)2 10 − (c + a)2 2

(Vasile C., 2006)

Solution. Let
s = a + b + c, s ≤ 3.
We need to show that
1 1 1 1
+ + ≤
10 − (s − a)2 10 − (s − b)2 10 − (s − c)2 2

for a + b + c = s and a2 + b2 + c 2 = 3. Apply Corollary 1 to the function

−1
f (u) = , 0 ≤ u ≤ s ≤ 3.
10 − (s − u)2

We have
2(s − x)
g(x) = f 0 (x) = ,
[10 − (s − x)2 ]2
24(s − x)[10 + (s − x)2 ]
g 00 (x) = .
[10 − (s − x)2 ]4
Since g 00 (x) > 0 for x ∈ [0, s), g is strictly convex on [0, s]. According to the
Corollary 1, if

a + b + c = s, a2 + b2 + c 2 = 3, 0 ≤ a ≤ b ≤ c,

then
S3 = f (a) + f (b) + f (c)
EV Method for Nonnegative Variables 409

is minimal for either a = 0 or 0 < a ≤ b = c. Therefore, we only need to prove the


homogeneous inequality
X 1 1

10(a + b + c ) − 3(b + c)
2 2 2 2 2(a + b2 + c 2 )
2

for a = 0 and for b = c = 1.


Case 1: a = 0. The homogeneous inequality becomes
1 1 1 1
+ + 2 ≤ .
7(b2 + c ) − 6bc 10b + 7c
2 2 2 7b + 10c 2 2(b + c 2 )
2

This is true since


1 1

7(b2 + c 2 ) − 6bc 4(b2 + c 2 )
and
1 1 17(b2 + c 2 )
+ =
10b2 + 7c 2 7b2 + 10c 2 70(b2 + c 2 ) + 149b2 c 2
17(b2 + c 2 )

70(b2 + c 2 ) + 140b2 c 2
17 1
= < .
70(b + c ) 4(b + c 2 )
2 2 2

Case 2: b = c = 1. The homogeneous inequality turns into


1 2 1
+ 2 ≤ ,
2(5a + 4) 7a − 6a + 17 2(a + 2)
2 2

2 2a2 + 1
≤ ,
7a2 − 6a + 17 (5a2 + 4)(a2 + 2)
4a4 − 12a3 + 13a2 − 6a + 1 ≥ 0,
(a − 1)2 (2a − 1)2 ≥ 0.
The equality holds for a = b = c = 1, and also for
2
2a = b = c = p
3
(or any cyclic permutation).

P 5.46. If a, b, c are nonnegative real numbers, no two of which are zero, so that
a4 + b4 + c 4 = 3, then
1 1 1 3
+ 5 + 5 ≥ .
a5 +b 5 b +c 5 c +a 5 2
(Vasile C., 2010)
410 Vasile Cîrtoaje

Solution. Using the substitution

x = a4 , y = b4 , z = c 4 , p = x 5/4 + y 5/4 + z 5/4 ,

we need to show that x + y + z = 3 and x 5/4 + y 5/4 + z 5/4 = p involve

3
f (x) + f ( y) + f (z) ≥ ,
2
where
1
f (u) = , 0 ≤ u < p4/5 .
p − u5/4
We will apply the EV-Theorem for k = 5/4. We have

5u1/4
f 0 (u) = ,
4(p − u5/4 )2
€ 1 Š 5x
g(x) = f 0 x k−1 = f 0 (x 4 ) = ,
4(p − x 5 )2
75x 4 (2p + 3x 5 )
g 00 (x) = .
2(p − x 5 )4
Since g 00 (x) ≥ 0, g is strictly convex. According to the EV-Theorem and Note 3, if

x + y + z = 3, x 5/4 + y 5/4 + z 5/4 = p = const ant, 0 ≤ x ≤ y ≤ z,

then
S3 = f (x) + f ( y) + f (z)
is minimal for either x = 0 or 0 < x ≤ y = z. Thus, we only need to prove the
homogeneous inequality
‹5/4
1 1 1 3 3

+ + ≥
a5 + b5 b5 + c 5 c 5 + a5 2 a4 + b4 + c 4

for a = 0 and 0 < a ≤ b = c = 1.


Case 1: a = 0. The homogeneous inequality becomes
‹5/4
1 1 1 3 3

+ + ≥ ,
b5 c 5 b5 + c 5 2 b4 + c 4
 5/4
 ‹5/2  5/2  ‹9/4
b c 1 3 2
+ + ≥   ,
c b b 5/2 c 5/2 2 b 2 c 2
 
+
 
c + b c b
 ‹9/4  ‹5/4
1 3 2
t 5/2
+t −5/2
+ 5/2 ≥ ,
t + t −5/2 2 t 2 + t −2
EV Method for Nonnegative Variables 411

 ‹9/4
1 3 1
2A 5/2
+ 5/2 ≥ · 5/2 ,
2A 2 B
where 2/5 ‹1/2
t 5/2 + t −5/2 t 2 + t −2
 
b
A= , B= , t= .
2 2 c
By power mean inequality, we have A ≥ B ≥ 1. Since

1 1 1
 ‹  ‹
2A + 5/2 − 2B + 5/2 = A − B
5/2 5/2 5/2 5/2
2 − 5/2 5/2 ≥ 0,
2A 2B 2A B

it suffices to show that


 ‹9/4
1 3 1
2B 5/2
+ 5/2 ≥ · 5/2 ,
2B 2 B
‹1/4
39

4B + 1 ≥
5
,
25
which is true if
‹1/4
39

5≥ ,
25
32 · 54 ≥ 39 .
This inequality follows by multiplying the inequalities

54 > 23 · 33

and
32 · 23 > 36 .
Case 2: 0 < a ≤ 1 = b = c. The homogeneous inequality becomes
‹5/4
a5 + 5 3

≥3 4 ,
a5 + 1 a +2

which is true if g(a) ≥ 0, where

5 9 ln 3
g(a) = ln(a5 + 5) − ln(a5 + 1) + ln(a4 + 2) − ,
4 4
with

g 0 (a) a a 1 a10 + 2a5 − 8a + 5


= 5 − + =
5a3 a + 5 a5 + 1 a4 + 2 (a4 + 5)(a5 + 1)(a4 + 2)
(a − 1)(a9 + a8 + a7 + a6 + a5 + 3a4 + 3a3 + 3a2 + 3a − 5)
= .
(a4 + 5)(a5 + 1)(a4 + 2)
412 Vasile Cîrtoaje

There exists d ∈ (0, 1) so that g 0 (d) = 0, g 0 (a) > 0 for a ∈ [0, d) and g 0 (a) < 0 for
a ∈ (d, 1). Therefore, g is increasing on [0, d] and is decreasing on [d, 1]. Since
g(1) = 0, we only need to show that g(0) ≥ 0. Indeed,

1 54 · 25
g(0) = ln > 0.
4 39

The equality holds for a = b = c = 1.

P 5.47. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
v 
1
q q Æ t ‹
a1 + 1+ a2 + 1+· · ·+ an + 1 ≥ 2 1 −
2 2 2 (a12 + a22 + · · · + an2 ) + 2(n2 − n + 1).
n

(Vasile C., 2014)

Solution. For n = 2, we need to show that a1 + a2 = 2 involves


q q q
a12 + 1 + a22 + 1 ≥ a12 + a22 + 6.

By squaring, the inequality becomes


q
(a12 + 1)(a22 + 1) ≥ 2,

which follows immediately from the Cauchy-Schwarz inequality:

(a12 + 1)(a22 + 1) = (a12 + 1)(1 + a22 ) ≥ (a1 + a2 )2 = 4.

Assume further that n ≥ 3 and a1 ≤ a2 ≤ · · · ≤ an . We will apply Corollary 1 to the


function p
f (u) = − u2 + 4, u ≥ 0.
We have
−x
g(x) = f 0 (x) = p ,
x2 + 4
12x
g 00 (x) = 2 .
(x + 4)5/2
Since g 00 (x) > 0 for x > 0, g(x) is strictly convex for x ≥ 0. By Corollary 1, if
a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a12 + a22 + · · · + an2 = const ant,

then the sum


Sn = f (a1 ) + f (a2 ) + · · · + f (an )
EV Method for Nonnegative Variables 413

is maximal for a1 = a2 = · · · = an−1 . Thus, we only need to show that


v 
1
p p t ‹
a + 1 + (n − 1) b + 1 ≥ 2 1 −
2 2 [a2 + (n − 1)b2 ] + 2(n2 − n + 1).
n

for
a + (n − 1)b = n.
By squaring, the inequality becomes
Æ
2n(n − 1) (a2 + 1)(b2 + 1) ≥ (n − 2)a2 − (n − 2)(n − 1)2 b2 + n3 ,

which is equivalent to
Æ
(b2 + 1)[(n − 1)2 b2 − 2n(n − 1)b + n2 + 1] ≥ n − (n − 2)b.

This is true if

(b2 + 1)[(n − 1)2 b2 − 2n(n − 1)b + n2 + 1] ≥ [n − (n − 2)b]2 ,

which is equivalent o

(n − 1)2 b4 − 2n(n − 1)b3 + (n2 + 2n − 2)b2 − 2nb + 1 ≥ 0,

(b − 1)2 [(n − 1)b − 1]2 ≥ 0.


The equality holds for a1 = a2 = · · · = an = 1, and also for

1
a1 = a2 = · · · = an−1 = , an = n − 1
n−1
(or any cyclic permutation).

P 5.48. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
Xq q
(3n − 4)a12 + n ≥ (3n − 4)(a12 + a22 + · · · + an2 ) + n(4n2 − 7n + 4).

(Vasile C., 2009)

Solution. The proof is similar to the one of the preceding P 5.47. Thus, it suf-
fices to prove the inequality for a1 = a2 = · · · = an−1 . Write the inequality in the
homogeneous form
Xq q
n(3n − 4)a12 + S 2 ≥ n(3n − 4)(a12 + a22 + · · · + an2 ) + (4n2 − 7n + 4)S 2 ,
414 Vasile Cîrtoaje

where S = a1 + a2 + · · · + an . We only need to prove this inequality for a1 = a2 =


· · · = an−1 = 1, that is
Æ q
(n − 1) n(3n − 4) + (n − 1 + an )2 + n(3n − 4)an2 + (n − 1 + an )2 ≥
q
≥ n(3n − 4)(n − 1 + an2 ) + (4n2 − 7n + 4)(n − 1 + an )2 ,
which is equivalent to
q q
(n − 1)[an2 + 2(n − 1)an + 4n2 − 6n + 1] + (3n − 1)an2 + 2an + n − 1 ≥
q
≥ (7n − 4)an2 + 2(4n2 − 7n + 4)an + 4n3 − 8n2 + 7n − 4.
By squaring, the inequality turns into
q
2 (n − 1)[(3n − 1)an2 + 2an + n − 1][an2 + 2(n − 1)an + 4n2 − 6n + 1] ≥

(3n − 2)an2 + 2(n − 1)(3n − 2)an + 2n2 − n − 2.


Squaring again, we get

(an − 1)2 (an − 2n + 3)2 ≥ 0.

The equality holds for a1 = a2 = · · · = an = 1, and also for


an n
a1 = a2 = · · · = an−1 = =
2n − 3 3n − 4
(or any cyclic permutation).
Remark. For n = 3, we get the inequality
p p p Æ
5a2 + 3 + 5b2 + 3 + 5c 2 + 3 ≥ 5(a2 + b2 + c 2 ) + 57,

where a, b, c are nonnegative real numbers so that a + b + c = 3. By squaring, the


inequality turns into
Æ Æ Æ
(5a2 + 3)(5b2 + 3) + (5b2 + 3)(5c 2 + 3) + (5c 2 + 3)(5a2 + 3) ≥ 24,

with equality for a = b = c = 1, and also for

c 3
a=b= =
3 5
(or any cyclic permutation).
EV Method for Nonnegative Variables 415

P 5.49. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


v
p p p t8
a2 + 4 + b2 + 4 + c 2 + 4 ≤ (a2 + b2 + c 2 ) + 37.
3

(Vasile C., 2009)

Solution. Assume that a ≤ b ≤ c, and apply Corollary 1 to the function a


p
f (u) = − u2 + 4, u ≥ 0.

We have
−x
g(x) = f 0 (x) = p ,
x2 + 4
12x
g 00 (x) = 2 .
(x + 4)5/2
Since g 00 (x) > 0 for x > 0, g(x) is strictly convex for x ≥ 0. By Corollary 1, if

a + b + c = 3, a2 + b2 + c 2 = const ant , a ≤ b ≤ c,

then the sum


S3 = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c. Thus, we only need to prove the
desired inequality for these cases.
Case 1: a = 0. We need to prove that b + c = 3 involves
v
p p t8
b2 + 4 + c 2 + 4 ≤ (b2 + c 2 ) + 37 − 2.
3
Substituting
3x 3y
b= , c= ,
2 2
we need to prove that x + y = 2 involves
p p Æ
9x 2 + 16 + 9 y 2 + 16 ≤ 2 6(x 2 + y 2 ) + 37 − 4.

By squaring, the inequality becomes


Æ Æ
2 (9x 2 + 16)(9 y 2 + 16) ≤ 15(x 2 + y 2 ) + 132 − 16 6(x 2 + y 2 ) + 37.

Denoting
p = x y, 0 ≤ p ≤ 1,
we have

x 2 + y 2 = 4 − 2p, (9x 2 + 16)(9 y 2 + 16) = 81p2 − 288p + 832,


416 Vasile Cîrtoaje

and the inequality becomes


p p
81p2 − 288p + 832 ≤ −15p + 96 − 8 61 − 12p,
v
t 81 15 p
p2 − 72p + 208 ≤ − p + (48 − 4 61 − 12p),
4 2
By squaring again (the right hand side is positive), the inequality becomes
81 2 225 2 p p
p − 72p + 208 ≤ p − 15p(48 − 4 61 − 12p) + (48 − 4 61 − 12p)2 ,
4 4
p
3p2 − 70p + 256 ≥ (32 − 5p) 61 − 12p.
Since
p 61 − 12p 2(55 − 6p)
2 61 − 12p ≤ 7 + = ,
7 7
it suffices to show that
7(3p2 − 70p + 256) ≥ (32 − 5p)(55 − 6p),
which is equivalent to
(1 − p)(32 + 9p) ≥ 0.
Case 2: b = c. We need to prove that
a + 2b = 3
implies v
p p t8
a2 + 4 + 2 b2 + 4 ≤ (a2 + 2b2 ) + 37.
3
By squaring, the inequality becomes
Æ
12 (a2 + 4)(b2 + 4) ≤ 5a2 + 4b2 + 51,
which is equivalent to
Æ
(4b2 − 12b + 13)(b2 + 4) ≤ 2b2 − 5b + 8.
By squaring again, the inequality becomes
2b3 − 7b2 + 8b − 3 ≤ 0,
(b − 1)2 (2b − 3) ≤ 0,
(b − 1)2 a ≥ 0.
The equality holds for a = b = c = 1, and also for
3
a = 0, b=c=
2
(or any cyclic permutation).
EV Method for Nonnegative Variables 417

P 5.50. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


p p p Æ
32a2 + 3 + 32b2 + 3 + 32c 2 + 3 ≤ 32(a2 + b2 + c 2 ) + 219.

(Vasile C., 2009)

Solution. The proof is similar to the one of P 5.49. Thus, it suffices to prove the
homogeneous inequality
XÆ Æ
96a2 + (a + b + c)2 ≤ 96(a2 + b2 + c 2 ) + 73(a + b + c)2

for a = 0 and for b = c = 1.


Case 1: a = 0. We have to show that
p p Æ
b + c + 97b2 + 2bc + c 2 + b2 + 2bc + 97c 2 ≤ 169(b2 + c 2 ) + 146bc.

Since 2bc ≤ b2 + c 2 , it suffices to prove that


p p Æ
b + c + 98b2 + 2c 2 + 2b2 + 98c 2 ≤ 169(b2 + c 2 ) + 146bc.

By squaring, we get
€p p Š Æ
(b + c) 98b2 + 2c 2 + 2b2 + 98c 2 + 2 (49b2 + c 2 )(b2 + 49c 2 ) ≤

≤ 34(b2 + c 2 ) + 72bc.
Since
p p Æ Æ
98b2 + 2c 2 + 2b2 + 98c 2 ≤ 2(98b2 + 2c 2 + 2b2 + 98c 2 ) = 10 2(b2 + c 2 )

and Æ
10(b + c) 2(b2 + c 2 ) ≤ 20(b + c)2 ,
it suffices to show that
Æ
(49b2 + c 2 )(b2 + 49c 2 ) ≤ 7(b2 + c 2 ) + 36bc.

Squaring again, the inequality becomes

bc(b − c)2 ≥ 0.

Case 2: b = c = 1. The homogeneous inequality turns into


p p p
97a2 + 4a + 4 + 2 a2 + 4a + 100 ≤ 169a2 + 292a + 484.

By squaring, we get
Æ
(97a2 + 4a + 4)(a2 + 4a + 100) ≤ 17a2 + 68a + 20.
418 Vasile Cîrtoaje

Squaring again, the inequality reduces to

a(a − 1)2 (a + 12) ≥ 0.

The equality holds for a = b = c = 1, and also for a = 0 and b = c = 3/2 (or any
cyclic permutation).
Remark. By squaring, we deduce the inequality
Æ Æ Æ
(32a2 + 3)(32b2 + 3) + (32b2 + 3)(32c 2 + 3) + (32c 2 + 3)(32a2 + 3) ≤ 105,

with equality for a = b = c = 1, and also for


3
a = 0, b=c=
2
(or any cyclic permutation).

P 5.51. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then


p
1 1 1 2n n − 1 p
+ + ··· + + 2 ≥ n + 2 n − 1.
a1 a2 an a1 + a22 + · · · + an2

(Vasile C., 2009)

Solution. For n = 2, the inequality reduces to

(a1 a2 − 1)2 ≥ 0.

Consider further that n ≥ 3 and a1 ≤ a2 ≤ · · · ≤ an . By Corollary 5 (case k = 2 and


m = −1), if a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a12 + a22 + · · · + an2 = const ant,

then the sum


1 1 1
Sn = + + ··· +
a1 a2 an
is minimal for a1 = · · · = an−1 ≤ an . Therefore, we only need to prove that
p
n−1 1 2n n − 1 p
+ + ≥ n + 2 n − 1,
a1 an (n − 1)a12 + an2

for (n − 1)a1 + an = n. The inequality is equivalent to


 ‹2
n
(a1 − 1) 2
a1 − p ≥ 0.
n−1+ n−1
EV Method for Nonnegative Variables 419

The equality holds for a1 = a2 = · · · = an = 1, and also for


an
a1 = a2 = · · · = an−1 = p
n−1
(or any cyclic permutation).

P 5.52. If a, b, c ∈ [0, 1], then

(1 + 3a2 )(1 + 3b2 )(1 + 3c 2 ) ≥ (1 + ab + bc + ca)3 .

Solution. Since

2(ab + bc + ca) = (a + b + c)2 − (a2 + b2 + c 2 ),

we may apply Corollary 1 to the function

f (u) = − ln(1 + 3u2 ), u ∈ [0, 1],

to prove the inequality

f (a) + f (b) + f (c) + 3 ln(1 + ab + bc + ca) ≤ 0.

We have
−6x
g(x) = f 0 (x) = ,
1 + 3x 2
108x(1 − x 2 )
g (x) =
00
.
(1 + 3x 2 )3
Since g 00 (x) > 0 for x ∈ (0, 1), g is strictly convex on [0, 1]. According to Corollary
1 and Note 2, if

a + b + c = const ant, a2 + b2 + c 2 = const ant, 0 ≤ a ≤ b ≤ c ≤ 1,

then
S3 = f (a) + f (b) + f (c)
is maximal for a = b ≤ c. or for c = 1. Thus, we only need to prove the original
inequality for these cases.
Case 1: a = b ≤ c. We need to show that

(1 + 3a2 )2 (1 + 3c 2 ) ≥ (1 + a2 + 2ac)3 .

For c = 0, the inequality is an equality. For fixed c, 0 < c ≤ 1, we need to show that
h(a) ≥ 0, where

h(a) = 2 ln(1 + 3a2 ) + ln(1 + 3c 2 ) − 3 ln(1 + a2 + 2ac), a ∈ [0, c].


420 Vasile Cîrtoaje

From
12a 6(a + c) 6(1 − a2 )(a − c)
h0 (a) = − = ≤ 0,
1 + 3a2 1 + a2 + 2ac (1 + 3a2 )(1 + a2 + 2ac)
it follows that h is decreasing on [0, c], hence h(a) ≥ h(c) = 0.
Case 2: c = 1. We need to show that

4(1 + 3a2 )(1 + 3b2 ) ≥ (1 + a)3 (1 + b)3 .

This is true because

2(1 + 3a2 ) ≥ (1 + a)3 , 2(1 + 3b2 ) ≥ (1 + b)3 .

The first inequality is equivalent to

(1 − a)3 ≥ 0.

The proof is completed. The equality holds for a = b = c.

P 5.53. If a, b, c are nonnegative real numbers so that a + b + c = ab + bc + ca, then


1 1 1 1
+ + ≥ .
4 + 5a 2 4 + 5a 2 4 + 5a 2 3

(Vasile C., 2007)

Solution. By expanding, the inequality becomes

4(a2 + b2 + c 2 ) + 15 ≥ 25a2 b2 c 2 + 5(a2 b2 + b2 c 2 + c 2 a2 ).

Let p = a + b + c. Since

a2 + b2 + c 2 = p2 − 2p, a2 b2 + b2 c 2 + c 2 a2 = p2 − 2pabc,

the inequality becomes


(2p − 4)2 ≥ (p − 5abc)2 ,
(3p − 4 − 5abc)(p + 5abc − 4) ≥ 0.
We will show that 3p ≥ 4 + 5abc and p + 5abc ≥ 4. According to Corollary 4 (case
n = 3, k = 2) or P 3.57 in Volume 1, if

a + b + c = const ant, ab + bc + ca = const ant, 0 ≤ a ≤ b ≤ c ≤ d,

then the product abc is maximal for a = b, and is minimal for a = 0 or b = c. Thus,
we only need to prove that 3p ≥ 4 + 5abc for a = b, and p + 5abc ≥ 4 for a = 0
and for b = c.
EV Method for Nonnegative Variables 421

For a = b, from a + b + c = ab + bc + ca we get

a(2 − a) 1
c= , < a ≤ 2,
2a − 1 2
hence
(a − 1)2 (5a2 + 4)
3p − 4 − 5abc = (3 − 5a2 )c + 6a − 4 = ≥ 0.
2a − 1
For a = 0, from a + b + c = ab + bc + ca we get
b
c= , b > 1,
b−1
hence
(b − 2)2
p + 5abc − 4 = b + c − 4 = ≥ 0.
b−1
For b = c, from a + b + c = ab + bc + ca we get

b(2 − b) 1
a= , < b ≤ 2,
2b − 1 2
hence
(2 − b)(5b3 − 3b + 2)
p + 5abc − 4 = a(5b2 + 1) + 2b − 4 =
2b − 1
(2 − b)[4b3 + (b − 1)2 (b + 2)]
= ≥ 0.
2b − 1
The equality holds for a = b = c = 1, and also for a = 0 and b = c = 2 (or any
cyclic permutation).

P 5.54. If a, b, c, d are positive real numbers so that a + b + c + d = 4abcd, then


1 1 1 1
+ + + ≥ 1.
1 + 3a 1 + 3b 1 + 3c 1 + 3d

(Vasile C., 2007)

Solution. By expanding, the inequality becomes

1 + 3(ab + ac + ad + bc + bd + cd) ≥ 19abcd,

2 + 3(a + b + c + d)2 ≥ 3(a2 + b2 + c 2 + d 2 ) + 38abcd.


According to Corollary 5 (case n = 4, k = 0, m = 2), if

a + b + c + d = const ant, abcd = const ant, 0 < a ≤ b ≤ c ≤ d,


422 Vasile Cîrtoaje

then the sum


S4 = a2 + b2 + c 2 + d 2
is maximal for a = b = c ≤ d. Thus, we only need to prove that

3a 1
3a + d = 4a3 d, d= , a> p
3
,
4a3 − 1 4
involves
3 1
+ ≥ 1,
3a + 1 3d + 1
3 4a3 − 1
+ 3 ≥ 1,
3a + 1 4a + 9a − 1
4a3 − 9a2 + 6a − 1 ≥ 0,
(a − 1)2 (4a − 1) ≥ 0.
The equality holds for a = b = c = d = 1.

Open problem. If a1 , a2 , . . . , an (n ≥ 3) are positive real numbers so that

a1 + a2 + · · · + an = na1 a2 · · · an ,

then
1 1 1
+ + ··· + ≥ 1.
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an

P 5.55. If a1 , a2 , . . . , an are positive real numbers so that

1 1 1
a1 + a2 + · · · + an = + + ··· + ,
a1 a2 an

then
1 1 1
+ + ··· + ≥ 1.
1 + (n − 1)a1 1 + (n − 1)a2 1 + (n − 1)an
(Vasile C., 1996)

Solution. For n = 2, the inequality is an identity. For n ≥ 3, we consider

a1 ≤ a2 ≤ · · · ≤ an ,

and apply Corollary 2 to the function

1
f (u) = , u > 0.
1 + (n − 1)u
EV Method for Nonnegative Variables 423

We have
−(n − 1)
f 0 (u) = ,
[1 + (n − 1)u]2
1 −(n − 1)x
 ‹
g(x) = f p
0
= p ,
x [ x + n − 1]2
3(n − 1)2
g 00 (x) = p p .
2 x( x + n − 1)4
Since g 00 (x) > 0 for x > 0, g is strictly convex on [0, ∞). By Corollary 2, if
0 < a1 ≤ a2 ≤ · · · ≤ an and
1 1 1
a1 + a2 + · · · + an = const ant, + + ··· + = const ant,
a1 a2 an
then the sum
Sn = f (a1 ) + f (a2 ) + · · · + f (an )
is minimal for a2 = · · · = an . Therefore, we only need to show that
1 n−1
+ ≥1
1 + (n − 1)a 1 + (n − 1)b
for
1 n−1
a + (n − 1)b = + , 0 < a ≤ b.
a b
Write the hypothesis as
1 1
 ‹
− a = (n − 1) b − ,
a b
which involves a ≤ 1 ≤ b and
1 1
−a≥ b− , ab ≤ 1.
a b
Write the desired inequality as
n−1 1
≥1− ,
1 + (n − 1)b 1 + (n − 1)a
which is equivalent to
n−1 (n − 1)a
≥ ,
1 + (n − 1)b 1 + (n − 1)a
1 − a ≥ (n − 1)a(b − 1).
For the nontrivial case b 6= 1, we have
b(1 − a2 ) (1 − a)(1 − ab)
1 − a − (n − 1)a(b − 1) = 1 − a − a(b − 1) = ≥ 0.
a(b2 − 1) b+1
If n ≥ 3, then the equality holds for a1 = a2 = · · · = an = 1.
424 Vasile Cîrtoaje

P 5.56. If a, b, c, d, e are nonnegative real numbers so that a4 + b4 + c 4 + d 4 + e4 = 5,


then
7(a2 + b2 + c 2 + d 2 + e2 ) ≥ (a + b + c + d + e)2 + 10.

(Vasile C., 2008)

Solution. According to Corollary 5 (case n = 5, k = 4, m = 2), if

a + b + c + d + e = const ant, a4 + b4 + c 4 + d 4 + e4 = 5, 0 ≤ a ≤ b ≤ c ≤ d ≤ e,

then the sum


S4 = a2 + b2 + c 2 + d 2 + e2
is minimal for a = b = c = d ≤ e. Thus, we only need to prove the homogeneous
inequality

[7(a2 + b2 + c 2 + d 2 + e2 ) − (a + b + c + d + e)2 ]2 ≥ 20(a4 + b4 + c 4 + d 4 + e4 )

for a = b = c = d = 0 and a = b = c = d = 1. The first case is trivial. In the second


case, the inequality becomes

[7(4 + e2 ) − (4 + e)2 ]2 ≥ 20(4 + e4 ),

(3e2 − 4e + 6)2 ≥ 5e4 + 20,


e4 − 6e3 + 13e2 − 12e + 4 ≥ 0,
(e − 1)2 (e − 2)2 ≥ 0.
The equality holds for a = b = c = d = e = 1, and also for
e 1
a=b=c=d= =p
2 2
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• If a1 , a2 , . . . , an are nonnegative real numbers so that

a14 + a24 + · · · + an4 = n,

then p
(n + n − 1)(a12 + a22 + · · · + an2 − n) ≥ (a1 + a2 + · · · + an )2 − n2 ,
with equality for a1 = a2 = · · · = an = 1, and also for
an 1
a1 = · · · = an−1 = p =p
4
n−1 n−1
(or any cyclic permutation).
EV Method for Nonnegative Variables 425

P 5.57. If a1 , a2 , . . . , an are nonnegative real numbers so that a1 + a2 + · · · + an = n,


then
n(n − 1) 
(a12 + a22 + · · · + an2 )2 − n2 ≥ 2 a14 + a24 + · · · + an4 − n .
n −n+1
(Vasile C., 2008)

Solution. For n = 2, the inequality reduces to (a1 a2 −1)2 ≥ 0. For n ≥ 3, we apply


Corollary 5 for k = 2 and m = 4 : if 0 ≤ a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a12 + a22 + · · · + an2 = const ant,

then
Sn = a14 + a24 + · · · + an4
is maximal for a1 = · · · = an−1 ≤ an . Thus, we only need to prove the homogeneous
inequality

n2 (n2 − n + 1)(a12 + a22 + · · · + an2 )2 ≥ (n2 − 2n + 2)(a1 + a2 + · · · + an )4 + n3 (n − 1)Sn ,

for a1 = · · · = an−1 = 0 and for a1 = · · · = an−1 = 1. For the nontrivial case


a1 = · · · = an−1 = 1, the inequality becomes

n2 (n2 − n + 1)(n − 1 + an2 )2 ≥ (n2 − 2n + 2)(n − 1 + an )4 + n3 (n − 1)(n − 1 + an4 ),

(an − 1)2 [an − (n − 1)2 ]2 ≥ 0.


The equality holds for a1 = a2 = · · · = an = 1, and also for

1
a1 = · · · = an−1 = , an = n − 1
n−1
(or any cyclic permutation).

P 5.58. If a1 , a2 , . . . , an are nonnegative real numbers so that a12 + a22 + · · · + an2 = n,


then v
1
t  ‹
a1 + a2 + · · · + an ≥ n − n + 1 + 1 −
3 3 3 2 (a16 + a26 + · · · + an6 ).
n
(Vasile C., 2008)

Solution. For n = 2, the inequality is equivalent to

a16 + a26 + 4a13 a23 ≥ 6,

(a12 + a22 )3 − 3a12 a22 (a12 + a22 ) + 4a13 a23 ≥ 6,


2a13 a23 − 3a12 a22 + 1 ≥ 0,
426 Vasile Cîrtoaje

(a1 a2 − 1)2 (2a1 a2 + 1) ≥ 0.


For n ≥ 3, we apply Corollary 5 for k = 3/2 and m = 3 : if 0 ≤ x 1 ≤ x 2 ≤ · · · ≤ x n
and
3/2 3/2
x 1 + x 2 + · · · + x n = n, x 1 + x 2 + · · · + x n3/2 = const ant,
then
Sn = x 13 + x 23 + · · · + x n3
is maximal for x 1 = · · · = x n−1 ≤ x n . Thus, we only need to prove the homogeneous
inequality

n2 − n + 1 2 1
 ‹
(a13 + a23 + · · · + an3 )2 ≥ 3
(a1 + a2 + · · · + an ) + 1 −
2 2 3
(a16 + a26 + · · · + an6 )
n n

for a1 = · · · = an−1 = 0 and for a1 = · · · = an−1 = 1. For the nontrivial case


a1 = · · · = an−1 = 1, the inequality becomes

n3 (n − 1 + an3 )2 ≥ (n2 − n + 1)(n − 1 + an2 )3 + n2 (n − 1)(n − 1 + an6 ),

(an − 1)2 (an − n + 1)2 (an2 + 2nan + n − 1) ≥ 0.


The equality holds for a1 = a2 = · · · = an = 1, and also for

an 1
a1 = · · · = an−1 = =p
n−1 n−1
(or any cyclic permutation).

P 5.59. If a, b, c are positive real numbers so that abc = 1, then

1 1 1 50
 ‹
4 + + + ≥ 27.
a b c a+b+c

(Vasile C., 2012)

Solution. According to Corollary 5 (case k=0 and m = −1, if

a + b + c = const ant, abc = 1, 0 < a ≤ b ≤ c,

then
1 1 1
+ + S3 =
a b c
is minimal for 0 < a = b ≤ c. Thus, we only need to prove that

2 1 50
 ‹
4 + + ≥ 27
a c 2a + c
EV Method for Nonnegative Variables 427

for
a2 c = 1, a ≤ 1.
The inequality is equivalent to

8a6 − 54a4 − 26a3 − 27a + 8 ≥ 0,

(2a − 1)2 (2a4 + 2a3 − 12a2 + 5a + 8) ≥ 0.


It is true for a ∈ (0, 1] because

2a4 + 2a3 − 12a2 + 5a + 8 > −12a2 + 4a + 8 = 4(1 − a)(2 + 3a) ≥ 0.

The equality holds for


1
a=b= , c=4
2
(or any cyclic permutation).

Remark. In the same manner, we can prove the following generalization:


• If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1 1 (2n + n − 1)2
 ‹
2 n
+ + ··· + + ≥ 2n(2n + 1),
a1 a2 an a1 + a2 + · · · + a n

with equality for


1
a1 = · · · = an−1 = , an = 2n−1
2
(or any cyclic permutation).
For
a1 = · · · = an−1 = a ≤ 1, a n−1 an = 1,
the inequality is equivalent to f (a) ≥ 0, where

n−1 (2n + n − 1)2 a n−1


 ‹
f (a) = 2 n
+a n−1
+ − 2n(2n + 1).
a (n − 1)a + 1
n

We have

f 0 (a) 2n (a n − 1) (2n + n − 1)2 a n−2 (a n − 1)


= −
n−1 a2 [(n − 1)a n + 1]2
(a n − 1)(2n a n − 1)[(n − 1)2 a n − 2n ]
= .
a2 [(n − 1)a n + 1]2

Since
(n − 1)2 a n − 2n ≤ (n − 1)2 − 2n < 0,
428 Vasile Cîrtoaje

1 1
 ‹  ‹
it follows that f (a) < 0 for a ∈ 0,
0
, and f (a) > 0 for a ∈
0
, 1 . Therefore,
2 2
1 1
 ˜ • ˜
f is decreasing on 0, and increasing on , 1 , hence
2 2
1
 ‹
f (a) ≥ f = 0.
2

P 5.60. If a, b, c are positive real numbers so that abc = 1, then


1 1 1
 ‹
a3 + b3 + c 3 + 15 ≥ 6 + + .
a b c
(Michael Rozenberg, 2006)
Solution. Replacing a, b, c by their reverses 1/a, 1/b, 1/c, we need to show that
a bc = 1 involves
1 1 1
+ + + 15 ≥ 6(a + b + c).
a3 b3 c 3
According to Corollary 5 (case k=0 and m = −3, if
a + b + c = const ant, abc = 1, 0 < a ≤ b ≤ c,
then
1 1 1
S3 =
3
+ 3+ 3
a b c
is minimal for 0 < a = b ≤ c. Thus, we only need to prove that
2 1
3
+ 3 + 15 ≥ 6(2a + c)
a c
for
a2 c = 1, a ≤ 1.
The inequality is equivalent to
2 1
 ‹
+ a + 15 ≥ 6 2a + 2 ,
6
a3 a
a9 − 12a4 + 15a3 − 6a + 2 ≥ 0,
(1 − a)2 (2 − 2a − 6a2 + 5a3 + 4a4 + 3a5 + 2a6 + a7 ) ≥ 0.
It suffices to show that
2 − 2a − 6a2 + 5a3 + 3a4 ≥ 0.
Indeed, we have
3 2 3
 ‹  ‹
2(2 − 2a − 6a + 5a + 3a ) = (2 − 3a) 1 + 2a + a + a 1 − a ≥ 0.
2 3 4 2 3
4 4
The equality holds for a = b = c = 1.
EV Method for Nonnegative Variables 429

P 5.61. Let a1 , a2 , . . . , an be positive numbers so that a1 a2 · · · an = 1. If k ≥ n − 1,


then
1 1 1
 ‹
a1 + a2 + · · · + an + (2k − n)n ≥ (2k − n + 1)
k k k
+ + ··· + .
a1 a2 an

(Vasile C., 2008)

Solution. For n = 2 and k = 1, the inequality is an identity. For n = 2 and k > 1,


we need to show that f (a) ≥ 0 for a > 0, where

f (a) = a k + a−k + 4(k − 1) − (2k − 1)(a + a−1 ).

We have
f 0 (a) = k(a k−1 − a−k−1 ) − (2k − 1)(1 − a−2 ),
f 00 (a) = k[(k − 1)a k−2 + (k + 1)a−k−2 ] − 2(2k − 1)a−3 .
By the weighted AM-GM inequality, we get
(k−1)(k−2)+(k+1)(−k−2)
(k − 1)a k−2 + (k + 1)a−k−2 ≥ 2ka 2k = 2ka−3 ,

hence
f 00 (a) ≥ 2k2 a−3 − 2(2k − 1)a−3 = 2(k − 1)2 a−3 > 0,
f 0 is strictly increasing. Since f 0 (1) = 0, it follows that f 0 (a) < 0 for a < 1 and
f 0 (a) > 0 for a > 1, f is decreasing on (0, 1] and increasing on [1, ∞), hence
f (a) ≥ f (1) = 0.
Consider further that n ≥ 3. Replacing a1 , a2 , . . . , an by 1/a1 , 1/a2 , . . . , 1/an , we
need to show that a1 a2 · · · an = 1 involves

1 1 1
+ + ··· + + (2k − n)n ≥ (2k − n + 1)(a1 + a2 + · · · + an ).
a1k a2k ank

According to Corollary 5, if 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a1 a2 · · · an = 1,

then
1 1 1
Sn = + + ··· +
a1k a2k ank
is minimal for 0 < a1 = · · · = an−1 ≤ an . Thus, we only need to prove the original
inequality for a1 = · · · = an−1 ≥ 1; that is, to show that t ≥ 1 involves f (t) ≥ 0,
where
1 n−1
 ‹
f (t) = (n − 1)t + k(n−1) + (2k − n)n − (2k − n + 1)
k
+t n−1
.
t t
430 Vasile Cîrtoaje

We have
(n − 1)g(t)
f 0 (t) = , g(t) = k(t kn − 1) − (2k − n + 1)t kn−k−1 (t n − 1),
t kn−k+1

g 0 (t) = t kn−k−2 h(t), h(t) = k2 nt k+1 − (2k − n + 1)[(k + 1)(n − 1)t n − kn + k + 1],
h0 (t) = (k + 1)nt n−1 [k2 t k−n+1 − (2k − n + 1)(n − 1)].
If k = n − 1, then h(t) = n(n − 1)(n − 2) > 0. If k > n − 1, then

k2 t k−n+1 − (2k − n + 1)(n − 1) ≥ k2 − (2k − n + 1)(n − 1) = (k − n + 1)2 > 0,

h0 (t) > 0 for t ≥ 1, h is strictly increasing on [1, ∞), hence

h(t) ≥ h(1) = n[(k − 1)2 + n − 2] > 0.

From h > 0, we get g 0 > 0, g is strictly increasing, g(t) ≥ g(1) = 0 for t ≥ 1,


f 0 (t) > 0 for t > 1, f is strictly increasing, f (t) ≥ f (1) = 0 for t ≥ 1.
The equality holds for a1 = a2 = · · · = an = 1. If n = 2 and k = 1, then the
equality holds for a1 a2 = 1.

P 5.62. Let a1 , a2 , . . . , an (n ≥ 3) be nonnegative numbers so that a1 +a2 +· · ·+an = n,


and let k be an integer satisfying 2 ≤ k ≤ n + 2. If
 n k−1
r= − 1,
n−1
then
a1k + a2k + · · · + ank − n ≥ nr(1 − a1 a2 · · · an ).

(Vasile C., 2005)

Solution. According to Corollary 4, if 0 ≤ a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a1k + a2k + · · · + ank = const ant,

then the product


P = a1 a2 · · · an
is minimal for either a1 = 0 or 0 < a1 ≤ a2 = · · · = an .
Case 1: a1 = 0. We need to show that

nk
a2k + · · · + ank ≥
(n − 1)k−1
EV Method for Nonnegative Variables 431

for a2 + · · · + an = n. This follows by Jensen’s inequality


 a + · · · + a k
2 n
a2k + · · · + ank ≥ (n − 1) .
n−1
Case 2: 0 < a1 ≤ a2 = · · · = an . Denoting a1 = x and a2 = y (x ≤ y), we only need
to show that
f (x) ≥ 0,
where
n− x
f (x) = x k + (n − 1) y k + nr x y n−1 − n(r + 1), y= , 0 < x ≤ 1 ≤ y.
n−1
It is easy to check that
f (0) = f (1) = 0.
Since
−1
y0 = ,
n−1
we have
f 0 (x) = k(x k−1 − y k−1 ) + nr y n−2 ( y − x)
= ( y − x)[nr y n−2 − k( y k−2 + y k−3 x + · · · + x k−2 )]
= ( y − x) y n−2 [nr − kg(x)],
where
1 x x k−2
g(x) = + + ··· + .
y n−k y n−k+1 y n−2
We see that f 0 (x) has the same sign as
h(x) = nr − kg(x).
Since the function
n− x
y(x) =
n−1
is strictly decreasing, g is strictly increasing for 2 ≤ k ≤ n. Also, g is strictly
increasing for k = n + 1, when
x2 x n−1
g(x) = y + x + + · · · + n−2
y y
(n − 2)x + n x 2
x n−1
= + + · · · + n−2 ,
n−1 y y
and for k = n + 2, when
x3 xn
g(x) = y 2 + y x + x 2 + + · · · + n−2
y y
(n − 3n + 3)x + n(n − 3)x + n2 x 3
2 2
xn
= + + · · · + n−2 .
(n − 1)2 y y
432 Vasile Cîrtoaje

Therefore, the function h(x) is strictly decreasing for x ∈ [0, 1]. Since f (0) =
f (1) = 0, there exists x 1 ∈ (0, 1) so that f (x) is increasing on [0, x 1 ] and decreasing
on [x 1 , 1]. As a consequence, f (x) ≥ 0 for x ∈ [0, 1].
The equality holds for a1 = a2 = · · · = an = 1, and also for
n
a1 = 0, a2 = · · · = a n =
n−1
(or any cyclic permutation).
Remark. For the particular case k = n, the inequality has been posted in 2004 on
Art of Problem Solving website by Gabriel Dospinescu and Calin Popa.

1 1 1
P 5.63. If a, b, c are positive real numbers so that + + = 3, then
a b c
4(a2 + b2 + c 2 ) + 9 ≥ 21abc.

(Vasile C., 2006)


Solution. Replacing a, b, c by their reverses 1/a, 1/b, 1/c, we need to show that
a + b + c = 3 involves
1 1 1 21
 ‹
4 2 + 2 + 2 +9≥ .
a b c abc
According to Corollary 5 (case k=0 and m = −2), if

a + b + c = 3, abc = const ant, 0 < a ≤ b ≤ c,

then
1 1 1
S3 =2
+ 2+ 2
a b c
is minimal for 0 < a = b ≤ c. Thus, we only need to prove that
2 1 21
 ‹
4 2 + 2 +9≥ 2
a c a c
for 2a + b = 3. The inequality is equivalent to

(9a2 + 8)c 2 − 21c + 4a2 ≥ 0,

4a4 − 12a3 + 13a2 − 6a + 1 ≥ 0,


(a − 1)2 (2a − 1)2 ≥ 0.
The equality holds for a = b = c = 1, and also for
1
a = b = 2, c=
2
(or any cyclic permutation).
EV Method for Nonnegative Variables 433

1 1 1
P 5.64. If a1 , a2 , . . . , an are positive real numbers so that + + ··· + = n,
a1 a2 an
then
a1 + a2 + · · · + an − n ≤ en−1 (a1 a2 · · · an − 1),
where ‹n−1
1

en−1 = 1 + .
n−1

(Gabriel Dospinescu and Calin Popa, 2004)

Solution. For n = 2, the inequality is an identity. For n ≥ 3, replacing a1 , a2 , . . . , an


by 1/a1 , 1/a2 , . . . , 1/an , we need to show that a1 + a2 + · · · + an = n involves

1 1 1
 ‹
a1 a2 · · · an + + ··· + − n + en−1 ≤ en−1 .
a1 a2 an

According to Corollary 5 (case k = 0 and m = −1), if 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a1 a2 · · · an = const ant,

then
1 1 1
Sn = + + ··· +
a1 a2 an
is maximal for 0 < a1 ≤ a2 = · · · = an . Using the notation a1 = x and a2 = y, we
only need to show that f (x) ≤ 0 for

x + (n − 1) y = n, 0 < x ≤ 1,

where
1 n−1
 ‹
f (x) = x y n−1
+ − n + en−1 − en−1
x y
= y n−1 + (n − 1)x y n−2 − (n − en−1 )x y n−1 − en−1 .

Since
−1
y0 = ,
n−1
we get
f 0 (x)
= ( y − x)h(x),
y n−3
where
n− x
h(x) = n − 2 − (n − en−1 ) y = n − 2 − (n − en−1 )
n−1
is a linear increasing function. Since

2
 ‹
n
h(0) = en−1 − 3 + <0
n−1 n
434 Vasile Cîrtoaje

and
h(1) = en−1 − 2 > 0,
there exists x 1 ∈ (0, 1) so that h(x 1 ) = 0, h(x) < 0 for x ∈ [0, x 1 ), and h(x) > 0 for
x ∈ (x 1 , 1]. Consequently, f is strictly decreasing on [0, x 1 ] and strictly increasing
on [x 1 , 1]. From
f (0) = f (1) = 0,
it follows that f (x) ≤ 0 for x ∈ [0, 1].
The equality holds for a1 = a2 = · · · = an = 1. If n = 2, then the equality holds
for a1 + a2 = 2a1 a2 .

P 5.65. If a1 , a2 , . . . , an are positive real numbers, then


a1n + a2n + · · · + ann 1 1 1
 ‹
+ n(n − 1) ≥ (a1 + a2 + · · · + an ) + + ··· + .
a1 a2 · · · an a1 a2 an

(Vasile C., 2004)


Solution. For n = 2, the inequality is an identity. For n ≥ 3, according to Corollary
5 (case k = 0 and m ∈ {−1, n}), if 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a1 a2 · · · an = const ant,


1 1 1
then the sum + +···+ is maximal and the sum a1n + a2n + · · · + ann is minimal
a1 a2 an
for
0 < a1 ≤ a2 = · · · = an .
Consequently, we only need to prove the desired homogeneous inequality for a2 =
· · · = an = 1, when it becomes

a1n + (n − 2)a1 ≥ (n − 1)a12 .

Indeed, by the AM-GM inequality, we have


q
a1n + (n − 2)a1 ≥ (n − 1) a1n · a1n−2 = (n − 1)a12 .
n−1

For n ≥ 3, the equality holds when a1 = a2 = · · · = an .

P 5.66. If a1 , a2 , . . . , an are nonnegative real numbers, then

(n−1)(a1n + a2n +· · ·+ ann )+ na1 a2 · · · an ≥ (a1 + a2 +· · ·+ an )(a1n−1 + a2n−1 +· · ·+ ann−1 ).

(Janos Suranyi, MSC-Hungary)


EV Method for Nonnegative Variables 435

Solution. For n = 2, the inequality is an identity. For n ≥ 3, according to Corollary


5 (case k = n and m = n − 1), if 0 ≤ a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a1n + a2n + · · · + ann = const ant,

then the sum a1n−1 +a2n−1 +· · ·+ann−1 is maximal and the product a1 a2 · · · an is minimal
for either a1 = 0 or 0 < a1 ≤ a2 = · · · = an . Consequently, we only need to consider
these cases.
Case 1: a1 = 0. The inequality reduces to

(n − 1)(a2n + · · · + ann ) ≥ (a2 + · · · + an )(a2n−1 + · · · + ann−1 ),

which follows immediately from Chebyshev’s inequality.


Case 2: 0 < a1 ≤ a2 = · · · = an . Due to homogeneity, we may set a2 = · · · = an = 1,
when the inequality becomes

(n − 2)a1n + a1 ≥ (n − 1)a1n−1 .

Indeed, by the AM-GM inequality, we have


Ç
n−1 n(n−2)
(n − 2)a1n + a1 ≥ (n − 1) a1 · a1 = (n − 1)a1n−1 .

For n ≥ 3, the equality holds when a1 = a2 = · · · = an , and also when

a1 = 0, a2 = · · · = a n

(or any cyclic permutation).

P 5.67. If a1 , a2 , . . . , an are nonnegative real numbers, then

(n − 1)(a1n+1 + a2n+1 + · · · + ann+1 ) ≥ (a1 + a2 + · · · + an )(a1n + a2n + · · · + ann − a1 a2 · · · an ).

(Vasile C., 2006)

Solution. For n = 2, the inequality is an identity. For n ≥ 3, according to Corollary


5 (case k = n + 1 and m = n), if 0 ≤ a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a1n+1 + a2n+1 + · · · + ann+1 = const ant,

then the sum a1n + a2n + · · · + ann is maximal and the product a1 a2 · · · an is minimal for
either a1 = 0 or 0 < a1 ≤ a2 = · · · = an . Consequently, we only need to consider
these cases.
Case 1: a1 = 0. The inequality reduces to

(n − 1)(a2n+1 + · · · + ann+1 ) ≥ (a2 + · · · + an )(a2n + · · · + ann ),


436 Vasile Cîrtoaje

which follows immediately from Chebyshev’s inequality.


Case 2: 0 < a1 ≤ a2 = · · · = an . Due to homogeneity, we may set a2 = · · · = an = 1,
when the inequality becomes

(n − 2)a1n+1 + a12 ≥ (n − 1)a1n .

Indeed, by the AM-GM inequality, we have


Ç
n−1 (n+1)(n−2)
(n − 2)a1n+1 + a12 ≥ (n − 1) a1 · a12 = (n − 1)a1n .

For n ≥ 3, the equality holds when a1 = a2 = · · · = an , and also when

a1 = 0, a2 = · · · = a n

(or any cyclic permutation).

P 5.68. If a1 , a2 , . . . , an are positive real numbers, then


1 1 1 1
 ‹
(a1 + a2 + · · · + an − n) + + ··· + − n + a1 a2 · · · an + ≥ 2.
a1 a2 an a1 a2 · · · an

(Vasile C., 2006)


Solution. For n = 2, the inequality reduces to

(1 − a1 )2 (1 − a2 )2 ≥ 0.

Consider further that n ≥ 3. Since the inequality remains unchanged by replacing


each ai with 1/ai , we may consider a1 a2 · · · an ≥ 1. By the AM-GM inequality, we
get
a1 + a2 + · · · + an ≥ n n a1 a2 · · · an ≥ n.
p

According to Corollary 5 (case k = 0 and m = −1), if 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a1 a2 · · · an = const ant,

then the sum


1 1 1
Sn = + + ··· +
a1 a2 an
is minimal for 0 < a1 = a2 = · · · = an−1 ≤ an . Consequently, we only need to
consider
a1 = a2 = · · · = an−1 = x, an = y, x ≤ y.
The inequality becomes
n−1 1 1
 ‹
[(n − 1)x + y − n] + − n + x n−1 y + n−1 ≥ 2,
x y x y
EV Method for Nonnegative Variables 437

n−1 1 1 n(n − 1)(x − 1)2


 ‹ • ˜
x n−1 + − n y + n−1 + (n − 1)x − n ≥ .
x x y x
Since
n−1 x − 1  n−1 
x n−1 + −n= (x − 1) + (x n−2 − 1) + · · · + (x − 1)
x x
(x − 1)2  n−2 
= x + 2x n−3 + · · · + (n − 1) ,
x
and
1 (x − 1)2 1 2
• ˜
+ (n − 1)x − n = + + · · · + (n − 1) ,
x n−1 x x n−2 x n−3
it is enough to prove the inequality

1 2 1
  • ˜
x n−2
+ 2x n−3
+ · · · + (n − 1) y + n−2 + n−3 + · · · + (n − 1) ≥ n(n − 1),
x x y

which is equivalent to

1 1 1
 ‹  ‹  ‹
x n−2
y + n−2 − 2 + 2 x n−3
y + n−3 − 2 + · · · + (n − 1) y + − 2 ≥ 0,
x y x y y

(x n−2 y − 1)2 2(x n−3 y − 1)2 (n − 1)( y − 1)2


+ + · · · + ≥ 0.
x n−2 y x n−3 y y

The equality holds if n − 1 of the numbers ai are equal to 1.

P 5.69. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

1 1
−Ç < 1.
a1 + a2 + · · · + an − n
p
1 1 1
a1 + a2 + ··· + an −n

(Vasile C., 2006)

Solution. Let
1 1 1
A = a1 + a2 + · · · + an − n, B= + + ··· + − n.
a1 a2 an

By the AM-GM inequality, it follows that A > 0 and B > 0. According to the pre-
ceding P 5.68, the following inequality holds

1 1 1
 ‹
(a1 + · · · + an+1 − n − 1) + ··· + − n − 1 + a1 · · · an+1 + ≥ 2,
a1 an+1 a1 · · · an+1
438 Vasile Cîrtoaje

which is equivalent to
1 1
 ‹
(A − 1 + an+1 ) B − 1 + + an+1 + ≥ 2,
an+1 an+1
A
+ Ban+1 + AB − A − B ≥ 0.
an+1
Choosing v
tA
an+1 = ,
B
we get p
2 AB + AB − A − B ≥ 0,
p p 2
AB ≥ A− B ,
1 1
1≥ p −p .
A B

P 5.70. If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then


n2 (n − 2) 1 1 1
 ‹
a1 + a2 + · · · + an +
n−1 n−1 n−1
≥ (n − 1) + + ··· + .
a1 + a2 + · · · + an a1 a2 an
Solution. For n = 2, the inequality is an identity. Consider further that n ≥ 3.
According to Corollary 5 (case k = 0), if 0 < a1 ≤ a2 ≤ · · · ≤ an and
a1 + a2 + · · · + an = const ant, a1 a2 · · · an = 1,
1 1 1
then the sum a1n−1 + a2n−1 + · · · + ann−1 is minimal and the sum + + ··· + is
a1 a2 an
maximal for 0 < a1 ≤ a2 = · · · = an . Thus, we only need to prove the homogeneous
inequality

n−1 n (n − 2)a1 a2 · · · an
2
1 1 1
 ‹
a1 +a2 +· · ·+an +
n−1 n−1
≥ (n−1)a1 a2 · · · an + + ··· +
a1 + a2 + · · · + an a1 a2 an
for a2 = · · · = an = 1; that is, to show that f (x) ≥ 0 for x ∈ [0, 1], where
n2 (n − 2)
f (x) = x n−2 + − (n − 1)2 ,
x +n−1
f 0 (x) n2
=x n−3
− .
n−2 (x + n − 1)2
Since f 0 is increasing, we have f 0 (x) ≤ f 0 (1) = 0 for x ∈ [0, 1], f is decreasing on
[0, 1], hence f (x) ≥ f (1) = 0.
The equality holds for a1 = a2 = · · · = an = 1. If n = 2, then the equality holds
for a1 a2 = 1.
EV Method for Nonnegative Variables 439

P 5.71. If a, b, c are nonnegative real numbers, then

abc − 1 2
(a + b + c − 3)2 ≥ (a + b2 + c 2 − 3).
abc + 1

(Vasile C., 2006)

Solution. For a = 0, the inequality reduces to

b2 + c 2 + bc + 3 ≥ 3(b + c),

which is equivalent to
(b − c)2 + 3(b + c − 2)2 ≥ 0.

For a bc > 0, according to Corollary 5 (case k = 0 and m = 2), if

a + b + c = const ant, abc = const ant,

then
S3 = a2 + b2 + c 2

is minimal and maximal when two of a, b, c are equal. Thus, we only need to prove
the desired inequality for a = b; that is,

a2 c − 1
(2a + c − 3)2 ≥ (2a2 + c 2 − 3),
a2 c + 1

which is equivalent to

(a − 1)2 [ca2 + 2c(c − 2)a + c 2 − 3c + 3] ≥ 0.

For c ≥ 2, the inequality is clearly true. It is also true for c ≤ 2, because

ca2 + 2c(c − 2)a + c 2 − 3c + 3 = c(a + c − 2)2 + (1 − c)2 (3 − c) ≥ 0.

The equality holds if two of a, b, c are equal to 1.

P 5.72. If a1 , a2 , . . . , an are positive real numbers so that a1 + a2 + · · · + an = n, then

p1
(a1 a2 · · · an ) n−1 (a12 + a22 + · · · + an2 ) ≤ n.

(Vasile C., 2006)


440 Vasile Cîrtoaje

Solution. For n = 2, the inequality is equivalent to


(a1 a2 − 1)2 ≥ 0.
For n ≥ 3, according to Corollary 5 (case k = 0, m = 2), if 0 < a1 ≤ a2 ≤ · · · ≤ an
and
a1 + a2 + · · · + an = n, a1 a2 · · · an = const ant,
then the sum
Sn = a12 + a22 + · · · + an2
is maximal for a1 = a2 = · · · = an−1 . Therefore, we only need to prove the homoge-
neous inequality
a12 + a22 + · · · + an2
n
 a + a + · · · + a 2+ pn−1
p1 1 2 n
(a1 a2 · · · an ) n−1 · ≤
n n
for a1 = a2 = · · · = an−1 = 1. The inequality is equivalent to f (x) ≥ 0 for x ≥ 1,
where
x +n−1 ln x x2 + n − 1
 ‹
n
f (x) = 2 + p ln −p − ln .
n−1 n n−1 n
Let
1
p= p .
n−1
Since
2 + np p 2x
f 0 (x) = − − 2
x +n−1 x  x +n−1
(n − 1)(x − 1) p 2
‹
= −
x +n−1 x x2 + n − 1
p
p(n − 1)(x − 1)(x − n − 1)2
= ≥ 0,
x(x + n − 1)(x 2 + n − 1)
f (x) is increasing for x ≥ 1, hence
f (x) ≥ f (1) = 0.
The equality holds for a1 = a2 = · · · = an = 1.
Remark. For n = 5, from the homogeneous inequality above, we get the following
nice results:
• If a, b, c, d, e are positive real numbers so that
a2 + b2 + c 2 + d 2 + e2 = 5,
then
(a) abcde(a4 + b4 + c 4 + d 4 + e4 ) ≤ 5;
p9
(b) a + b + c + d + e ≥ 5 abcde.
EV Method for Nonnegative Variables 441

P 5.73. If a1 , a2 , . . . , an are positive real numbers such that a1 + a2 + · · · + an = n − 1,


then v
t a1 + a22 + · · · + an2
v u 2
tn n−1
≥ 4 .
a1 a2 · · · an n(n − 1)
(Vasile Cîrtoaje and KaiRain, 2020)
Solution. For n = 2, we need to show that a1 + a2 = 1 involves
1
≥ 8(a12 + a2 )2 ,
a1 a2
which is equivalent to
(4a1 a2 − 1)2 ≥ 0.
For n ≥ 3, write the inequality in the homogeneous form
v
t a1 + a22 + · · · + an2
 a + a + · · · + a 2 v u 2
1 2 n
t
n n−1
≥4 .
n−1 a1 a2 · · · an n(n − 1)

According to Corollary 4, for a1 + a2 + · · · + an = const ant and a12 + a22 + · · · + an2 =


const ant, the product a1 a2 · · · an is maximal for a1 = a2 = · · · = an−1 ≤ an . Due to
homogeneity, we may set a1 = a2 = · · · = an−1 = 1, when the inequality becomes
A(x + n − 1)2 p 2
pn
≥ x + n − 1,
x
where p
n
A= , x ≥ 1.
4(n − 1)(3n−2)/(2n)
The inequality is true if f (x) ≥ 0, where
1 1
f (x) = ln A + 2 ln(x + n − 1) − ln x − ln(x 2 + n − 1).
n 2
From
2 1 x
f 0 (x) = − − 2
x + n − 1 nx x + n − 1
 3 
(n − 1) x − (n + 1)x 2 + (2n − 1)x − n + 1
=
nx(x + n − 1)(x 2 + n − 1)
(n − 1)(x − 1)2 (x − n + 1)
= ,
nx(x + n − 1)(x 2 + n − 1)
it follows that f is decreasing on [1, n − 1] and increasing on [n − 1, ∞), therefore
f (x) ≥ f (n − 1) = 0.
1 n−1
The equality occurs for a1 = a2 = · · · = an−1 = and an = (or any cyclic
2 2
permutation).
442 Vasile Cîrtoaje

P 5.74. If a1 , a2 , . . . , an are positive real numbers so that a13 + a23 + · · · + an3 = n, then

a1 + a2 + · · · + an ≥ n n+1 a1 a2 · · · an .
p

(Vasile C., 2007)

Solution. For n = 2, we need to show that a13 + a23 = 2 involves (a1 + a2 )3 ≥ 8a1 a2 .
Let
x = a1 + a2 .
From
2 = a13 + a23 = x 3 − 3a1 a2 x,
we get
x3 − 2
a1 a2 = .
3x
Thus,

8(x 3 − 2) (x − 2)2 (3x 2 + 4x + 4)


(a1 + a2 )3 − 8a1 a2 = x 3 − = ≥ 0.
3x 3x
For n ≥ 3, according to Corollary 4, if 0 < a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = const ant, a13 + a23 + · · · + an3 = n,

then the product


P = a1 a2 · · · an
is maximal for a1 = a2 = · · · = an−1 . Therefore, we only need to prove the homoge-
neous inequality
v
3 a + a + ··· + a
 a + a + · · · + a n+1 t 3 3 3
1 2 n 1 2 n
≥ a1 a2 · · · a n
n n
for a1 = a2 = · · · = an−1 = 1. The inequality is equivalent to f (x) ≥ 0 for x ≥ 1,
where
x +n−1 1 x3 + n − 1
f (x) = (n + 1) ln − ln x − ln .
n 3 n
Since
n+1 1 x2
f 0 (x) = − − 3
x +n−1 x x +n−1
(n − 1)(x − 1)(x 3 − x 2 − x + n − 1)
=
x(x + n − 1)(x 3 + n − 1)
(n − 1)(x − 1)(x 3 − x 2 − x + 1)

x(x + n − 1)(x 3 + n − 1)
(n − 1)(x − 1)3 (x + 1)
= ,
x(x + n − 1)(x 3 + n − 1)
EV Method for Nonnegative Variables 443

f (x) is increasing for x ≥ 1, hence

f (x) ≥ f (1) = 0.

The equality holds for a1 = a2 = · · · = an = 1.

P 5.75. Let a, b, c be nonnegative real numbers so that ab + bc + ca = 3. If


ln 4
k ≥2− ≈ 0.738,
ln 3
then
a k + b k + c k ≥ 3.
(Vasile C., 2004)
Solution. Let
ln 4
r =2− .
ln 3
By the power mean inequality, we have

ak + bk + c k a + b r + c r k/r
 r ‹
≥ .
3 3
Thus, it suffices to show that

a r + b r + c r ≥ 3.

Since
2(ab + bc + ca) = (a + b + c)2 − (a2 + b2 + c 2 ),
according to Corollary 5 (case k = 2, m = r), if a ≤ b ≤ c and

a + b + c = const ant, a2 + b2 + c 2 = const ant,

then
S3 = a r + b r + c r
is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. We need to show that bc = 3 implies b r + c r ≥ 3. Indeed, by the
AM-GM inequality, we have
Æ
b r + c r ≥ 2 (bc) r = 2 · 3 r/2 = 3.

Case 2: 0 < a ≤ b = c. We only need to show that the homogeneous inequality

ab + bc + ca r/2
 ‹
a +b +c ≥3
r r r
3
444 Vasile Cîrtoaje

holds for b = c = 1; that is, to show that a ∈ (0, 1] involves

2a + 1
 ‹ r/2
a +2≥3
r
,
3
which is equivalent to f (a) ≥ 0, where

a r + 2 r 2a + 1
f (a) = ln − ln .
3 2 3
The derivative
r a r−1 r r g(a)
f 0 (a) = − = 1−r r ,
a + 2 2a + 1 a (a + 2)(2a + 1)
r

where
g(a) = a − 2a1−r + 1.
From
2(1 − r)
g 0 (a) = 1 − ,
ar
it follows that g 0 (a) < 0 for a ∈ (0, a1 ), and g 0 (a) > 0 for a ∈ (a1 , 1], where

a1 = (2 − 2r)1/r ≈ 0.416.

Then, g is strictly decreasing on [0, a1 ] and strictly increasing on [a1 , 1]. Since
g(0) = 1 and g(1) = 0, there exists a2 ∈ (0, 1) so that g(a2 ) = 0, g(a) > 0 for
a ∈ [0, a2 ), and g(a) < 0 for a ∈ (a2 , 1]. Consequently, f is increasing on [0, a2 ]
and decreasing on [a2 , 1]. Since f (0) = f (1) = 0, we have f (a) ≥ 0 for 0 < a ≤ 1.
ln 4
The equality holds for a = b = c = 1. If k = 2 − , then the equality holds also
ln 3
for p
a = 0, b=c= 3
(or any cyclic permutation).
Remark. For k = 3/4, we get the following nice results (see P 3.33 in Volume 1):
• Let a, b, c be positive real numbers.
(a) If a4 b4 + b4 c 4 + c 4 a4 = 3, then

a3 + b3 + c 3 ≥ 3.

(b) If a3 + b3 + c 3 = 3, then

a4 b4 + b4 c 4 + c 4 a4 ≤ 3.
EV Method for Nonnegative Variables 445

P 5.76. Let a, b, c be nonnegative real numbers so that a + b + c = 3. If

ln 9 − ln 8
k≥ ≈ 0.29,
ln 3 − ln 2
then
a k + b k + c k ≥ ab + bc + ca.

(Vasile C., 2005)

Solution. For k ≥ 1, by Jensen’s inequality, we get

a+b+c
‹k
1

a +b +c ≥3
k k k
= 3 = (a + b + c)2 ≥ ab + bc + ca.
3 3
Let
ln 9 − ln 8
r= .
ln 3 − ln 2
Assume further that
r ≤ k < 1,
and write the inequality as

2(a k + b k + c k ) + a2 + b2 + c 2 ≥ 9.

By Corollary 5, if a ≤ b ≤ c and

a + b + c = 3, a2 + b2 + c 2 = const ant,

then the sum


S3 = a k + b k + c k
is minimal for either a = 0 or 0 < a ≤ b = c. Thus, we only need to prove the
desired inequality for these cases.
Case 1: a = 0. We need to show that b + c = 3 involves b k + c k ≥ bc. Indeed, by
the AM-GM inequality, we have
 
b k + c k − bc ≥ 2(bc)k/2 − bc = (bc)k/2 2 − (bc)1−k/2
b + c 2−k
 ‹    ‹2−k 
3

≥ (bc) k/2
2− = (bc) k/2
2−
2 2
  ‹2−r 
3
≥ (bc)k/2 2 − = 0.
2

Case 2: 0 < a ≤ b = c. We only need to show that the homogeneous inequality


‹2−k
a+b+c

(a + b + c )
k k k
≥ ab + bc + ca
3
446 Vasile Cîrtoaje

holds for b = c = 1; that is, to show that a ∈ (0, 1] involves


‹2−k
a+2

(a + 2)
k
≥ 2a + 1,
3
which is equivalent to f (a) ≥ 0, where
a+2
f (a) = ln(a k + 2) + (2 − k) ln − ln(2a + 1).
3
We have
ka k−1 2 − k 2 2g(a)
f (a) = k
0
+ − = 1−k k ,
a + 2 a + 2 2a + 1 a (a + 2)(2a + 1)
where
g(a) = a2 + (2k − 1)a + k + 2(1 − k)a2−k − (k + 2)a1−k ,
with
g 0 (a) = 2a + 2k − 1 + 2(1 − k)(2 − k)a1−k − (k + 2)(1 − k)a−k ,
g 00 (a) = 2 + 2(1 − k)2 (2 − k)a−k + k(k + 2)(1 − k)a−k−1 .
Since g 00 > 0, g 0 is strictly increasing. From g 0 (0+ ) = −∞ and g 0 (1) = 3(1 −
k) + 3k2 > 0, it follows that there exists a1 ∈ (0, 1) so that g 0 (a1 ) = 0, g 0 (a) < 0
for a ∈ (0, a1 ) and g 0 (a) > 0 for a ∈ (a1 , 1]. Therefore, g is strictly decreasing
on [0, a1 ] and strictly increasing on [a1 , 1]. Since g(0) = k > 0 and g(1) = 0,
there exists a2 ∈ (0, a1 ) so that g(a2 ) = 0, g(a) > 0 for a ∈ [0, a2 ) and g(a) < 0
for a ∈ (a2 , 1]. Consequently, f is increasing on [0, a2 ] and decreasing on [a2 , 1].
Since
2 2
f (0) = ln 2 + (3 − k) ln ≥ ln 2 + (3 − r) ln = 0
3 3
and f (1) = 0, we get f (a) ≥ 0 for 0 ≤ a ≤ 1.
ln 9 − ln 8
The equality holds for a = b = c = 1. If k = , then the equality holds
ln 3 − ln 2
also for
3
a = 0, b=c=
2
(or any cyclic permutation).

P 5.77. If a1 , a2 , . . . , an (n ≥ 4) are nonnegative numbers so that a1 +a2 +· · ·+an = n,


then
1 1 1
+ + ··· + ≤ 1.
n + 1 − a2 a3 · · · an n + 1 − a3 a4 · · · a1 n + 1 − a1 a2 · · · an−1

(Vasile C., 2004)


EV Method for Nonnegative Variables 447

Solution. Let a1 ≤ a2 ≤ · · · ≤ an and


‹n−1
1

en−1 = 1 + .
n−1
By the AM-GM inequality, we have
 a + a + · · · + a n−1  a + a + · · · + a n−1
2 3 n 1 2 n
a2 a3 · · · an ≤ ≤ = en−1 ,
n−1 n−1
hence
n + 1 − a2 a3 · · · an ≥ n + 1 − en−1 = (n − 2) + (3 − en−1 ) > 0.
Consider the cases a1 = 0 and a1 > 0.

Case 1: a1 = 0. We need to show that a2 + a3 + · · · + an = n involves

1 n−1
+ ≤ 1,
n + 1 − a2 a3 · · · an n + 1

which is equivalent to
n+1
a2 a3 · · · an ≤ .
2
Since  a + a + · · · + a n−1
2 3 n
a2 a3 · · · an ≤ = en−1 ,
n−1
it suffices to show that
n+1
en−1 ≤ .
2
For n = 4, we have
n+1 7
− en−1 = > 0.
2 54
For n ≥ 5, we get
n+1
≥ 3 > en−1 .
2
Case 2: 0 < a1 ≤ a2 ≤ · · · ≤ an . Denote

a1 a2 · · · an = (n + 1)r, r > 0.

From a2 a3 · · · an ≤ en−1 , we get

(n + 1)r
a1 ≥ a, a= > r.
en−1

Write the inequality as follows


a1 a2 an
+ + ··· + ≤ n + 1,
a1 − r a2 − r an − r
448 Vasile Cîrtoaje

1 1 1 1
+ + ··· + ≤ ,
a1 − r a2 − r an − r r
1
f (a1 ) + f (a2 ) + · · · + f (an ) + ≥ 0,
r
where
−1
f (u) =
, u ≥ a.
u−r
We will apply Corollary 3 to the function f . We have
1
f 0 (u) = ,
(u − r)2

1 x2 4r x + 2
 ‹
g(x) = f 0
= , g 00 (x) = > 0.
x (1 − r x)2 (1 − r x)4
According to Corollary 3, if a ≤ a1 ≤ a2 ≤ · · · ≤ an and

a1 + a2 + · · · + an = n, a1 a2 · · · an = (n + 1)r = const ant,

then the sum S3 = f (a1 ) + f (a2 ) + · · · + f (an ) is minimal for a ≤ a1 ≤ a2 = · · · = an .


Thus, we only need to prove the homogeneous inequality
1 1 1
a2 a3 · · · an + a3 a4 · · · a1 + · · · + a1 a2 · · · an−1 ≤ 1
n+1− n+1− n+1−
s n−1 s n−1 s n−1
for 0 < a1 ≤ a2 = a3 = · · · = an = 1, where
a1 + a2 + · · · + an
s= ;
n
that is,
s n−1 (n − 1)s n−1 a1 + n − 1
+ ≤ 1, s= ,
(n + 1)s n−1 − 1 (n + 1)s n−1 − a1 n
which is equivalent to
f (s) ≥ 0, s1 < s ≤ 1,
where
n−1
s1 =
n
and
f (s) = (n + 1)s2n−2 − n2 s n + (n + 1)(n − 2)s n−1 + ns − n + 1.
We have
f 0 (s) = 2(n2 − 1)s2n−3 − n3 s n−1 + (n2 − 1)(n − 2)s n−2 + n,
f 00 (s) = (n − 1)s n−3 g(s),
where
g(s) = 2(2n − 3)(n + 1)s n−1 − n3 s + (n − 2)2 (n + 1),
EV Method for Nonnegative Variables 449

g 0 (s) = 2(2n − 3)(n2 − 1)s n−2 − n3 .


Since
2n(2n − 3)(n + 1)
g 0 (s) ≥ g 0 (s1 ) = − n3
en−1
2n(2n − 3)(n + 1) n(n2 − 2n − 6)
> − n3 = > 0,
3 3
g is increasing. There are two cases to consider: g(s1 ) ≥ 0 and g(s1 ) < 0.
Subcase A: g(s1 ) ≥ 0. Then, g(s) ≥ 0, f 00 (s) ≥ 0, f 0 is increasing. Since f 0 (1) = 0,
it follows that f 0 (s) ≤ 0 for s ∈ [s1 , 1], f is decreasing, hence f (s) ≥ f (1) = 0.
Subcase B: g(s1 ) < 0. Then, since g(1) = n2 −2n+4 > 0, there exists s2 ∈ (s1 , 1) so
that g(s2 ) = 0, g(s) < 0 for s ∈ [s1 , s2 ) and g(s) > 0 for s ∈ (s2 , 1], f 0 is decreasing
on [s1 , s2 ] and increasing on [s2 , 1]. We see that f 0 (1) = 0. If f 0 (s1 ) ≤ 0, then
f 0 (s) ≤ 0 for s ∈ [s1 , 1], f is decreasing, hence f (s) ≥ f (1) = 0. If f 0 (s1 ) > 0, then
there exists s3 ∈ (s1 , s2 ) so that f 0 (s3 ) = 0, f 0 (s) > 0 for s ∈ [s1 , s3 ) and g(s) < 0
for s ∈ (s3 , 1], hence f is increasing on [s1 , s3 ] and decreasing on [s3 , 1]. Since
f (1) = 0, it suffices to show that f (s1 ) ≥ 0. This is true since s = s1 involves
a1 = 0, and we have shown that the desired inequality holds for a1 = 0.
The equality occurs for a1 = a2 = · · · = an = 1.

P 5.78. If a, b, c are nonnegative real numbers so that

a + b + c ≥ 2, ab + bc + ca ≥ 1,

then p p p
3
a+ b+
3 3
c ≥ 2.

(Vasile C., 2005)

Solution. According to Corollary 5 (case k = 2 and m = 1/3), if 0 ≤ a ≤ b ≤ c and

a + b + c = const ant, ab + bc + ca = const ant,


p p3 p
then the sum S3 = 3 a + b + 3 c is minimal for either a = 0 or 0 < a ≤ b = c.
Case 1: a = 0. The hypothesis ab + bc + ca ≥ 1 implies bc ≥ 1; consequently,
p p
3 p p
3 p p6
a + b + c = b + c ≥ 2 bc ≥ 2.
3 3 3

Case 2: 0 < a ≤ b = c. If c ≥ 1, then


p p
3 p p
a + b + c ≥ 2 c ≥ 2.
3 3 3
450 Vasile Cîrtoaje

If c < 1, then
p p
3 p
a+ b+ c ≥ a + b + c ≥ 2.
3 3

The equality holds for


a = 0, b=c=1
(or any cyclic permutation).

P 5.79. If a, b, c, d are positive real numbers so that abcd = 1, then


p
(a + b + c + d)4 ≥ 36 3 (a2 + b2 + c 2 + d 2 ).

(Vasile C., 2008)

Solution. According to Corollary 5 (case k = 0 and m = 2), if a ≤ b ≤ c ≤ d and

a + b + c + d = const ant, abcd = 1,

then the sum


S4 = a2 + b2 + c 2 + d 2
is maximal for a = b = c ≤ d. Thus, we only need to show that
p
(3a + d)4 ≥ 36 3 (3a2 + d 2 )

for a3 d = 1. Write this inequality as f (a) ≥ 0, where

1 1 p
 ‹  ‹
f (a) = 4 ln 3a + 3 − ln 3a + 6 − ln 36 3,
2
0 < a ≤ 1.
a a
Since
12(a4 − 1) 6(a8 − 1) 6(a4 − 1)2 (3a4 − 1)
f (a) =
0
− = ,
a(3a4 + 1) a(3a8 + 1) a(3a4 + 1)(3a8 + 1)
p
4 p 4
f is decreasing on [0, 1/ 3] and increasing on [1/ 3, 1]; therefore,

1
 ‹
f (a) ≥ f p 4
= 0.
3
The equality holds for

1 p
4
a=b=c= p
4
, d= 27
3
(or any cyclic permutation).

Remark. In the same manner, we can prove the following generalization:


EV Method for Nonnegative Variables 451

• If a1 , a2 , . . . , an are positive real numbers so that a1 a2 · · · an = 1, then

16 Æ
(a1 + a2 + · · · + an )4 ≥ (n − 1)3n−2 (a12 + a22 + · · · + an2 ),
n

n
with equality for
1 Æ
a1 = a2 = · · · = an−1 = p an = (n − 1)n−1
n
n
,
n−1
(or any cyclic permutation).

P 5.80. If a, b, c are nonnegative real numbers so that ab + bc + ca = 1, then


p p p
33a2 + 16 + 33b2 + 16 + 33c 2 + 16 ≤ 9(a + b + c).

(Vasile C., 2006)

Solution. Write the inequality as

f (a) + f (b) + f (c) + 297(a + b + c) ≥ 0,

where
1p
f (u) = − 33u2 + 16, u ≥ 0.
33
We have
−x
g(x) = f 0 (x) = p ,
33x 2 + 16
33 · 48x
g 00 (x) = .
(33x 2 + 16)5/2
Since g 00 (x) > 0 for x > 0, g is strictly convex on [0, ∞). According to Corollary
1, if 0 ≤ a ≤ b ≤ c and

a + b + c = const ant, a2 + b2 + c 2 = const ant,

then the sum


Sn = f (a) + f (b) + f (c)
is minimal for either a = 0 or 0 < a ≤ b = c.

Case 1: a = 0. We need to show that bc = 1 involves


p p
33b2 + 16 + 33c 2 + 16 ≤ 9(b + c) − 4.

We see that p
9(b + c) − 4 ≥ 18 bc − 4 = 14 > 0.
452 Vasile Cîrtoaje

By squaring, the inequality becomes


p
528t 2 + 289 ≤ 24t 2 − 36t + 25,

where
t = b + c ≥ 2.
Since
24t 2 − 36t + 25 ≥ 6t 2 + 25,
it suffices to show that
528t 2 + 289 ≤ (6t 2 + 25)2 ,
which is equivalent to
(t 2 − 4)(3t 2 − 7) ≥ 0.

Case 2: 0 < a ≤ b = c. Write the inequality in the homogeneous form



33a2 + 16(ab + bc + ca) ≤ 9(a + b + c).

Without loss of generality, assume that b = c = 1, when the inequality becomes


p p
33a2 + 32a + 16 + 2 32a + 49 ≤ 9a + 18.

By squaring twice, the inequality turns as follows:


Æ
(33a2 + 32a + 16)(32a + 49) ≤ 12a2 + 41a + 28,

72a(2a3 − a2 − 4a + 3) ≥ 0,
72a(a − 1)2 (2a + 3) ≥ 0.
1
The equality holds for a = b = c = p , and also for
3
a = 0, b=c=1

(or any cyclic permutation).

P 5.81. If a, b, c are positive real numbers so that a + b + c = 3, then

3
a2 b2 + b2 c 2 + c 2 a2 ≤ p
3
.
abc

(Vasile C., 2006)


EV Method for Nonnegative Variables 453

Solution. Write the inequality in the homogeneous form


‹15 ‹3
a+b+c a2 b2 + b2 c 2 + c 2 a2
 
≥ abc .
3 3

Since

a2 b2 + b2 c 2 + c 2 a2 = (ab + bc + ca)2 − 2abc(a + b + c)


1
= (9 − a2 − b2 − c 2 ) − 6abc,
4

we will apply Corollary 5 (case k = 0 and m = 2):


• If 0 ≤ a ≤ b ≤ c and

a + b + c = 3, abc = const ant,

them the sum


S3 = a2 + b2 + c 2

is minimal for 0 < a ≤ b = c.

Therefore, we only need to prove the homogeneous inequality for 0 < a ≤ 1 and
b = c = 1. Taking logarithms, we have to show that f (a) ≥ 0, where

a+2 2a2 + 1
f (a) = 15 ln − ln a − 3 ln .
3 3

Since the derivative

15 1 12a 2(a − 1)(2a − 1)(4a − 1)


f 0 (a) = − − 2 =
a + 2 a 2a + 1 a(a + 2)(2a2 + 1)

1 1 1 1
 ‹  ‹  ‹
is negative for a ∈ 0, ∪ , 1 and positive for a ∈ , , f is decreasing
4 2 4 2
1 1 1 1
 ˜ • ˜ • ˜
on 0, ∪ , 1 and increasing on , . Therefore, it suffices to show that
 ‹4 2 4 2
1
f ≥ 0 and f (1) ≥ 0. Indeed, we have f (1) = 0 and
4

1 312
 ‹
f = ln 19 > 0.
4 2

The equality holds for a = b = c = 1.


454 Vasile Cîrtoaje

P 5.82. If a1 , a2 , . . . , an (n ≤ 81) are nonnegative real numbers so that

a12 + a22 + · · · + an2 = a15 + a25 + · · · + an5 ,

then
a16 + a26 + · · · + an6 ≤ n.

(Vasile C., 2006)

Solution. Setting an = 1, we obtain the statement for n − 1 numbers ai . Conse-


quently, it suffices to prove the inequality for n = 81. We need to show that the
following homogeneous inequality holds:

81(a15 + a25 + · · · + a81


5 2
) ≥ (a16 + a26 + · · · + a81
6
)(a12 + a22 + · · · + a81
2 2
) .

According to Corollary 5 (case k = 3 and m = 5/2), if 0 ≤ a1 ≤ a2 ≤ · · · ≤ a81 and

a12 + a22 + · · · + a81


2
= const ant, a16 + a26 + · · · + a81
6
= const ant,

then the sum a15 + a25 + · · · + a81


5
is minimal for a1 = a2 = · · · = a80 ≤ a81 . Therefore,
we only need to prove the homogeneous inequality for a1 = a2 = · · · = a80 = 0 and
for a1 = a2 = · · · = a80 = 1. The first case is trivial. In the second case, denoting
a81 by x, the homogeneous inequality becomes as follows:

81(80 + x 5 )2 ≥ (80 + x 6 )(80 + x 2 )2 ,

x 10 − 2x 8 − 80x 6 + 162x 5 − x 4 − 160x 2 + 80 ≥ 0,

(x − 1)2 (x − 2)2 (x 6 + 6x 5 + 21x 4 + 60x 3 + 75x 2 + 60x + 20) ≥ 0.


Thus, the proof is completed. The equality holds for a1 = a2 = · · · = an = 1. If
n = 81, then the equality holds also for
v
a81 t3 3
a1 = a2 = · · · = a80 = =
2 4

(or any cyclic permutation).

P 5.83. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


p Æ
1+ 1 + a3 + b3 + c 3 ≥ 3(a2 + b2 + c 2 ).

(Vasile C., 2006)


EV Method for Nonnegative Variables 455

Solution. Write the inequality as


p Æ
1 + a3 + b3 + c 3 ≥ 3(a2 + b2 + c 2 ) − 1.

By squaring, we may rewrite the inequality in the homogeneous form


‹2 Æ
a+b+c

a + b +c +2
3 3 3
3(a2 + b2 + c 2 ) ≥ (a + b + c)(a2 + b2 + c 2 ).
3

According to Corollary 5 (case k = 2 and m = 3), if 0 ≤ a ≤ b ≤ c and

a + b + c = const ant, a2 + b2 + c 2 = const ant,

then the sum


S3 = a3 + b3 + c 3
is minimal for either a = 0 or 0 < a ≤ b = c. Thus, we only need to prove the
homogeneous inequality for a = 0 and for b = c = 1.

Case 1: a = 0. We need to show that

b+c 2Æ
 ‹
b +c +2
3 3
3(b2 + c 2 ) ≥ (b + c)(b2 + c 2 ).
3

Simplifying by b + c, it remains to show that


p
p 3 3
(b + c) b + c ≥
2 2 bc.
2
Indeed, p
p € p Šp 3 3
(b + c) b2 + c 2 ≥ 2 bc 2bc ≥ bc.
2
Case 2: b = c = 1. We need to prove that
Æ
(a + 2)2 3(a2 + 2) ≥ 9(a2 + a + 1).

By squaring, the inequality becomes

a6 + 8a5 − a4 − 6a3 − 17a2 + 10a + 5 ≥ 0,

(a − 1)2 (a4 + 10a3 + 18a2 + 20a + 5) ≥ 0.

The equality holds for a = b = c = 1.


456 Vasile Cîrtoaje

P 5.84. If a, b, c are nonnegative real numbers so that a + b + c = 3, then


v
p p p t 2
a + b + b + c + c + a ≤ 16 + (ab + bc + ca).
3

(Lorian Saceanu, 2017)

Solution. Write the inequality in the form


v
t 2
f (a) + f (b) + f (c) + 16 + (ab + bc + ca) ≥ 0,
3
where p
f (u) = − 3 − u, 0 ≤ u ≤ 3.
We have
1
g(x) = f 0 (x) = p ,
2 3− x
3
g 00 (x) = (3 − x)−5/2 .
8
Since g (x) > 0 for x ∈ [0, 3), g is strictly convex on [0, 3]. According to Corollary
00

1, if 0 ≤ a ≤ b ≤ c and

a + b + c = 3, ab + bc + ca = const ant,

then the sum S3 = f (a) + f (b) + f (c) is minimal for either a = 0 or 0 < a ≤ b = c.
Therefore, we only need to prove the homogeneous inequality
v
p p p t 16 2(ab + bc + ca)
a+b+ b+c+ c+a≤ (a + b + c) +
3 a+b+c
for a = 0 and b = c = 1.

Case 1: a = 0. We need to show that


v
p p p t 16 2bc
b+ c+ b+c ≤ (b + c) + .
3 b+c
Consider the nontrivial case b, c > 0, use the substitution
v
tb sc
x= + , x ≥ 2,
c b
and write the inequality as
v
q p p t 16 2bc
b + c + 2 bc + b + c ≤ (b + c) + ,
3 b+c
EV Method for Nonnegative Variables 457

v
p p t 16
2
x +2+ x≤ . x+
3 x
By squaring twice, the inequality becomes as follows:
Æ 5 1
x(x + 2) ≤ x −1+ ,
3 x
16x 4 − 48x 3 + 39x 2 − 18x + 9 ≥ 0,
(x − 2)[16x 2 (x − 1) + 7x − 4] + 1 ≥ 0.

Case 2: b = c = 1. We need to prove that


v
p p t 16 2(2a + 1)
2 a+1+ 2≤ (a + 2) +
3 a+2
By squaring twice, the inequality becomes as follows:
Æ
6(a + 2) 2(a + 1) ≤ 2a2 + 17a + 17,

4a4 − 4a3 − 3a2 + 2a + 1 ≥ 0,


(a − 1)2 (2a + 1)2 ≥ 0.

The equality holds for a = b = c = 1.

P 5.85. If a, b, c ∈ [0, 4] and ab + bc + ca = 4, then


p p p p
a + b + b + c + c + a ≤ 3 + 5.

(Vasile Cîrtoaje, 2019)

First Solution. Denote s = a + b + c, consider s fixed and write the inequality as


p
f (a) + f (b) + f (c) ≥ −3 − 5,

where p
f (x) = − s − x. 0 ≤ x < s.
From
1 3
g(x) = f 0 (x) = (s − x)−1/2 , g 00 (x) = (s − x)−5/2 > 0,
2 8
it follows that g is strictly convex. Thus, by Corollary 1 and Note 2, the sum f (a) +
f (b) + f (c) is minimal for either a ≤ b = c or a = 0.
Case 1: a ≤ b = c. We need to show that 2ac + c 2 = 4 yields
p p p
2 a + c + 2c ≤ 3 + 5,
458 Vasile Cîrtoaje

that is v
t 2(c 2 + 1) p p
+ 2c ≤ 3 + 5.
c
From 2ac + c = 4,it follows that
2

2
p ≤ c ≤ 2.
3
p
Since 2c ≤ 2, it is enough to show that
v
t 2(c 2 + 1) p
≤ 1 + 5,
c
that is p
c 2 − (3 + 5)c + 4 ≤ 0.
Indeed, p
c 2 − (3 + 5)c + 4 ≤ c 2 − 5c + 4 = (c − 1)(c − 4) < 0.

Case 2: a = 0. We need to show that bc = 4 yields


p p p p
b + c + b + c ≤ 3 + 5.

From (4 − b)(4 − c) ≥ 0, we get b + c ≤ 5. Thus,


p
p p q p p
b + c + b + c ≤ b + c + 2 bc + b + c
Æ p p p
≤ 5 + 2 4 + 5 = 3 + 5.
The equality occurs for a = 0, b = 1 and c = 4 (or any permutation).
Second Solution(by Kiyoras-2001) Assume that a ≥ b ≥ c, denote

S = ab + bc + ca

and show that


p
 ‹
S
f (a, b, c) ≤ f a, , 0 ≤ 3 + 5,
a
where p p p
f (a, b, c) = a+b+ b+c+ c + a.
The left homogeneous inequality is true because
 ‹
S
f a, , 0 − f (a, b, c) =
a
v v
t S p tS p p p
= a+ − a+b+ − b+c+ a− c+a
a a
EV Method for Nonnegative Variables 459

a (a
+ b)
c bc
a c
=q p +q p −p p
(a+b)(a+c) a+ c+a
a + b+c
S
a + a+b
 p
a(a + b)

c a
≥ p p −p p ≥ 0.
a a+c+ a a+ c+a
Also, the right inequality is true for S = 4 and a, b, c ∈ [0, 4] since a > 1 and

4 p
 ‹
f a, , 0 − 3 − 5 =
a
v
t 4 p 2 p
= a+− 5+ p + a−3
a a

(a − 1) 1 − 4a

p 2
 ‹
= q p + ( a − 1) 1 − pa ≤ 0.
a + 4a + 5

P 5.86. If a, b, c are positive real numbers so that abc = 1, then


v
a+b+c t 3 2+ a + b + c
2 2 2
(a) ≥ ;
3 5
p
(b) a3 + b3 + c 3 ≥ 3(a4 + b4 + c 4 ).
(Vasile C., 2006)

Solution. (a) According to Corollary 5 (case k = 0 and m = 2), if a ≤ b ≤ c and

a + b + c = const ant, abc = 1,

the sum S3 = a2 + b2 + c 2 is maximal for 0 < a = b ≤ c. Thus, we only need to show


that a2 c = 1 involves v
2a + c t 3 2 + 2a + c
2 2
≥ ,
3 5
which is equivalent to

1 3 1
 ‹  ‹
5 2a + 2 ≥ 27 2 + 2a + 4 ,
2
a a

40a9 − 54a8 + 6a6 + 30a3 − 27a2 + 5 ≥ 0,

(a − 1)2 (40a7 + 26a6 + 12a5 + 4a4 − 4a3 − 12a2 + 10a + 5) ≥ 0.


460 Vasile Cîrtoaje

The inequality is true since

12a5 + 4a4 − 4a3 − 12a2 + 10a + 5 > 2a5 + 4a4 − 4a3 − 12a2 + 10a
= 2a(a − 1)2 (a2 + 4a + 5) ≥ 0.

The equality holds for a = b = c = 1.

(b) According to Corollary 5 (case k = 0 and m = 4/3), if a ≤ b ≤ c and

a3 + b3 + c 3 = const ant, a3 b3 c 3 = 1,

the sum S3 = a4 + b4 + c 4 is maximal for 0 < a = b ≤ c. Thus, we only need to show


that Æ
2a3 + c 3 ≥ 3(2a4 + c 4 )
for a2 c = 1, a ≤ 1. The inequality is equivalent to

1 2 1
 ‹  ‹
2a + 6 ≥ 3 2a + 8 .
3 4
a a

Substituting a = 1/t, t ≥ 1, the inequality becomes


‹2
2 2
  ‹
3
+ t6 ≥3 4 +t ,
8
t t

which is equivalent to f (t) ≥ 0, where

f (t) = t 18 − 3t 14 + 4t 9 − 6t 2 + 4.

We have
f 0 (t) = 6t g(t), g(t) = 3t 16 − 7t 12 + 6t 7 − 2,
g 0 (t) = 6t 6 h(t), h(t) = 8t 9 − 14t 5 + 7,
h0 (t) = 2t 4 (36t 2 − 35).
Since h0 (t) > 0 for t ≥ 1, h is increasing, h(t) ≥ h(1) = 1 for t ≥ 1, g is increasing,
g(t) ≥ g(1) = 0 for t ≥ 1, f is increasing, hence f (t) ≥ f (1) = 0 for t ≥ 1.
The equality holds for a = b = c = 1.

P 5.87. If a, b, c, d are nonnegative real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 18) ≤ 10(a3 + b3 + c 3 + d 3 − 4).

(Vasile Cîrtoaje, 2010)


EV Method for Nonnegative Variables 461

Solution. Apply Corollary 2 for n = 4, k = 2, m = 3:


• If a, b, c, d are real numbers so that 0 ≤ a ≤ b ≤ c ≤ d and

a + b + c + d = 4, a2 + b2 + c 2 + d 2 = const ant,

then
S4 = a3 + b3 + c 3 + d 3
is minimal for either 0 < a ≤ b = c = d or a = 0.
Case 1: 0 < a ≤ b = c = d. We need to show that a + 3d = 4 involves

(a2 + 3d 2 − 4)(a2 + 3d 2 + 18) ≤ 10(a3 + 3d 3 − 4).

This inequality is equivalent to

(1 − d)2 (1 + d)(4 − 3d) ≥ 0,

(1 − d)2 (1 + d)a ≥ 0.

Case 2: a = 0. Let
s = b2 + c 2 + d 2 .
We need to show that b + c + d = 4 involves

(s − 4)(s + 18) ≤ 10(b3 + c 3 + d 3 − 4).

By the Cauchy-Schwarz inequality, we have


1 16
s ≥ (b + c + d)2 =
3 3
and
s2
(b + c + d)(b3 + c 3 + d 3 ) ≥ (b2 + c 2 + d 2 )2 , b3 + c 3 + d 3 ≥ .
4
Thus, it suffices to show that

s2
 ‹
(s − 4)(s + 18) ≤ 10 −4 ,
4
which is equivalent to the obvious inequality

(s − 4)(3s − 16) ≥ 0.

The equality holds for a = b = c = d = 1, and also for


4
a = 0, b=c=d=
3
(or any cyclic permutation).
462 Vasile Cîrtoaje

P 5.88. If a, b, c, d are nonnegative real numbers such that

a + b + c + d = 4,

then
(a4 + b4 + c 4 + d 4 )2 ≥ (a2 + b2 + c 2 + d 2 )(a5 + b5 + c 5 + d 5 ).
(Vasile C., 2020)

Proof. Consider the inequality

(a14 + a24 + · · · + an4 )2 ≥ (a12 + a22 + · · · + an2 )(a15 + a25 + · · · + an5 ),

where a1 , a2 , . . . , an are nonnegative real numbers such that a1 + a2 + · · · + an = n.


Write this inequality in the homogeneous form

n(a14 + a24 + · · · + an4 )2 ≥ (a1 + a2 + · · · + an )(a12 + a22 + · · · + an2 )(a15 + a25 + · · · + an5 ).

1/4 1/4
Replacing a1 , a2 , . . . , an with x 1 , x 2 , . . . , x n1/4 , the inequality becomes

n(x 1 + x 2 + · · · + x n )2 ≥
1/4 1/4 1/2 1/2 5/4 5/4
  
≥ x 1 + x 2 + · · · + x n1/4 x 1 + x 2 + · · · + x n1/2 x 1 + x 2 + · · · + x n5/4 .
By Corollary 5 (case k = 5/4), if
5/4 5/4
x 1 + x 2 + · · · + x n = const ant, x 1 + x 2 + · · · + x n5/4 = const ant,

1/4 1/4 1/2 1/2


then the sums x 1 + x 2 + · · · + x n1/4 and x 1 + x 2 + · · · + x n1/2 are maximal for

0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n .

Since the case a1 = a2 = · · · = an−1 = 0 is trivial, it suffices to consider the case


a1 = a2 = · · · = an−1 = 1, when the required inequality becomes f (a) ≥ 0, where

f (a) = (a4 + n − 1)2 − (a + n − 1)(a2 + n − 1)(a5 + n − 1), a ≥ 1.

We have
f (a)
= a8 − a7 − a6 − (n − 1)a5 + 2na4 − a3 − (n − 1)a2 − (n − 1)a + n − 1
n−1

= a3 A − (n − 1)B,
where
A = a5 − a4 − a3 + 2a − 1, B = a5 − 2a4 + a2 + a − 1.
Since
A = (a − 1)2 (a3 + a2 − 1), B = (a − 1)2 (a3 − a − 1),
EV Method for Nonnegative Variables 463

we have
f (a) = (n − 1)(a − 1)2 g(a),
where
g(a) = a6 + a5 − na3 + (n − 1)a + n − 1.
The inequality is true if g(a) ≥ 0. For n = 4, we have
g(a) = a6 + a5 − 4a3 + 3a + 3 > 2a5 − 4a3 + 2a = 2a(a2 − 1)2 ≥ 0.
The equality occurs for a = b = c = d = 1.
Remark 1. Since g(a) ≥ 0 for n ≤ 16, the homogeneous inequality is true for all
n ≤ 16.
Remark 2. Since
(a1 + a2 + · · · + an )(a15 + a25 + · · · + an5 ) ≤ |(a1 + a2 + · · · + an )(a15 + a25 + · · · + an5 )|
≤ (|a1 | + |a2 | + · · · + |an |)(|a1 |5 + |a2 |5 + · · · + |an |5 ),
the homogeneous inequality is true for n ≤ 16 and real a1 , a2 , . . . , an .

P 5.89. If a, b, c, d are nonnegative real numbers such that


a + b + c + d = 4,
then
13(a2 + b2 + c 2 + d 2 )2 ≥ 12(a4 + b4 + c 4 + d 4 ) + 160.
(Vasile Cîrtoaje, 2020)
Solution. Write the inequality in the homogeneous form
104(a2 + b2 + c 2 + d 2 )2 ≥ 96(a4 + b4 + c 4 + d 4 ) + 5(a + b + c + d)4 .
According to Corollary 5, for a + b + c + d = const ant and a2 + b2 + c 2 + d 2 =
const ant, the sum
S = a4 + b4 + c 4 + d 4
is maximal when a ≥ b = c = d. Therefore, it suffices to consider this case. Due
to homogeneity, for the nontrivial case b = c = d 6= 0, we may consider that
b = c = d = 1. Thus we only need to prove that
104(a2 + 3)2 ≥ 96(a4 + 3) + 5(a + 3)4 ,
which is equivalent to
(a − 1)2 (a − 9)2 ≥ 0.
1
The equality occurs for a = b = c = d = 1, and also for a = 3 and b = c = d =
3
(or any cyclic permutation).
464 Vasile Cîrtoaje

P 5.90. If a1 , a2 , . . . , a8 are nonnegative real numbers, then

19(a12 + a22 + · · · + a82 )2 ≥ 12(a1 + a2 + · · · + a8 )(a13 + a23 + · · · + a83 ).

(Vasile C., 2007)

Solution. By Corollary 5 (case n = 8, k = 2, m = 3), if 0 ≤ a1 ≤ a2 ≤ · · · ≤ a8 and

a1 + a2 + · · · + a8 = const ant, a12 + a22 + · · · + a82 = const ant,

then the sum


S8 = a13 + a23 + · · · + a83
is maximal for a1 = a2 = · · · = a7 ≤ a8 . Due to homogeneity, we only need to
consider the cases a1 = a2 = · · · = a7 = 0 and a1 = a2 = · · · = a7 = 1. For the
second case (nontrivial), we need to show that

19(7 + a82 )2 ≥ 12(7 + a8 )(7 + a83 ),

which is equivalent to

a84 − 12a83 + 38a82 − 12a8 + 49 ≥ 0,

(a82 − 6a8 + 1)2 + 48 ≥ 0.

The equality holds for a1 = a2 = · · · = a8 = 0.

P 5.91. If a, b, c are nonnegative real numbers so that

5(a2 + b2 + c 2 ) = 17(ab + bc + ca),

then v v p
t3 s a 1+ 7
t b s
c
3 ≤ + + ≤ p .
5 b+c c+a a+b 2
(Vasile C., 2006)

Solution. Due to homogeneity, we may assume that a + b + c = 9. From the


hypothesis 5(a2 + b2 + c 2 ) = 17(ab + bc + ca), which is equivalent to

27(a2 + b2 + c 2 ) = 17(a + b + c)2 ,

we get
a2 + b2 + c 2 = 51.
Also, from 2(b2 + c 2 ) ≥ (b + c)2 and

b + c = 9 − a, b2 + c 2 = 51 − a2 ,
EV Method for Nonnegative Variables 465

we get a ≤ 7. Write the desired inequality in the form


v p
t3 1+ 7
3 ≤ f (a) + f (b) + f (c) ≤ p .
5 2
where s
u
f (u) = , 0 ≤ u ≤ 7.
9−u
We have
9
g(x) = f 0 (x) = ,
2x 1/2 (9 − x)3/2
27(8x 2 − 36x + 81)
g 00 (x) = .
8x 5/2 (9 − x)7/2
Since g 00 (x) > 0 for x ∈ (0, 7], g is strictly convex on (0, 7]. According to Corollary
1, if 0 ≤ a ≤ b ≤ c and

a + b + c = 9, a2 + b2 + c 2 = 51,

then the sum S3 = f (a) + f (b) + f (c) is maximal for a = b ≤ c, and is minimal for
either a = 0 or 0 < a ≤ b = c.
(a) To prove the right inequality, it suffices to consider the case a = b ≤ c.
From
a + b + c = 9, a2 + b2 + c 2 = 51,
we get a = b = 1 and c = 7, therefore
v p
1+ 7
s s
a t b c
+ + = p .
b+c c+a a+b 2
The original right inequality is an equality for a = b = c/7 (or any cyclic permuta-
tion).
(b) To prove the left inequality, it suffices to consider the cases a = 0 and
0 < a ≤ b = c. For a = 0, from

a + b + c = 9, a2 + b2 + c 2 = 51,

we get
b c 17
+ = ,
c b 5
therefore
v v v v
s
a t b s
c tb sc tb c t3
+ + = + = + +2=3 .
b+c c+a a+b c b c b 5
The case 0 < a ≤ b = c is not possible, because from

a + b + c = 9, a2 + b2 + c 2 = 51,
466 Vasile Cîrtoaje

we get a = 7 and b = c = 1, which don’t satisfy the condition a ≤ b. The original


left inequality is an equality for

b c 17
a = 0, + =
c b 5
(or any cyclic permutation).

P 5.92. If a, b, c are nonnegative real numbers so that

8(a2 + b2 + c 2 ) = 9(ab + bc + ca),

then
19 a b c 141
≤ + + ≤ .
12 b+c c+a a+b 88
(Vasile C., 2006)

Solution. The proof is similar to the one of the preceding P 5.91. Assume that
a + b + c = 15, which involves a2 + b2 + c 2 = 81 and a ∈ [3, 7], then write the
inequality in the form

19 141
≤ f (a) + f (b) + f (c) ≤ ,
12 88
where
u
f (u) = , 3 ≤ u ≤ 7.
15 − u
We have
1 90
g(x) = f 0 (x) = (15 − x)2 , g 00 (x) = .
5 (15 − x)4
Since g is strictly convex on [3, 7], according to Corollary 1, if 0 ≤ a ≤ b ≤ c and

a + b + c = 15, a2 + b2 + c 2 = 81,

then the sum S3 = f (a) + f (b) + f (c) is maximal for a = b ≤ c, and is minimal for
either a = 0 or 0 < a ≤ b = c.
(a) To prove the right inequality, it suffices to consider the case a = b ≤ c,
which involves
a = b = 4, c = 7,
and
a b c 141
+ + = .
b+c c+a a+b 88
The original right inequality is an equality for a = b = 4c/7 (or any cyclic permu-
tation).
EV Method for Nonnegative Variables 467

(b) To prove the left inequality, it suffices to consider the cases a = 0 and
0 < a ≤ b = c. The first case is not possible, while the second case involves

a = 3, b = c = 6,

and
a b c 19
+ + = .
b + c c + a a + b 12
The original left inequality is an equality for 2a = b = c (or any cyclic permutation).

P 5.93. If a, b, c ∈ (0, 2] such that a + b + c = 3, then


v v v
t 2(b + c) t 2(c + a) t 2(a + b) 9
−1 + −1 + −1≥ p .
a b c ab + bc + ca

(Vasile C., 2020)

Solution. Write the inequality in the form


p
−3 3
f (a) + f (b) + f (c) ≤ p ,
ab + bc + ca
where v
t2
f (u) = − − 1, 0 < u ≤ 2.
u
We have f (0+) = −∞ and

g(x) = f 0 (x) = x −3/2 (2 − x)−1/2 , g 0 (x) = (2x − 3)x −5/2 (2 − x)−3/2 ,

g 00 (x) = (7x 2 − 20x + 15)x −7/2 (2 − x)−5/2 > 0.


Since g is strictly convex on (0, 2), according to Corollary 1, Note 1 and Note 2, if
a ≥ b ≥ c > 0 and

a + b + c = 3, ab + bc + ca = const ant,

then the sum S3 = f (a) + f (b) + f (c) is maximal for a = 2 or a ≥ b = c. Thus, it


suffices to prove the desired inequality for these cases.
Case 1: a = 2. We need to prove the homogeneous inequality
v v v
t 2(b + c) t 2(c + a) t 2(a + b) 3(a + b + c)
−1 + −1 + −1≥ p
a b c ab + bc + ca
for
a = 2(b + c).
468 Vasile Cîrtoaje

The inequality is equivalent to


v v p
t 2b t 2c 3 3(b + c)
+1+ +1≥ p .
c b 2(b + c)2 + bc
Let
(b + c)2
x= , x ≥ 1.
4bc
Since v v v
t 2c t 2b 2b 2c p
t ‹ ‹
4 4
+1+ +1≥2 +1 + 1 = 2 8x + 1,
b c c b
the inequality becomes p
p
4 3 3x
8x + 1 ≥ p ,
8x + 1
(8x + 1)3 ≥ 729x 2 .
Since
8x + 1 ≥ 3(2x + 1),
it suffices to show that
(2x + 1)3 ≥ 27x 2 .
This is true because p
3
2x + 1 = x + x + 1 ≥ 3 x 2 .
Case 2: a ≥ b = c. We need to show that a + 2c = 3 implies
v v
t 4c t 2(a + c) 9
−1+2 −1≥ p ,
a c 2ac + c 2
that is v v
t2 − a t1 + a 6
+2 ≥p ,
a 3−a (1 + a)(3 − a)
v
t2 − a 2(2 − a)
≥p .
a (1 + a)(3 − a)
It is true if p
1 2 2−a
p ≥p ,
a (1 + a)(3 − a)
which, by squaring, reduces to

(a − 1)2 ≥ 0.
1
The equality occurs for a = b = c = 1, and also for a = b = and c = 2 (or any
2
cyclic permutation).
EV Method for Nonnegative Variables 469

P 5.94. Let a, b, c and x, y, z be nonnegative real numbers such that

x 3 + y 3 + z 3 = a3 + b3 + c 3 .

Then,
(a + b + c)(x + y + z) p
3
≥ 3.
ab + bc + ca + x y + yz + z x
(Vasile Cîrtoaje, 2019)

Solution. Assume that


x + y +z ≥ a+ b+c
and denote
x + y +z a+b+c
t= , t≥ .
3 3
Since v v
a+b+c x + y +z t3 x + y +z
3 3 3
3 a + b + c
t 3 3 3
≤ ≤ = ,
3 3 3 3
we have
t1 ≤ t ≤ t2,
where v
a+b+c 3 a + b + c
t 3 3 3
t1 = , t2 = .
3 3
It is enough to prove the inequality

1 1
p
3
(a + b + c)(x + y + z) ≥ ab + bc + ca + (x + y + z)2 .
3 3

For fixed a, b, c, we may write the required inequality as f (t) ≤ 0, where


p
3
f (t) = 3t 2 − 9 (a + b + c)t + ab + bc + ca

is a quadratic convex function. Thus, it is enough to show that f (t 1 ) ≤ 0 and


f (t 2 ) ≤ 0. We have
€p
3
Š
3 f (t 1 ) = 3(ab + bc + ca) − 9 − 1 (a + b + c)2
€ p
3
Š
≤ 3 2 − 9 (ab + bc + ca) ≤ 0.
To prove the inequality f (t 2 ) ≤ 0, we write it as
p3
3t 22 − 9 (a + b + c)t 2 + ab + bc + ca ≤ 0.

According to Corollary 5, for a + b + c = const ant and a n + b n + c n = const ant,


the sum a2 + b2 + c 2 is minimal (hence the sum ab + bc + ca is maximal) for a ≥
470 Vasile Cîrtoaje

b = c. Thus, due to homogeneity, it is enough to prove the inequality for a = 1 and


b = c ≤ 1. So, we need to prove that g(u) ≤ 0, where

c 2 + 2c
g(u) = u2 − (2c + 1)u + p3
,
3
with p
3
u= 2c 3 + 1, c ∈ [0, 1].
Consider two cases: c ∈ [0, 4/5] and c ∈ [4/5, 1].
p3
Case 1: c ∈ [0, 4/5]. Since 3 > 4/3, we have

3(c 2 + 2c) (2u − 3c)(2u − c − 2)


g(u) ≤ u − (2c + 1)u +
2
= .
4 4
Thus, we need to show that
3c c+2
≤u≤ .
2 2
The left inequality is equivalent to
v
t8
c≤ .
11
This is true since v
4 t8
c≤ < .
5 11
The right inequality is equivalent to

c(2c + 6 − 5c 2 ) ≥ 0.
p
3
Case 2: c ∈ [4/5, 1]. Since 3 > 7/5, we have g(u) < h(u), where

5(c 2 + 2c)
h(u) = u2 − (2c + 1)u + .
7
It suffices to prove that h(u) ≤ 0. From

h0 (u) = 2u − 2c − 1

and
64 24 −9
(2u)3 − (2c + 1)3 = 7 + 8c 3 − 12c 2 − 6c ≤ 7 − 4c 2 − 6c ≤ 7 − − = < 0,
25 5 25
it follows that h0 (u) < 0, hence h(u) is a decreasing function. Since

c3
u>1+ ,
3
EV Method for Nonnegative Variables 471

it follows that

c3 5c c 2 c 5 4 2c 3
 ‹  ‹
h(u) < h 1 + =c + + − − .
3 7 3 9 7 3
Since
5c c 2 c 5 5c c c 3 22c c 3
+ + ≤ + + = + ,
7 3 9 7 3 9 21 9
it suffices to show that
22c c 3 4 2c 3
+ − − ≤ 0,
21 9 7 3
that is
22c 4 5c 3
− − ≤ 0.
21 7 9
Indeed, we have
v
4 5c 3 2 2 5c 3 t3 20c 3 22c
+ = + + ≥3 > .
7 9 7 7 9 49 · 9 21

Thus, the proof is completed. If a ≥ b ≥ c and x ≥ y ≥ z, then the equality occurs


x a
for a = b = c = pn
and y = z = 0, and for x = y = z = pn
and b = c = 0.
3 3

P 5.95. If a, b, c, d are positive numbers such that

1 1 1 1
a+b+c+d = + + + ,
a b c d
then
ab + ac + ad + bc + bd + cd + 3abcd ≥ 9.

(Vasile Cîrtoaje, 2019)

Solution. Write the inequality as

(a + b + c + d)2 + 6abcd ≥ 18 + a2 + b2 + c 2 + d 2

and apply Corollary 4 for k = −1, and Corollary 5 for k = −1 and m = 2:


• If a, b, c, d are positive numbers such that

1 1 1 1
a + b + c + d = const ant , + + + = const ant, a ≤ b ≤ c ≤ d,
a b c d

then the product abcd is minimal and the sum a2 + b2 + c 2 + d 2 is maximal for
a = b = c ≤ d.
472 Vasile Cîrtoaje

Thus, it suffices to consider this case. We need to show that


3 1
3a + d = +
a d
involve
a2 + ad + a3 d ≥ 3.
From the hypothesis, we get
p
9a4 − 14a2 + 9
3(1 − a2 ) +
d= .
2a
So, the required inequality becomes as follows:
a2 + (a2 + 1)ad ≥ 3,
p
(a2 + 1) 9a4 − 14a2 + 9 ≥ 3a4 − 2a2 + 3,
(a2 + 1)2 (9a4 − 14a2 + 9) ≥ (3a4 − 2a2 + 3)2 ,
16a2 (a2 − 1)2 ≥ 0.
The equality occurs for a = b = c = d = 1.

P 5.96. If a1 , a2 , a3 , a4 , a5 are nonnegative real numbers, then

(a13 + a23 + a33 + a43 + a53 )2 1X


≥ ai a j .
a14 + a24 + a34 + a44 + a54 2 i< j

(Vasile Cîrtoaje, 2019)


Solution. Write the inequality in the form
4(a13 + a23 + a33 + a43 + a53 )2
+ a12 + a22 + a32 + a42 + a52 ≥ (a1 + a2 + a3 + a4 + a5 )2 .
a14 + a24 + a34 + a44 + a54
According to Corollary 5, for a1 + a2 + a3 + a4 + a5 = const ant and a13 + a23 +
a33 + a43 + a53 = const ant, the sum a12 + a22 + a32 + a42 + a52 is minimal and the sum
a14 + a24 + a34 + a44 + a54 is maximal for a1 = a2 = a3 = a4 ≤ a5 .Thus, it is enough to
show that
4(4x 3 + y 3 )2
+ 4x 2 + y 2 ≥ (4x + y)2 ,
4x 4 + y 4
which can be written as
4x 6 − 8x 5 y + 8x 3 y 3 − 3x 2 y 4 − 2x y 5 + y 6 ≥ 0,
(x − y)2 (2x 2 − y 2 )2 ≥ 0.
The proof is completed. The equality occurs for a1 = a2 = a3 = a4 = a5 , and also
a5
for a1 = a2 = a3 = a4 = p (or any cyclic permutation).
2
EV Method for Nonnegative Variables 473

P 5.97. If a1 , a2 , . . . , an ≥ 0 such that

a1 + a2 + · · · + an = n,

then v
1 X
 ‹
p u
a1 + a2 + · · · + an ≤ 2n − 1 + 2 1 −
p p t ai a j .
n i< j

(Vasile C., 2018)


Proof. Since
X
2 ai a j = (a1 + a2 + · · · + an )2 − a12 − a22 − · · · − an2 = n2 − a12 − a22 − · · · − an2 ,
i< j

we can write the inequality as


1
 ‹
p p 2
a1 + a2 + · · · + a n ≤ n + n − 1 − 1 −
2
(a12 + a22 + · · · + an2 ).
p
n
Now, we can apply Corollary 5 for k = 2 and m = 1/2:
• If a1 , a2 , . . . , an are nonnegative real numbers so that

a1 + a2 + · · · + an = n , a12 + a22 + · · · + an2 = const ant, a1 ≤ a2 ≤ · · · ≤ an ,

then the sum


p
a1 + a2 + · · · +
p p
an
is maximal for 0 ≤ a1 = · · · = an−1 ≤ an .
Thus, it suffices to show that
1
 ‹
[(n − 1)x + y] ≤ n + n − 1 − 1 −
2 2
[(n − 1)x 4 + y 4 ].
n
for
(n − 1)x 2 + y 2 = n, 0≤ x ≤ y
Write this inequality in the homogeneous form
[(n−1)x 2 + y 2 ]2
(n2 + n − 1) n − (n − 1)[(n − 1)x 4 + y 4 ]
[(n − 1)x + y] ≤ 2
,
(n − 1)x 2 + y 2
which is equivalent to

(n − 1)2 x 4 − 2n(n − 1)x 3 y + (n2 + 2n − 2)x 2 y 2 − 2nx y 3 + y 4 ≥ 0,

(x − y)2 [(n − 1)x − y]2 ≥ 0.


The inequality is an equality for a1 = a2 = · · · = an = 1, and also for a1 = · · · =
1
an−1 = and an = n − 1 (or any cyclic permutation).
n−1
474 Vasile Cîrtoaje

P 5.98. If a1 , a2 , . . . , an ≥ 0 such that


X
a1 + a2 + · · · + a n = ai a j > 0,
i< j

then
(n − 1)(n − 2) Xp
(a1 + a2 + · · · + an ) + ai a j ≥ n(n − 1).
2 i< j

(Vasile C., 2020)

Proof. For n = 2, we need to show that a1 + a2 = a1 a2 involves a1 a2 ≥ 4. Indeed,


this follows from
a1 a2 = a1 + a2 ≥ 2 a1 a2 ,
p

Since X
2 ai a j = (a1 + a2 + · · · + an )2 − a12 − a22 − · · · − an2
i< j

and Xp p
ai a j = ( a1 + a2 + · · · + an )2 − a1 − a2 − · · · − an ,
p p
2
i< j

we can apply Corollary 5 for k = 2 and m = 1/2:


• If a1 , a2 , . . . , an are nonnegative real numbers so that

a1 + a2 +· · ·+ an = const ant , a12 + a22 +· · ·+ an2 = const ant, a1 ≤ a2 ≤ · · · ≤ an ,

then the sum


p
a1 + a2 + · · · +
p p
an
is minimal for either 0 < a1 ≤ a2 = · · · = an or a1 = 0.
Thus, it suffices to consider the case a1 = x 2 , a2 = · · · = an = y 2 , 0 < x ≤ y, and
the case a1 = 0. In addition, we will use the induction method.
Case 1: a1 = x 2 , a2 = · · · = an = y 2 . We need to show that

(n − 1)(n − 2) 4
x 2 + (n − 1) y 2 = (n − 1)x 2 y 2 + y
2
implies
(n − 2) 2 (n − 2) 2
[x + (n − 1) y 2 ] + x y + y ≥ n,
2 2
which can be written in the homogeneous form

2(n − 1)x 2 y 2 + (n − 1)(n − 2) y 4


(n − 2)x 2 + 2x y + n(n − 2) y 2 ≥ n .
x 2 + (n − 1) y 2
For y = 1, the inequality becomes

(x 2 + n − 1)[(n − 2)x 2 + 2x + n(n − 2)] ≥ 2n(n − 1)x 2 + n(n − 1)(n − 2),


EV Method for Nonnegative Variables 475

(n − 2)x 4 + 2x 3 − (3n − 2)x 2 + 2(n − 1)x ≥ 0,

x(x − 1)2 [(n − 2)x + 2(n − 1)] ≥ 0.


Case 2: a1 = 0. We need to show that
X
a2 + a3 + · · · + a n = ai a j > 0 (1)
2≤i< j

involves

(n − 1)(n − 2) X p
(a2 + a3 + · · · + an ) + ai a j ≥ n(n − 1). (2)
2 2≤i< j

From
(a2 + a3 + · · · + an )2 ≤ (n − 1)(a22 + a23 + · · · + an2 )
X
= (n − 1)(a2 + a3 + · · · + an ) − 2(n − 1)
2
ai a j ,
2≤i< j

we get
X
(n − 2)(a2 + a3 + · · · + an )2 ≥ 2(n − 1) ai a j = 2(n − 1)(a2 + a3 + · · · + an ),
2≤i< j

hence
2(n − 1)
a2 + a3 + · · · + an ≥ . (3)
n−2
On the other hand, by the induction hypothesis, (1) involves

(n − 2)(n − 3) X p
(a2 + a3 + · · · + an ) + ai a j ≥ (n − 1)(n − 2).
2 2≤i< j

According to this inequality, (2) is true if

(n − 1)(n − 2) (n − 2)(n − 3)
(a2 + a3 + · · · + an ) + (n − 1)(n − 2) − (a2 + a3 + · · · + an )
2 2

≥ n(n − 1),
which is equivalent to (3).
2
The inequality is an equality for a1 = a2 = · · · = an = , and also for a1 = 0
n−1
2
and a2 = a3 = · · · = an = (or any cyclic permutation).
n−2
476 Vasile Cîrtoaje

P 5.99. Let

F (a1 , a2 , . . . , an ) = n(a12 + a22 + · · · + an2 ) − (a1 + a2 + · · · + an )2 ,

where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and

a12 (a22 + a32 + · · · + an2 ) ≥ n − 1.

Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an
(Vasile C., 2020)

Proof. For n = 2, we need to show that a1 a2 ≥ 1 involves

(a12 a22 − 1)(a1 − a2 )2 ≥ 0,

which is clearly true. For n ≥ 3, write the inequality as

1 2
  
1 1 1 1 1
‹
n(a1 +a2 +· · ·+an )−(a1 +a2 +· · ·+an ) ≥ n 2 + 2 + · · · + 2 −
2 2 2 2
+ + ··· + .
a1 a2 an a1 a2 an

According to Corollary 5 (case k = −1), we have:


• If a2 , a3 , . . . , an are positive real numbers so that
1 1 1
a2 +a3 +· · ·+an = const ant , + +· · ·+ = const ant, a2 ≤ a3 ≤ · · · ≤ an ,
a2 a3 an
1 1 1
then the sum a22 + a32 + · · · + an2 is minimal and the sum 2
+ 2 + · · · + 2 is maximal
a2 a3 an
for a2 ≤ a3 = · · · = an .
Thus, it suffices to consider the case a2 ≤ a3 = · · · = an . We need to show that if
x, y, z are positive real numbers such that x ≤ y ≤ z and

x 2 [ y 2 + (n − 2)z 2 ] ≥ n − 1,

then
1 1 n−2 1 1 n−2 2
‹   ‹
n[x + y +(n−2)z ]−[x+ y+(n−2)z] ≥ n
2 2 2
+ + 2 − 2
+ + ,
x2 y2 z x y z
which is equivalent to

(x − y)2 (n − 2)( y − z)2 (n − 2)(z − x)2


(x− y)2 +(n−2)( y−z)2 +(n−2)(z−x)2 ≥ + + ,
x2 y2 y 2z2 z2 x 2

1 1 1
 ‹  ‹  ‹
(x − y) 1 − 2 2 + (n − 2)( y − z) 1 − 2 2 + (n − 2)(z − x) 1 − 2 2 ≥ 0.
2 2 2
x y y z z x
EV Method for Nonnegative Variables 477

From
n − 1 ≤ x 2 [ y 2 + (n − 2)z 2 ] ≤ (n − 1)x 2 z 2 ,
it follows that
xz ≥ 1, yz ≥ 1.
Thus, suffices to show that
1 1
 ‹  ‹
(x − y)2 1 − + (n − 2)(z − x)2
1 − ≥ 0,
x2 y2 z2 x 2
that is
x 2 2 1 x 2 1
  ‹  ‹  ‹
(n − 2) 1 − z − 2 ≥ 1− 2
−y .
z x y x2
Since
x x
1− ≥ 1 − ≥ 0,
z y
it suffices to show that
1 1
 ‹
(n − 2) z − 2 ≥ 2 − y 2 ,
2
x x
that is equivalent to the hypothesis
n−1
y 2 + (n − 2)z 2 ≥ .
x2
1
The equality occurs for a1 = a2 = · · · = an ≥ 1 and for = a2 = · · · = an ≥ 1.
a1
Remark. Since a1 (a2 + a3 + · · · + an ) ≥ n − 1 yields a12 (a22 + a32 + · · · + an2 ) ≥ n − 1,
the inequality is also true for
a1 (a2 + a3 + · · · + an ) ≥ n − 1.
In addition, it is true in the particular case
a1 , a2 , . . . , an ≥ 1.

P 5.100. Let
F (a1 , a2 , . . . , an ) = a1 + a2 + · · · + an − n n a1 a2 · · · an ,
p

where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and


a1 (a2 + a3 + · · · + an ) ≥ n − 1.
Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an
(Vasile C., 2020)
478 Vasile Cîrtoaje

Solution. For n = 2, we need to show that a1 a2 ≥ 1 involves


p 2
(a1 a2 − 1)
p
a1 − a2 ≥ 0,

which is true. For n ≥ 3, the inequality has the form


1 1 1 n
a1 + a2 + · · · + an − n n a1 a2 · · · an ≥ + + ··· +
p
−p .
a1 a2 an n
a1 a2 · · · an
According to Corollary 5 (case k = 0 and m = −1), we have:
• If a2 , a3 , . . . , an are positive real numbers so that

a2 + a3 + · · · + an = const ant , a2 a3 · · · an = const ant, a2 ≤ a3 ≤ · · · ≤ a n ,


1 1 1
then the sum + + ··· + is maximal for a2 ≤ a3 = · · · = an .
a2 a3 an
Thus, we only need to show that
p 1 1 n−2 n
x + y + (n − 2)z − n + +
n
x yz n−2 ≥ −p
x y z n
x yz n−2
for 0 < x ≤ y ≤ z and x[ y + (n − 2)z] ≥ n − 1. Since both sides of the inequality
are nonnegative, it suffices to prove the homogeneous inequality
— x[ y + (n − 2)z] 1 1 n − 2
 
” p n
x + y + (n − 2)z − n x yz n−2 ≥ + +
n
−p ,
n−1 x y z n
x yz n−2
that is ” p —
(n − 1) x + y + (n − 2)z − n x yz n−2 ≥
n

v
[ y + (n − 2)z][(n − 2) y + z] t x n−1
≥ y + (n − 2)z + x − n[ y + (n − 2)z]
n
.
yz yz n−2
For fixed y and z, write this inequality as f (x) ≥ 0, x ∈ (0, y]. We will show that

f (x) ≥ f ( y) ≥ 0.

To prove that f (x) ≥ f ( y), we show that f 0 (x) ≤ 0, which is equivalent to


v
n yz
t n−2 [ y + (n − 2)z][(n − 2) y + z] y + (n − 2)z
n − 1 − (n − 1) n−1
− + (n − 1) p ≤0,
x yz n
x yz n−2
v
n−2 y + (n − 2)z
 ‹
y z n yz
t
(n − 2) + + n − 3 + (n − 1) ≥ (n − 1) p .
z y x n−1 n
x yz n−2
By the AM-GM inequality, we have
v
n−2
 ‹
y z n yz
t
(n − 2) · + + n − 3 + (n − 1) ≥
z y x n−1
EV Method for Nonnegative Variables 479

v v

t y ‹n−2 n−2
z n yz
n−1
t
≥ (n − 1) + +n−3 · (n − 1) .
z y x n−1
Thus, it suffices to show that
v v
y + (n − 2)z

t y z ‹n−2 n−2
n yz
n−1
t
+ +n−3 · (n − 1) n−1
≥ p ,
z y x n
x yz n−2
which is equivalent to
 ‹n−2
y z
(n − 1) + +n−3 yz n−2 ≥ [ y + (n − 2)z]n−1 .
z y
Due to homogeneity, we may set z = 1, when the inequality becomes

(n − 1)Ay ≥ y + n − 2,

where ‹n−2
y + 1/ y + n − 3

A= , 0 < y ≤ 1.
y +n−2
By Bernoulli’s inequality, we have

1/ y − 1 n−2 (n − 2)(1/ y − 1) y2 + n − 2
 ‹
A= 1+ ≥1+ = ,
y +n−2 y +n−2 y( y + n − 2)
hence
(n − 1)( y 2 + n − 2)
(n − 1)Ay − ( y + n − 2) ≥ − ( y + n − 2)
y +n−2
(n − 2)( y − 1)2
= ≥0.
y +n−2
The inequality f ( y) ≥ 0 has the form

+ (n
 
p y[ y − 2)z] 2 n − 2 n
2 y + (n − 2)z − n +
n
y 2 z n−2 ≥ −p .
n−1 y z n
y 2 z n−2
Due to homogeneity, we may set z = 1 (hence 0 < y ≤ 1), when the inequality
becomes
+
 
p y( y n − 2) 2 n
2 y + n − 2 − n y2 ≥ +n−2− p
n
.
n−1 y n
y2
Denoting
p
t= n
y, 0 < t ≤ 1,
we need to show that g(t) ≥ 0, where

g(t) = (n − 1)(2t n − nt 2 + n − 2) − (t n + n − 2)[(n − 2)t n − nt n−2 + 2]


480 Vasile Cîrtoaje

= −(n − 2)t 2n + nt 2n−2 − (n − 2)(n − 4)t n + n(n − 2)t n−2 − n(n − 1)t 2 + (n − 2)(n − 3) .
For n = 3, we have
g(t) = t(1 − t)3 (3 + 3t + t 2 ) ≥ 0,
and for n = 4, we have

g(t) = 2(1 − t 2 )3 (1 + t 2 ) ≥ 0.

For n ≥ 5, we have
g 0 (t) = nt g1 (t),
g1 (t) = −2(n − 2)t 2n−2 + 2(n − 1)t 2n−4 − (n − 2)(n − 4)t n−2 + (n − 2)2 t n−4 − 2(n − 1),
g10 (t) = (n − 2)t n−5 (1 − t 2 )[4(n − 1)t n + n − 2] ≥ 0 ,
hence g1 (t) is increasing, g1 (t) ≤ g1 (1) = 0, g 0 (t) ≤ 0, g(t) is decreasing, g(t) ≥
g(1) = 0. Thus, the proof is completed. The equality holds for a1 = a2 = · · · =
an ≥ 1.
Remark 1. Since a1n−1 a2 a3 · · · an ≥ 1 yields a1 (a2 + a3 + · · · + an ) ≥ n − 1, the
inequality
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,...,
a1 a2 an
is also valid if a1 , a2 , . . . , an are positive real numbers such that

a1 ≤ a2 ≤ · · · ≤ an , a1n−1 a2 a3 · · · an ≥ 1.

Also, it is valid in the particular case

a1 , a2 , . . . , an ≥ 1.

Remark 2. Since a1 a2 · · · an ≥ 1, from P 5.100 it follows that


1 1 1
a1 + a2 + · · · + an ≥ + + ··· +
a1 a2 an
for
a1 (a2 + a3 + · · · + an ) ≥ n − 1.

P 5.101. Let
v
t a2 + a2 + · · · + a2
1 2 n a1 + a2 + · · · + an
F (a1 , a2 , . . . , an ) = − ,
n n
where a1 , a2 , . . . , an are positive real numbers such that a1 ≤ a2 ≤ · · · ≤ an and

a1n−1 (a2 + a3 + · · · + an ) ≥ n − 1.
EV Method for Nonnegative Variables 481

Then,
1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,..., .
a1 a2 an
(Vasile C., 2020)
Solution. For n = 2, we need to show that a1 a2 ≥ 1 involves
q
(a1 a2 − 1)( 2(a12 + a22 ) − a1 − a2 ) ≥ 0,

which is true. For n ≥ 3, write the inequality in the form


q
n(a12 + a22 + · · · + an2 ) − (a1 + a2 + · · · + an )
v  
1 1 1 1 1 1
u
≥ n 2 + 2 + ··· + 2 − + + ··· +
t
≥0.
a1 a2 an a1 a2 an
According to Corollary 5 (case k = −1), we have:
• If a2 , a3 , . . . , an are positive real numbers so that
1 1 1
a2 +a3 +· · ·+an = const ant , + +· · ·+ = const ant, a2 ≤ a3 ≤ · · · ≤ an ,
a2 a3 an
1 1 1
then the sum a22 + a32 + · · · + an2 is minimal and the sum 2
+ 2 + · · · + 2 is maximal
a2 a3 an
for a2 ≤ a3 = · · · = an .
Thus, it suffices to consider the case a2 ≤ a3 = · · · = an . We need to show that if
x, y, z are positive real numbers such that x ≤ y ≤ z and

x n−1 [ y + (n − 2)z] ≥ n − 1,

then E(x, y, z) ≥ 0, where


Æ x + y + (n − 2)z
E(x, y, z) = x 2 + y 2 + (n − 2)z 2 −
p
n
v
t1 1 n−2 1 1 1 n−2
 ‹
− + + 2 +p + + .
x2 y2 z n x y z
We will show that
E(x, y, z) ≥ E(x, w, w) ≥ 0,
where
y + (n − 2)z
w= , x ≤ y ≤ w ≤ z.
n−1
Write the inequality E(x, y, z) ≥ E(x, w, w) as follows:

y 2 + (n − 2)z 2 − (n − 1)w2 1 1 n−2 n−1


 ‹
p + p + −
x 2 + y 2 + (n − 2)z 2 + x 2 + (n − 1)w2
p
n y z w
482 Vasile Cîrtoaje

1
y2 + n−2
z2 + n−1
w2
≥Ç q ,
1 1 1
x2 + y2 + n−2
z2 + x2 + n−1
w2

(n − 2)( y − z)2 1 (n − 2)( y − z)2


·p + p
n−1 n yz[ y + (n − 2)z]
p
x 2 + y 2 + (n − 2)z 2 + x 2 + (n − 1)w2
(n − 2)( y − z)2 [ y 2 + 2(n − 1) yz + (n − 2)z 2 ] 1
≥ ·Ç ,
y 2 z 2 [ y + (n − 2)z]2
q
1 1 1
x2 + + + +
n−2 n−1
y2 z2 x2 w2

which is true if
1 1 1
·p +p
n−1 n yz[ y + (n − 2)z]
p
x 2 + y 2 + (n − 2)z 2 + x 2 + (n − 1)w2

y 2 + 2(n − 1) yz + (n − 2)z 2 1
≥ ·Ç .
y 2 z 2 [ y + (n − 2)z]2
q
1 1 1
x2 + + + +
n−2 n−1
y2 z2 x2 w2

Since x ≤ y, it is enough to show that


1 1 1
·p +p
n−1 2 y 2 + (n − 2)z 2 + y 2 + (n − 1)w2 n yz[ y + (n − 2)z]
p

y 2 + 2(n − 1) yz + (n − 2)z 2 1
≥ ·Ç .
y 2 z 2 [ y + (n − 2)z]2
Ç
2 1
y2 + n−2
z2 + y2 + n−1
w2

In addition, since w ≤ z, it suffices to show that


1 1 1
·p +p
n−1 2 y 2 + (n − 2)z 2 + y 2 + (n − 1)z 2 n yz[ y + (n − 2)z]
p

y 2 + 2(n − 1) yz + (n − 2)z 2 1
≥ ·Ç .
y 2 z 2 [ y + (n − 2)z]2
Ç
2 1
y2 + n−2
z2 + y2 + n−1
z2

Since
y 2 + 2(n − 1) yz + (n − 2)z 2 = [ y 2 + (n − 2)z 2 ] + 2(n − 1) yz,
we rewrite the inequality as
A + B ≥ C + D,
where
1 1
A= ·p ,
n−1 2 y 2 + (n − 2)z 2 + y 2 + (n − 1)z 2
p

1
B=p ,
n yz[ y + (n − 2)z]
y 2 + (n − 2)z 2 1
C= ·Ç ,
y 2 z 2 [ y + (n − 2)z]2
Ç
2 1
y2 + n−2
z2 + y2 + n−1
z2
EV Method for Nonnegative Variables 483

2(n − 1) yz 1
D= ·Ç .
y 2 z 2 [ y + (n − 2)z]2
Ç
2 1
y2 + n−2
z2 + y2 + n−1
z2

We will show that


A ≥ C, B ≥ D.
Since the inequality B ≥ D is homogeneous, we may consider y = 1 and z ≥ 1,
when it becomes
”p p — p
[(n − 2)z + 1] 2z 2 + n − 2 + z 2 + n − 1 ≥ 2 n(n − 1)z .

Since
p p 2z + n − 2 z + n − 1 3z + 2n − 3
2z 2 + n − 2 + z2 + n − 1 ≥ p + p = p ,
n n n
it is sufficient to show that

[(n − 2)z + 1](3z + 2n − 3) ≥ 2n(n − 1),

which is equivalent to

(z − 1)[3(n − 2)z + 2n2 − 4n + 3] ≥ 0.

To show that A ≥ C, we see that x n−1 [ y + (n − 2)z] ≥ n − 1 yields

y n−1 [ y + (n − 2)z] ≥ n − 1.

Thus, it suffices to prove the homogeneous inequality


2/n
y n−1 [ y + (n − 2)z]

A ≥ C0 C, C0 = ,
n−1

that is
1

2 y 2 + (n − 2)z 2 + y 2 + (n − 1)z 2
p p

(n − 1)[ y 2 + (n − 2)z 2 ] C0
≥ ·Ç ,
y 2 z 2 [ y + (n − 2)z]
Ç
2 1
y2 + n−2
z2 + y2 + n−1
z2

Due to homogeneity, we may set y = 1, hence z ≥ 1. The inequality becomes


p p
2z 2 + n − 2 + z 2 + n − 1 ≥

(n − 1)[1 + (n − 2)z 2 ]C1 ”Æ Æ —


≥ 2 + (n − 2)z 2+ 1 + (n − 1)z 2 ,
z[1 + (n − 2)z]2
where ˜2/n
1 + (n − 2)z
•
C1 = .
n−1
484 Vasile Cîrtoaje

By Bernoulli’s inequality, we have


˜2/n
(n − 2)(z − 1) 2(n − 2)(z − 1) 2(n − 2)z + n2 − 3n + 4)
•
C1 = 1 + ≤1+ = .
n−1 n(n − 1) n(n − 1)

Thus, it suffices to show that


p p
2z 2 + n − 2 + z 2 + n − 1 ≥

[1 + (n − 2)z 2 ][2(n − 2)z + n2 − 3n + 4] ”Æ Æ —


≥ 2 + (n − 2)z 2+ 1 + (n − 1)z 2 .
nz[1 + (n − 2)z]2
We will show that
p [1 + (n − 2)z 2 ][2(n − 2)z + n2 − 3n + 4] Æ
2z 2 + n − 2 ≥ (n − 1)z 2 + 1
nz[1 + (n − 2)z]2

and
p [1 + (n − 2)z 2 ][2(n − 2)z + n2 − 3n + 4] Æ
z2 + n − 1 ≥ (n − 2)z 2 + 2 .
nz[1 + (n − 2)z]2

Since

2z 2 + n − 2 z2 + n − 1 (n − 3)(z 2 − 1)2
− = ≥0,
(n − 1)z 2 + 1 (n − 2)z 2 + 2 [n − 1)z 2 + 1][(n − 2)z 2 + 2]

it suffices to prove the second inequality. After squaring and making many cal-
culations, this inequality can be written as (z − 1)P(z) ≥ 0, where P(z) ≥ 0 for
z ≥ 1.
To complete the proof, we need to show that E(x, w, w) ≥ 0 for x n−1 w ≥ 1.
Write the required inequality as follows:
v •
1 n−1 1 n−1
Æ t ˜  ‹
n[x + (n − 1)w ] − [x + (n − 1)w] ≥ n 2 +
2 2 − + ,
x w2 x w

(n − 1)(x − w)2 1 (n − 1)(x − w)2


p x+(n−1)w
≥ ·p (n−1)x+w
.
x 2 + (n − 1)w2 + p
n
xw (n − 1)x 2 + w2 + p n

This is true if
(n − 1)x + w 1 x + (n − 1)w
Æ •Æ ˜
(n − 1)x 2 + w2 + p ≥ · x + (n − 1)w +
2 2 p .
n xw n

Since x n−1 w ≥ 1, it suffices to prove the homogeneous inequality

(n − 1)x + w (x n−1 w)2/n Æ 2 x + (n − 1)w


Æ • ˜
(n − 1)x 2 + w2 + p ≥ · x + (n − 1)w +
2 p .
n xw n
EV Method for Nonnegative Variables 485

Due to homogeneity, we may set w = 1, which yields x ≤ 1. The inequality becomes

(n − 1)x + 1 + − 1
•p ˜
Æ n−2 x n
(n − 1)x 2 + 1 + p ≥x n x2 + n − 1 + p .
n n

We can get this by summing the inequalities


n−2
Æ p
(n − 1)x 2 + 1 ≥ x n · x2 + n − 1

and
(n − 1)x + 1 n−2 x +n−1
p ≥x n · p .
n n
Replacing x with x 2 in the second inequality gives the first inequality. Thus,it suf-
fices to prove the second inequality, which can be rewritten as f (x) ≥ 0, where

n−2
f (x) = ln[(n − 1)x + 1] − ln(x + n − 1) − ln x .
n
From

n−1 1 n−2 −(n − 1)(n − 2)(x − 1)2


f 0 (x) = − − = ≤0,
(n − 1)x + 1 x + n − 1 nx nx[(n − 1)x + 1] x + n − 1)

it follows that f is decreasing, hence f (x) ≥ f (1) = 0.


The proof is completed. The equality holds for a1 = a2 = · · · = an ≥ 1.
Remark. The inequality

1 1 1
 ‹
F (a1 , a2 , . . . , an ) ≥ F , ,...,
a1 a2 an

is also valid in the particular case

a1 , a2 , . . . , an ≥ 1.

P 5.102. If a1 , a2 , . . . , an (n ≥ 4) are positive real numbers such that

a1 + a2 + · · · + an = n, an = max{a1 , a2 , . . . , an },

then
1 1 1
 ‹
n + + ··· + ≥ 4(a12 + a22 + · · · + an2 ) + n(n − 5).
a1 a2 an−1
(Vasile C., 2021)
486 Vasile Cîrtoaje

Solution. Assume that an is fixed and a1 ≤ a2 ≤ · · · ≤ an . According to Corollary 5


(case k = 2 and m = −1), we have:
• If a1 , a2 , . . . , an−1 are positive real numbers so that

a1 +a2 +· · ·+an−1 = const ant, a12 +a22 +· · ·+an−1


2
= const ant, a1 ≤ a2 ≤ · · · ≤ an−1 ,

1 1 1
then the sum + + ··· + is minimal for a1 = a2 = · · · = an−2 ≤ an−1 .
a1 a2 an−1
Therefore, it suffices to consider the case a1 = a2 = · · · = an−2 , that is to show that
F (a, b) ≥ 0, where
n−2 1
 ‹
F (a, b) = n + − 4(n − 2)a2 − 4b2 − 4c 2 − n(n − 5), c = n − (n − 2)a − b,
a b
with a, b positive real numbers such that a ≤ b ≤ c. From c ≥ b, we get

(n − 2)a + 2b ≤ n.

We will show that


F (a, b) ≥ F (t, t) ≥ 0,
where
(n − 2)a + b
t= , t ≤ 1.
n−1
Since
n−2 1 n−1
 ‹  
F (a, b) − F (t, t) = n + − − 4 (n − 2)a2 + b2 − (n − 1)t 2
a b t

n(n − 2)(a − b)2 4(n − 2)(a − b)2


= −
(n − 1)abt n−1
n(n − 2)(a − b)2 4(n − 2)(a − b)2
≥ −
(n − 1)ab n−1
(n − 2)(a − b)2 (n − 4ab)
= ,
(n − 1)ab
it suffices to show that 4ab ≤ n. From
Æ
n ≥ (n − 2)a + 2b ≥ 2 2(n − 2)ab,

we get
n2 n(4 − n)
4ab − n ≤ −n= ≤ 0.
2(n − 2) n−2
In addition,
n(n − 1)
F (t, t) = − 4(n − 1)t 2 − 4[n − (n − 1)t]2 − n(n − 5)
t
EV Method for Nonnegative Variables 487

n(n − 1)(1 − t)(1 − 2t)2


= ≥ 0.
t
The equality occurs for a1 = a2 = · · · = an = 1, and also for
1 n+1
a1 = a2 = · · · = an−1 = , an = .
2 2

P 5.103. If a, b, c are nonnegative real numbers so that ab + bc + ca = 3, then


1 1 1
+ + ≤ 1.
a+ b+1 b+c+1 c+a+1
(Vasile C., 2021)
Solution. Using the substitution
m = a + b + c + 1,
we have to show that
f (a) + f (b) + f (c) ≤ 1
for
a + b + c = m − 1, a2 + b2 + c 2 = (m − 1)2 − 6,
1
f (u) = , 0 ≤ u < m − 1.
m−u
From
1 6
g(x) = f 0 (x) = , g 00
(x) = ,
(m − u)2 (m − u)4
it follows that g 00 (x) > 0, hence g is strictly convex. For fixed m, by Corollary 1, if
a + b + c = f i x ed, a2 + b2 + c 2 = f i x ed,
then the sum
S3 = f (a) + f (b) + f (c)
is maximal for a = b ≤ c. Thus, we only need to prove the inequality for a = b ≤ c;
that is, to show that a2 + 2ac = 3 involves
2 1
+ ≤ 1.
a + c + 1 2a + 1
Write this inequality as follows
4a 1
+ ≤ 1,
a2 + 2a + 3 2a + 1
a(a − 1)2 ≥ 0.
The equality holds for a = b = c = 1.
488 Vasile Cîrtoaje
Chapter 6

EV Method for Real Variables

6.1 Theoretical Basis


The Equal Variables Method may be extended to solve some difficult symmetric
inequalities in real variables.

EV-Theorem (Vasile Cirtoaje, 2010). Let a1 , a2 , . . . , an (n ≥ 3) be fixed real numbers,


and let
x1 ≤ x2 ≤ · · · ≤ x n
so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,

where k is an even positive integer. If f is a differentiable function on R so that the


joined function g : R → R defined by
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 )

is minimum for x 2 = x 3 = · · · = x n , and is maximum for x 1 = x 2 = · · · = x n−1 .


To prove this theorem, we will use EV-Lemma and EV-Proposition below.

EV-Lemma. Let a, b, c be fixed real numbers, not all equal, and let x, y, z be real
numbers satisfying

x ≤ y ≤ z, x + y + z = a + b + c, x k + y k + z k = ak + bk + c k ,

where k is an even positive integer. Then, there exist two real numbers m and M so
that m < M and
(1) y ∈ [m, M ];

489
490 Vasile Cîrtoaje

(2) y = m if and only if x = y;


(3) y = M if and only if y = z.

Proof. We show first, by contradiction method, that x < z. Indeed, if x = z, then


 x + y + z k
x = z ⇒ x = y = z ⇒ x k + y k + zk = 3
3
+ +
 ‹k
a b c
⇒ ak + bk + c k = 3 ⇒ a = b = c,
3
which is false. Notice that the last implication follows from Jensen’s inequality

a+b+c k
 ‹
a +b +c ≥3
k k k
,
3
with equality if and only if a = b = c.
According to the relations

x + z = a + b + c − y, x k + z k = ak + bk + c k − y k ,

we may consider x and z as functions of y. From

x 0 + z 0 = −1, x k−1 x 0 + z k−1 z 0 = − y k−1 ,

we get
y k−1 − z k−1 y k−1 − x k−1
x0 = , z 0
= . (*)
z k−1 − x k−1 x k−1 − z k−1
The two-sided inequality
x( y) ≤ y ≤ z( y)
is equivalent to the inequalities f1 ( y) ≤ 0 and f2 ( y) ≥ 0, where

f1 ( y) = x( y) − y, f2 ( y) = z( y) − y.

Using (*), we get


y k−1 − z k−1
f10 ( y) = −1
z k−1 − x k−1
and
y k−1 − x k−1
f20 ( y) = − 1.
x k−1 − z k−1
Since f10 ( y) ≤ −1 and f20 ( y) ≤ −1, f1 and f2 are strictly decreasing. Thus, the
inequality f1 ( y) ≤ 0 involves y ≥ m, where m is the root of the equation x( y) = y,
while the inequality f2 ( y) ≥ 0 involves y ≤ M , where M is the root of the equation
z( y) = y. Moreover, y = m if and only if x = y, and y = M if and only if y = z.

EV-Proposition. Let a, b, c be fixed real numbers, and let x, y, z be real numbers


satisfying

x ≤ y ≤ z, x + y + z = a + b + c, x k + y k + z k = ak + bk + c k ,
EV Method for Real Variables 491

where k is an even positive integer. If f is a differentiable function on R so that the


joined function g : R → R defined by
p 
g(x) = f 0
k−1
x

is strictly convex on R, then the sum

S = f (x) + f ( y) + f (z)

is minimum if and only if y = z, and is maximum if and only if x = y.

Proof. If a = b = c, then

a+b+c k
‹ 
a= b=c ⇒ a +b +c =3 k k k
3
 x + y + z k
⇒ x k + y k + zk = 3 ⇒ x = y = z.
3
Consider further that a, b, c are not all equal. As it is shown in the proof of EV-
Lemma, we have x < z. According to the relations

x + z = a + b + c − y, x k + z k = ak + bk + c k − y k ,

we may consider x and z as functions of y. Thus, we have

S = f (x( y)) + f ( y) + f (z( y)) := F ( y).

According to EV-Lemma, it suffices to show that F is maximum for y = m and is


minimum for y = M . Using (*), we have

F 0 ( y) = x 0 f 0 (x) + f 0 ( y) + z 0 f 0 (z)
y k−1 − z k−1 y k−1 − x k−1
= k−1 g(x ) + g( y ) + k−1
k−1 k−1
g(z k−1 ),
z −x k−1 x −z k−1

which, for x < y < z, is equivalent to

F 0 ( y) g(x k−1 )
=
( y k−1 − x k−1 )( y k−1 − z k−1 ) (x k−1 − y k−1 )(x k−1 − z k−1 )
g( y k−1 ) g(z k−1 )
+ k−1 + .
(y − z k−1 )( y k−1 − x k−1 ) (z k−1 − x k−1 )(z k−1 − y k−1 )

Since g is strictly convex, the right hand side is positive. Moreover, since

( y k−1 − x k−1 )( y k−1 − z k−1 ) < 0,

we have F 0 ( y) < 0 for y ∈ (m, M ), hence F is strictly decreasing on [m, M ]. There-


fore, F is maximum for y = m and is minimum for y = M .
492 Vasile Cîrtoaje

Proof of EV-Theorem. For n = 3, EV-Theorem follows immediately from EV-


Proposition. Consider next that n ≥ 4. Since X = (x 1 , x 2 , . . . , x n ) is defined in
EV-Theorem as a compact set in Rn , Sn attains its minimum and maximum values.
Using this property and EV-Proposition, we can prove EV-Theorem via contradic-
tion. Thus, for the sake of contradiction, assume that Sn attains its maximum at
(b1 , b2 , . . . , bn ), where b1 ≤ b2 ≤ · · · ≤ bn and b1 < bn−1 . Let x 1 , x n−1 and x n be
real numbers so that

x 1 ≤ x n−1 ≤ x n , x 1 + x n−1 + x n = b1 + bn−1 + bn , x 1k + x n−1


k
+ x nk = b1k + bn−1
k
+ bnk .

According to EV-Proposition, the sum f (x 1 ) + f (x n−1 ) + f (x n ) is maximum for x 1 =


x n−1 , when

f (x 1 ) + f (x n−1 ) + f (x n ) > f (b1 ) + f (bn−1 ) + f (bn ).

This result contradicts the assumption that Sn attains its maximum value at (b1 , b2 , . . . , bn )
with b1 < bn−1 . Similarly, we can prove that Sn is minimum for x 2 = x 3 = · · · = x n .
Taking k = 2 in EV-Theorem, we obtain the following corollary.
Corollary 1. Let a1 , a2 , . . . , an (n ≥ 3) be fixed real numbers, and let x 1 , x 2 , . . . , x n
be real variables so that
x1 ≤ x2 ≤ · · · ≤ x n,
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an ,
x 12 + x 22 + · · · + x n2 = a12 + a22 + · · · + an2 .
If f is a differentiable function on R so that the derivative f 0 is strictly convex on R,
then the sum
Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )
is minimum for x 2 = x 3 = · · · = x n , and is maximum for x 1 = x 2 = · · · = x n−1 .

Corollary 2. Let a1 , a2 , . . . , an (n ≥ 3) be fixed real numbers, and let x 1 , x 2 , . . . , x n


be real variables so that
x1 ≤ x2 ≤ · · · ≤ x n,
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an ,
x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,
where k is an even positive integer. For any positive odd number m, m > k, the power
sum
Sn = x 1m + x 2m + · · · + x nm
is minimum for x 2 = x 3 = · · · = x n , and is maximum for x 1 = x 2 = · · · = x n−1 .
Proof. We apply the EV-Theorem the function f (u) = um . The joined function
p  p
k−1
g(x) = f 0 x =m
k−1
x m−1
EV Method for Real Variables 493

is strictly convex on R because its derivative

m(m − 1) k−1
p
g 0 (x) = x m−k
k−1
is strictly increasing on R.

Theorem 1. Let a1 , a2 , . . . , an (n ≥ 3) be fixed real numbers, and let x 1 , x 2 , . . . , x n be


real variables so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + a n ,

x 12 + x 22 + · · · + x n2 = a12 + a22 + · · · + an2 .


The power sum
Sn = x 14 + x 24 + · · · + x n4
is minimum and maximum when the set (x 1 , x 2 , . . . , x n ) has at most two distinct val-
ues.
To prove this theorem, we will use Proposition 1 below.

Proposition 1. Let a, b, c be fixed real numbers, and let x, y, z be real numbers so


that
x + y + z = a + b + c, x 2 + y 2 + z 2 = a2 + b2 + c 2 .
The power sum
S = x 4 + y 4 + z4
is minimum and maximum when two of x, y, z are equal

Proof. The proof is based on EV-Lemma. Without loss of generality, assume that
x ≤ y ≤ z. For the nontrivial case when a, b, c are not all equal (which involves
x < z), consider the function of y

F ( y) = x 4 ( y) + y 4 + z 4 ( y).

According to (*), we have


y −z y−x
+ 4 y 3 + 4z 3
F 0 ( y) = 4x 3 x 0 + 4 y 3 + 4z 3 z 0 = 4x 3
z−x x −z
= 4(x + y + z)( y − x)( y − z) = 4(a + b + c)( y − x)( y − z).

There are three cases to consider.


Case 1: a + b + c < 0. Since F 0 ( y) > 0 for x < y < z, F is strictly increasing on
[m, M ].
Case 2: a + b + c > 0. Since F 0 ( y) < 0 for x < y < z, F is strictly decreasing on
[m, M ].
Case 3: a + b + c = 0. Since F 0 ( y) = 0, F is constant on [m, M ].
494 Vasile Cîrtoaje

In all cases, F is monotonic on m, M ]. Therefore, F is minimum and maximum for


y = m or y = M ; that is, when x = y or y = z (see EV-Lemma). Notice that for
a+ b+c 6= 0, F is strictly monotonic on [m, M ], hence F is minimum and maximum
if and only if y = m or y = M ; that is, if and only if x = y or y = z.

Proof of Theorem 1. For n = 3, Theorem 1 follows from Proposition 1. In order


to prove Theorem 1 for any n ≥ 4, we will use the contradiction method. For the
sake of contradiction, assume that (b1 , b2 , . . . , bn ) is an extreme point having at
least three distinct components; let us say b1 < b2 < b3 . Let x 1 , x 2 and x 3 be real
numbers so that

x1 ≤ x2 ≤ x3, x 1 + x 2 + x 3 = b1 + b2 + b3 x 12 + x 22 + x 32 = b12 + b22 + b32 .

We need to consider two cases.


Case 1: b1 + b2 + b3 6= 0. According to Proposition 1, the sum x 14 + x 24 + x 34 is
extreme only when two of x 1 , x 2 , x 3 are equal, which contradicts the assumption
that the sum x 14 + x 24 + · · · + x n4 attains its extreme value at (b1 , b2 , . . . , bn ) with
b1 < b2 < b3 .
Case 2: b1 + b2 + b3 = 0. There exist three real numbers x 1 , x 2 , x 3 so that x 1 = x 2
and
x 1 + x 2 + x 3 = b1 + b2 + b3 = 0, x 12 + x 22 + x 32 = b12 + b22 + b32 .
Letting x 1 = x 2 := x and x 3 := y, we have 2x + y = 0, x 6= y. According to
Proposition 1, the sum x 14 + x 24 + x 34 is constant (equal to b14 + b24 + b34 ). Thus,
(x, x, y, b4 , . . . , bn ) is also an extreme point. According to our hypothesis, this ex-
treme point has at least three distinct components. Therefore, among the numbers
b4 , . . . , bn there is one, let us say b4 , so that x, y and b4 are distinct. Since

x + y + b4 = −x + b4 6= 0,

we have a case similar to Case 1, which leads to a contradiction.

Theorem 2. Let a1 , a2 , . . . , an (n ≥ 3) be fixed real numbers, and let x 1 , x 2 , . . . , x n be


real variables so that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an ,

x 12 + x 22 + · · · + x n2 = a12 + a22 + · · · + an2 .


For m ∈ {6, 8}, the power sum

Sn = x 1m + x 2m + · · · + x nm

is maximum when the set (x 1 , x 2 , . . . , x n ) has at most two distinct values.

Theorem 2 can be proved using Proposition 2 below, in a similar way as the


EV-Theorem.
EV Method for Real Variables 495

Proposition 2. Let a, b, c be fixed real numbers, let x, y, z be real numbers so that

x + y + z = a + b + c, x 2 + y 2 + z 2 = a2 + b2 + c 2 .

For m ∈ {6, 8}, the power sum

Sm = x m + y m + z m

is maximum if and only if two of x, y, z are equal.

Proof. Consider the nontrivial case where a, b, c are not all equal. Let

p = a + b + c, q = ab + bc + ca, r = x yz.

Since x + y + z = p and x y + yz + z x = q, from

(x − y)2 ( y − z)2 (z − x)2 ≥ 0,

which is equivalent to

27r 2 + 2(2p3 − 9pq)r − p2 q2 + 4q3 ≤ 0,

we get r ∈ [r1 , r2 ], where

9pq − 2p3 + 2(p2 − 3q) p2 − 3q


p p
9pq − 2p3 − 2(p2 − 3q) p2 − 3q
r1 = , r2 = .
27 27
From
−27(r − r1 )(r − r2 ) = (x − y)2 ( y − z)2 (z − x)2 ≥ 0,
it follows that the product r = x yz attains its minimum value r1 and its maximum
value r2 only when two of x, y, z are equal. For fixed p and q, we have

S6 = 3r 2 + f6 (p, q)r + h6 (p, q) := g6 (r),

S8 = 4(3p2 − 2q)r 2 + f8 (p, q)r + h8 (p, q) := g8 (r).


Since
7 2 2 2
3p2 − 2q = p + (p − 3q) > 0,
3 3
the functions g6 and g8 are strictly convex, hence are maximum only for r = r1 or
r = r2 ; that is, only when two of x, y, z are equal.

Open problem. Theorem 2 is valid for any integer number m ≥ 3.

Note. The EV-Theorem for real variables and Corollary 1 are also valid under the
conditions in Note 2 and Note 3 from the preceding chapter 5, where m, M ∈ R.
496 Vasile Cîrtoaje
EV Method for Real Variables 497

6.2 Applications

6.1. If a, b, c, d are real numbers so that a + b + c + d = 4, then


‹2
8 64
  ‹
a +b +c +d +
2 2 2 2
≥4 a +b +c +d +
3 3
.3 3
3 9

6.2. If a, b, c, d are real numbers so that a + b + c + d = 4, then

76
 ‹
(a + b + c + d − 4) a + b + c + d +
2 2 2 2 2 2 2 2
≥ 8(a3 + b3 + c 3 + d 3 − 4).
3

6.3. If a, b, c are real numbers so that a + b + c = 3, then

(a2 + b2 + c 2 − 3)(a2 + b2 + c 2 + 93) ≥ 24(a3 + b3 + c 3 − 3).

6.4. If a, b, c, d are real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 116) ≥ 24(a3 + b3 + c 3 + d 3 − 4).

6.5. Let a, b, c, d be real numbers so that a + b + c + d = 4, and let

E = a2 + b2 + c 2 + d 2 − 4, F = a3 + b3 + c 3 + d 3 − 4.

Prove that ‚v Œ
tE
E + 3 ≥ F.
3

6.6. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1).

If m is an odd number (m ≥ 3), then

n − 1 − (n − 1)m ≤ a1m + a2m + · · · + anm ≤ (n − 1)m − n + 1.


498 Vasile Cîrtoaje

6.7. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1.

If m is an odd number (m ≥ 3), then

2 m 2 m
 ‹  ‹
(n − 1) 1 + − n− ≤ a1m + a2m + · · · + anm ≤ nm − n + 1.
n n

6.8. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 − 3n + 3.

If m is an odd number (m ≥ 3), then


‹m
2 2 m
  ‹
n − 1 − (n − 2) ≤
m
a1m + a2m + ··· + anm ≤ n−2+ − (n − 1) 1 − .
n n

6.9. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = a12 + a22 + · · · + an2 = n − 1.

If m is an odd number (m ≥ 3), then

2 m 2 m
 ‹  ‹
n−1≤ a1m + a2m + ··· + anm ≤ (n − 1) 1 − + 2− .
n n

6.10. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = n + 3.

If m is an odd number (m ≥ 3), then


 ‹m
2 2 m
 ‹
+ (n − 1) 1 + ≤ a1m + a2m + · · · + anm ≤ 2m + n − 1.
n n

6.11. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = a14 + a24 + · · · + an4 = n − 1,

then
a15 + a25 + · · · + an5 ≥ n − 1.
EV Method for Real Variables 499

6.12. If a, b, c are real numbers so that a2 + b2 + c 2 = 3, then

a3 + b3 + c 3 + 3 ≥ 2(a + b + c).

6.13. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a14 + a24 + · · · + an4 ≤ n(n − 1)(n2 − 3n + 3).

6.14. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = 4n2 + n − 1,

then
a14 + a24 + · · · + an4 ≤ 16n4 + n − 1.

6.15. If n is an odd number and a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n2 − 1),

then
a14 + a24 + · · · + an4 ≥ n(n2 − 1)(n2 + 3).

6.16. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − n − 1, a12 + a22 + · · · + an2 = n3 + 2n2 − n − 1,

then
a14 + a24 + · · · + an4 ≥ n4 + (n − 1)(n + 1)4 .

6.17. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − 2n − 1, a12 + a22 + · · · + an2 = n3 + 2n + 1,

then
a14 + a24 + · · · + an4 ≥ (n + 1)4 + (n − 1)n4 .
500 Vasile Cîrtoaje

6.18. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − 3n − 2, a12 + a22 + · · · + an2 = n3 + 2n2 − 3n − 2,

then
a14 + a24 + · · · + an4 ≥ 2n4 + (n − 2)(n + 1)4 .

6.19. If a, b, c, d are real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 36) ≤ 12(a4 + b4 + c 4 + d 4 − 4).

6.20. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a16 + a26 + · · · + an6 ≤ (n − 1)6 + n − 1.

6.21. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1,

then
a16 + a26 + · · · + an6 ≤ n6 + n − 1.

6.22. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a18 + a28 + · · · + an8 ≤ (n − 1)8 + n − 1.

6.23. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1,

then
a18 + a28 + · · · + an8 ≤ n8 + n − 1.
EV Method for Real Variables 501

6.24. Let a1 , a2 , . . . , an (n ≥ 2) be real numbers (not all equal), and let


a1 + a2 + · · · + an a12 + a22 + · · · + an2 a13 + a23 + · · · + an3
A= , B= , C= .
n n n
Then, ‚ v Œ ‚ v Œ
1 t 2n2 B 2 − AC 1 t 2n2
1− 1+ ≤ 2 ≤ 1+ 1+ .
4 n−1 B − A4 4 n−1

6.25. If a, b, c, d are real numbers so that


a + b + c + d = 2,
then
3
a4 + b4 + c 4 + d 4 ≤ 40 + (a2 + b2 + c 2 + d 2 )2 .
4

6.26. If a, b, c, d, e are real numbers, then


p
31 + 18 3 3
a +b +c +d +e ≤
4 4 4 4 4
(a + b + c + d + e)4 + (a2 + b2 + c 2 + d 2 + e2 )2 .
8 4

−5
6.27. Let a, b, c, d, e 6= be real numbers so that a + b + c + d + e = 5. Then,
4
a(a − 1) b(b − 1) c(c − 1) d(d − 1) e(e − 1)
+ + + + ≥ 0.
(4a + 5)2 (4b + 5)2 (4c + 5)2 (4d + 5)2 (4e + 5)2

6.28. If a, b, c are real numbers so that


a + b + c = 9, ab + bc + ca = 15,
then
19 1 1 1 7
≤ 2 + 2 + 2 ≤ .
175 b + bc + c 2 c + ca + a 2 a + ab + b 2 19

6.29. If a, b, c are real numbers so that


8(a2 + b2 + c 2 ) = 9(ab + bc + ca),
then
419 a2 b2 c2 311
≤ 2 + + ≤ .
175 b + bc + c 2 c + ca + a
2 2 a + ab + b
2 2 19

6.30. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 10,


then
2(a12 + a22 + · · · + an2 )2 − n(a13 + a23 + · · · + an3 ) ≥ n2 .
502 Vasile Cîrtoaje
EV Method for Real Variables 503

6.3 Solutions

P 6.1. If a, b, c, d are real numbers so that a + b + c + d = 4, then


8 2 64
 ‹  ‹
a +b +c +d +
2 2 2 2
≥4 a +b +c +d +
3 3 3 3
.
3 9
(Vasile Cîrtoaje, 2010)
Solution. Apply Corollary 2 for n = 4, k = 2, m = 3:
• If a, b, c, d are real numbers so that a ≤ b ≤ c ≤ d and

a + b + c + d = 4, a2 + b2 + c 2 + d 2 = const ant,

then
S4 = a3 + b3 + c 3 + d 3
is maximum for a = b = c ≤ d.
Thus, we only need to show that 3a + d = 4 involves
8 2 64
 ‹  ‹
3a + d +
2 2
≥ 4 3a + d +
3 3
.
3 9
This inequality is equivalent to

(a − 1)2 (3a − 2)2 ≥ 0.

The equality holds for a = b = c = d = 1, and also for


2
a=b=c= , d =2
3
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n,

then
‹2  n4 (n2 + 16n − 16)
n3

a12 + a22 + ··· + an2 + ≥ n a13 + a23 + · · · + an3 + ,
8n − 8 64(n − 1)2
with equality for a1 = a2 = · · · = an = 1, and also for
n n
a1 = a2 = · · · = an−1 = , an =
2n − 2 2
(or any cyclic permutation).
504 Vasile Cîrtoaje

P 6.2. If a, b, c, d are real numbers so that a + b + c + d = 4, then


76
 ‹
(a + b + c + d − 4) a + b + c + d +
2 2 2 2 2 2 2 2
≥ 8(a3 + b3 + c 3 + d 3 − 4).
3

(Vasile Cîrtoaje, 2010)

Solution. As shown in the preceding P 6.1, we only need to show that

3a + d = 4

involves
76
 ‹
(3a + d − 4) 3a + d +
2 2 2
2
≥ 8(3a3 + d 3 − 4).
3
This inequality is equivalent to

(a − 1)2 (3a − 1)2 ≥ 0.

The equality holds for a = b = c = d = 1, and also for


1
a=b=c= , d =3
3
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n,

then
• 2 n(n2 + n − 1)
˜ 
a12 + ··· + an2 − n a1 + · · · + an +
2
≥ 2n a13 + · · · + an3 − n ,
n−1
with equality for a1 = a2 = · · · = an = 1, and also for
1
a1 = a2 = · · · = an−1 = , an = n − 1
n−1
(or any cyclic permutation).

P 6.3. If a, b, c are real numbers so that a + b + c = 3, then

(a2 + b2 + c 2 − 3)(a2 + b2 + c 2 + 93) ≥ 24(a3 + b3 + c 3 − 3).

(Vasile Cîrtoaje, 2010)


EV Method for Real Variables 505

Solution. As shown in the proof of P 6.1, we only need to show that

2a + c = 3

involves
(2a2 + c 2 − 3)(2a2 + c 2 + 93) ≥ 24(2a3 + c 3 − 3).
This inequality is equivalent to

(a2 − 1)2 ≥ 0.

The equality holds for a = b = c = 1, and also for

a = b = −1, c=5

(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• Let a, b, c be real numbers so that a + b + c = 3. For any real k, the following
inequality holds

(a2 + b2 + c 2 − 3)(a2 + b2 + c 2 + 6k2 + 36k − 3) ≥ 12k(a3 + b3 + c 3 − 3),

with equality for a = b = c = 1, and also for

a = b = 1 − k, c = 1 + 2k

(or any cyclic permutation).

P 6.4. If a, b, c, d are real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 116) ≥ 24(a3 + b3 + c 3 + d 3 − 4).

(Vasile Cîrtoaje, 2010)

Solution. As shown in the proof of P 6.1, we only need to show that

3a + d = 4

involves
(3a2 + d 2 − 4)(3a2 + d 2 + 116) ≥ 24(3a3 + d 3 − 4).
This inequality is equivalent to

(a2 − 1)2 ≥ 0.
506 Vasile Cîrtoaje

The equality holds for a = b = c = d = 1, and also for

a = b = c = −1, d =7

(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = n.

If k is a real number, then

k(a13 + · · · + an3 − n) a12 + · · · + an2 + n(n − 1)(n − 2)2 k2 + 6n(n − 1)k − n


≤ ,
a12 + · · · + an2 − n 2n(n − 1)

with equality for

a1 = · · · = an−1 = 1 − (n − 2)k, an = 1 + (n − 1)(n − 2)k

(or any cyclic permutation).


−6
For k = , we get the following nice inequality
n−2
2 12n(n − 1) 3 
a12 + a22 + · · · + an2 − n + a1 + a23 + · · · + an3 − n ≥ 0,
n−2
with equality for a1 = a2 = · · · = an = 1, and also for

a1 = · · · = an−1 = 7, an = 7 − 6n

(or any cyclic permutation).

P 6.5. Let a, b, c, d be real numbers so that a + b + c + d = 4, and let

E = a2 + b2 + c 2 + d 2 − 4, F = a3 + b3 + c 3 + d 3 − 4.

Prove that ‚v Œ
tE
E + 3 ≥ F.
3

(Vasile Cîrtoaje, 2016)


EV Method for Real Variables 507

Solution. As shown in the proof of P 6.1, we only need to prove the desired in-
equality for 3a + d = 4 and

E = 3a2 + d 2 − 4, F = 3a3 + d 3 − 4.

Since
E = 12(1 − a)2 , F = 12(5 − 2a)(1 − a)2 ,
we get
‚v Œ
tE
E + 3 − F = 12(1 − a)2 (2|1 − a| + 3) − 12(5 − 2a)(1 − a)2
3
= 24(1 − a)2 [|1 − a| − (1 − a)] ≥ 0.

The equality holds for


4−d
a=b=c= ≤1
3
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , an be real numbers so that a1 + a2 + · · · + an = n, and let

E = a12 + a22 + · · · + an2 − n, F = a13 + a23 + · · · + an3 − n.

Then,  v 
t E
E (n − 2) + 3 ≥ F,
n(n − 1)
with equality for
n − an
a1 = · · · = an−1 = ≤1
n−1
(or any cyclic permutation).

P 6.6. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1).

If m is an odd number (m ≥ 3), then

n − 1 − (n − 1)m ≤ a1m + a2m + · · · + anm ≤ (n − 1)m − n + 1.

(Vasile Cîrtoaje, 2010)


508 Vasile Cîrtoaje

Solution. Without loss of generality, assume that

a1 ≤ a2 ≤ · · · ≤ an .

(a) Consider the right inequality. For n = 2, we need to show that

a1 + a2 = 0, a12 + a22 = 2

implies
a1m + a2m ≤ 0.
We have
a1 = −1, a2 = 1,
therefore a1m + a2m = 0. Assume now that n ≥ 3. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is maximum for a1 = a2 = · · · = an−1 . Thus, we only need to show that

(n − 1)a + b = 0, (n − 1)a2 + b2 = n(n − 1), a≤b

involve
(n − 1)a m + b m ≤ (n − 1)m − n + 1.
From the equations above, we get

a = −1, b = n − 1;

therefore,

(n − 1)a m + b m = (n − 1)(−1)m + (n − 1)m = (n − 1)m − n + 1.

The equality holds for

a1 = · · · = an−1 = −1, an = n − 1

(or any cyclic permutation).


(b) The left inequality follows from the right inequality by replacing a1 , a2 , . . . , an
with −a1 , −a2 , . . . , −an , respectively. The equality holds for

a1 = −n + 1, a2 = a3 = · · · = an = 1

(or any cyclic permutation).


EV Method for Real Variables 509

P 6.7. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1.

If m is an odd number (m ≥ 3), then

2 m 2 m
 ‹  ‹
(n − 1) 1 + − n− ≤ a1m + a2m + · · · + anm ≤ nm − n + 1.
n n

(Vasile Cîrtoaje, 2010)

Solution. Without loss of generality, assume that

a1 ≤ a2 ≤ · · · ≤ an .

For n = 2, we need to show that

a1 + a2 = 1, a12 + a22 = 5,

implies
2m − 1 ≤ a1m + a2m ≤ 2m − 1.
We have
a1 = −1, a2 = 2,
for which a1m + a2m = 2m − 1. Assume now that n ≥ 3.
(a) Consider the right inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is maximum for a1 = a2 = · · · = an−1 . Thus, we only need to show that

(n − 1)a + b = 1, (n − 1)a2 + b2 = n2 + n − 1, a≤b

involve
(n − 1)a m + b m ≤ nm − n + 1.
From the equations above, we get

a = −1, b = n;

therefore,
(n − 1)a m + b m = (n − 1)(−1)m + nm = nm − n + 1.
The equality holds for

a1 = a2 = · · · = an−1 = −1, an = n

(or any cyclic permutation).


510 Vasile Cîrtoaje

(b) Consider the left inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is minimum for a2 = a3 = · · · = an . Thus, we only need to show that

a + (n − 1)b = 1, a2 + (n − 1)b2 = n2 + n − 1, a≤b

involve
2 m 2 m
 ‹  ‹
a + (n − 1)b ≥ (n − 1) 1 +
m m
− n− .
n n
From the equations above, we get
2 2
a = −n + , b =1+ ;
n n
therefore,
2 m 2 m
 ‹  ‹
a + (n − 1)b = −n +
m m
+ (n − 1) 1 +
n n
‹m  ‹m
2 2

= (n − 1) 1 + − n− .
n n
The equality holds for
2 2
a1 = −n + , a2 = a3 = · · · = an = 1 +
n n
(or any cyclic permutation).

P 6.8. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 − 3n + 3.

If m is an odd number (m ≥ 3), then


‹m
2 2 m
  ‹
n − 1 − (n − 2) ≤
m
a1m + a2m + ··· + anm ≤ n−2+ − (n − 1) 1 − .
n n

(Vasile Cîrtoaje, 2010)

Solution. Without loss of generality, assume that

a1 ≤ a2 ≤ · · · ≤ an .

For n = 2, we need to show that

a1 + a2 = 1, a12 + a22 = 1,
EV Method for Real Variables 511

implies
1 ≤ a1m + a2m ≤ 1.
We have
a1 = 0, a2 = 1,
when a1m + a2m = 1. Assume now that n ≥ 3.
(a) Consider the left inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is minimum for a2 = a3 = · · · = an . Thus, we only need to show that

a + (n − 1)b = 1, a2 + (n − 1)b2 = n2 − 3n + 3, a≤b

involve
a m + (n − 1)b m ≤ n − 1 − (n − 2)m .
From the equations above, we get

a = 2 − n, b = 1;

therefore,

a m + (n − 1)b m = (2 − n)m + n − 1 = n − 1 − (n − 2)m .

The equality holds for

a1 = 2 − n, a2 = a3 = · · · = an = 1

(or any cyclic permutation).


(b) Consider the right inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is maximum for a1 = a2 = · · · = an−1 . Thus, we only need to show that

(n − 1)a + b = 1, (n − 1)a2 + b2 = n2 − 3n + 3, a≤b

involve ‹m
2 2 m
  ‹
(n − 1)a + b ≤ n − 2 +
m m
− (n − 1) 1 − .
n n
From the equations above, we get

2 2
a = −1 + , b = n−2+ ;
n n
512 Vasile Cîrtoaje

therefore,
2 m 2 m
 ‹  ‹
(n − 1)a + b = (n − 1) −1 +
m m
+ n−2+
n n
‹m
2 2 m
  ‹
= n−2+ − (n − 1) 1 − .
n n
The equality holds for
2 2
a1 = · · · = an−1 = −1 + , an = n − 2 +
n n
(or any cyclic permutation).

P 6.9. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = a12 + a22 + · · · + an2 = n − 1.

If m is an odd number (m ≥ 3), then

2 m 2 m
 ‹  ‹
n−1≤ a1m + a2m + ··· + anm ≤ (n − 1) 1 − + 2− .
n n

(Vasile Cîrtoaje, 2010)

Solution. Without loss of generality, assume that

a1 ≤ a2 ≤ · · · ≤ an .

For n = 2, we need to show that

a1 + a2 = 1, a12 + a22 = 1,

implies
1 ≤ a1m + a2m ≤ 1.
The above equations involve

a1 = 0, a2 = 1,

hence a1m + a2m = 1. Assume now that n ≥ 3.


(a) Consider the left inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is minimum for a2 = a3 = · · · = an . Thus, we only need to show that

a + (n − 1)b = n − 1, a2 + (n − 1)b2 = n − 1, a≤b


EV Method for Real Variables 513

involve
a m + (n − 1)b m ≥ n − 1.
From the equations above, we get

a = 0, b = 1;

therefore,
a m + (n − 1)b m = n − 1.
The equality holds for
a1 = 0, a2 = · · · = a n = 1
(or any cyclic permutation).
(b) Consider the right inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is maximum for a1 = a2 = · · · = an−1 . Thus, we only need to show that

(n − 1)a + b = n − 1, (n − 1)a2 + b2 = n − 1, a≤b

involve
2 m 2 m
 ‹  ‹
(n − 1)a m + b m ≤ (n − 1) 1 − + 2− .
n n
From the equations above, we get
2 2
a =1− , b =2− ,
n n
when
2 m 2 m
 ‹  ‹
(n − 1)a + b = (n − 1) 1 −
m m
+ 2− .
n n

The equality holds for


2 2
a1 = a2 = · · · = an−1 = 1 − , an = 2 −
n n
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


• Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = k, a12 + a22 + · · · + an2 = n2 + (2k − 1)n + k(k − 2),

where k is a real number, k ≥ −n. If m is an odd number (m ≥ 3), then


‹m ‹m
2k 2k
 
+ 1 − n − k + (n − 1) + 1 ≤ a1m + a2m + · · · + anm ≤ (n + k − 1)m − n + 1.
n n
514 Vasile Cîrtoaje

The left inequality is an equality for


2k 2k
a1 = + 1 − n − k, a2 = · · · = an = +1
n n
(or any cyclic permutation). The right inequality is an equality for

a1 = · · · = an−1 = −1, an = n + k − 1

(or any cyclic permutation).


For k = 0 and k = 1, we get the inequalities in P 6.6 and P 6.7, respectively. For k =
−1 and k = −n+1, by replacing k with −k and a1 , a2 , . . . , an with −a1 , −a2 , . . . , −an ,
we get the inequalities in P 6.8 and P 6.9, respectively.

P 6.10. Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = n + 3.

If m is an odd number (m ≥ 3), then


 ‹m
2 2 m
 ‹
+ (n − 1) 1 + ≤ a1m + a2m + · · · + anm ≤ 2m + n − 1.
n n

(Vasile Cîrtoaje, 2010)

Solution. Without loss of generality, assume that

a1 ≤ a2 ≤ · · · ≤ an .

For n = 2, we need to show that

a1 + a2 = 3, a12 + a22 = 5,

implies
2m + 1 ≤ a1m + a2m ≤ 2m + 1.
We get
a1 = 1, a2 = 2,
when a1m + a2m = 2 + 1. Assume now that n ≥ 3.
m

(a) Consider the left inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is minimum for a2 = a3 = · · · = an . Thus, we only need to show that

a + (n − 1)b = n + 1, a2 + (n − 1)b2 = n + 3, a≤b


EV Method for Real Variables 515

involve  ‹m
2 2 m
 ‹
a + (n − 1)b ≥
m m
+ (n − 1) 1 + .
n n
From the equations

a + (n − 1)b = n + 1, a2 + (n − 1)b2 = n + 3,

we get
2 2
a= , b =1+ ;
n n
therefore,  ‹m
2 2 m
 ‹
a + (n − 1)b =
m m
+ (n − 1) 1 + .
n n
The equality holds for

2 2
a1 = , a2 = · · · = a n = 1 +
n n
(or any cyclic permutation).
(b) Consider the right inequality. According to Corollary 2, the sum

Sn = a1m + a2m + · · · + anm

is maximum for a1 = a2 = · · · = an−1 . Thus, we only need to show that

(n − 1)a + b = n + 1, (n − 1)a2 + b2 = n + 3, a≤b

involve
(n − 1)a m + b m ≤ 2m + n − 1.
From the equations

(n − 1)a + b = n + 1, (n − 1)a2 + b2 = n + 3,

we get
a = 1, b = 2;
therefore,
(n − 1)a m + b m = n − 1 + 2m .
The equality holds for

a1 = · · · = an−1 = 1, an = 2

(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization:


516 Vasile Cîrtoaje

• Let a1 , a2 , . . . , an be real numbers so that

a1 + a2 + · · · + an = k, a12 + a22 + · · · + an2 = n2 − (2k + 1)n + k(k + 2),

where k is a positive number, k > n. If m is an odd number (m ≥ 3), then


‹m ‹m
2k 2k
 
− 1 + n − k + (n − 1) − 1 ≤ a1m + a2m + · · · + anm ≤ (k − n + 1)m + n − 1.
n n

The left inequality is an equality for

2k 2k
a1 = − 1 + n − k, a2 = · · · = an = −1
n n
(or any cyclic permutation). The right inequality is an equality for

a1 = · · · = an−1 = 1, an = k − n + 1

(or any cyclic permutation).

For k = n + 1, we get the inequalities in P 6.10.

P 6.11. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = a14 + a24 + · · · + an4 = n − 1,

then
a15 + a25 + · · · + an5 ≥ n − 1.

(Vasile Cîrtoaje, 2010)

Solution. For n = 2, we need to show that

a1 + a2 = 1, a14 + a24 = 1,

implies
a15 + a25 ≥ 1.
We have
a1 = 0, a2 = 1,
or
a1 = 1, a2 = 0.
For each of these cases, the inequality is an equality. Assume now that n ≥ 3 and

a1 ≤ a2 ≤ · · · ≤ an .
EV Method for Real Variables 517

According to Corollary 2, the sum

Sn = a15 + a25 + · · · + an5

is minimum for a2 = a3 = · · · = an . Thus, we only need to show that

a + (n − 1)b = a4 + (n − 1)b4 = n − 1, a≤b

involve
a5 + (n − 1)b5 ≥ n − 1.
The equations

a + (n − 1)b = n − 1, a4 + (n − 1)b4 = n − 1,

are equivalent to

(1 − b)[(n − 1)3 (1 − b)3 − 1 − b − b2 − b3 ] = 0, a = (n − 1)(1 − b);

that is,
b = 1, a = 0,
and
a3 = 1 + b + b2 + b3 , a = (n − 1)(1 − b).
For the second case, the condition a ≤ b involves

b3 ≥ 1 + b + b2 + b3 ,

which is not possible. Therefore, it suffices to show that

a5 + (n − 1)b5 ≥ n − 1

for a = 0 and b = 1, that is clearly true. Thus, the proof is completed. The equality
holds for
a1 = 0, a2 = · · · = a n = 1
(or any cyclic permutation).

P 6.12. If a, b, c are real numbers so that

a2 + b2 + c 2 = 3,

then
a3 + b3 + c 3 + 3 ≥ 2(a + b + c).

(Vasile Cîrtoaje, 2010)


518 Vasile Cîrtoaje

Solution. Assume that


a ≤ b ≤ c.
According to Corollary 2, for a ≤ b ≤ c and

a + b + c = const ant, a2 + b2 + c 2 = 3,

the sum
S3 = a3 + b3 + c 3
is minimum for a ≤ b = c. Thus, we only need to show that

a2 + 2b2 = 3, a ≤ b,

involves
a3 + 2b3 + 3 ≥ 2(a + 2b).
We will show this by two methods. From a2 + 2b2 = 3 and a ≤ b, it follows that
v v
p t3 t3
− 3 ≤ a ≤ 1, − <b≤ .
2 2

Method 1. Write the desired inequality as

a3 + b(3 − a2 ) + 3 ≥ 2(a + 2b),

a3 − 2a + 3 ≥ b(a2 + 1).
For a ≥ 0, we have
a3 − 2a + 3 ≥ −2a + 3 > 0,
and for a ≤ 0, we have

a3 − 2a + 3 = a(a2 − 3) + a + 3 = −2ab2 + a + 3 ≥ a + 3 > 0.

Thus, it suffices to show that

(a3 − 2a + 3)2 ≥ b2 (a2 + 1)2 ,

which is equivalent to

2(a3 − 2a + 3)2 ≥ (3 − a2 )(a2 + 1)2 ,

(a − 1)2 f (a) ≥ 0,
where
f (a) = a4 + 2a3 + 2a + 5.
We need to prove that f (a) ≥ 0. For a ≥ −1, we have

f (a) = (a + 2)(a3 + 2) + 1 > 0.


EV Method for Real Variables 519

For a ≤ −1, we have


f (a) = (a + 1)2 (a + 2)2 + g(a), g(a) = −4a3 − 13a2 − 10a + 1.
It suffices to show that g(a) ≥ 0. Since
7 2 13 53
 ‹
g(a) = −(a + 1) 2a + + 5h(a), h(a) = a2 + a+
2 4 20
and
13 2 3
 ‹
h(a) = a + + > 0,
8 320
the conclusion follows. The equality holds for a = b = c = 1.
Method 2. Write the desired inequality as follows:
2(a3 − 2a + 1) + 4(b3 − 2b + 1) ≥ 0,
2(a3 − 2a + 1) + 4(b3 − 2b + 1) ≥ a2 + 2b2 − 3,
(2a3 − a2 − 4a + 3) + 2(b3 − b2 − 4b + 3) ≥ 0,
(a − 1)2 (2a + 3) + 2(b − 1)2 (2b + 3) ≥ 0.
Since 2b + 3 > 0, the inequality is true for a ≥ −3/2. Consider further that
p −3
− 3≤a≤ ,
2
and rewrite the desired inequality as follows:
2(a3 − 2a + 1) + 4(b3 − 2b + 1) + 4(a2 + 2b2 − 3) ≥ 0,
(2a3 + 4a2 − 4a − 2) + 2(2b3 + 4b2 − 4b − 2) ≥ 0,
33 9
 ‹  ‹
2a + 4a − 4a −
3 2
+ 4b + 8b − 8b +
3 2
≥ 0,
4 4
1 11
 ‹
(2a + 3) a + a −
2
+ f (b) ≥ 0,
2 4
where
9
f (b) = 4b3 + 8b2 − 8b + .
4
Since 2a + 3 ≤ 0 and
1 11 1 11 1
a2 + a − ≤3+ a− = (2a + 1) < 0,
2 4 2 4 4
it suffices to show that f (b) ≥ 0. For b ≥ 0, we have
f (b) > 8b2 − 8b + 2 = 2(2b − 1)2 ≥ 0,
and for b ≤ 0, we have
f (b) > 4b3 + 8b2 = 4b2 (b + 2) ≥ 0.
520 Vasile Cîrtoaje

P 6.13. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a14 + a24 + · · · + an4 ≤ n(n − 1)(n2 − 3n + 3).

(Vasile Cîrtoaje, 2010)

Solution. For n = 2, we need to show that

a1 + a2 = 0, a12 + a22 = 2,

implies
a14 + a24 ≤ 2.
We have
a1 = −1, a2 = 1,
or
a1 = 1, a2 = −1.
For each of these cases, the desired inequality is an equality. Assume now that
n ≥ 3. According to Theorem 1, the sum

Sn = a14 + a24 + · · · + an4

is maximum for
a1 = · · · = a j , a j+1 = · · · = an ,
where j ∈ {1, 2, . . . , n − 1}. Thus, we only need to show that

ja1 + (n − j)an = 0, ja12 + (n − j)an2 = n(n − 1)

involve
ja14 + (n − j)an4 ≤ n(n − 1)(n2 − 3n + 3).
From the equations above, we get

(n − j)(n − 1) j(n − 1)
a12 = , an2 = ;
j n− j

therefore,

(n − j)3 + j 3 n2
• ˜
ja14 + (n − j)an4 = (n − 1) =
2
− 3 n(n − 1)2 .
j(n − j) j(n − j)

Since
j(n − j) − (n − 1) = ( j − 1)(n − j − 1) ≥ 0,
EV Method for Real Variables 521

we get

n2
• ˜
ja14 + (n − j)an4 ≤ − 3 n(n − 1)2 = n(n − 1)(n2 − 3n + 3).
n−1

The equality holds for

a1 = −n + 1, a2 = · · · = an = 1

and for
a1 = n − 1, a2 = · · · = an = −1
(or any cyclic permutation).

P 6.14. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n + 1, a12 + a22 + · · · + an2 = 4n2 + n − 1,

then
a14 + a24 + · · · + an4 ≤ 16n4 + n − 1.

(Vasile Cîrtoaje, 2010)

Solution. Replacing n by 2n + 1 in the preceding P 6.13, we get the following


statement:

• If a1 , a2 , . . . , a2n+1 are real numbers so that

a1 + a2 + · · · + a2n+1 = 0, a12 + a22 + · · · + a2n+1


2
= 2n(2n + 1),

then
a14 + a24 + · · · + a2n+1
4
≤ 2n(2n + 1)(4n2 − 2n + 1),
with equality for
a1 = −2n, a2 = · · · = a2n+1 = 1
and for
a1 = 2n, a2 = · · · = a2n+1 = −1
(or any cyclic permutation).

Putting
an+1 = · · · = a2n+1 = −1,
it follows that

a1 + a2 + · · · + an − n − 1 = 0, a12 + a22 + · · · + an2 + n + 1 = 2n(2n + 1)


522 Vasile Cîrtoaje

involve
a14 + a24 + · · · + an4 + n + 1 ≤ 2n(2n + 1)(4n2 − 2n + 1).
This is equivalent to the desired statement. The equality holds for

a1 = 2n, a2 = · · · = an = −1

(or any cyclic permutation).

P 6.15. If n is an odd number and a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n2 − 1),

then
a14 + a24 + · · · + an4 ≥ n(n2 − 1)(n2 + 3).

(Vasile Cîrtoaje, 2010)

Solution. According to Theorem 1, the sum

Sn = a14 + a24 + · · · + an4

is minimum for
a1 = · · · = a j , a j+1 = · · · = an ,
where j ∈ {1, 2, . . . , n − 1}. Thus, we only need to show that

ja1 + (n − j)an = 0, ja12 + (n − j)an2 = n(n2 − 1)

involve
ja14 + (n − j)an4 ≤ n(n2 − 1)(n2 + 3).
From the equations above, we get

(n − j)(n2 − 1) j(n2 − 1)
a12 = , an2 = ;
j n− j
therefore,

(n − j)3 + j 3 2 n2
• ˜
ja14 + (n − j)an4 = (n − 1) =
2
− 3 n(n2 − 1)2 .
j(n − j) j(n − j)
Since
n2 − 1 (n − 2 j)2 − 1
− j(n − j) = ≥ 0,
4 4
we get

4n2
 ‹
ja14 + (n − j)an4 ≥ 2
− 3 n(n2 − 1)2 = n(n2 − 1)(n2 + 3).
n −1
EV Method for Real Variables 523

n−1
The equality holds when of a1 , a2 , . . . , an are equal to −n − 1 and the other
2
n+1 n−1
are equal to n − 1, and also when of a1 , a2 , . . . , an are equal to n + 1
2 2
n+1
and the other are equal to −n + 1.
2

P 6.16. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − n − 1, a12 + a22 + · · · + an2 = n3 + 2n2 − n − 1,

then
a14 + a24 + · · · + an4 ≥ n4 + (n − 1)(n + 1)4 .

(Vasile Cîrtoaje, 2010)

Solution. Replacing a1 , a2 , . . . , an by 2a1 , 2a2 , . . . , 2an and then n by 2n + 1, the


preceding P 6.15 becomes as follows:
• If a1 , a2 , . . . , a2n+1 are real numbers so that

a1 + a2 + · · · + a2n+1 = 0, a12 + a22 + · · · + a2n+1


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

then
a14 + a24 + · · · + a2n+1
4
≥ n(n + 1)(2n + 1)(n2 + n + 1),
with equality when n of a1 , a2 , . . . , a2n+1 are equal to −n − 1 and the other n + 1 are
equal to n, and also when n of a1 , a2 , . . . , a2n+1 are equal to n + 1 and the other n + 1
are equal to −n.
Putting
an+1 = · · · = a2n = −n, a2n+1 = n + 1,
it follows that
a1 + a2 + · · · + an + n(−n) + (n + 1) = 0
and
a12 + a22 + · · · + an2 + n(−n)2 + (n + 1)2 = n(n + 1)(2n + 1)
involve

a14 + a24 + · · · + an4 + n(−n)4 + (n + 1)4 ≤ n(n + 1)(2n + 1)(n2 + n + 1).

This is equivalent to the desired statement. The equality holds for

a1 = · · · = an−1 = n + 1, an = −n

(or any cyclic permutation).


524 Vasile Cîrtoaje

P 6.17. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − 2n − 1, a12 + a22 + · · · + an2 = n3 + 2n + 1,

then
a14 + a24 + · · · + an4 ≥ (n + 1)4 + (n − 1)n4 .
(Vasile Cîrtoaje, 2010)
Solution. As shown in the proof of the preceding P 6.16, the following statement
holds:
• If a1 , a2 , . . . , a2n+1 are real numbers so that

a1 + a2 + · · · + a2n+1 = 0, a12 + a22 + · · · + a2n+1


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

then
a14 + a24 + · · · + a2n+1
4
≥ n(n + 1)(2n + 1)(n2 + n + 1),
with equality when n of a1 , a2 , . . . , a2n+1 are equal to −n − 1 and the other n + 1 are
equal to n, and also when n of a1 , a2 , . . . , a2n+1 are equal to n + 1 and the other n + 1
are equal to −n.
Putting
an+1 = · · · = a2n−1 = −n − 1, a2n = a2n+1 = n,
it follows that
a1 + a2 + · · · + an + (n − 1)(−n − 1) + 2n = 0
and
a12 + a22 + · · · + an2 + (n − 1)(−n − 1)2 + 2n2 = n(n + 1)(2n + 1)
involve

a14 + a24 + · · · + an4 + (n − 1)(−n − 1)4 + 2n4 ≤ n(n + 1)(2n + 1)(n2 + n + 1),

which is equivalent to the desired statement. The equality holds for

a1 = −n − 1, a2 = · · · = a n = n

(or any cyclic permutation).

P 6.18. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = n2 − 3n − 2, a12 + a22 + · · · + an2 = n3 + 2n2 − 3n − 2,

then
a14 + a24 + · · · + an4 ≥ 2n4 + (n − 2)(n + 1)4 .
(Vasile Cîrtoaje, 2010)
EV Method for Real Variables 525

Solution. As shown in the proof of P 6.16, the following statement holds:

• If a1 , a2 , . . . , a2n+1 are real numbers so that

a1 + a2 + · · · + a2n+1 = 0, a12 + a22 + · · · + a2n+1


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

then
a14 + a24 + · · · + a2n+1
4
≥ n(n + 1)(2n + 1)(n2 + n + 1),
with equality when n of a1 , a2 , . . . , a2n+1 are equal to −n − 1 and the other n + 1 are
equal to n, and also when n of a1 , a2 , . . . , a2n+1 are equal to n + 1 and the other n + 1
are equal to −n.

Putting
an+1 = · · · = a2n−1 = −n, a2n = a2n+1 = n + 1,
it follows that

a1 + a2 + · · · + an + (n − 1)(−n) + 2(n + 1) = 0

and
a12 + a22 + · · · + an2 + (n − 1)(−n)2 + 2(n + 1)2 = n(n + 1)(2n + 1)
involve

a14 + a24 + · · · + an4 + (n − 1)(−n)4 + 2(n + 1)4 ≤ n(n + 1)(2n + 1)(n2 + n + 1),

which is equivalent to the desired statement. The equality holds for

a1 = a2 = −n, a3 = · · · = an = n + 1

(or any permutation).

P 6.19. If a, b, c, d are real numbers so that a + b + c + d = 4, then

(a2 + b2 + c 2 + d 2 − 4)(a2 + b2 + c 2 + d 2 + 36) ≤ 12(a4 + b4 + c 4 + d 4 − 4).

(Vasile Cîrtoaje, 2010)

Solution. By Theorem 1, for a + b + c + d = 4 and a2 + b2 + c 2 + d 2 = const ant, the


sum a4 + b4 + c 4 + d 4 is maximum when the set (a, b, c, d) has at most two distinct
values. Therefore, it suffices to consider the following two cases.
Case 1: a = b and c = d. We need to show that a + c = 2 involves

(a2 + c 2 − 2)(a2 + c 2 + 18) ≤ 6(a4 + c 4 − 2).


526 Vasile Cîrtoaje

Since

a2 + c 2 − 2 = (a + c)2 − 2ac − 2 = 2(1 − ac), a2 + c 2 + 18 = 2(11 − ac),

a4 + c 4 − 2 = (a2 + c 2 )2 − 2a2 c 2 − 2 = 2(1 − ac)(7 − ac),


the inequality becomes

(1 − ac)(11 − ac) ≤ 3(1 − ac)(7 − ac),

(1 − ac)(5 − ac) ≥ 0.
It is true because
1
ac ≤ (a + c)2 = 1.
4
Case 2: b = c = d. We need to show that a + 3b = 4 involves

(a2 + 3b2 − 4)(a2 + 3b2 + 36) ≤ 12(a4 + 3b4 − 4).

Since

a2 + 3b2 − 4 = 12(b − 1)2 , a2 + 3b2 + 36 = 4(3b2 − 6b + 13),

a4 + 3b4 − 4 = (4 − 3b)4 + 3b4 − 4 = 12(b − 1)2 (7b2 − 22b + 21),


the inequality becomes

(b − 1)2 [(3b2 − 6b + 13) ≤ 3(b − 1)2 (7b2 − 22b + 21),

(b − 1)2 (3b − 5)2 ≥ 0.


The equality holds for a = b = c = d = 1, and also for

5
a = −1, b=c=d=
3
(or any cyclic permutation).

P 6.20. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a16 + a26 + · · · + an6 ≤ (n − 1)6 + n − 1.

(Vasile Cîrtoaje, 2010)


EV Method for Real Variables 527

Solution. For n = 2, we need to show that

a1 + a2 = 0, a12 + a22 = 2,

implies
a16 + a26 ≤ 2.
We have
a1 = −1, a2 = 1,
or
a1 = 1, a2 = −1.
For each of these cases, the desired inequality is an equality. According to Theorem
2, the sum
Sn = a16 + a26 + · · · + an6
is maximum for
a1 = · · · = a j , a j+1 = · · · = an ,
where j ∈ {1, 2, . . . , n − 1}. Thus, we only need to show that

ja1 + (n − j)an = 0, ja12 + (n − j)an2 = n(n − 1)

involve
ja16 + (n − j)an6 ≤ (n − 1)6 + n − 1.
From the equations above, we get

(n − j)(n − 1) j(n − 1)
a12 = , an2 = .
j n− j

Thus, the desired inequality becomes

(n − j)5 + j 5 (n − 1)5 + 1
≤ ,
j 2 (n − j)2 (n − 1)2

(n − j)4 − (n − j)3 j + (n − j)2 j 2 − (n − j) j 3 + j 4



j 2 (n − j)2
(n − 1)4 − (n − 1)3 + (n − 1)2 − (n − 1) + 1
≤ ,
(n − 1)2
(n − j)2 n − j j j2 1 1
− − + ≤ (n − 1)2
− (n − 1) − + ,
j2 j n − j (n − j)2 n − 1 (n − 1)2
which can be written as
f (a) ≥ f (b),
where
1 1
f (x) = x 2 − x − + 2,
x x
528 Vasile Cîrtoaje

n
a = n − 1, b= − 1.
j
Since a ≥ b and
n−1
‹  ‹
n
ab − 1 = (n − 1) −1 −1= n − 1 ≥ 0,
j j
we have
1 a+b
 ‹
f (a) − f (b) = (a − b) a + b − 1 + −
ab a2 b2
1 1
 ‹•  ‹ ˜
= (a − b) 1 − (a + b) 1 + − 1 ≥ 0.
ab ab
The equality holds for

a1 = −n + 1, a2 = · · · = an = 1,

and for
a1 = n − 1, a2 = · · · = an = −1
(or any cyclic permutation).

P 6.21. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1,

then
a16 + a26 + · · · + an6 ≤ n6 + n − 1.
(Vasile Cîrtoaje, 2010)
Solution. The inequality follows from the preceding P 6.20 by replacing n with
n + 1, and then making an+1 = −1. The equality holds for

a1 = n, a2 = · · · = an = −1

(or any cyclic permutation).

P 6.22. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 0, a12 + a22 + · · · + an2 = n(n − 1),

then
a18 + a28 + · · · + an8 ≤ (n − 1)8 + n − 1.
(Vasile Cîrtoaje, 2010)
EV Method for Real Variables 529

Solution. For n = 2, we need to show that

a1 + a2 = 0, a12 + a22 = 2,

implies
a18 + a28 ≤ 2.
We have
a1 = −1, a2 = 1,
or
a1 = 1, a2 = −1.
For each of these cases, the desired inequality is an equality. According to Theorem
2, the sum
Sn = a18 + a28 + · · · + an8
is maximum for
a1 = · · · = a j , a j+1 = · · · = an ,
where j ∈ {1, 2, . . . , n − 1}. Thus, we only need to show that

ja1 + (n − j)an = 0, ja12 + (n − j)an2 = n(n − 1)

involve
ja18 + (n − j)an8 ≤ (n − 1)8 + n − 1.
From the equations above, we get

(n − j)(n − 1) j(n − 1)
a12 = , an2 = .
j n− j

Thus, the desired inequality becomes

(n − j)7 + j 7 (n − 1)7 + 1
≤ ,
j 3 (n − j)3 (n − 1)4

(n − j)3 (n − j)2 n − j j j2 j3
− + + − + ≤
j3 j2 j n − j (n − j)2 (n − j)3
1 1 1
≤ (n − 1)3 − (n − 1)2 + (n − 1) + − + ,
n − 1 (n − 1)2 (n − 1)3

f (a) ≥ f (b),
where
n
a = n − 1, b= − 1,
j
1 1 1
f (x) = x 3 − x 2 + x + − 2 + 3, x > 0.
x x x
530 Vasile Cîrtoaje

Since
1
f (x) = (t − 1)(t 2 − 2), t=x+ ≥ 2,
x
it suffices to show that
1 1
a+ ≥ b+ .
a b
We have a ≥ b,

n−1
‹  ‹
n
ab − 1 = (n − 1) −1 −1= n − 1 ≥ 0,
j j

therefore
1 1 1
 ‹
a + − b − = (a − b) 1 − ≥ 0.
a b ab
The equality holds for

a1 = −n + 1, a2 = · · · = an = 1

and for
a1 = n − 1, a2 = · · · = an = −1

(or any cyclic permutation).

P 6.23. If a1 , a2 , . . . , an are real numbers so that

a1 + a2 + · · · + an = 1, a12 + a22 + · · · + an2 = n2 + n − 1,

then
a18 + a28 + · · · + an8 ≤ n8 + n − 1.

(Vasile Cîrtoaje, 2010)

Solution. The inequality follows from the preceding P 6.22 by replacing n with
n + 1, and making an+1 = −1. The equality holds for

a1 = n, a2 = · · · = an = −1

(or any cyclic permutation).


EV Method for Real Variables 531

P 6.24. Let a1 , a2 , . . . , an (n ≥ 2) be real numbers (not all equal), and let

a1 + a2 + · · · + an a12 + a22 + · · · + an2 a13 + a23 + · · · + an3


A= , B= , C= .
n n n
Then, ‚ v Œ ‚ v Œ
1 t 2n2 B 2 − AC 1 t 2n2
1− 1+ ≤ 2 ≤ 1+ 1+ .
4 n−1 B − A4 4 n−1

(Vasile Cîrtoaje, 2010)

Solution. It is well-known that B > A2 , hence B 2 > A4 .


(a) For n = 2, the right inequality reduces to (a12 − a22 )2 ≥ 0. Consider further
that n ≥ 3. Since the right inequality remains unchanged by replacing a1 , a2 , . . . , an
with −a1 , −a2 , . . . , −an , we may suppose that A ≥ 0. Assuming that

A = const ant, B = const ant,

we only need to consider the case when C is minimum. Thus, according to Corollary
2, it suffices to prove the required inequality for a1 < a2 = a3 = · · · = an . Setting

a1 := a, a2 = a3 = · · · = an := b, a < b,

the inequality becomes


˜2
a2 + (n − 1)b2 a + (n − 1)b a3 + (n − 1)b3
•
− · ‚ v Œ
n n n 1 t 2n2
˜2 • ≤ 1+ 1+ ,
a + (n − 1)b2 a + (n − 1)b 4 4 n−1
• 2 ˜

n n

After dividing the numerator and denominator of the left fraction by (a − b)2 , the
inequality reduces to
v
−4n2 ab t 2n2
≤ 1 + 1 + ,
(n + 1)a2 + 2(n − 1)ab + (2n2 − 3n + 1)b n−1

−2ab 1
≤p ,
(n + 1)a2 + 2(n − 1)ab + (2n − 3n + 1)b
2
(n2 − 1)(2n − 1) − n + 1
‚ v Œ2
t 2n2 − 3n + 1
a+ b ≥ 0.
n+1
The equality holds for
v
t n+1
− a1 = a2 = · · · = an
(n − 1)(2n − 1)
532 Vasile Cîrtoaje

(or any cyclic permutation).


(b) For n = 2, the left inequality reduces to (a1 − a2 )4 ≥ 0. For n ≥ 3, the proof
is similar to the one of the right inequality. The equality holds for
v
t n+1
a1 = a2 = · · · = a n
(n − 1)(2n − 1)
(or any cyclic permutation).

P 6.25. If a, b, c, d are real numbers so that

a + b + c + d = 2,

then
3
a4 + b4 + c 4 + d 4 ≤ 40 + (a2 + b2 + c 2 + d 2 )2 .
4
(Vasile Cîrtoaje, 2010)
Solution. Write the inequality in the homogeneous form

10(a + b + c + d)4 + 3(a2 + b2 + c 2 + d 2 )2 ≥ 4(a4 + b4 + c 4 + d 4 ).

By Theorem 1, for a + b + c + d = const ant and a2 + b2 + c 2 + d 2 = const ant, the


sum a4 + b4 + c 4 + d 4 is maximum when the set (a, b, c, d) has at most two distinct
values. Therefore, it suffices to consider the following two cases.
Case 1: a = b and c = d. The inequality reduces to

41(a2 + c 2 )2 + 160ac(a2 + c 2 ) + 164a2 c 2 ≥ 0,

which can be written in the obvious form

(a2 + c 2 )2 + 40(a2 + c 2 + 2ac)2 + 4a2 c 2 ≥ 0.

Case 2: b = c = d. The inequality reduces to the obvious form

(a + 5b)2 (3a2 + 10ab + 11b2 ) ≥ 0.

Since the homogeneous inequality becomes an equality for


−a
=b=c=d
5
(or any cyclic permutation), the original inequality is an equality for

a = 5, b = c = d = −1

(or any cyclic permutation).


EV Method for Real Variables 533

P 6.26. If a, b, c, d, e are real numbers, then


p
31 + 18 3 3
a +b +c +d +e ≤
4 4 4 4 4
(a + b + c + d + e)4 + (a2 + b2 + c 2 + d 2 + e2 )2 .
8 4

(Vasile Cîrtoaje, 2010)

Solution. We proceed as in the proof of the preceding P 6.25. Taking into account
Theorem 1, it suffices to consider the cases b = c = d = e, and a = b and c = d = e.
Case 1: b = c = d = e. Due to homogeneity, we may consider b = c = d = e = 0
and b = c = d = e = 1. The first case is trivial. In the second case, the inequality
becomes p
31 + 18 3 3
a +4≤
4
(a + 4)4 + (a2 + 4)2 ,
8 4
p 2  p 
a+2+2 3 f (a) + 2 3 g(a) ≥ 0,
where
f (a) = 29a2 + 164a + 272, g(a) = 9a2 + 50a + 76.
It suffices to show that f (a) ≥ 0 and g(a) ≥ 0. Indeed, we have

82 2 1
 ‹
f (a) > 25a + 164a + 269 = 5a +
2
+ > 0,
5 25

25 2 5
 ‹
g(a) > 9a + 50a + 70 = 3a +
2
+ > 0.
3 9

Case 2: a = b and c = d = e. It suffices to show that


3
a4 + b4 + c 4 + d 4 + e4 ≤ (a2 + b2 + c 2 + d 2 + e2 )2 ,
4
which reduces to
3
2a4 + 3c 4 ≤ (2a2 + 3c 2 )2 ,
4
3(2a2 + 3c 2 )2 ≥ 4(2a4 + 3c 4 ),
4a4 + 36a2 c 2 + 15c 4 ≥ 0.

The equality holds for


−a
p =b=c=d=e
2(1 + 3)

(or any cyclic permutation).


534 Vasile Cîrtoaje

−5
P 6.27. Let a, b, c, d, e 6= be real numbers so that a + b + c + d + e = 5. Then,
4
a(a − 1) b(b − 1) c(c − 1) d(d − 1) e(e − 1)
+ + + + ≥ 0.
(4a + 5)2 (4b + 5)2 (4c + 5)2 (4d + 5)2 (4e + 5)2

(Vasile Cîrtoaje, 2010)

Solution. Write the inequality as


X • 180a(a − 1) ˜
+ 1 ≥ 5,
(4a + 5)2
X (14a − 5)2
≥ 5.
(4a + 5)2
By the Cauchy-Schwarz inequality, we have
P 2
X (14a − 5)2 (4a + 5)(14a − 5)
≥ .
(4a + 5)2 (4a + 5)4
P

Therefore, it suffices to show that


€ X Š2 X
56 a2 + 125 ≥ 5 (4a + 5)4 .

Using the substitution

4a + 5 4b + 5 4e + 5
a1 = , a2 = , . . . , a5 = ,
9 9 9
we need to prove that a1 + a2 + a3 + a4 + a5 = 5 involves
‚ 5
Œ2 5
X X
7 ai2 − 25 ≥ 20 ai4 .
i=1 i=1

Rewrite this inequality in the homogeneous form


 2
5
‚ 5
Œ2 5
X X X
7 ai2 − ai  ≥ 20 ai4 .
i=1 i=1 i=1

By Theorem 1, for a1 + a2 + a3 + a4 + a5 = 5 and a12 + a22 + a32 + a42 + a52 = const ant,
the sum a14 + a24 + a34 + a44 + a54 is maximum when the set (a1 , a2 , a3 , a4 , a5 ) has at
most two distinct values. Therefore, we need to consider the following two cases.
Case 1: a1 = x and a2 = a3 = a4 = a5 = y. The homogeneous inequality reduces
to
(3x 2 + 6 y 2 − 4x y)2 ≥ 5(x 4 + 4 y 4 ),
EV Method for Real Variables 535

which is equivalent to the obvious inequality

(x − y)2 (x − 2 y)2 ≥ 0.

Case 2: a1 = a2 = x and a3 = a4 = a5 = y. The homogeneous inequality becomes

(5x 2 + 6 y 2 − 6x y)2 ≥ 5(2x 4 + 3 y 4 ),

which is equivalent to the obvious inequality

(x − y)2 [5(x − y)2 + 2 y 2 ] ≥ 0.

The equality holds for a = b = c = d = e = 1, and also for

5 5
a= , b=c=d=e=
2 8
(or any cyclic permutation).

Remark. Similarly, we can prove the following generalization.


• Let x 1 , x 2 , . . . , x n 6= −k be real numbers so that x 1 + x 2 + · · · + x n = n, where
n
k≥ p .
2 n−1

Then,
x 1 (x 1 − 1) x 2 (x 2 − 1) x n (x n − 1)
+ + ··· + ≥ 0,
(x 1 + k) 2 (x 2 + k)2 (x n + k)2
n
with equality for x 1 = x 2 = · · · = x n = 1. If k = p , then the equality holds
2 n−1
also for
n n
x1 = , x2 = · · · = x n =
2 2(n − 1)
(or any cyclic permutation).

P 6.28. If a, b, c are real numbers so that

a + b + c = 9, ab + bc + ca = 15,

then
19 1 1 1 7
≤ 2 + 2 + 2 ≤ .
175 b + bc + c 2 c + ca + a 2 a + ab + b 2 19
(Vasile C., 2011)
536 Vasile Cîrtoaje

Solution. From
(b + c)2 ≥ 4bc
and
b + c = 9 − a, bc = 15 − a(b + c) = 15 − a(9 − a) = a2 − 9a + 15,
we get a ≤ 7. Since
b2 + bc + c 2 = (a + b + c)(b + c) − (ab + bc + ca) = 9(9 − a) − 15 = 3(22 − 3a),
we may write the inequality in the form
57 21
≤ f (a) + f (b) + f (c) ≤ .
175 19
where
1
f (u) = , u ≤ 7.
22 − 3u
We have
3
g(x) = f 0 (x) = ,
(22 − 3x)2
162
g 00 (x) = .
(22 − 3x)4
Since g 00 (x) > 0 for x ≤ 7, g is strictly convex on (−∞, 7]. According to Corollary
1, if a ≤ b ≤ c and
a + b + c = 9, a2 + b2 + c 2 = 51,
then the sum S3 = f (a) + f (b) + f (c) is maximum for a = b ≤ c, and is minimum
for a ≤ b = c.
(a) To prove the right inequality, it suffices to consider the case a = b ≤ c.
From
a + b + c = 9, ab + bc + ca = 15,
we get a = b = 1 and c = 7, therefore
1 1 1 7
+ 2 + 2 = .
b2 + bc + c 2 c + ca + a 2 a + ab + b 2 19
The original right inequality is an equality for a = b = 1 and c = 7 (or any cyclic
permutation).
(b) To prove the left inequality, it suffices to consider the case a ≤ b = c, which
involves a = −1 and b = c = 5, hence
1 1 1 19
+ + = .
b2 + bc + c 2 c 2 + ca + a2 a2 + ab + b2 175
The original left inequality is an equality for a = −1 and b = c = 5 (or any cyclic
permutation).
EV Method for Real Variables 537

P 6.29. If a, b, c are real numbers so that

8(a2 + b2 + c 2 ) = 9(ab + bc + ca),

then
419 a2 b2 c2 311
≤ 2 + + ≤ .
175 b + bc + c 2 c 2 + ca + a2 a2 + ab + b2 19
(Vasile C., 2011)

Solution. Due to homogeneity, we may assume that

a + b + c = 9, a2 + b2 + c 2 = 51.

Next, the proof is similar to the one of the preceding P 6.28. Write the inequality
in the form
1257 933
≤ f (a) + f (b) + f (c) ≤ .
175 19
where
u2
f (u) = , u ≤ 7.
22 − 3u
We have
−3x 2 + 44x 8712
g(x) = f 0 (x) = , g 00 (x) = .
(22 − 3x) 2 (22 − 3x)4
Since g is strictly convex on (−∞, 7], according to Corollary 1, the sum S3 =
f (a) + f (b) + f (c) is maximum for a = b ≤ c, and is minimum for a ≤ b = c.
(a) To prove the right inequality, it suffices to consider the case a = b ≤ c,
which involves
a = b = 1, c = 7,
and
a2 b2 c2 311
+ + = .
b + bc + c
2 2 c + ca + a
2 2 a + ab + b
2 2 19
The original right inequality is an equality for a = b = c/7 (or any cyclic permuta-
tion).
(b) To prove the left inequality, it suffices to consider the case a ≤ b = c, which
involves a = −1 and b = c = 5, hence

a2 b2 c2 419
+ + = .
b + bc + c
2 2 c + ca + a
2 2 a + ab + b
2 2 175
The original left inequality is an equality for −5a = b = c (or any cyclic permuta-
tion).
538 Vasile Cîrtoaje

P 6.30. Let a1 , a2 , . . . , an be real numbers such that a1 + a2 + · · · + an = n. If n ≤ 10,


then
2(a12 + a22 + · · · + an2 )2 − n(a13 + a23 + · · · + an3 ) ≥ n2 .
(Vasile Cîrtoaje, 2020)

Solution. Write the inequality in the homogeneous form

2n2 (a12 + a22 +· · ·+ an2 )2 − n2 (a1 + a2 +· · ·+ an )(a13 + a23 +· · ·+ an3 ) ≥ (a1 + a2 +· · ·+ an )4 .

According to Corollary 2, for a1 +a2 +· · ·+an = const ant > 0 and a12 +a22 +· · ·+an2 =
const ant, the sum
S = a13 + a23 + · · · + an3
is maximal when n − 1 of a1 , a2 , . . . , an are equal. Therefore, it suffices to consider
the case a2 = a3 = · · · = an . Due to homogeneity, for the nontrivial case a2 = a3 =
· · · = an 6= 0, we may consider that a2 = a3 = · · · = an = 1. Thus we only need to
prove that

2n2 (a12 + n − 1)2 − n2 (a1 + n − 1)(a13 + n − 1) ≥ (a1 + n − 1)4 ,

which is equivalent to

(a1 − 1)2 (Aa12 − Ba1 + C) ≥ 0,

where

A = n(n + 1), B = n(n2 − 2n + 2), C = n(n − 1)(2n − 1).

The inequality is true because

4AC − B 2 = n4 (−n2 + 12n − 12) ≥ 0.

The equality occurs for a1 = a2 = · · · = an = 1.


Appendix A

Glosar

1. AM-GM (ARITHMETIC MEAN-GEOMETRIC MEAN) INEQUALITY

If a1 , a2 , . . . , an are nonnegative real numbers, then

a1 + a2 + · · · + an ≥ n n a1 a2 · · · an ,
p

with equality if and only if a1 = a2 = · · · = an .

2. WEIGHTED AM-GM INEQUALITY

Let p1 , p2 , . . . , pn be positive real numbers satisfying

p1 + p2 + · · · + pn = 1.

If a1 , a2 , . . . , an are nonnegative real numbers, then


p p
p1 a1 + p2 a2 + · · · + pn an ≥ a1 1 a2 2 · · · anpn ,

with equality if and only if a1 = a2 = · · · = an .

3. AM-HM (ARITHMETIC MEAN-HARMONIC MEAN) INEQUALITY

If a1 , a2 , . . . , an are positive real numbers, then

1 1 1
 ‹
(a1 + a2 + · · · + an ) + + ··· + ≥ n2 ,
a1 a2 an

with equality if and only if a1 = a2 = · · · = an .

539
540 Vasile Cîrtoaje

4. POWER MEAN INEQUALITY

The power mean of order k of positive real numbers a1 , a2 , . . . , an ,


  1
a1k +a2k +···+ank k

 n , k 6= 0
Mk = p ,
 n a1 a2 · · · an ,
 k=0

is an increasing function with respect to k ∈ R. For instant, M2 ≥ M1 ≥ M0 ≥ M−1


is equivalent to
v
t a2 + a2 + · · · + a2
1 2 n a1 + a2 + · · · + an p n
≥ ≥ n a1 a2 · · · an ≥ .
n n 1 1 1
+ + ··· +
a1 a2 an

5. BERNOULLI’S INEQUALITY

For any real number x ≥ −1, we have


a) (1 + x) r ≥ 1 + r x for r ≥ 1 and r ≤ 0;
b) (1 + x) r ≤ 1 + r x for 0 ≤ r ≤ 1.
If a1 , a2 , . . . , an are real numbers such that either a1 , a2 , . . . , an ≥ 0 or

−1 ≤ a1 , a2 , . . . , an ≤ 0,

then
(1 + a1 )(1 + a2 ) · · · (1 + an ) ≥ 1 + a1 + a2 + · · · + an .

6. SCHUR’S INEQUALITY

For any nonnegative real numbers a, b, c and any positive number k, the inequality
holds
a k (a − b)(a − c) + b k (b − c)(b − a) + c k (c − a)(c − b) ≥ 0,
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
a3 + b3 + c 3 + 3abc ≥ ab(a + b) + bc(b + c) + ca(c + a),
(a + b + c)3 + 9abc ≥ 4(a + b + c)(ab + bc + ca),
9abc
a2 + b2 + c 2 + ≥ 2(ab + bc + ca),
a+b+c
Glosar 541

(b − c)2 (b + c − a) + (c − a)2 (c + a − b) + (a − b)2 (a + b − c) ≥ 0.


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

a4 + b4 + c 4 + abc(a + b + c) ≥ ab(a2 + b2 ) + bc(b2 + c 2 ) + ca(c 2 + a2 ),

a4 + b4 + c 4 − a2 b2 − b2 c 2 − c 2 a2 ≥ (ab + bc + ca)(a2 + b2 + c 2 − ab − bc − ca),


(b − c)2 (b + c − a)2 + (c − a)2 (c + a − b)2 + (a − b)2 (a + b − c)2 ≥ 0,
6abcp ≥ (p2 − q)(4q − p2 ), p = a + b + c, q = ab + bc + ca.
A generalization of the fourth degree Schur’s inequality, which holds for any
real numbers a, b, c and any real number m, is the following (Vasile Cirtoaje, 2004)
X
(a − mb)(a − mc)(a − b)(a − c) ≥ 0,

with equality for a = b = c, and also for a/m = b = c (or any cyclic permutation).
This inequality is equivalent to
X X X X
a + m(m + 2)
4
a b + (1 − m )abc
2 2 2
a ≥ (m + 1) ab(a2 + b2 ),
X
(b − c)2 (b + c − a − ma)2 ≥ 0.

7. CAUCHY-SCHWARZ INEQUALITY
If a1 , a2 , . . . , an and b1 , b2 , . . . , bn are real numbers, then

(a12 + a22 + · · · + an2 )(b12 + b22 + · · · + bn2 ) ≥ (a1 b1 + a2 b2 + · · · + an bn )2 ,

with equality for


a1 a an
= 2 = ··· = .
b1 b2 bn
Notice that the equality conditions are also valid for ai = bi = 0, where 1 ≤ i ≤ n.

8. HÖLDER’S INEQUALITY
If x i j (i = 1, 2, · · · , m; j = 1, 2, · · · n) are nonnegative real numbers, then
‚ n Œ v !m
m
Y X n uY
X m
m
≥ .
t
xi j xi j
i=1 j=1 j=1 i=1
542 Vasile Cîrtoaje

9. CHEBYSHEV’S INEQUALITY
Let a1 ≥ a2 ≥ · · · ≥ an be real numbers.

a) If b1 ≥ b2 ≥ · · · bn , then
n
‚ n
Œ‚ n
Œ
X X X
n ai bi ≥ ai bi ;
i=1 i=1 i=1

b) If b1 ≤ b2 ≤ · · · ≤ bn , then
n
‚ n
Œ‚ n
Œ
X X X
n ai bi ≤ ai bi .
i=1 i=1 i=1

10. REARRANGEMENT INEQUALITY


(1) If (a1 , a2 , . . . , an ) and (b1 , b2 , . . . , bn ) are two increasing (or decreasing) real
sequences, and (i1 , i2 , · · · , in ) is an arbitrary permutation of (1, 2, · · · , n), then

a1 b1 + a2 b2 + · · · + an bn ≥ a1 bi1 + a2 bi2 + · · · + an bin

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

a1 b1 + a2 b2 + · · · + an bn ≤ a1 bi1 + a2 bi2 + · · · + an bin

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 + · · · + bi ≥ c1 + · · · + ci , i = 1, 2, · · · , n.

If a1 ≥ a2 ≥ · · · ≥ an ≥ 0, then

a1 b1 + a2 b2 + · · · + an bn ≥ a1 c1 + a2 c2 + · · · + an cn .

Notice that all these inequalities follow immediately from the identity
n n
‚ i i
Œ
X X X X
ai (bi − ci ) = (ai − ai+1 ) bj − cj , an+1 = 0.
i=1 i=1 j=1 j=1
Glosar 543

11. SQUARE PRODUCT INEQUALITY

Let a, b, c be real numbers, and let

p = a + b + c, q = ab + bc + ca, r = abc,
p p
s = p2 − 3q = a2 + b2 + c 2 − ab − bc − ca.
From the identity

(a − b)2 (b − c)2 (c − a)2 = −27r 2 + 2(9pq − 2p3 )r + p2 q2 − 4q3 ,

it follows that

−2p3 + 9pq − 2(p2 − 3q) p2 − 3q −2p3 + 9pq + 2(p2 − 3q) p2 − 3q


p p
≤r≤ ,
27 27
which is equivalent to

p3 − 3ps2 − 2s3 p3 − 3ps2 + 2s3


≤r≤ .
27 27
Therefore, for constant p and q, the product r is minimum and maximum when
two of a, b, c are equal.

12. KARAMATA’S MAJORIZATION INEQUALITY


Let f be a convex function on a real interval I. If a decreasingly ordered sequence

A = (a1 , a2 , . . . , an ), ai ∈ I,

majorizes a decreasingly ordered sequence

B = (b1 , b2 , . . . , bn ), bi ∈ I,

then
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ f (b1 ) + f (b2 ) + · · · + f (bn ).
We say that a sequence A = (a1 , a2 , . . . , an ) with a1 ≥ a2 ≥ · · · ≥ an majorizes a
sequence B = (b1 , b2 , . . . , bn ) with b1 ≥ b2 ≥ · · · ≥ bn , and write it as

A  B,

if
a1 ≥ b1 ,
a1 + a2 ≥ b1 + b2 ,
·····················
a1 + a2 + · · · + an−1 ≥ b1 + b2 + · · · + bn−1 ,
a1 + a2 + · · · + an = b1 + b2 + · · · + bn .
544 Vasile Cîrtoaje

13. CONVEX FUNCTIONS


A function f defined on a real interval I is said to be convex if

f (αx + β y) ≤ α f (x) + β f ( y)

for all x, y ∈ I and any α, β ≥ 0 with α + β = 1. If the inequality is reversed, then


f is said to be concave.
If f is differentiable on I, then f is (strictly) convex if and only if the derivative f 0
is (strictly) increasing. If f 00 ≥ 0 on I, then f is convex on I. Also, if f 00 ≥ 0 on (a,
b) and f is continuous on [a, b], then f is convex on [a, b].

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

p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) p1 a1 + p2 a2 + · · · + pn an


 ‹
≥f .
p1 + p2 + · · · + pn p1 + p2 + · · · + pn
For p1 = p2 = · · · = pn , Jensen’s inequality becomes
a + a + ··· + a 
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f .
n

Right Half Convex Function Theorem (Vasile Cîrtoaje, 2004). Let f be a real
function defined on an interval I and convex on I≥s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I such that x ≤ s ≤ y and x + (n − 1) y = ns.

Left Half Convex Function Theorem (Vasile Cîrtoaje, 2004). Let f be a real function
defined on an interval I and convex on I≤s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
Glosar 545

for all x, y ∈ I such that x ≥ s ≥ y and x + (n − 1) y = ns.

Left Convex-Right Concave Function Theorem (Vasile Cîrtoaje, 2004). Let a ≤ c


be real numbers, let f be a continuous function defined on I = [a, ∞), strictly convex
on [a, c] and strictly concave on [c, ∞), and let

E(a1 , a2 , . . . , an ) = f (a1 ) + f (a2 ) + · · · + f (an ).

If a1 , a2 , . . . , an ∈ I such that

a1 + a2 + · · · + an = S = const ant,

then
(a) E is minimum for a1 = a2 = · · · = an−1 ≤ an ;
(b) E is maximum for either a1 = a or a < a1 ≤ a2 = · · · = an .

Right Half Convex Function Theorem for Ordered Variables (Vasile Cîrtoaje,
2008). Let f be a real function defined on an interval I and convex on I≥s , where
s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≤ a2 ≤ · · · ≤ am ≤ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I such that

x ≤ s ≤ y, x + (n − m) y = (1 + n − m)s.

Left Half Convex Function Theorem for Ordered Variables (Vasile Cîrtoaje, 2008).
Let f be a real function defined on an interval I and convex on I≤s , where s ∈ int(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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≥ a2 ≥ · · · ≥ am ≥ s, m ∈ {1, 2, . . . , n − 1},
546 Vasile Cîrtoaje

if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I such tht

x ≥ s ≥ y, x + (n − m) y = (1 + n − m)s.

Right Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval I and convex on [s, s0 ], where s, s0 ∈ I, s < s0 . In
addition, f is decreasing on I≤s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I such that x ≤ s ≤ y and x + (n − 1) y = ns.

Left Partially Convex Function Theorem (Vasile Cîrtoaje, 2012). Let f be a real
function defined on an interval I and convex on [s0 , s], where s0 , s ∈ I, s0 < s. In
addition, f is increasing on I≥s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

if and only if
f (x) + (n − 1) f ( y) ≥ n f (s)
for all x, y ∈ I such that x ≥ s ≥ y and x + (n − 1) y = ns.

Right Partially Convex Function Theorem for Ordered Variables (Vasile Cirtoaje,
2014). Let f be a real function defined on an interval I and convex on [s, s0 ], where
s, s0 ∈ I, s < s0 . In addition, f is decreasing on I≤s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns
Glosar 547

and
a1 ≤ a2 ≤ · · · ≤ am ≤ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I such that x ≤ s ≤ y and x + (n − m) y = (1 + n − m)s.

Left Partially Convex Function Theorem for Ordered Variables (Vasile Cirtoaje,
2014). Let f be a real function defined on an interval I and convex on [s0 , s], where
s0 , s ∈ I, s0 < s. In addition, f is increasing on I≥s0 and f (u) ≥ f (s0 ) for u ∈ 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 satisfying

a1 + a2 + · · · + an = ns

and
a1 ≥ a2 ≥ · · · ≥ am ≥ s, m ∈ {1, 2, . . . , n − 1},
if and only if
f (x) + (n − m) f ( y) ≥ (1 + n − m) f (s)
for all x, y ∈ I such that x ≥ s ≥ y and x + (n − m) y = (1 + n − m)s.

Equal Variables Theorem for Nonnegative Variables (Vasile Cirtoaje, 2005). Let
a1 , a2 , . . . , an (n ≥ 3) be fixed nonnegative real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

such that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,

where k is a real number (k 6= 1); for k = 0, assume that

x 1 x 2 · · · x n = a1 a2 · · · an .

Let f be a real-valued function, continuous on [0, ∞) and differentiable on (0, ∞),


such that the associated function
€ 1 Š
g(x) = f 0 x k−1

is strictly convex on (0, ∞). Then, the sum

Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )

is maximum for
x 1 = x 2 = · · · = x n−1 ≤ x n ,
548 Vasile Cîrtoaje

and is minimum for


0 < x1 ≤ x2 = x3 = · · · = x n
or
0 = x 1 = · · · = x j ≤ x j+1 ≤ x j+2 = · · · = x n , j ∈ {1, 2, . . . , n − 1}.

Equal Variables Theorem for Real Variables (Vasile Cirtoaje, 2010). Let a1 , a2 , . . . , an
(n ≥ 3) be fixed real numbers, and let

0 ≤ x1 ≤ x2 ≤ · · · ≤ x n

such that

x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an , x 1k + x 2k + · · · + x nk = a1k + a2k + · · · + ank ,

where k is an even positive integer. If f is a differentiable function on R such that the


associated function g : R → R defined by
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 )

is minimum for x 2 = x 3 = · · · = x n , and is maximum for x 1 = x 2 = · · · = x n−1 .

Best Upper Bound of Jensen’s Difference Theorem (Vasile Cirtoaje, 1990). Let
p1 , p2 , . . . , pn (n ≥ 3) be fixed positive real numbers, and let f be a convex function
on 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


 ‹
−f
p1 + p2 + · · · + pn p1 + p2 + · · · + pn

is maximum when all ai ∈ {a, b}.


Appendix B

Bibliography

[1] Andreescu T., Cîrtoaje V., Dospinescu G., Lascu M., Old and New Inequalities,
GIL Publishing House, 2004.
[2] Bin X., Boreico I., Can V.Q.B., Bulj A., Lascu M., Opympiad Inequalities, GIL
Publishing House, 2015.
[3] Bin X., Boreico I., Can V.Q.B., Cîrtoaje V., Lascu M., An Introduction to Inequali-
ties, GIL Publishing House, 2015.
[4] Can V.Q.B., Pohoaţă C., Old and New Inequalities, GIL Publishing House, 2008.
[5] Can V.Q.B., Anh T.Q., Su Dung Phuong Phap Cauchy-Schwarz De Chung Minh
Bat Dang Thuc, Nha Xuat Ban Dai Hoc Su Pham, 2010.
[6] Cîrtoaje V., Asupra unor inegalitati cu restrictii, Revista matematica din Timisoara,
Nr. 1, 1990.
[7] Cîrtoaje V., Two Generalizations of Popoviciu’s Inequality, Crux Mathematicorum,
Issue 5, 2005.
[8] Cîrtoaje V., A Generalization of Jensen’s Inequality, Gazeta Matematica-A, Nr. 2,
2005.
[9] Cîrtoaje V., Algebraic Inequalities-Old and New Methods, GIL Publishing House,
2006.
[10] Cîrtoaje V., The Equal Variable Method, Journal of Inequalities In Pure and
Applied Mathematics, Volume 8, Issue 1, 2007.
[11] Cîrtoaje V., On Jensen Type Inequalities with Ordered Variables, Journal of In-
equalities In Pure and Applied Mathematics, Volume 9, Issue 1, 2008.
[12] Cîrtoaje V., The Proof of Three Open Inequalities, Crux Mathematicorum, Vol-
ume 34, Issue 4, 2008.
[13] Cîrtoaje V., Can V.Q.B., Anh T.Q., Inequalities with Beautiful Solutions, GIL Pub-
lishing House, 2009.
[14] Cîrtoaje V., The Best Lower Bound Depended on Two Fixed Variables for Jensen’s
Inequality with Ordered Variables, Journal of Inequalities and Applications, Volume
2010.

549
550 Vasile Cîrtoaje

[15] Cîrtoaje V., The Best Upper Bound for Jensen’s Inequality, Australian Journal of
Mathematical Analysis and Aplications, Volume 7, Issue 2, Art. 22, 2011.
[16] Cîrtoaje V., Baiesu A., An Extension of Jensen’s discrete inequality to half convex
functions, Journal of Inequalities and Applications, Volume 2011.
[17] 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.
[18] Cîrtoaje V., An Extension of Jensen’s discrete inequality to partially convex func-
tions, Journal of Inequalities and Applications, Volume 2013:54.
[19] Cîrtoaje V., On the Equal Variables Method Applied to Real Variables, Creative
Mathematics and Informatics, no. 2, 2015.
[20] Cîrtoaje V., Three extensions of HCF and PCF theorems, Advances in Inequalities
and Applications, no. 2, 2016.
[21] Cîrtoaje V., Extensions of Jensen’s discrete inequality with ordered variables to
half and partially convex functions, Journal of Inequalities and Special Functions,
Volume 8, Issue 3, 2017.
[22] Cîrtoaje V., Mathematical Inequalities - Volume 1, Symmetric Polynomial In-
equalities, Lambert Academic Publishing, 2018.
[23] Cîrtoaje V., Mathematical Inequalities - Volume 2, Symmetric Rational and Non-
rational Inequalities, Lambert Academic Publishing, 2018.
[24] Cîrtoaje V., Mathematical Inequalities - Volume 3, Cyclic and Noncyclic Inequal-
ities, Lambert Academic Publishing, 2018.
[25] Cvetkovski, Inequalities: Theorems, Techniques and Selected Problems, Springer-
Verlag Berlin Heidelberg, 2012.
[26] Hung P.K., Secrets in Inequalities, Volume 1: Basic Inequalities, GIL Publishing
House, 2007.
[27] Hung P.K., Secrets in Inequalities, Volume 2: Advanced Inequalities, GIL Pub-
lishing House, 2008.
[28] Littlewood G.H., Polya J.E., Inequalities, Cambridge University Press, 1967.
[29] Mitrinovic̆ D.S., Pecaric̆ J.E., Fink A.M., Classical and New Inequalities in Anal-
ysis, Kluwer, 1993.

You might also like