Sunteți pe pagina 1din 26

Algebra in Mathematical Olympiads

IMOTC 2019

Sutanay Bhattacharya
2
Introduction

0.1 What is this?


In this note, we are going to briefly sketch the necessary prerequisites related
to algebra that you are expected to be familiar with before coming to attend
IMOTC. Algebra, like any other topic in olympiad mathematics, ideally requires
little or no prerequisites, which is why we are going to spend most of this book
either recapping super-duper high school level math trivia1 or giving arcane-
sounding names to simple ideas.

0.2 Prerequisites
It would be rather ironic if a prerquisites’ book required prerequisites to begin
with, so I have decided to start from ground zero. Of course, as ancient wisdom
sayeth, you can’t just start reading a book with absolutely zero pre-knowledge;
so keep in mind the following:

1. You are gonna need some basic set theory notations. In case you are rusty
with them, just look them up on Wikipedia or something.

2. You should know how mathematical proofs works, and be able to follow
logical chains of reasoning.

3. You must be a human capable of reading English.

In short, unless you are a middle-school dropout, a neanderthal, an engineer or


a hamster, you should be fine2 .

0.3 How to use it


This books contains quite a lot of facts, and I haven’t bothered including the
proofs for most of them (except for sketches at times). You are expected to try
and think of a proof in case it is absent, or try to complete the given sketch
1 Any Danganronpa fans out there who got that reference?
2 In case you happen to be all four of these, please contact us as soon as possible

i
ii

whenever applicable. The treatment of the topics are nowhere near comprehen-
sive; it is highly encouraged that you read up them from external sources on
your own. In addition, I’ve included quite a few problems with varying diffi-
culty. Try as many of them as you feel like, but don’t get discouraged if you
happen to get stuck; many of them are hard by any standards.

0.4 Credits, or lack thereof


The problems in this note have been taken from a variety of sources, including
IMO ShortLists, several national and internation contests, and posts on Art of
Problem Solving. I apologize for the lack of specific information on the sources
for the individual problems; this will be fixed in the later versions.
Chapter 1

Fantastic polynomials and


how to interpolate them

1.1 The identity


Chances are, you are already familiar with what polynomials are; they are, let’s
say, “things” of the form an xn + an−1 xn−1 + · · · + a0 . But there are two ways of
looking at these “things”; one, they are functions that map, say reals to reals,
and two, they are finite sequences of reals that act as coefficients of the powers
of the variable. Let’s limit our focus to reals, then we can say

Definition 1.1.1. A function f : R → R is a polynomial if it can expressed as

f (x) = an xn + · · · + a1 x + a0

for some non-negative integer n.

The other way of looking at it is:

Definition 1.1.2. A polynomial f is a sequence a0 , a1 , · · · of reals such that


an = 0 for all n > N for some NP. Two sequences can be added as (a+b)n = an +
bn , and multiplied as (ab)n = i+j=n ai aj . This polynomial is often expressed
as a0 + a1 x + · · · + an xn .

For reals, these two viewpoints are one and the same, so for the rest of this
note, we won’t worry about the fine distinction between these. But that is not
the case for exotic number systems, say the field1 of integers modulo some prime
p. Clearly x and xp are same as functions, but they are different polynomials
1 Informally speaking, a “system” of number where we can add, multiply, subtarct and

divide by non-zero elements is called a field. Reals form a field, but there are lots of others.
If you ever hear a mathematician is working in a “field”, it’s probably one of those and not
actual fieldwork with practical applications, because pfft, who wants to do that. Same goes
for a mathematician working in a group.

1
2CHAPTER 1. FANTASTIC POLYNOMIALS AND HOW TO INTERPOLATE THEM

because they correspond to different sequences. Thus in general, the second


definition is the “right” way to think about polynomials.
Now, a bunch of terminologies:

Definition 1.1.3. In the polynomial f (x) = an xn + · · · + a1 x + a0 with an 6= 0,

1. The integer n is called the degree of f . The degree of the zero polynomials
is usually left undefined or defined to be −∞, whatever that is.

2. The reals an , · · · a0 are called the coefficients of f .

3. an is called the leading coefficient.

4. a0 is the constant term.

5. If an = 1, f is called a monic polynomial.

6. If all the terms other than an are zero, f is called a monomial.

7. The number x is a root of the polynomial if f (x) = 0.

1.2 Divide and conquer


One key property of the real polynomials is that you can perform the so-called
Euclidean division on them (similar to how we do it in number theory). To
formalize, we say that:

Fact 1.2.1. Given any polynomial A and any non-zero polynomial B, there exist
unique polynomials Q, R so that A = BQ + R and R = 0 or deg R < deg B. If
R = 0, B is called a factor of A.

The existence of divison algorithm means that polynomials have a lot of


properties resembling integers. For example,

Fact 1.2.2. (Bézout for polynomials) For any two real polynomial f, g with
GCD h, there exist polynomials x, y so that xf + yf = h.

Fact 1.2.3. Every polynomial can be written as a product of irreducible2 factors


uniquely upto reordering and scaling of the factors.3

The above facts have been phrased in terms of reals, but they hold for ra-
tional polynomials as well.

The fact that in Euclidean division algorithm, the degree of R is always less
than B means that if B is a one-degree polynomial, R has to be a constant
polynomial. This simple observation gives rise to the following:
2A polynomial that can’t be factored into smaller degree polynomials.
3 We will see later that for reals, these irreducible factors have degree at most 2.
1.2. DIVIDE AND CONQUER 3

Fact 1.2.4. The remainder of a polynomial P when divided by x − c is P (c).


In particular, x − c is a factor of P if and only if c is a root of P .

If P has root c, then P (x) = (x − c)Q(x) for some polynomial Q with degree
less than P (unless P = 0). Again, if Q has a root c1 , then Q(x) = (x − c1 )R(x)
for some even smaller R. This can’t go on forv er since at each step, the degree
reduces by 1, which leads to the following:

Fact 1.2.5. Any polynomial of degree n has at most n roots.

All this while, we have considering only real numbers, but there’s nothing
stopping us from considering polynomials in complex numbers too. In fact if we
do, we can say something much stronger:

Fact 1.2.6. A polynomial f has exactly n complex roots, counting multiplici-


ties. In other word, we write f as

f (x) = k(x − a1 )n1 · · · (x − ak )nk

with n1 + · · · nk = deg f .

This result might seem like a trivial generalisation of the previous fact, but
is surprisingly profound; in fact, profound enough to warrant the name Fun-
ndamental Theorem of Algebra. The proof is way beyond the scope of this
puny volume, so we will not attempt that.
So we saw that a real polynomial can easily have complex roots. But as
it turns out, the fact that those complex fcator mulitply out to produce some-
thing real introduces some rather stringent conditions on the roots themselves.
Specifically, the roots must occur in conjugate pairs:

Fact 1.2.7. If f is real polynomial with a complex root z, then z is also a root
of f .

(Hint for proof: Show in geneal that f (z) = f (z)) Using this, we can get a
better picture of how real polynomial factors over reals:

Fact 1.2.8. A real polynomial can written as a product of real polynomials


with degree at most 2.

Now we know how to express a given polynomial as product of linear factors,


so it is natural to compare this with the canonical format for polynomials. Say
f (x) = an xn +· · ·+a1 x+a0 is a polynomial with roots r1 , · · · , rn (not necessarily
all different). Then

an xn + · · · a1 x + a0 = k(x − r1 ) · · · (x − rn ).

If we multiply out the right hand side and compare the coefficients, we will
arrive at the following:
4CHAPTER 1. FANTASTIC POLYNOMIALS AND HOW TO INTERPOLATE THEM

Fact 1.2.9. (Vieta’s relations) For the above scenario, we have


n
X an−1 X an−2 a0
ri = − , ri rj = , · · · , r1 r2 · · · rn = (−1)n .
i=1
an an an
1≤i<j≤n

In keeping with the theme of relating coefficients to roots, here’s one more
useful result. Although it is technically more of a number theory result then
algebra, it’s included here.
Fact 1.2.10. If a rational number p/q with (p, q) = 1 is a root of the integer
polynomial an xn + · · · + a0 , then p|a0 , q|an .

1.3 Interpolation
Given any polynomial, we can easily evaluate it at any point as we wish to. But
how about the other way round? Say we are given the values of a polynomial at
a bunch of points, can we reconstruct the polynomial? One natural question to
ask is how many data points we need to get back, say, some n degree polynomial.
It is easy to see that n+1 data points are enough; if two n degree polynomials
match up in n + 1 points, their difference has n + 1 roots and thus has to be
zero. On the other hand, n data points is too few; if the points are x1 , · · · , xn
and p is any n degree polynomial that fits the given data, then so does k(x −
x1 ) · · · (x − xn ) + p for any k.
But how do we actually go about finding the said polynomial? The task is so
complicated and delicate that we often need helps from internation organisations
such as Interpol, which is why the process is dubbed interpolation. The most
useful interpolation method is the following:
Fact 1.3.1. (Lagrange interpolation formula) If p is a polynomial with degree
n or less, then and x0 1, · · · xn are distinct reals, then
n
X Y x − xj
p(x) = p(xi ) .
i=0
xi − xj
0≤j≤n
i6=j

Verify that the polynomial found by the above formula actually does satisfy
the given data set. By the discussion in the preceeding paragraphs, it has to be
unique.

1.4 Analytic properties


Strictly speaking, concepts like limits and continuity don’t belong to olympiad
math; but it is helpful to have some intuition about these, especially their
application to nice functions such as polynomials. Since it is impossible to
introduce analytic ideas rigorously within the miniscule scope of this note, we
will be very informal in a lot of places.
1.4. ANALYTIC PROPERTIES 5

Polynomials are so-called continuous functions. But what does that mean?
In high school, continuity is handwaved as “drawable without lifting your pen-
cil”, but that is hardly satisfactory. It is not hard to come up with erratic
examples where the intuition in the “not-lifting-your pencil” idea fails, not to
mention erratic situation such as when the student decides use a pen to draw.
So what do we mean when we say a function, say reals to reals, is continuous?
Definition 1.4.1. (Very informal) A function f : R → R is continuous at a
if no matter how small an interval you choose around f (a), there is a small
interval around a that is mapped into that interval by f .4
In other words, if you gradually “close in on” the point a, f (x) will also
“close in on” f (a); f doesn’t have a sudden “jump” at a. If f happens to be
continuous at every point in R, we say that f is a continuous function (on R).
Continuity, among other things, gives a very useful property:
Fact 1.4.1. (Intermediate value property) If p is a polynomial so that p(x) = a
and p(y) = b (x < y), then p attains all values strictly between a and b on the
interval (x, y). In particular, if p has different signs at x and y, then it has a
root between them.
Now we will look at behaviour of polynomials for large inputs.
Look at the polynomial f (x) = 2x3 + 3x + 1. For really large values of x,
the terms 3x and 1 will be miniscule compared to the leading term, so the 2x3
will dominate. If we ignore the scaling factor of 2, the function, in some sense,
“behaves” roughly like x3 . We express this as f (x) = Θ(x3 ). What this means
is that for huge x, this will grow arbitraily large. Of course, this also means
that quadratic polynomials grow like x2 , and in general n degree polynomials
grow like xn . So if f, g are two polynomials so that deg f > deg g, then for large
enough x, |f (x)| will outgrow |g(x)|. This is expressed as f (x) = O(g(n)), or
g(x) = o(f (x)).
Finally, note that if the leading coefficient is positive, then f (x) will approach
+∞ as x → +∞, whereas if the leading coefficient is negative, f (x) → −∞ for
x → +∞. The behaviour when x → −∞ can be deduced similarly, except it
happens to also depend on whether deg f is odd or even. Can you figure out
how exactly?

Now we look at another aspect of polynomials: derivatives. In calculus,


derivatives are rate of change of functions; specifically, the derivative of a func-
tion at a point is the slope of the tangent to that curve ther. But here, we will
treat derivative purely symbolically.
Pn
Definition 1.4.2.P For a polynomial p(x) = i=0 ai xi , the derivative of p is
n
defined as p0 (x) = i=1 ian xi−1 .
Clearly that is also a polynomial. It is routine to verify that derivatives
satisfy the following important properties:
4 Does this agree with your “can’t-lift-your-pencil” intuition?
6CHAPTER 1. FANTASTIC POLYNOMIALS AND HOW TO INTERPOLATE THEM

Fact 1.4.2. Let f, g be two polynomials. Then


1. (f + g)0 (x) = f 0 (x) + g 0 (x)
2. (cf )0 (x) = cf 0 (x) for any c ∈ R
3. (f g)0 (x) = g · f 0 (x) + f · g 0 (x)
4. (f ◦ g)0 (x) = f 0 (g(x)) · g 0 (x)
These let us easily compute derivatives of expressions involving polynomials.
Another result which is often useful is
Fact 1.4.3. The polynomial f is divisible by a square of a non-constant poly-
nomial if and only if f and f 0 have a non-constant GCD.

1.5 Problems
Problem 1.5.1 Find all real polynomials P (x) that satisfy

P (x3 − 2) = P (x)3 − 2.

Problem 1.5.2 If the polynomials f (x) and g(x) are written on a blackboard
then we can also write down the polynomials f (x) ± g(x), f (x)g(x), f (g(x)) and
cf (x), where c is an arbitrary real constant. The polynomials x3 − 3x2 + 5 and
x2 − 4x are written on the blackboard. Can we write a nonzero polynomial of
the form xn − 1 after a finite number of steps? Justify your answer.
Problem 1.5.3 Let f be a polynomial with real coefficients and suppose f
has no nonnegative real root. Prove that there exists a polynomial h with real
coefficients such that the coefficients of f h are nonnegative.
Problem 1.5.4 A real polynomial of odd degree has all positive coefficients.
Prove that there is a (possibly trivial) permutation of the coefficients such that
the resulting polynomial has exactly one real zero.
Problem 1.5.5 N different numbers are written on blackboard and one of
these numbers is equal to 0. One may take any polynomial such that each
of its coefficients is equal to one of written numbers (there may be some equal
coefficients) and write all its roots on blackboard. After some of these operations
all integers between −2016 and 2016 were written on blackboard (and some other
numbers possibly). Find the smallest possible value of N .
Problem 1.5.6 Suppose f, g ∈ R[x] are non constant polynomials. Suppose
neither of f, g is the square of a real polynomial but f (g(x)) is. Prove that
g(f (x)) is not the square of a real polynomial.
Problem 1.5.7 Let f be a monic real polynomial of degree n, and let x0 <
x2 < · · · < xn be some integers. Prove that there is an integer k ∈ {0, 1, · · · , n}
so that |f (xk )| ≥ 2n!n .
1.5. PROBLEMS 7

Problem 1.5.8 Let f and g be two polynomials with integer coefficients such
that the leading coefficients of both the polynomials are positive. Suppose
deg(f ) is odd and the sets {f (a) | a ∈ Z} and {g(a) | a ∈ Z} are the same.
Prove that there exists an integer k such that g(x) = f (x + k).
Problem 1.5.9 Determine whether there exist non-constant polynomials P (x)
and Q(x) with real coefficients satisfying

P (x)10 + P (x)9 = Q(x)21 + Q(x)20 .


8CHAPTER 1. FANTASTIC POLYNOMIALS AND HOW TO INTERPOLATE THEM
Chapter 2

Inequalities and
discrimination

2.1 Introduction
In this chapter, we will look into how to derive and solve various inequalities.
We’ll focus mostly on mathematical inequalities and less on social inequalities
and discriminations; those are prerequisites for a different camp known as the
concentration camp.

2.2 Means of derivation and derivation of means


Note that no matter how far we go with bizzare and obscure ressults in the
theory of inequalities, it always boils down to a few simple facts:

1. Positive numbers are greaters than zero (duh);

2. Squares of reals are non-negative;

3. You can add inequalities;

4. You can scale inequalities by positive reals (a > b =⇒ ac > bc for c > 0);

5. You can multiply inequalities as long as things are positive.

Let’s look at the first item here; using that, we can take squares of any random
expression and get an inequality out of it. What if we decide to look at (a − b)2 ?
The fact that this is non-negative leads to a2 + b2 ≥ 2ab, and via a change of
variables, we see that for non-negative x, y,
x+y √
≥ xy.
2

9
10 CHAPTER 2. INEQUALITIES AND DISCRIMINATION

The left side is the normal average of x, y; the right side, on the other hand, is a
different kind of average. It also gives some number between x, y, but “between”
in a multiplicative sense. The above inequality states that for any non-negative
numbers, the normal average (arithmetic mean) is always greater than the
“multiplicative” average (geometric mean.). In fact, this generalises to n
numbers:
Fact 2.2.1. (AM-GM inequality) Given any n non-negative numbers x1 , · · · , xn ,
we have
x1 + x2 + · · · + xn √
≥ n x1 x2 · · · xn .
n
Equality holds if and only if x1 = x2 = · · · = xn .
One of the many ways to prove this is to use the so-called Cauchy induc-
tion; you prove this for n = 2k by induction, then generalise to any n < 2k by
adding “dummy points” to the data set.
Of course, the xi ’s need not be distinct. In fact, if our data set has ki
instances of xi , the above inequality becomes
k1 x1 + · · · + kn xn
q
xk11 · · · xknn .
k1 +···+kn

k1 + · · · + kn
In the above, we took ki ’s to be integers. By some continuity argument, we can
in fact generalise them to be anything:
Fact 2.2.2. (Weighted AM-GM inequality) Given non-negative reals x1 , · · · , xn
and positive reals k1 , · · · , kn , we have
k1 x1 + · · · + kn xn
q
xk11 · · · xknn .
k +···+kn
≥ 1
k1 + · · · + kn
Often, the special case of the above fact where k1 + · · · + kn = 1 is termed
as weighted AM-GM inequality. Of course, these two forms are equivalent.

2.3 CS:Go
In this section, we look at another interesting result in inequalities. As with a
lot of important results in math, this has several names with equally confus-
ing spellings and pronunciation, including, Cauchy-Schwarz inequality, Cauchy-
Bunyakovsky inequality, and of course Cauchy-Schwarz-Bunyakovsky inequality.
So to save time and effort, we’ll henceforth call this CS inequality. For now,
we’ll using CS as a noun; once you attain divine mastery on the dark arts of
inequalities, you unlock the power to use it as a verb, for example, “I heard
arqady CSed that problem.”
Without further ado, here’s the inequality:
Fact 2.3.1. (Cauchy-Schwarz Inequality) Let a1 , · · · , an , b1 , · · · , bn be 2n reals,
then
(a1 b1 + a2 b2 + · · · + an bn )2 ≤ (a21 + · · · + a2n )(b21 + · · · + b2n ).
2.4. FROM REARRANGEMENT TO CHEBYSHEV 11

There are many proofs of this result; one popular approach involves looking
at the discriminant of the quadratic polynomial (a1 x − b1 )2 + · · · + (an x − bn )2 .
As an exercise, use this idea to figure out when equality holds in CS inequality.
Also, try to prove the QM-AM inequality using this (look up what QM-AM
inequality is).
Useful as this is, there’s another form of Cauchy-Schwarz inequality that is
more readily applied in many olympiad contexts:
Fact 2.3.2. (Titu’s Lemma, or Engel form of CS) For n reals x1 , · · · , xn and n
positive reals y1 , · · · , yn , we have
x21 x2 (x1 + · · · + xn )2
+ ··· + n ≥ .
y1 yn y1 + · · · + yn
Prove this from CS. Do you see why we need yi ’s to be positive?

2.4 From rearrangement to Chebyshev


Say you have notes of denominations y1 , · · · , yn and you take x1 of any kind
of note, x2 notes of a second kind, and so on. What should you do to get the
maximum money?
Of course, it would be great if you could just take all notes of the biggest
denomination, but that’s not a option. So you do the next best thing, take
the largest number of notes of the largest denomination, second largest number
of the next large denomination, and so one. This simple idea is formalized as
rearrangement inequality.
Fact 2.4.1. (Rearrangement inequality) If x1 ≤ x2 ≤ · · · ≤ xn and y1 ≤
y2 ≤ · · · ≤ yn are reals, and σ is any permutation of {1, · · · , n}, then the sum
xσ(1) y1 + · · · + xσ(n) yn is maximised when σ is the identity permutation and
minimised when σ(i) = n + 1 − i.
Exercise: prove this formally. The key idea is that if you swap two elements
in the permutations to make the corresponding xi ’s in right order, the sum in-
creases.

Note that rearrangement inequality is really a lot of inequalities in disguies;


n! inequalities, in fact, one for each permutation. When we have a lot of in-
equalities, the natural thing is to add them up. In this case, if we add the
corresponding inequalities when σ is one of the n cyclic shifts, we end up with
Fact 2.4.2. (Chebyshev’s inequality1 )Let x1 , · · · , xn and y1 , · · · , yn be two sim-
ilarly ordered sequences (both increasing or both decreasing). Then
  
x1 y1 + · · · + xn yn x1 + · · · + xn y1 + · · · + yn
≥ .
n n n
1 Also know as Chebysheff, Chebychov, Chebyshov, Tchebychev, Tchebycheff,
Tschebyschev, Tschebyschef, Tschebyscheff or Chebychev’s iequality. If you have trouble
with the spelling, write anything that sounds close; chances are it’ll be one of the alternate
spellings.
12 CHAPTER 2. INEQUALITIES AND DISCRIMINATION

The inequality is reversed when the sequences are oppsitely ordered.

2.5 Trivial by Jensen


Before we talk about Jensen, we need to introduce a certain class of functions,
known as convex functions (and also concave functions).
Roughly, a function is convex if it’s the region above its graph is a convex
set; i.e., any line joining two points on this curve lies strictly above the curve.
Definition 2.5.1. A function f : X → R is called convex2 if for any x1 , x2
in X and t ∈ [0, 1], we have f (tx1 + (1 − t)x2 ) ≤ tf (x1 ) + (1 − t)f (x2 ). If the
inequality is strict, f is called strictly convex. If the inequality is reversed, f is
called a concave function.
Convince yourself that the above definition matches with the intuition in the
previous paragraph.
You can easily tell the graphs of f (x) = x2 and f (x) = ex are convex visually,
but that’s far from a sure-fire method. That’s why we need the following:
Fact 2.5.1. If f, g are two convex functions, then so are their sum, positively
scaled versions, and composition. Further, if a function is twice differentiable,
then it’s convex if and only if it’s second derivative3 is positive.
Now that we know what convex functions are, we’re ready to state the key
result of this section:
Fact 2.5.2. (Jensen’s inequality, weighted version) If f is function that’s convex
on X, and x1 , · · · , xn are in X, and a1 , · · · , an are positive reals that sum up
to 1, then
a1 f (x1 ) + a2 f (x2 ) + · · · + an f (xn ) ≥ f (a1 x2 + a2 x2 + · · · + an xn ).
The version we will use most often is the speacial case when a1 = · · · =
an = 1/n. Do you see why this is a generalisation of the definition of convex
function?

2.6 Darker arts


There are many, many results that are often helpful in proving inequalities.
We’ll not get into details and will briefly mention some of them.
Fact 2.6.1. (Hölder’s inequality) If we have m squences of n positive numbers
a1,1 , a1,2 , · · · , a1,n ; a2,1 , · · · , a2,n ; · · · ; am,1 , · · · , am,n , then
   v m
m n n uY
Y X X um
 ai,j  ≥  m
t ai,j  .
i=1 j=1 j=1 i=1

2 For some mysterious reason, in amny older texts convex and concave are called concave

up and concave down respectively. Talk about terrible naming.


3 Look at any standard calculus textbook if you don’t know how to compute derivatives.
2.6. DARKER ARTS 13

Equality holds if and only if every two sequences are proprtional.


Notice how this generalises CS.
Fact 2.6.2. (Schur’s inequality) Suppose a, b, c are nonnegative reals, and r > 0.
Then
ar (a − c)(b − c) + br (b − c)(b − a) + cr (c − a)(c − b) ≥ 0.
Equality holds for a = b = c or when one of a, b, c is zero and the other two are
equal.
Definition 2.6.1. We say a sequence (a1 , · · · , an ) majorises (b1 , · · · , bn ) if a1 ≥
b1 , a1 + a2 ≥ b1 + b2 , · · · , a1 + · · · + an−1 ≥ b1 + · · · bn−1 and a1 + · · · + an =
b1 + · · · + bn , and it is denoted as (a1 , · · · , an )  (b1 , · · · , bn ).
Fact 2.6.3. (Karamata’s inequality) If f is a convex function in X ⊆ R and
(x1 , · · · , xn )  (y1 , · · · yn ) for xi , yi ∈ X, then

f (x1 ) + · · · + f (xn ) ≥ f (y1 ) + · · · + f (yn ).

Fact 2.6.4. (Muirhead) Let (a1 , · · · , an )  (b1 , · · · , bn ). Then for positive reals
x1 , · · · , xn , we have
X X
xa1 1 · · · xann ≥ xb11 · · · xbnn .
sym sym
P
Here sym means we are are taking the sum over all possible permutations of
x1 , · · · , xn .
To get a feel for Muirhead, try writing it out for, say, three variables x, y, z
and any two sequences (one majorizing the other) of your choice, Then prove it
using weighted AM-GM (you may need to choose your weights cleverly to apply
AM-GM).
Fact 2.6.5. (Power mean inequality) Let x1 , · · · , xn be positive reals with
weights w1 , · · · , wn adding up to 1. For r 6= 0, define the r−th power mean
1
Mr as (w1 xr1 + · · · + wn xrn ) r , and define M0 to be xw wn
1 · · · xn
1
(the weighted
geometric mean). Then for any two reals r > s, we have Mr > Ms .
We have barely scratched the surface the plethora of techniques used to prove
inequalities. Apart from general methods such as expressing sum of squares,
telescoping or induction, and those already mentioned here, the following are
worth mentioning:
1. Minkowski’s inequality
2. Tangent line trick
3. Mixing variable
4. Cauchy reverse technique
14 CHAPTER 2. INEQUALITIES AND DISCRIMINATION

5. n − 1 EV technique

6. the uvw method

7. Lagrange multipliers

and so on. Interested readers are advised to look them up on the internet.

2.7 Problems
√ √
7 √
Problem 2.7.1 Let a, b, c be positive reals satisfying a + b + c = 7
a+ b+ 7
c.
Prove that aa bb cc ≥ 1.

Problem 2.7.2 Let a, b, c, d be real numbers satisfying |a|, |b|, |c|, |d| > 1 and
abc + abd + acd + bcd + a + b + c + d = 0. Prove that
1 1 1 1
+ + + > 0.
a−1 b−1 c−1 d−1
Problem 2.7.3 Given non-negative reals x, y, z satisfying x + y + z = 1, find
the minimum possible value of
1 1 1
x x+1 y y+1 z z+1
+ + .
x+1 y+1 z+1

Problem 2.7.4 Suppose that a sequence a1 , a2 , . . . of positive real numbers


satisfies
kak
ak+1 ≥ 2
ak + (k − 1)
for every positive integer k. Prove that a1 + a2 + . . . + an ≥ n for every n ≥ 2.

Problem 2.7.5 Let n ≥ 3 be an integer, and let a2 , a3 , . . . , an be positive real


numbers such that a2 a3 · · · an = 1. Prove that

(1 + a2 )2 (1 + a3 )3 · · · (1 + an )n > nn .

Problem 2.7.6 Let a1 , a2 , . . . an , k, and M be positive integers such that


1 1 1
+ + ··· + =k and a1 a2 · · · an = M.
a1 a2 an
If M > 1, prove that the polynomial

P (x) = M (x + 1)k − (x + a1 )(x + a2 ) · · · (x + an )

has no positive roots.

Problem 2.7.7 Let x, y, and z be real numbers (not√ necessarily positive) such
that x4 + y 4 + z 4 + xyz = 4. Show that x ≤ 2 and 2 − x ≥ y+z
2 .
2.7. PROBLEMS 15

Problem 2.7.8 Let n be a positive integer and let (x1 , . . . , xn ), (y1 , . . . , yn )


be two sequences of positive real numbers. Suppose (z2 , . . . , z2n ) is a sequence
2
of positive real numbers such that zi+j ≥ xi yj for all 1 ≤ i, j ≤ n. Let M =
max{z2 , . . . , z2n }. Prove that
 2   
M + z2 + · · · + z2n x1 + · · · + xn y1 + · · · + yn
≥ .
2n n n

Problem 2.7.9 Let f (x) and g(x) be given by


1 1 1 1
f (x) = + + + ··· + ,
x x−2 x−4 x − 2018
1 1 1 1
g(x) = + + + ··· + .
x−1 x−3 x−5 x − 2017
Prove that |f (x) − g(x)| > 2 for any non-integer real number x satisfying 0 <
x < 2018.

Problem 2.7.10 Find the largest real constant a such that for all n ≥ 1 and
for all real numbers x0 , x1 , ..., xn satisfying 0 = x0 < x1 < x2 < · · · < xn we
have
 
1 1 1 2 3 n+1
+ + ··· + ≥a + + ··· +
x1 − x0 x2 − x1 xn − xn−1 x1 x2 xn
16 CHAPTER 2. INEQUALITIES AND DISCRIMINATION
Chapter 3

Functional equations

3.1 Introduction
Often times you’ll encounter a problem that starts with “find all functions...”.
More often than not, these will specify some properties of a mystery function
and ask to find all such functions. Sometimes, they’ll ask to prove some more
properties of that function. In this section, we’ll look at some tactics to tackle
these sort of problems.

3.2 Terms and conditions


Most likely you know what a function is; here goes the formal definition for the
sake of completeness:
Definition 3.2.1. If X and Y are two sets, then a function f : X → Y is
a subset of X × Y such that for every x ∈ X, there is unique t ∈ Y so that
(x, y) ∈ F . In that case, we write f (x) = y. X and Y are called respectively
the domain and co-domain of the function. The is called the range of f .
So a function takes an input from X and spits out a unique element in Y as
output. While typically the output depends on the input, it’s entirely possible
that two or more element get mapped to the same output element. Also, it’s
possible that some elements in y are not outputs of anything; i.e., they have no
pre-image. In case these nice conditions do hold, we give special names to f :
Definition 3.2.2. A function f : X → Y is called injective is f (x1 ) =
f (x2 ) =⇒ x1 = x2 for all x1 , x2 ∈ X, i.e., two different elements can’t
map to the same thing. A function f : X → Y is surjective if for every y ∈ Y ,
there is x ∈ X so that f (x) = y, i.e., every element in y has a pre-image. In case
f happens to be both injective and surjectve, it’s called a bijective function,
or a bijection.
In addition, the following terminology is often of use:

17
18 CHAPTER 3. FUNCTIONAL EQUATIONS

Definition 3.2.3. If function f defined on a suitable domain (usually R) is


even if f (x) = f (−x) for all x, and odd if f (−x) = −f (x) for all x.

As a simple exercise, try proving the following:

Fact 3.2.1. Any function f : R → R can be written as a sum of an odd function


and an even function.

As usual, you can compose functions; if f : X → Y and g : Y → Z are


functions, (g ◦ f )(x) = g(f (x)) defines a new function X → Z. In case the
the output space is equipped with addition and other nice operation, we can
perform those operation on the functions mapping to them; say f1 and f2 are
two functions from X to R, then (f1 + f2 )(x) = f1 (x) + f2 (x) is a new function.
You know, the usual stuff.

3.3 Strategies and tactics


With the basic odds and ends out of the way, we can now focus on how the
actual solving procedure takes place. Here’s the general gameplan to solve a
functional equation:

1. Look at the problem statement. It should consist of the name of one or


more mystery function(s), say f , their domain, co-domain, and one or
more conditions that the mystery function(s) must satisfy.

2. Use these properties to deduce more information about f , and use these
to gather even more information, until you have narrowed down f to a
few possibilities.

3. Check each of those possibilities and see which ones work.

The second step is usually the hardest and the key part of a solution; we’ll come
to that later. But now some words about the other two points are in order.

Point 1 This is where you get an idea of the general setting of the problem;
what facts are known and how to use them. For example, if the function is from
R → R, and it satisfies the condition f (x2 + y) = f (x)2 + y for all reals x, y,
you can use the condition for any specific real values of x and y. Note that
the domain of the function (where the function is defined) and domain of
the functional equation (for which values of the variables you can use the
given condition) are two separate things. Very commonly these coincide, but
that need not be the case always.

Point 3 The deductions obtained in step 1 are usually one-way implications;


they are of the form “If f satisfies blah blah, then f must also satisfy blah blah”.
With enough work, you’ll eventually narrow down the deductions to the form
“If f satisfies blah blah, then f is one of the function blah blah”. But the list
3.3. STRATEGIES AND TACTICS 19

of functions you have at this point are still coming from a one-way implication;
they need not satify the original conditions. This is why it is so important to
check back if they do.
One very common mistake is falling victim to the so called pointwise trap.
Suppose you’ve done some mumbo-jumbo with the equation to get something
like f (x)(f (x) − x2 ) = 0 for all x ∈ R. Can you conclude that the only solutions
are f (x) ≡ 0 and f (x) = x2 ? No. This is because the statemnet you obtained
implies
(f (x) = 0 or f (x) = x2 ) for all x ∈ R.
This is different form

(f (x) = 0 for all x ∈ R) or (f (x) = x2 for all x ∈ R).

To drive the point home, note that the function given by


(
f (x) = 0 ifx ∈ Q
f (x) =
f (x) = x2 ifx 6∈ Q

satisfies the condition you obtained. To actually avoid this trap, say in this
case, you will have to prove f (x) = 0 and f (y) = y 2 cannot happen at the same
time for non-zero x, y.

Now we will talk about actual solving strategies. The methods that are often
handy are:

Plugging in values and substitution Using the given condition cleverly for
specific values of the variables can give information about f at particular point;
at times they can give information about f at a lot more points. For example
if f : R → R satisfies f (x + y) = xf (y), then plugging in x = 0 immediately
gives f (y) = 0 for all y which gives away the answer. Plugging in 0, ±1 or other
nice numbers that make a lot of terms cancel out are often helpful. Sometimes
you may want to substitute expressions to cancel out stuff; say if the equation
is f (x2 + y) = f (x) + other stuff, subbing in y = x − x2 might not be a bad idea.
However, make sure the substitution is logical; for example, in f (x2 +3) = x2 +3,
you can’t just sub in y = x2 + 3 and conclude f (y) = y for all real y. The reason
is x2 + 3 does not cover the entirety of real line.

Proving injectivity/surjectivity Often, showing the function is injective


or surjevctive gives us room to work with. For example, if you know f in
injective and f (f (x) + x) = f (2f (x)), then you can conclude f (x) + x = 2f (x)
by definition of injectivity. Similarly, if a surjective function satisfies f (f (x)) =
f (x) + 2, then you know for any y there is x so that f (x) = y; so f (y) = y + 2
holds for all y. The generic way to show a function is injective is to assume
f (x1 ) = f (x2 ) and somehow use the known facts about f to show x1 = x2 .
For surjectivity, if you can obtain some equation like f (somthing) = x for all x,
you’ll be done.
20 CHAPTER 3. FUNCTIONAL EQUATIONS

Exploiting symmetry When an equation is “almost” symmetric, you can


swap some variables to get a new equation and compare it to the original to
get interesting stuff. For example, if f (xy) = xf (y), then swapping x, y and
comparing gives xf (y) = yf (x), which practically kills the problem (do you see
how?).

Introducing new variable Sometimes introducing a second variable can give


you room to work with. For example, if you have some expression for f (x + y),
then you can try evaluating f (x + y + z) in two ways (as f (x + (y + z)) and
f ((x + y) + z))) and comparing these to see if something pops out.

Fixed points An element x in the domain so that f (x) = x is called a fixed


point of f . Looking at these can be helpful in specific situations.

Induction and recurrence Induction is especially helpful for FEs on inte-


gers; so is recurrence relations (it is reccommedned you look up the basic theory
behind solving linear homogeneous recurrence relations). Sometimes even for
real function, recurrence comes in handy, say to evaluate f at a sequence of
points and looking at the limit cases, or to get increasingly tighter bounds on
some expression using previous bounds.

3.4 The Classics


The lack of any concrete preqrequisites in FEs means the process of solving
every single FE starts from scratch; you have only the absolute bare minuimum
to back you up. That’s rather discouraging; so in this section, we will take a
look at some standard FEs that people smarter than us have already solved for
us.

The most ubiquitous among these is doubtless Cauchy’s functional equa-


tion. This is concerned with characterizing all additive real functions; i.e., we
look at all functions f : R → R that satisfy f (x + y) = f (x) + f (y). Functions
of the form f (x) = kx clearly work; you’d except that’d be all. That’s indeed
true if you’re only concerned about rationals.
Fact 3.4.1. If f is any additive function on reals, then f (x) = f (1)x for all
rationals x.
It would seem this generalises to reals; but alas, the real line is a dark and
scary place.
Fact 3.4.2. There exist non-linear additive functions on R. Moreover, any such
function’s graph is dense in R2 ; in other words, any disk of positive radius in
R2 has some point of f ’s graph.
But thankfully, all hope is not lost. We can’t impose other “nice” conditions
on f to force it to be linear.
3.5. PROBLEMS 21

Fact 3.4.3. Let f : R → R be additive. Then the followng are equivalent:

1. f is linear;

2. f is continuous at some point;

3. f is bounded (from below or above) on some non-empty open interval;

4. f is monotonic on some non-empty open inteval.

Convince yourself that these follow from the previous fact. So to prove
a Cauchy function is linear, all you need is to prove one of monotonicity or
boundedness (continuity works too, but that’s usually harder unless it’s given).
There’s one more condition that works:

Fact 3.4.4. If f is an additive function on R and there is a polynomial P of


degree> 1 that satisfies f (P (x)) = P (f (x)), then f is linear.

All this while we’ve only talked about when the domain of f is the whole
of R; but these hold true even if it’s some other non-trivial interval in R. In
particular, any additive function f : R+ → R+ is necessarily linear (since the
co-domain is already bounded from below).

There is one variant of Cauchy that’s worthy of mention:

Fact 3.4.5. (Jensen’s Functional equation) If f : R → R satisfies


 
x+y f (x) + f (y)
f = ,
2 2

then f (x) = g(x) + f (0) for some additive function g.

So again, with any of those nice conditions like monotonicity and bounded-
ness, f has to ax + b in this case.

3.5 Problems
As always, here are some cute problems you may want to try:

Problem 3.5.1 Determine all functions f : R → R such that for all x, y ∈ R

f (xf (y) − yf (x)) = f (xy) − xy.

Problem 3.5.2 Find all strictly monotonic functions f : (0, +∞) → (0, +∞)
such that  2 
x
f =x
f (x)
for all x ∈ (0, +∞)
22 CHAPTER 3. FUNCTIONAL EQUATIONS

Problem 3.5.3 Find all functions f : R → R such that

f x2 + xf (y) = xf (x + y)


for all reals x, y.

Problem 3.5.4 Find all functions f : R 7→ R such that

f (x)f (yf (x) − 1) = x2 f (y) − f (x),

for all x, y ∈ R.
Problem 3.5.5 Find all surjective functions f : R → R such that for every
x, y ∈ R, we have

f (x + f (x) + 2f (y)) = f (2x) + f (2y).

Problem 3.5.6 Find all functions f : R+ → R+ such that

f x2 + xf (y) = xf (x + y)


for all x, y ∈ R+ .
Problem 3.5.7 Find all surjective functions f : N → N such that for all positive
integers a and b, exactly one of the following equations is true:

f (a) = f (b),
f (a + b) = min{f (a), f (b)}.

Problem 3.5.8 Find all functions f : R → R such that f (0) ∈ Q and


2
f (x + f (y)2 ) = f (x + y) .

Problem 3.5.9 Find all functions f : R → R such that for all real numbers
a, b, and c:
1. If a + b + c ≥ 0 then f (a3 ) + f (b3 ) + f (c3 ) ≥ 3f (abc).
2. If a + b + c ≤ 0 then f (a3 ) + f (b3 ) + f (c3 ) ≤ 3f (abc).

Problem 3.5.10 Find all functions f : R+ → R+ satisfying

f (xy + f (y)2 ) = f (x)f (y) + yf (y).

S-ar putea să vă placă și