0DWKHPDWLFDO2O\PSLDG'DUN$UWV_

Polynomials
The set of polynomials
f x  a0  a1 x  a2 x2  …  an xn , where n is a non-negative integer and
a0 , a1 , …, an   S, is denoted S
x (pronounced ‘S adjoin x’). In this chapter, we will explore the cases where S
is the set of complex numbers or real numbers.

Complex polynomials, &
x

Suppose we have a degree-n polynomial f x  a0  a1 x  a2 x2  …  an xn , where a0 , …, an are complex
constants and an is non-zero. According to the fundamental theorem of algebra, we can express it as a product of
linear factors of the form x  %i , where %i is a (complex) root of the polynomial.

 If we have a monic polynomial f x  a0  a1 x  a2 x2  …  an xn , where an  1, then we can express

f x  x  %1  x  %2  …x  %n , where %1 , …, %n are (not necessarily distinct) roots of the polynomial.
[Fundamental theorem of algebra]

For example, the polynomial x4  1 can be factorised as x  1 x  1 x   x  , where  1 is the

2
imaginary unit. In general, the polynomial xn  1  x  1 x  & x  &  …x  &
2 n1
, where &   n is a princi-
pal nth root of unity. The roots of unity are positioned at the vertices of a regular n-gon with centre 0 and a vertex
at 1. The example for x5  1 is shown below.

This means that the degree-n curve y  f x meets the degree-1 line y  0 in at most n points. There is nothing
special about the line y  0, and this also applies to any line. More remarkably, the polynomial curve can be
replaced with any algebraic curve (such as the unit circle, x2  y2  1, which has degree 2). Even more generally,
where the line is replaced with another algebraic curve, we have Bezout’s theorem.

 Suppose P and Q are two curves of degrees m and n, respectively. If they intersect in finitely many points, then they
intersect in at most m n points. [Bezout’s theorem]

Equality occurs if we consider intersections on the complex projective plane (instead of the real plane), and count
intersections with appropriate multiplicity (e.g. twice for tangency, thrice for osculation et cetera). The complex
projective plane is discussed in later chapters, and it is only necessary at this point to use the weak form of
Bezout’s theorem.
_0DWKHPDWLFDO2O\PSLDG'DUN$UWV

1. Show that two ellipses intersect in at most four points.

2. Consider the regular n-gon with vertices A1 , A2 , …, An , where n  5. Let P be a variable point on the
circumcircle of the n-gon. Show that the value of f P  A1 P4  A2 P4  …  An P4 remains constant.

Difference of three cubes

A particularly useful polynomial is x3  y3  z3  3 x y z. Over the reals, it factorises into
x  y  z x2  y2  z2  x y  y z  z x, and the quadratic term can be further factorised over the complex
numbers. This polynomial recurs in many situations, including olympiad problems.

x y z
 The polynomial x3  y3  z3  3 x y z  det z x y  x  y  z x   y  2 z x  2 y   z, where  is a
y z x
principal cube-root of unity. [Difference of three cubes]

This generalises to n variables, instead of merely three. Indeed, the name is derived from the n  2 case, known as
x y
the ‘difference of two squares’, x2  y2  det   x  y x  y.
y x

x0 x1 x2  xn1
xn1 x0 x1  xn2
n1 n1 k
   xr  n  .
2 r
 det xn2 xn1 x0  xn3
r0 r0
ž ž ž › ž
x1 x2 x3  x0

The sums in the product on the right-hand side are the terms of the discrete Fourier transform of
x0 , x1 , …, xn1 . (Continuous) Fourier transforms were originally discovered to explain how the sound of an
entire orchestra can be composed of basic sinusoidal waves. Today, this idea is used to analyse electrical circuits.
The discrete Fourier transform can be computed quickly using certain algorithms, forming the basis of the fastest
known algorithm for multiplying two large integers.

3. Find the minimum value of x2  y2  z2 , where x, y, z are real numbers such that x3  y3  z3  3 x y z  1.
[BMO2 2008, Question 1]

Cubic equations
For sufficiently small degree, it is possible to solve polynomial equations using radicals (nth roots). For the
general quadratic equation, the Babylonian technique of ‘completing the square’ suffices. Solving the cubic
equation is a more difficult, multi-step process. If we can immediately find a root %, it is possible to divide by
x  % to reduce the equation to a quadratic. Otherwise, more ingenious techniques are required.

4. Suppose we have an equation a y3  b y2  c y  d  0. Show that this can be converted to an equation of

the form x3  p x  q  0, where x is a linear function of y. [Reduction to a monic trinomial]

Hence, it is only necessary to consider the latter case, as all other cubics can be reduced to it.

5. Show that x3  p x  q  0 is equivalent to x3  a3  b3  3 a b x  0, where a3 and b3 are the roots of the

p3
quadratic equation z2  q z  27
 0.

6. Hence show that x3  p x  q  0 has a root

 0DWKHPDWLFDO2O\PSLDG'DUN$UWV_

1 1 1 1 1 1
x  a  b  3
2 q  4
q2  27
p3  3
2 q  4
q2  27
p3 . [General solution to cubic

equations]

The general solution to the cubic equation was the quest of many mediæval mathematicians. After a partial
solution by Omar Khayyam, the first complete solution was by Niccolò Tartaglia. However, when the quartic was
later solved by Lodovico Ferrari, Tartaglia’s solution of the cubic was mistakenly attributed to Gerolamo Car-
dano, and is thus referred to as Cardano’s formula. This displeased Tartaglia to a great extent.
Due to the existence of formulae for solving the general quadratic, cubic and quartic equations, people imagined
that there might be similar algebraic methods for solving any polynomials using radicals. However, this is not the
case. Galois theory demonstrates that it is impossible to solve the general quintic, and of course polynomial
equations of higher degree.

7. Two parallel planes cut the sphere of unit radius into three equal volumes. Find a cubic equation in rational
coefficients, one root of which is the separation d between the two planes.

Symmetric polynomials
So far, we have mainly considered polynomials in one variable. The difference of three cubes was an exception,
as it featured three variables. Indeed, it is what is known as a symmetric polynomial, as interchanging any two of
the variables leaves the polynomial unchanged. x2  y2 is not symmetric, since interchanging x and y negates the
value, rather than preserving it. x  y2 , however, is symmetric.

8. Suppose we have a symmetric polynomial in two variables, x and y. Show that it can be expressed as a
polynomial in s and p, where s  x  y and p  x y.

This is a special case of Newton’s theorem of symmetric polynomials:

 Any n-variable symmetric polynomial in x1 , x2 , …, xn can be expressed as a polynomial in the elementary symmetric
polynomials (ESPs), i.e. coefficients of x  x1  x  x2  …x  xn . [Newton’s theorem of symmetric polynomials]

Proof:
Note that any symmetric polynomial can be multiplied out to yield a sum of terms of the form
k  x1 a1 x2 a2 … xn an , where the sigma indicates a symmetric sum. We will represent this as k f a1 , a2 , …, an .
sym

We assume without loss of generality that a1  a2  …  an , and lexicographically order the terms. (Specifically,
f a1 , a2, …, an  precedes f b1 , b2 , …, bn  if a1  b1 , or a1  b1 and a2  b2 , or a1  b1 and a2  b2 and a3  b3 , et

_0DWKHPDWLFDO2O\PSLDG'DUN$UWV

cetera.) We then proceed via an inductive argument.

For a given term k f a1 , a2 , …, an  of the ordering, assume that all preceding terms can indeed be expressed as
polynomials in the ESPs. Suppose that a1  a2  …  ah  ah1 . We subtract the symmetric polynomial
k f 1, 1, …, 1, 0, …, 0 f a1  1, a2  1, …, ah  1, ah1 , ah2, …, an . Since f 1, 1, …, 1, 0, …, 0 is already
an elementary symmetric polynomial, and f a1  1, a2  1, …, ah  1, ah1 , ah2 , …, an  is a symmetric polyno-
mial of lower degree than the original expression, the term we have subtracted can be expressed as a polynomial
in ESPs. The remainder exclusively contains terms that precede f a1 , a2 , …, an , thus can also be expressed as a
polynomial in the ESPs.

9. Given real numbers a, b, c, with a  b  c  0, show that a3  b3  c3  0 if and only if a5  b5  c5  0.

[BMO2 2004, Question 3]
 0DWKHPDWLFDO2O\PSLDG'DUN$UWV_

Solutions
1. Ellipses can be represented by quadratic equations in x and y (like all conic sections). As a consequence of
Bezout’s theorem, they can intersect in no more than 2 2  4 points.

2. Consider an arbitrary point Q in general position on the circumcircle of the n-gon, and consider the curve
f P  f Q. It is a quartic curve (by definition) and must intersect the circumcircle in 2 n points (rotations
and reflections of Q). Due to Bezout’s theorem, a quartic and circle sharing no common component can
only intersect in at most 8 points; however, 2 n  8, so the quartic must contain the circle. Hence, all points
on the circle satisfy f P  f Q.

x y z
3. x3  y3  z3  3 x y z  det z x y is the volume of the equilateral parallelepiped with vertex
y z x
O  0, 0, 0 and adjacent vertices A  x, y, z, B  y, z, x and C  z, x, y. Suppose we fix
x2  y2  z2  r2 , the square of the side length, instead of fixing the volume. If O, A and B are constrained to
lie in a plane, then the volume is maximised when O C is a normal to this plane. Hence, the volume is
maximised relative to the side length when the parallelepiped is a cube with volume r3 . As such, the
minimum value of r2 for a parallelepiped with unit volume is 1, when x, y, z  1, 0, 0.

b c d b
4. Firstly, we divide by a to obtain x3  x2  x  0. Let y  x  3a
. Then, our cubic becomes
a a a
y   y  
b 3 b b 2 b d
3a
 a 3a
 y 3a
 a
 0. By using the binomial expansion of the first two terms, the
coefficients in y3 and y2 are one and zero, respectively.

5. We let q  a3  b3 and p  3 a b. a3 and b3 are the roots of z  a3  z  b3   0, which expands to

z2  a3  b3  z  a3 b3  0. We already have a3  b3  q, and a3 b3  a b3      27 .
p 3 p3
3

6. x3  a3  b3  3 a b x  x  a  b x   a  2 b x  2 a   b  0 has roots x  a  b,

x   a  2 b and x  2 a   b. Using the Babylonian formula for solving the quadratic equation
gives us the values of a3 and b3 , whence we can obtain a and b by cube-rooting.

d d

7. The volume enclosed by the two planes is given by   y2

x   1  x2 
x  d  .
2 2 d3
d d
12
For this to be
 
2 2

 9
1 d3 4 16
3
of the total volume of the sphere, we have d  12
. This simplifies to d 3  12 d  3
 0, one root
of which must be the separation between the planes.

8. This is a special case of Newton’s theorem of symmetric polynomials, which is proved later in the chapter.

9. We have a3  b3  c3  3 a b c, by the difference of three cubes. Also, a2  b2  c2  2 a b  2 b c  2 c a,

since a  b  c2  0. We can express a5  b5  c5 as a  b  c a4  b4  c4    a4 b. The first term is
sym
zero, so a5  b5  c5   a4 b  a3 b c  b3 c a  c3 a b  a3  b3  c3  a b  b c  c a. After factorisation,
sym

this is then equal to a b ca2  b2  c2   3 a b ca b  b c  c a, or a b ca2  b2  c2 . Except in the trivial
5
2
case where all variables are zero, a2  b2  c2  0, so a3  b3  c3 and a5  b5  c5 are positive real multiples
of a b c, thus both have the same sign.