. . . . . . . . . . . . . . . . . . . . . . . . . .

Olympiad Training for Individual Study

Fundamentals of Inequalities
Evan Chen《陳誼廷》
29 February 2024
BAW-INEQBASIC

》
廷
誼 se
陳 U
n《 al
h e rn
C nt e
n
a , I
Ev I S
y
B O T

OTIS, © Evan Chen, internal use only. Artwork contributed by Alan Cheng.

1
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

§1 Reading
Read §2.1-§2.4 of The OTIS Excerpts. Alternatively, you can use A Brief Introduction to
Olympiad Inequalities, §1, §3.

§1.1 A bit of philosophizing about inequalities

This section is repeated in all inequality units.
Competition inequalities occupy a bit of a strange place because there are sort of two
different, orthogonal forms of difficulty. I call them sharpness and tractability.

§1.1.1 Sharpness
The first is sharpness. Roughly, we say an inequality L ≥ R is “sharp”, “strong”, or

》
“tight” if L and R are close to each other for a lot of inputs. Conversely, if L and R are
far away from each other then we might say the inequality is “weak” or “loose”. This

廷
isn’t strictly formalizable, but a couple simple examples:

誼 se
• a3 + + ≥ 3abc is considered a pretty loose inequality, whereas
b3 c3 a3 + + c3 ≥
b3
a2 b + b2 c + c2 a is less loose. This is because in fact a2 b + b2 c + c2 a ≥ 3abc, always.

陳 U
• Schur’s inequality might be considered “sharper than AM-GM”, because it has

《 al
additional equality cases when a = b and c = 0.

n
e rn
• The inequality (a − b)2 (b − c)2 (c − a)2 ≥ 0 is considered really sharp, because it

h
has lots of equality cases.

C e
• Tongue-in-cheek example: x ≥ x is the sharpest inequality of all. (This is really

t
just saying, equalities are safer to use than estimates, as we’ll soon see.)

n
a , I n
v
§1.1.2 Sharpness continued — blackjack analogy

E I S
From this “sharpness” perspective, proving an inequality L ≥ R is sort of like playing a
game of blackjack, where you want to ideally hit 21, be okay with going a little under,

y T
but never overshoot the threshold.

B O
Let’s give a concrete example.

Example 1.1 (Nesbitt)

For a, b, c > 0 prove that a
b+c + b
c+a + c
a+b ≥ 32 .

Maybe, we start by trying to use Cauchy-Schwarz in some way (I’ll just pick one):
√ √ √
a b c C-S ( a + b + c)2
+ + ≥ .
b+c c+a a+b 2(a + b + c)

Then, it would suffice to prove the right-hand side is at least 32 .

This is like trying to hit in blackjack: we make some estimate. So rather than prove
L ≥ R directly, we are trying to show
C-S ?
L ≥ A≥R

for some intermediate quantity A. However, we don’t know whether the “?” inequality is
even true or not!

2
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

• If it is true, then actually it is a stronger inequality than the original one that have
reduced to proving. (Put another way, every irreversible move you do makes the
inequality you wish to prove sharper.)

• If it is not true, that’s like busting in Blackjack. You’re dead. Try something else.

Exercise 1.2. Show that the above method is “busted”.

Exercise 1.3. Come up with a different Cauchy-Schwarz application that does work.
Let’s summarize the blackjack analogy. Formally, an inequality proof can often be
written as having a chain of relations

L = A0 ≥ A1 ≥ A2 ≥ · · · ≥ Ak = R.

》
Each estimate (hit) makes the inequality tighter. So, if you use sharper estimates, you
are less likely to bust. In particular, equalities will never break anything, so it is good

廷
to do as much exact calculation as possible before making estimates. Moreover, any

誼 se
equality cases in the original inequality must be preserved.
Thus, an inequality might be difficult if it is extremely sharp; these means that most

陳 U
estimates you try to apply end up busting, and thus you are hindered by having few
feasible options. An example is the infamous inequality:

n《 al
Example 1.4 (Vasc)

e rn
For a, b, c > 0 show that (a2 + b2 + c2 )2 ≥ 3(a3 b + b3 c + c3 a).

C h e
This inequality is impervious to many standard methods. It is featured as a walkthrough

t
in ZAY-hardineq.

n
a , I n
§1.1.3 Sharpness advice for beginners

Ev S
If you are new to inequalities, a common mistake is to be too optimistic about which

I
inequalities might be true. That is, beginners tend to underestimate how often busting

y T
happens.

B O
When you are starting out, two defense mechanisms you can try are:

• Trying to err on playing more conservatively, using fewer inequalities and more
equalities when possible;

• More importantly, if you have a claim you think might be true, you should audit it
by plugging in some cases (such as a = b = 1 and c = 0 or a = b = 0 and c = 1)
to see if it actually is true in these situations. This will prevent you from wasting
time entertaining impossibilities.

As you get more experience, you’ll naturally have better instincts for which inequalities
could plausibly be true or not. For example, inequality veterans can tell by looking that

(a2/3 + b2/3 + c2/3 )3 ≥ 9(ab + bc + ca) ∀a, b, c > 0

is probably not true, without even having to plug in values. (See if you can find a simple
counterexample quickly. This is a real-life mistake from Evan.)

3
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

§1.1.4 Tractability
So, why not just always use the sharpest inequalities possible? In the Nesbitt example,
rather than play “let’s guess which Cauchy-Schwarz application will lead to a solution”,
one might argue that one should just expand — there is no risk of busting if you do this.
Exercise 1.5. Fully expand Nesbitt’s inequality and prove it.
This illustrates something strange about competition inequalities. If you have access
to a computer algebra system, then you can try to stick to using super sharp estimates,
or better yet, always use equalities up until the end.
Unfortunately, time and computational power are both precious resources in an
olympiad exam. Therefore, one has to also factor in tractability: whether the expressions
can be manipulated quickly enough to execute a certain solution during an exam.

》
Nesbitt’s inequality is a small enough example that you could expand it easily. In
practice, inequalities will frequently have longer denominators, square roots, or strange

廷
conditions that prevent such brute-force approaches.
Tractability is the other way that an inequality can be difficult even if on paper it is

誼 se
a pretty loose inequality. For example, TSTST 2012/6 is an inequality that is actually
quite weak, but the expressions involved are so awful that few people solved it anyways.

陳 U
《 al
Example 1.6 (TSTST 2012/6)

n
Positive real numbers x, y, z satisfy xyz + xy + yz + zx = x + y + z + 1. Prove that

h e rn
s ! 
x + y + z 5/8
r r
1 + x2 1 + y2 1 + z2

1
≤

e
+ + .

C
3 1+x 1+y 1+z 3

n I nt
a ,
§1.1.5 Sharpness vs. tractability

Ev S
Sharpness and tractability are the yin and yang of olympiad inequalities. When

I
considering different approaches, one is usually weighing the sharpness of the approach,

y T
versus its tractability. Approaches that are both sharp and tractable should be used, if

B O
possible. Barring that, one has to make more judgment calls.
Usually, I try to do estimates which are not that sharp but extremely tractable first,
because these are the quick to verify. If they work, great; if not, they shouldn’t take
much time, and they give me a better sense of how sharp the inequality is (by seeing
“how far off” the weak estimate is). As time progresses, I tend to turn the dial up on
sharpness and down on tractability. But it is hard to give a good general rule, since the
exact parameters depend much on the specific problem. As always, experience is the best
teacher.

§2 Lecture notes
Techniques covered this particular lecture:

• Homogenizing and standard substitutions (such as (x/y, y/z, z/x) to get rid of
the condition abc = 1). See USAMO 2011/1 for a key example of homogenizing.
(The OTIS Excerpts reading covers homogenization in more detail.)

• AM-GM / Power-Mean / Muirhead

4
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

• Cauchy / Hölder

§2.1 AM-GM review (from reading)

Example 2.1 (USAMO 2011/1)

Let a, b, c be positive real numbers such that a2 + b2 + c2 + (a + b + c)2 ≤ 4. Prove
that
ab + 1 bc + 1 ca + 1
2
+ 2
+ ≥ 3.
(a + b) (b + c) (c + a)2

Walkthrough. This is from the OTIS application, so you may remember it. If not, you
should make sure you can solve it easily.
The key is to homogenize the completely artificial condition away (as you always

》
should) and go from there. It’s crucial that you don’t try to be cute about this or take
unnecessary risks: follow the fail-safe mechanical procedure of homogenizing, simplifying,

廷
and seeing what the inequality becomes.

誼 se
Example 2.2

陳 U
Let a, b, c > 0. Prove by AM-GM:

《 al
a2 + b2 + c2 ≥ ab + bc + ca

n
a3 + b3 + c3 ≥ a2 b + b2 c + c2 a.

h e rn
Walkthrough. This should be straightforward. (The first inequality follows by Muirhead

C e
as well, but the second inequality does not, because a2 b + b2 c + c2 a is cyclic, not

t
symmetric.)

n
a , I n
v
Example 2.3 (Australia Training, added by Leonardo Wang)

E S
Suppose that x1 , x2 , . . . , xn are positive real numbers which satisfy x1 x2 . . . xn = 1.

I
If r > s > 0, prove that

y
B O T
xr1 + xr2 + · · · + xrn ≥ xs1 + xs2 + · · · + xsn .

21AUSTRAININGWalkthrough. There are two approaches here. The first approach is by Muirhead:

(a) Homogenize the inequality using x1 x2 . . . xn = 1.

(b) Solve using Muirhead.

Another approach is to directly compare with power mean:

(c) Use power mean to compare the r’th and s’th power mean.

(d) Prove unconditionally that

xr1 + · · · + xrn xs + · · · + xsn

≥ 1 · (x1 . . . xn )k
n n
for some real number k > 0, and conclude.

5
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

§2.2 Problems for lecture

Example 2.4
If abcd = 1 for a, b, c, d > 0, prove that

a4 b + b4 c + c4 d + d4 a ≥ a + b + c + d.

Z1422B10 Walkthrough. There are two possible solutions I know of, one by Hölder and one by
AM-GM. I find the latter much more natural.

(a) Homogenize the inequality to eliminate the condition (while keeping the inequality
fifth-degree).

(b) Fill in the blanks in the following AM-GM:

》 ? · a4 b + ? · b4 c + ? · c4 d + ? · d4 a ≥ a2 bcd.

廷
(c) Cyclically sum to finish.

誼 se
Depending on how you did this, the number 51 might appear.

陳 U
《 al
Example 2.5 (IMO 2001/2)

n
Let a, b, c be positive reals. Prove that

h e rn
a b c
√ +√ +√ ≥ 1.

e
a2 2 2

C
+ 8bc b + 8ca c + 8ab

nt
Walkthrough. There are a few ways to set up, but the general idea is to use Hölder in

n
01IMO2

I
the form

va ,
!p !q !p+q
X a X X

S
√ ? ≥ ?

E I
cyc a2 + 8bc cyc cyc

y
for some choice of weights p and q to eliminate the radicals and get a polynomial

T
inequality.

B O(a) Pick a choice of weights p, q > 0 eliminate the radicals.

(b) Decide on values to fill in the ? above. (You probably want to eliminate the
denominator, i.e. the left sum should be some multiple of a2 + 8bc.)

(c) Try to prove the resulting inequality. Depending on what choices you made in (a)
or (b), this may be relatively easy, or it may be impossible (because the inequality
may not even be true.)

Example 2.6 (IMO 2005/3)

Let x, y, z > 0 satisfy xyz ≥ 1. Prove that

x5 − x2 y5 − y2 z5 − z2
+ + ≥ 0.
x5 + y 2 + z 2 x2 + y 5 + z 2 x2 + y 2 + z 5

05IMO3 Walkthrough.

6
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

(a) By subtracting 3 from both sides and manipulating, rewrite the inequality in a
form that does not have any minus signs, and for which the fractions have constant
numerator.

(b) Show that we can assume WLOG that xyz = 1.

(c) Homogenize the resulting inequality.

(d) EXPAND! (Or try to find a nicer solution if you’re so inclined.)

》
廷
誼 se
陳 U
n《 al
h e rn
C nt e
n
a , I
Ev I S
y
B O T

7
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

§3 Practice Problems
Instructions: Solve [42♣]. If you have time, solve [50♣]. Problems with red weights are mandatory.

It’s psychosomatic. You need a lobotomy. I’ll get a saw.

Calvin in Calvin and Hobbes

83AMO2
[2♣] Problem 1 (USAMO 1983/2). Prove that the roots of

x5 + ax4 + bx3 + cx2 + dx + e = 0

cannot all be real if 2a2 < 5b.

KYLEWU

》
[2♣] Problem 2 (Kyle Wu). For positive real numbers a, b, and c with a + b + c = 1,
prove that

廷
b2 + c2 c2 + a2 a2 + b2 1
+ + ≥ .
1+a 1+b 1+c 2

誼 se
78AMO1
[3♣] Problem 3 (USAMO 1978/1). Given that a, b, c, d, e are real numbers such that

陳 U
a + b + c + d + e = 8 and a2 + b2 + c2 + d2 + e2 = 16, determine the maximum value of e.
ZB2F3EE8
[3♣] Problem 4. For a, b, c > 0 prove that

n《 al c2 + ab a2 + bc b2 + ca

e rn
+ + ≥ a + b + c.
a+b b+c c+a

h
98SLA3
[3♣] Problem 5 (Shortlist 1998 A3, added by Leonardo Wang). Let x, y, z be positive

C t e
real numbers which satisfy xyz = 1. Show that

n I n
x3

a ,
X 3
≥ .

v
cyc
(1 + y)(1 + z) 4

E S
12JMO3

I
[3♣] Problem 6 (JMO 2012/3). For a, b, c > 0 prove that

y T
a3 + 3b3 b3 + 3c3 c3 + 3a3 2

B O
+ + ≥ (a2 + b2 + c2 ).
5a + b 5b + c 5c + a 3
97AMO5
[3♣] Problem 7 (USAMO 1997/5). If a, b, c > 0 prove that
1 1 1 1
+ 3 + 3 ≤ .
a3 3 3 3
+ b + abc b + c + abc c + a + abc abc
00IMO2
[3♣] Problem 8 (IMO 2000/2). Let a, b, c be positive real numbers with abc = 1. Show
that    
1 1 1
a−1+ b−1+ c−1+ ≤ 1.
b c a
14TWNQ1J1
[3♣] Problem 9 (Taiwan Quiz 2014). For a, b, c > 0 prove that
r
√
3
3 3
3 a + b + c
3
3(a + b + c) ≥ 8 abc + .
3
LINLIU
[3♣] Problem 10 (Lin Liu). Prove that for any a, b, c > 0,
  X 3/2 ! X
a+b+c b
√ ≥ a(2b − a).
3 cyc
a cyc

8
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

Z148E702
[3♣] Problem 11 (Latvia Baltic Way TST 2019, added by Kims Georgs Pavlovs). Prove
that for all positive real numbers a, b, c with a1 + 1b + 1c = 1 the following inequality holds:
9 9abc
3(ab + bc + ca) + ≤ + 2(a2 + b2 + c2 ) + 1.
a+b+c a+b+c
MRO459
[2♣] Problem 12 (Mathematical Reflections O-459). Let a, b, x be real numbers such
that
(4a2 b2 + 1)x2 + 9(a2 + b2 ) ≤ 2018.
Prove that
20(4ab + 1)x + 9(a + b) ≤ 2018.
98IRN
[5♣] Problem 13 (Iran 1998). Let x, y, z be real numbers greater than 1. Prove that if

》
x + y + z = 2 then
1 1 1

廷
√ p √ √
x − 1 + y − 1 + z − 1 ≤ x + y + z.
12CGMO1

誼 se
[3♣] Problem 14 (CGMO 2012/1). Let n be a positive integer, and let a1 , a2 , . . . , an
be nonnegative real numbers. Show that

陳 U
1 a1 a1 a2 · · · an−1
+ + ··· + ≤ 1.

《 al
1 + a1 (1 + a1 )(1 + a2 ) (1 + a1 )(1 + a2 ) · · · (1 + an )

n
Moreover, determine all the cases of equality.

e rn
12IMO2

h
[5♣] Required Problem 15 (IMO 2012/2). Let a2 , a3 , . . . , an be positive reals with
product 1, where n ≥ 3. Show that

C nt e (1 + a2 )2 (1 + a3 )3 . . . (1 + an )n > nn .

n I
03ELMO4

a ,
[5♣] Required Problem 16 (ELMO 2003, Po-Ru Loh). Let x, y, z > 1 be real numbers

v
such that

E S
1 1 1

I
+ + = 1.
x2 − 1 y 2 − 1 z 2 − 1

y T
Prove that

B O
1 1 1
+ + ≤ 1.
x+1 y+1 z+1
04SLA5
[9♣] Problem 17 (Shortlist 2004 A5). If a, b, c are three positive real numbers such
that ab + bc + ca = 1, prove that
r r r
3 1 3 1 3 1 1
+ 6b + + 6c + + 6a ≤ .
a b c abc
04IMO4
[5♣] Required Problem 18 (IMO 2004/4). Let n ≥ 3 be an integer and t1 , t2 , . . . , tn
positive real numbers such that
 
2 1 1 1
n + 1 > (t1 + t2 + · · · + tn ) + + ··· + .
t1 t2 tn
Show that ti , tj , tk are the sides of a triangle for all i, j, k with 1 ≤ i < j < k ≤ n.
18RUS112
[3♣] Problem 19 (Russia 2018, added by Prithuj Sarkar). Let n ≥ 2 and let x1 , x2 ,
. . . , xn be positive real numbers. Prove that
1 + x21 1 + x22 1 + x2n
+ + ··· + ≥ n.
1 + x1 x2 1 + x2 x3 1 + xn x1

9
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

Z93A61C5
[3♣] Problem 20 (Latvia Baltic Way TST 2020, added by Kims Georgs Pavlovs). Prove
that for positive reals a, b, c satisfying a + b + c = 3 the following inequality holds:
a b c
3
+ 3
+ ≥ 1.
1 + 2b 1 + 2c 1 + 2a3
95IMO2
[5♣] Problem 21 (IMO 1995/2). For positive real numbers a, b, c satisfying abc = 1,
prove that
1 1 1 3
+ + ≥ .
a3 (b + c) b3 (c + a) c3 (a + b) 2
11MOPR42
[9♣] Required Problem 22 (MOP 2011 R4.2). For positive real numbers a, b, c with
a + b + c = 3 prove that

》
√
r r r
a3 + b3 b3 + c3 c3 + a3 3
+ + + 9 abc ≤ 12.

廷
a+b b+c c+a
Z2606BB8
[2♣] Problem 23 (Evan’s driving test 2012). Prove that if a, b, c > 0 satisfy a3 +b3 +c3 =

誼 se
3 then

陳 U
a3 (b + c) + b3 (c + a) + c3 (a + b) ≤ 6.

《 al
Remark (Etymology). I remember this problem all my life because I saw it on AoPS right

n
before my driver’s license test while surfing on my phone. The good news is that I solved it

e rn
in my head during that driving test. The bad news is that it caused me to run a red light

h
and not even notice until my examiner grilled me at the end of said driving test.
(I failed my second try as well by running a stop sign and colliding with the curb. I hate

C t e
driving so much.)

n
a , I n
02AMO2
[5♣] Problem 24 (USAMO 2002/2). Let ABC be a triangle such that

Ev I S
A 2 B 2 C 2
       2
6s
cot + 2 cot + 3 cot = ,

y T
2 2 2 7r

B O
where s and r denote its semiperimeter and its inradius, respectively. Prove that triangle
ABC is similar to a triangle T whose side lengths are all positive integers with no common
divisors and determine these integers.

[1♣] Mini Survey. Fill out feedback on the OTIS-WEB portal when submitting this
problem set. Any thoughts on problems (e.g. especially nice, instructive, easy, etc.) or
overall comments on the unit are welcome.
In addition, if you have any suggestions for problems to add, or want to write hints for
one problem you really liked, please do so in the ARCH system!

The maximum number of [♣] for this unit is [93♣], including the mini-survey.

10
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

§4 Solutions to the walkthroughs

§4.1 Solution 2.1, USAMO 2011/1
The condition becomes 2 ≥ a2 + b2 + c2 + ab + bc + ca. Therefore,
X 2ab + 2 X 2ab + (a2 + b2 + c2 + ab + bc + ca)
≥
cyc
(a + b)2 cyc
(a + b)2
X (a + b)2 + (c + a)(c + b)
=
cyc
(a + b)2
X (c + a)(c + b)
=3+
cyc
(a + b)2

》
v
(c + a)(c + b)
uY
≥ 3 + 3t
u
3 =3+3=6

廷
cyc
(a + b)2

with the last line by AM-GM. This completes the proof.

誼 se
§4.2 Solution 2.3, Australia Training, added by Leonardo Wang

陳 U
This problem and solution were suggested by Leonardo Wang.

《 al
Here are two approaches.

n
e rn
¶ Solution using Muirhead Homogenize both sides of the inequality by exploiting the

h
condition that x1 x2 . . . xn = 1, and apply Muirhead since it’s symmetric.

C t e
¶ Solution using power mean inequality By Power Mean,

n
n
a , I
r r
r r s s
r x1 + · · · + xn s x1 + · · · + xn

v
≥
n n

E I S
r
r r x1 + · · · + xsn xs1 + · · · + xsn s −1
s

x1 + · · · + xn
=⇒ ≥ .

y T
n n n

B O
xs +...xs
Notice that by condition, rs − 1 > 0; also by AM-GM 1 n n ≥ n xs1 . . . xsn = 1, which
p

finishes off the problem.

§4.3 Solution 2.4

We present two solutions.

¶ First solution by weighted AM-GM By AM-GM,

23a4 b + 7b4 c + 11c4 d + 10d4 a √
51
≥ a102 b51 c51 d51 = a2 bcd = a.
51
You could find this solution by searching for weights w, x, y, z with sum 1 for which
w · a4 b + x · b4 c + y · c4 d + z · d4 a ≥ a2 bcd holds; this amounts to the system of equations
4w + z = 2
4x + w = 1
4y + x = 1
4z + y = 1
which when solved gives the weights above.

11
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

¶ Second solution by Hölder By Hölder,

! ! ! ! !4
X X X X X√
4
4
a b a c d ≥ a4 · abcd = (a + b + c + d)4 .
cyc cyc cyc cyc cyc

Thus done.

§4.4 Solution 2.5, IMO 2001/2

By Holder, we have
!2 !
X a X
√ a(a2 + 8bc) ≥ (a + b + c)3 .
2
a + 8bc

》
cyc cyc

So it suffices to show (a + b + c)3 ≥ a3 + b3 + c3 + 24abc which is clear by expanding.

廷
§4.5 Solution 2.6, IMO 2005/3

誼 se
Negating both sides and adding 3 eliminates the minus signs:

陳 U
X 1 3

《 al
≤ 2 .
cyc
x5 2
+y +z 2 x + y2 + z2

n
e rn
Thus we only need to consider the case xyz = 1.

h
Direct expansion and Muirhead works now! As advertised, once we show it suffices to

e
analyze if xyz = 1 the inequality becomes more economically written as

n C nt
?
X
x2 (x2 − yz)(y 4 + x3 z + xz 3 )(z 4 + x3 y + xy 3 ) ≥ 0.

I
S=

a ,
cyc

v S
So, clearing all the denominators gives

y E T I
X
x2 (x2 − yz) y 4 z 4 + x3 y 5 + xy 7 + x3 z 5 + x6 yz + x4 y 3 z + xz 7 + x4 yz 3 + x2 y 3 z 3
 
S=
cyc

B O
X
x4 y 4 z 4 + x7 y 5 + x5 y 7 + x7 z 5 + x10 yz + x8 y 3 z + x5 z 7 + x8 yz 3 + x6 y 3 z 3

=
cyc
X
x2 y 5 z 5 + x5 y 6 z + x3 y 8 z + x5 yz 6 + x8 y 2 z 2 + x6 y 4 z 2 + x3 yz 8 + x6 y 2 z 4 + x4 y 4 z 4

−
cyc
X
x7 y 5 + x5 y 7 + x7 z 5 + x10 yz + x5 z 7 + x6 y 3 z 3

=
cyc
X
x2 y 5 z 5 + x5 y 6 z + x5 yz 6 + x8 y 2 z 2 + x6 y 4 z 2 + x6 y 2 z 4

−
cyc

In other words we need to show

X 1 10 1 6 3 3
 X
1 8 2 2 1 5 5 2

7 5 6 4 2 6 5
2x y + x yz + x y z ≥ x y z + x y z +x y z +x y z .
sym
2 2 sym
2 2

which follows by summing

X x10 yz + x6 y 3 z 3 X
≥ x8 y 2 z 2
sym
2 sym

12
 Evan Chen《陳誼廷》 (OTIS, updated 2024-02-29) Fundamentals of Inequalities

1X 8 2 2 1X 6 4 2
x y z ≥ x y z
2 sym 2 sym
1X 7 5 1X 5 5 2
x y ≥ x y z
2 sym 2 sym
1X 7 5 1X 6 4 2
x y ≥ x y z
2 sym 2 sym
X X
x7 y 5 ≥ x6 y 5 z.
sym sym

The first line here comes from AM-GM, the rest come from Muirhead.

Remark. More elegant approach is to use Cauchy in the form

》
1 x−1 + y 2 + z 2
≤ 2

廷
.
x5 2
+y +z 2 (x + y 2 + z 2 )2

誼 se
陳 U
n《 al
h e rn
C nt e
n
a , I
Ev I S
y
B O T

13