The pqr - method and its application in inequalities

A.Doledenok, M.Fadin, A.Menshchikov, A.Semchankau

23 июля 2017 г.

Содержание
1 Introduction 1

2 What is the pqr – method 1

3 Unusual conditions 4

4 Special Cases 5

5 n variables 7

6 Basises 8

7 Appendix 9

1 Introduction
A bit of history. As far as we know, pqr-method performed originally in Vietnam, where it was known as ABC-method
(i.e. abstract-concreteness method). That is the earliest (but seems to be not the origin) paper we found ([1]). Later
this method of solving inequalities arrived to Europe, where it’s been popularized by Michael Rozenberg (check out
[2]). Also, the abc-method became a uvw-method, and the use of method was modified by M. Rozenberg. It seems,
that [3] is the most popular introduction to uvw-method for today.
In this article we’re going to give the introduction to the pqr-method (section 2, that is reworked uvw-method) and
give some generalizations (sections 3 − 6). We will demonstrate all ideas on examples. This paper is based on Summer
Conference of Tournament of Towns project, which we proposed this year. You can find all materials, problems and
solutions at
http://www.turgor.ru/lktg/2016/3/index.html. If you can’t find some solution in English version, try Russian.

2 What is the pqr – method

We call expression f (a, b, c) symmetric, if its value does not change after swapping any two variables (i.e. f (a, b, c) =
f (b, a, c) = · · · ). If f is a polynomial, than it can be rewritten in terms of p, q, r (prove that) by convinience:

p = a + b + c, q = ab + bc + ca, r = abc.

Obviously, a, b, c are roots of the cubic equation x3 − px2 + qx − r = 0 (two of a, b, c can be equal or complex). You
can find some useful expressions in Appendix. Rewriting in terms of p, q, r can be helpful while solving inequalities,
but there is a problem: for example, obvious inequality a2 + b2 + c2 ≥ 0 will be transformed into p2 ≥ 2q, which is not
true for some random p and q. So, we have a problem: if we know values of p, q, r, how to check that all of a, b, c are real?

Theorem. Numbers a, b, c are real ⇔ numbers p, q, r are real and T (p, q, r) ≥ 0, where

T (p, q, r) = −4p3 r + p2 q 2 + 18pqr − 4q 3 − 27r2 .

Proof. We‘ll prove that numbers a, b, c are real ⇔ p, q, r are real and (a − b)2 (b − c)2 (c − a)2 ≥ 0.
LHS ⇒ RHS: obvious
RHS ⇒ LHS: We know that p, q, r are real. It‘s well known that polynomial x3 − px2 + qx − r = 0 has either exactly

1
three real roots or exactly one. If it has three roots theorem is obvious. Otherwise, there is exactly one real root
(w.l.o.g. we say that a is real). Then 2 other roots are complex (and conjugated). Let b = x + yi and c = x − yi.
(y 6= 0). It is clear that

(a − b)2 (b − c)2 (c − a)2 = ((a − b)(a − c))2 (b − c)2 = = (a2 − (b + c)a + bc)2 (b − c)2 =

= (a2 − 2xa + x2 + y 2 )(−2yi)2 = = −4y 2 ((a − x)2 + y 2 ) < 0,

and we are done.
Let l12 = a2 b + b2 c + c2 a, l21 = ab2 + bc2 + ca2 . Then:

(a − b)2 (b − c)2 (c − a)2 = (l12 − l21 )2 = (l12 + l21 )2 − 4l12 l21 =

= (a2 b + b2 c + c2 a + b2 a + c2 b + a2 c)2 − 4((a3 b3 + b3 c3 + c3 a3 ) + abc(a3 + b3 + c3 ) + 3(abc)2 ) =

= (pq − 3r)2 − 4((q 3 − 3pqr + 3r2 ) + r(p3 − 3pq + 3r) + 3r2 ) = − 4p3 r + p2 q 2 + 18pqr − 4q 3 − 27r2 = T (p, q, r).
Q.E.D.

It is clear that T (p, q, r) = 0 if and only if some of a, b, c are equal because of relation T (p, q, r) = (a−b)2 (b−c)2 (c−a)2
The reader can find this formula for T (p, q, r) too complicated; almost everywhere in this article we only use that T
is a polynomial.
For example, what are possible values of abc, when a+b+c = 9, ab+bc+ca = 24? T (9, 24, r) ≥ 0 ⇔ −27(r−16)(r−20) ≥
0 ⇔ r ∈ [16, 20]. r = 16 ⇔ {a, b, c} = {1, 4, 4}, r = 20 ⇔ {a, b, c} = {2, 2, 5}.
For solving inequalities, we need next result:

Non-negavity lemma. If p, q, r ≥ 0 and T (p, q, r) ≥ 0 then a, b, c are real and non-negative.

Proof. It’s easy to see that x3 − px2 + qx − r = 0 can’t have negative roots, so a, b, c ≥ 0. Q.E.D.

Obviously, if there are zeros among non-negative p, q, r then there are zeros among a, b, c.

A triple (p, q, r) is called acceptable if p, q, r ≥ 0 and T (p, q, r) ≥ 0, i.e., if the non-negativity lemma can be applied.
In other words, polynomial x3 − px2 + qx − r = 0 has three (perhaps, equal) real non-negative roots. The following
lemmas are very useful for solving inequalities:

r-lemma. Fix some values p = p0 and q = q0 such that there exists at least one value of r for which the triple
(p0 , q0 , r) is acceptable. Then there exists such triple with the minimal possible value of r and it corresponds to a
triple (a, b, c) in which either two numbers are equal, or abc = 0. Moreover, such triple with the maximal value of r
corresponds to a triple (a, b, c) containing two equal numbers.

q-lemma. Fix some values p = p0 and r = r0 such that there exists at least one value of q for which the triple
(p0 , q, r0 ) is acceptable. Then there exist such triples with minimal and maximal possible values of q and each of them
corresponds to a triple (a, b, c) containing two equal numbers.

p-lemma. Fix some values q = q0 and r = r0 > 0 such that there exists at least one value of p for which the
triple (p, q0 , r0 ) is acceptable. Then there exist such triples with minimal and maximal possible values of p and each
of them corresponds to a triple (a, b, c) containing two equal numbers.
Does the same statement holds true when r0 = 0?
Proofs are quite similar to each other.
Proof of r-lemma. For triple (p, q, r) to be acceptable it is necessary to have T (p0 , q0 , r) ≥ 0 and r ≥ 0 (p0 , q0
are non-negative, by the condition). From fact that T (p0 , q0 , r) is a quadratic function in r with non-negative leading
coefficient follows that inequality T (p0 , q0 , r) ≥ 0 defines a segment. Therefore the maximal value of r is right end-point
of that segment and the minimal one is either left end-point or 0. Since T (p0 , q0 , r) = 0 in the end-points, it follows
that two numbers of a, b, c are equal (T (p, q, r) = (a − b)2 (b − c)2 (c − a)2 ). If r = 0, then one number of a, b, c equals
0.
Proof of q-lemma. T (p0 , q, r0 ) is a third degree polynomial in q with negative leading coefficient. Constant term
equals −4p30 r0 − 27r02 , so it is negative. Therefore T (p0 , q, r0 ) ≥ 0 defines a union of a segment with non-negative
end-points and a ray (−∞, q 0 ], where q 0 ≤ 0. The rest is trivial.

2
 √
Proof of p-lemma. The proof is similar to the proof of q-lemma when r0 6= 0. If r0 = 0, then p ∈ [2 q0 , +∞), i.e. p
is unbounded.
1. Example (Iran TST 1996). For all a, b, c ≥ 0 with ab + bc + ca 6= 0 prove:
 
1 1 1 9
f (a, b, c) = (ab + bc + ca) + + ≥ .
(a + b)2 (b + c)2 (c + a)2 4

Proof. The desired inequality can be written as

4qpr + q(p4 − 2p2 q + q 2 ) 9

2
≥
(pq − r) 4

Fix p and q. Since r ≤ pq, it follows that left hand side attains its minimal value when r attains either maximal or
minimal value.
More formally, let’s suppose that it doesn’t hold for some a, b, c with a + b + c = p0 , ab + bc + ca = q0 , abc = r. Let’s
take such a0 , b0 , c0 so that a0 + b0 + c0 = p0 , a0 b0 + b0 c0 + c0 a0 = q0 , a0 b0 c0 = rmin , where rmin is minimal possible. Then
it’s easy to see, that f (a, b, c) ≥ f (a0 , b0 , c0 ).
Anyway, we may consider, that now r is minimal possible for fixed p, q, so there are two cases:

• If a = 0, then the desired inequality can be written as

(b − c)2 (b2 + bc + c2 )
≥ 0.
2bc(b + c)2

Equality is attained when a = 0, b = c.

• If a = b, then the desired inequality can be written as

t(t − 1)2 ≥ 0,

where t = cb . Equality is attained when a = b = c. Q.E.D.

You can prove next problems, using the same ideas:

2. Let a, b, c be non-negative real numbers such that a + b + c = 3. Prove that
1 1 1 3
+ + ≤ .
9 − ab 9 − bc 9 − ca 8

3. Let P be a symmetric polynomial of degree not greater than 5. Prove that if P (a, a, c) ≥ 0 and P (0, b, c) ≥ 0 for
all non-negative real numbers a, b, c, then P (a, b, c) ≥ 0 for all non-negative real numbers a, b, c.
4. (Russia TST, 2015) Let a, b, c be non-negative real numbers such that 1 + a + b + c = 2abc. Prove that

ab bc ca 3
+ + ≥ .
1+a+b 1+b+c 1+c+a 2

5. Let a, b, c be non-negative real numbers such that ab + bc + ca 6= 0. Prove that

1 1 1 10
+ 2 + 2 ≥ .
a2 + b2 b + c2 c + a2 (a + b + c)2

6. a) Let P (a, b, c) be a homogeneous symmetric polynomial of degree not greater than 8. Find an algorithm checking
whether P is non-negative when a, b, c are non-negative. We assume that we are ale to find extrema and zeroes of an
arbitrary function in one variable.
b) Find an analogous algorithm for a homogeneous symmetric polynomial of degree not greater than 11.
c) Find an analogous algorithm for a homogeneous symmetric polynomial of degree not greater than 17.
Note. Check out [4] for some interesting implementations of this algorithm in solving problems.

3
3 Unusual conditions
Some inequalities contain conditions like a, b, c ≥ 1, and it isn‘t correct to use pqr - method in this case. Let’s solve
next problem:
7. Find conditions on numbers p, q, r necessary and sufficient for
a) numbers a, b, c to be not less than 1;
b) numbers a, b, c to be side lengths of a triangle (perhaps degenerate);
c) non-negative real numbers a, b, c to satisfy 2 min(a, b, c) ≥ max(a, b, c).
Solution. a) At first, a, b, c are real, so p, q, r must be real and T (p, q, r) ≥ 0 must hold. At second, we should use
non-negativity lemma for numbers a − 1, b − 1, c − 1 :

(a − 1) + (b − 1) + (c − 1) ≥ 0 ⇔ p ≥ 3,
(a − 1)(b − 1) + (b − 1)(c − 1) + (c − 1)(a − 1) ≥ 0 ⇔ q − 2p + 3 ≥ 0,
(a − 1)(b − 1)(c − 1) ≥ 0 ⇔ r − q + p − 1 ≥ 0.
b), c) Use the same idea.

Now we’ll demonstrate how to solve a problem with unusual condition:

8. Let a, b, c be real numbers such that a, b, c ∈ 31 , 3 . Prove that
 

a b c 7
+ + ≥ .
a+b b+c c+a 5

Proof. Perform the substitution x = ab , y = cb , z = ac . Then inequality is equivalent to

1 1 1 7
+ + ≥ ,
1+x 1+y 1+z 5
1 
where xyz = 1, and x, y, z ∈ 9, 9 . If we prove this inequality, than old inequality holds too. In terms of p, q, r that
is equivalent to:
8 + 3p − 2q − 7r ≥ 0.
2q0 −1
Let’s fix q = q0 и r = r0 = 1, then LHS is equivalent to p ≥ 3 .
Let’s discover all possible values of p.
• x, y, z ≥ 0 and T (p, q0 , r0 ) ≥ 0 ⇔ p ∈ [pmin , pmax ].
1
• x, y, z ≥ 9 ⇔ is equivalent for x − 19 , y − 19 , z − 1
9 to be non-negative, so, we have

1 2 1 1 1 1
p≥ , q0 − p + ≥ 0, r0 − q0 + p − ≥ 0,
3 9 27 3 27 729
which gives      
1 9 1 1
p∈ , +∞ ∩ −∞, q0 + ∩ 9q0 − 27 + , +∞
3 2 6 27
We wouldn’t use use this awful formulas, just conclude that x, y, z ≥ 1
9 is equivalent to p ∈ [p0min , p0max ].
• x, y, z ≥ 9 ⇔ 9 − x, 9 − y, 9 − z ≥ 0 and using non-negativity lemma gives us p ∈ [p00min , p00max ].

So, we conclude, that p ∈ [pmin , pmax ] ∩ [p0min , p0max ] ∩ [p00min , p00max ] for fixed q = q0 , r = r0 = 1.

Let’s return to inequality p ≥ 2q03−1 . We just have to check lower bound of p. From previous discussion we have,
that if p is minimal possible, then either T (p, q, r) = 0, or some of x − 91 , y − 19 , z − 19 is zero, or some of 9 − x, 9 − y, 9 − z
is zero, so we have to check one of cases:

• p = pmin ⇔ T (p, q0 , r0 ) = 0. Thus we have x = y and inequality goes to

(3 − x)(2x2 − 2x + 1) ≥ 0,
1 1
which is true cause x ≤ 3. The case of equality is x = y = 3, z = 9, i.e. a = 3, b = 1, c = 3 and cyclic
permutations.

4
 • p = p0min which means that there exist a zero among x − 91 , y − 19 , z − 19 . WLOG let x = 1
9. Using z = 9
y,
inequality is equivalent to (y − 3)2 ≥ 0. That gives the same case of equality.
• p = p00min which means that there exist a zero among 9 − x, 9 − y, 9 − z. WLOG x = 9. Using z = 1
9y , we get

−27y 2 + 50y − 3 ≥ 0,

which is true cause y ≤ 1. Q.E.D.

9. Let a, b, c be real numbers such that a, b, c ≥ 1 and a + b + c = 9. Prove that

√ √ √ √
ab + bc + ca ≥ a + b + c.

10. Let a, b, c be side lengths of a triangle. Prove that

   
1 1 1 a b c
(a + b + c) + + ≥6 + + .
a b c b+c a+c b+a

4 Special Cases
Consider an inequality: Let a, b, c be non-negative numbers such that a2 + b2 + c2 + abc = 4. Prove that
p
a + b + c > 2 + abc(4 − a − b − c).

Unfortunately, pqr-lemmas don’t help here. Indeed, if we fix two numbers of the triple p, q, r, then the third one
will almost certainly be uniquely defined. What is to be done?

2
Lets fix only r = r0 ≥ 0. Condition says that q = p −4+r 2
0
= h(p) i.e q is some polynomial of p. We need to
p
prove that f (p) = p − 2 − r0 (4 − p) ≥ 0. It is obvious that for any fixed r = r0 f is monotonic thus we only
2
need to check inequality when p is maximal (with condition q = p −4+r 2
0
and for each fixed r0 ). So we need to
find out when p attains maximum. We know that (p, q, r) is allowable if and only if p, q, r, T (p, q, r) ≥ 0 which is
p, h(p), r0 , T (p, h(p), r0 ) ≥ 0.By definition r0 ≥ 0 and p, h(p), T (p, h(p), r0 ) are all polynomials of p. It is well-known
that if Q is polynimal then {x ∈ R|Q(x) ≥ 0} is union of finite number of non-intersecting segments and closed rays.
Then {x ∈ R|x ≥ 0, h(x) ≥ 0, T (x, h(x), r0 ) ≥ 0} is an intersection of three finite unions of closed rays and segments
which means that it is finite union of segments and closed rays itself. Since it is limited it is finite union of segments.
It is clear that each edge in this union is some edge in one of three original unions.
Fix one number of the numbers p, q, r (to be precise, r). Then our condition will describe the relationship between
the remaining two numbers. If we express one of them by the other, then our condition states that the first elementary
symmetric polynomial (to be precise, p) is a function of the second one (to be precise, q). Thus, the triple (p, q, r) is
uniquely determined by q. This means that our inequality can be naturally written as h(q) > 0, where h is a function
of real non-negative variable. If h is monotonic or concave, then it suffices to prove our inequality only for extremal q.
So, the problem is reduced to the problem of finding extremal q (when condition and r are fixed).

Let us spend some time setting the scene.

Let n denote a set of ordered groups of n real numbers. For small n the set n has a natural geometric interpretation.
Namely, 1 is a line, 2 is a plane. Hence we say that the elements of n are points.
p
The distance between two points x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈n is defined by ||x−y|| = (x1 − y1 )2 + . . . + (xn − yn )2 .
In particular, for n = 1 distance is equal to the modular of the difference between corresponding numbers, for n = 2
distance is the well-known distance between points.

Let M be a subset of n . The function f : M → is continuous in the point x ∈ M if for any ε > 0 there exists
δ = δ(x, ε) such that if for y ∈n holds ||x − y|| < δ, then |f (x) − f (y)| < ε.

Example. Let us prove that the function f (x) = x2 is continuous for any x ∈. Note that

|x2 − y 2 | = |x − y| · |(y − x) + 2x| 6 |x − y| · (|x − y| + |2x|) < δ 2 + 2|x|δ.

5
 ε
Suppose δ = min( 2ε , 4|x| ). Since both items are not greater than 2ε , it follows that |x2 − y 2 | < ε.
p

You may use results of the next three problems in the sequel without proof.
√
11. Prove that the function f (x) = x1 is continuous for any x ∈ (0, +∞), the function f (x) = x is continuous for
any x ∈ (0, +∞), the function f (x, y) = x + y is continuous for any (x, y) ∈2 .
12. Let the function g(x) be continuous for any x ∈ M ⊂n , let range of g(x) belongs to N ⊂. Let the real-valued
function f (y) be continuous for any y ∈ N . Prove that the function f (g(x)) is continuous for any x ∈ M .
13. a) Prove that the polynomial P (x) in one variable is continuous for any x ∈.
b) Prove that the polynomial P (x1 , . . . , xn ) in n variables is continuous for any (x1 , . . . , xn ) ∈n .

We say that the real numbers a, b, c obey the symmetric relation G, if the equality g(p, q, r) = 0 holds, where g
is a continuous function in 3 variables.
14. Let the variables a, b, c obey the symmetric condition G. Fix some value r > 0 such that there exist p, q > 0 for
which the triple (p, q, r) is acceptable and G(p, q, r) = 0. Assume that the condition G is equivalent to p = f (q) (while
r is fixed), where f is a function defined below, and a set of admissible q is bounded. Prove that q attains its maximal
and minimal value. Moreover, in both cases two variables of a, b, c are equal when
a) f is a linear function;
b) f is a polynomial.
15. Let the variables a, b, c obey the symmetric condition G. Let (x, y, z) be an arbitrary permutation of (p, q, r).
Fix some value z > 0 such that there exist x, y > 0 for which the triple (p, q, r) is acceptable and G(p, q, r) = 0.
Assume that the condition G is equivalent to x = f (y) (while z is fixed), where f is a polynomial, and a set of
admissible y is bounded. Prove that y attains its maximal and minimal values. Moreover, if z = r, then the equality
(a − b)(b − c)(c − a) = 0 holds for any extreme point, otherwise abc(a − b)(b − c)(c − a) = 0 holds for any extreme
point.

Remark. Let I be a union of segment containing admissible y. Statement of the previous problem is also true
when f is continuous function for any point of I. If you solve the previous problem, you may use this fact in the sequel
without proof.

Example. Let a, b, c be non-negative numbers such that a2 + b2 + c2 + abc = 4. Prove that

p
a + b + c > 2 + abc(4 − a − b − c).

Proof. The desired inequality can be written as

p
p>2+ r(4 − p), (∗)
2
while the relation a2 + b2 + c2 + abc = 4 can be written as p −4+r
2 = q. If variable r is fixed, we can write the desired
p
inequality as f (p) > 0, where f (x) = x − 2 − r(4 − x) is a monotonic function. Therefore, it suffices to check (∗)
for the smallest value of p. From the problem 54 it follows that it suffices to check the desired inequality for a = b. If
a = b, then c = 2 − a2 . The desired inequality can be written as

a4 (a − 1)2 > 0.

Equality holds if a = b = c = 1; a = b = 0, c = 2; a = c = 0, b = 2; b = c = 0, a = 2.
16. Let a, b, c be non-negative real numbers such that (a + b + c)( a1 + 1
b + 1c ) = 10. Prove that

9 a2 + b2 + c2 6
6 6 .
8 ab + bc + ca 5

1 1
17. Let a, b, c be positive real numbers such that a + b + c = a + b + 1c . Prove that

(a + b + c)(1 + abc) > 6.

18. Let a, b, c be non-negative real numbers such that ab + bc + ca = a3 + b3 + c3 . Prove that

ab + bc + ca > a2 b2 + b2 c2 + c2 a2 .

6
19. Let a, b, c be non-negative real numbers, p, q, r are defined naturally. Prove that if q + r = 4, then
p3 − 27r > 7(p2 − 3q).

20. Let a, b, c be positive real numbers such that a2 + b2 + c2 = ab + bc + ca + (abc − 1)2 . Prove that
ab + bc + ca + 3 > 2(a + b + c).

5 n variables
Up to this point we considered symmetric inequalities in 3 variables. What can we say if there are more than 3 variables?

The ball (the sphere) in n of radius r and center x is a set of points y such that ||x − y|| 6 r (||x − y|| = r,
respectively). In particular, if n = 2, then ball is a cirсle and sphere is a circumference.

Consider a sequence z1 , z2 , . . . in n . We say that this sequence converges to the point z, if for any ε > 0 the exterior
of the ball of radius ε and center z contains only finitely many elements of this sequence.

1
Example. Since there are only finitely many elements of the sequence zn = n, where n ∈, in the exterior of the
segment [−ε, ε], it follows that zn converges to z = 0.

Consider a set of points M in n . The point z ∈ M is called a limit point of M if there exists a sequence of points
z1 , z2 , . . ., where zi ∈ M and zi 6= z, converging to z. We say that the set M in n is closed, if it contains all its limit
points.

Example. Since 0 is a limit point of the interval (0, 1) but does not belong to it, this interval is not a closed
set. Conversely, the segment [0, 1] is a closed set.

We say that the set M in n is bounded, if there exists constant C such that for any x = (x1 , x2 , . . . , xn ) ∈ M
and for any i, 1 6 i 6 n, |xi | < C. The set M in n is called a compact set if M is closed and bounded.
21. Is the segment [0, 1] a compact set? The ray [0, +∞)? The point (1, 1, . . . , 1)? The ball of radius 1 and center
(1, 1, . . . , 1)? The same ball without the point (1, 1, . . . , 1, 0)? The sphere of radius 1 and center (1, 1, . . . , 1)?

22. Consider a number of relations

P1 (x1 , . . . , xn ) = 0, . . . , Ps (x1 , . . . , xn ) = 0, Ps+1 (x1 , . . . , xn ) > 0, . . . , Pm (x1 , . . . , xn ) > 0,
where m ∈, s ∈0 , P1 , . . . , Pm are polynomials in n variables. Let M be a set of points obeying this relations. Prove
that if M is bounded, then M is a compact set.

We need the following classical theorem of calculus.

The Weierstrass theorem. Continuous function on a compact set attains its maximum and minimum.

Example. The function x2 on the segment [0, 1] attains its maximum and minimum. However the Weierstrass theorem
doesn’t imply an analogous statement for x2 on the half-interval (0, 1].

Example. Let a, b, c, d be non-negative real numbers such that a + b + c + d = 4. Prove that

3(a2 + b2 + c2 + d2 ) + 4abcd > 16.
Proof. It is easy to see the set {a, b, c, d > 0, a + b + c + d = 4} is a compact — it’s closed and bounded. By definition,
put
f (a, b, c, d) = 3(a2 + b2 + c2 + d2 ) + 4abcd.
It is easy to see f is a continuous function. By Weierstrass theorem we may find a point (a0 , b0 , c0 , d0 ) where f
attains its minimal value. Fix d = d0 > 0. By definition, put p = a + b + c, q = ab + bc + ac, r = abc. Consider the
function
h(a, b, c) = f (a, b, c, d0 ) = 3(a2 + b2 + c2 + d20 ) + 4abcd0 = 3p2 − 6q + 4d0 r + 3d20 =

7
 = 3p2 − 6q + 4d0 r + 3d20 .
Fix p and r. Since h linear in q, it follows that h attains its minimum only when (a − b)(b − c)(c − a) = 0, so we may
conclude that (a0 − b0 )(b0 − c0 )(c0 − a0 ) = 0. That’s why we can state there are less than 3 different numbers among
a, b, c, d.

• Considering a = b and c = d = 2 − a, we obtain

4(a − 1)2 ((a − 1)2 + 1) > 0.

Equality holds if a = b = c = d = 1.

• Considering a = b = c and d = 4 − 3a, we obtain

4(a − 1)2 (4 − 3a)(a + 2) > 0.

Since a 6 43 , if follows that inequality holds true. Equality holds if a = b = c = 43 , d = 0; a = b = d = 43 , c = 0;

a = c = d = 43 , b = 0; b = c = d = 43 , a = 0.

23. Let a, b, c, d be non-negative real numbers such that a + b + c + d = 1. Prove that

(1 − a2 )2 + (1 − b2 )2 + (1 − c2 )2 + (1 − d2 )2 > 3.

24. (Russia TST, 2015) Let a, b, c, d be non-negative real numbers such that a2 + b2 + c2 + d2 = 1. Prove that

a3 + b3 + c3 + d3 + abc + bcd + cda + dab 6 1.

25. (IMO Shortlist, 2010) Let a, b, c, d be real numbers such that a + b + c + d = 6 and a2 + b2 + c2 + d2 = 12. Prove
that
a) abcd 6 3;
b) 36 6 4(a3 + b3 + c3 + d3 ) − (a4 + b4 + c4 + d4 ) 6 48.
26. Let a, b, c, d be non-negative real numbers. Prove that
√
4
a3 + b3 + c3 + d3 + 4 a3 b3 c3 d3 > 2(abc + bcd + cda + dab).

27. (Russia, 2016) Let a, b, c, d be positive real numbers such that a + b + c + d = 3.

a) Prove that
1 1 1 1 1
+ 2 + 2 + 2 6 2 2 2 2.
a2 b c d a b c d
b) Prove that
1 1 1 1 1
+ 3 + 3 + 3 6 3 3 3 3.
a3 b c d a b c d

6 Basises
Up to this moment, we’ve always solved inequalities in the next way: let’s rewrite everything in terms of p, q, r, fix
something and play with T (p, q, r). Actually, this approach states on next observations:
• Any symmetric polynomial can be rewritten in terms of p, q, r

• a, b, c are real ⇔ p, q, r are real and T (p, q, r) ≥ 0, where T is continuous function

• a, b, c are non-negative ⇔ some inequalities with p, q, r hold.
Using this observations we solved a lot of inequalities with very trivial speculations. It’s time to realize, that p, q, r are
not the only possible symmetric polynomials with such properties. We call polynomials x = x(a, b, c), y = y(a, b, c), z =
z(a, b, c) a basis, if this 3 properties hold. Let’s demonstrate, how this idea works in next famous inequality:
28. Example (USAMO 2001, 6). Let a, b, c ≥ 0 and satisfy

a2 + b2 + c2 + abc = 4

8
Prove:
ab + bc + ca − abc ≤ 2
2 2 2
Proof. Let x = a + b + c, y = a + b + c , z = abc. Now we prove, that (x, y, z) is a basis.
2
It is easy to see, that p =x, q = x 2−y
 , r = z, so, any symmetric polynomial can be rewritten in terms of x, y, z.
x2 −y
Let’s take S(x, y, z) = T x, 2 , z , then T (p, q, r) ≥ 0 ⇔ S(x, y, z) ≥ 0, where S is a polynomial too.
Non-negativity lemma transforms as p ≥ 0 ⇔ x ≥ 0, q ≥ 0 ⇔ x2 − y ≥ 0, r ≥ 0 ⇔ z ≥ 0.

Let’s return to our inequality. It can be rewritten as y + z = 4, prove that x2 − y − 2z ≤ 4.

Fix y = y0 , z = z0 . From S(x, y0 , z0 ) ≥ 0 we get that x lies in some set of segments and rays. Obviously, x
is nonnegative, so for proving inequality we only need to check upper bound of x. It exists because of inequality
x2 = (a + b + c)2 ≤ 3(a2 + b2 + c2 ) ≤ 3(a2 + b2 + c2 + abc) = 12. Non-negativity lemma for x, y, z gives only lower
bounds for x, so x reaches its upper bound when S(x, y, z) = 0, i.e. b = c.

a2 + ab2 + 2b2 − 4 = 0 ⇒ (a + 2)(a + b2 − 2) = 0 ⇒ a = 2 − b2 . Desired inequality transforms into: ab + bb + ba − abb ≤

2 ⇔ a(2b − b2 ) + b2 − 2 ≤ 0 ⇔ a(2b −√b2 ) √ − a ≤ 0√ ⇔ −a(b 2
√ − 1) ≤ √0 which
√ is obvious.
Equality holds iff (a, b, c) = (1, 1, 1), ( 2, 2, 0), ( 2, 0, 2) or (0, 2, 2). Q.E.D.
29. Prove, that triple (x, y, z) = (a2 + b2 + c2 + abc, a + b + c, abc) is a basis too, and give an analagous proof of
previous problem using this basis.
30. Let a, b, c be non-negative real numbers such that a2 + b2 + c2 + nabc = n + 3 for some real n.
a) Assume that 0 ≤ n ≤ 23 . Prove that a + b + c ≤ 3.
p
b) Assume that 32 ≤ n ≤ 2. Prove that a + b + c ≤ 2(n + 3).
c) Assume that n = 2. Prove that ab + bc + ac − abc ≤ 25 .
31. Let a, b, c be non-negative numbers such that a2 + b2 + c2 + abc = 4. Prove that
√
a2 b2 + b2 c2 + c2 a2 ≥ abc(2 + 4 − 3abc).

32. (China-West, 2004) Let a, b, c be positive real numbers. Prove that

a b c 3√
1< √ +√ +√ ≤ 2.
a2 +b 2 2
b +c 2 2
c +a 2 2

7 Appendix
Here you can find some useful expressions.

a2 + b2 + c2 = p2 − 2q,
(a + b)(b + c)(c + a) = pq − r,
a3 + b3 + c3 = p3 − 3pq + 3r,
(ab)2 + (bc)2 + (ca)2 = q 2 − 2pr,
l12 + l21 = pq − r,
(ab)3 + (bc)3 + (ca)3 = q 3 − 3pqr + 3r2 .

Список литературы
[1] Nguyen Anh Cuong , ABC concreteness method, 2005
[2] Michael Rozenberg , uvw-method в доказательстве неравенств, Математическое Образование, 3-4 (59-
60), 2011.
[3] M.B.T. Knudsen , The uvw method, http://www.artofproblemsolving.com/community/c6h278791.
[4] Bach Ngoc Thanh Cong, Nguyen Vu Tuan, Nguyen Trung Kien, Three-variable inequalities, 2007