Enrichment Lectures 2010

Some facts and problems about polynomials

Finbarr Holland,
Department of Mathematics,
University College Cork,
[email protected];
February 18, 2010

1 A quick review of complex numbers

Although people had used complex numbers long before him, it was the Irish mathe-
matician Hamilton (18051865) who axiomatised them. He defined them as ordered
pairs (a, b) of real numbers that obeyed certain operations of addition and multi-
plication. Using the usual laws of addition and multiplication for real numbers, he
defined the sum and product of two such pairs (a, b), (c, d) by

(a, b) + (c, d) = (a + c, b + d), (a, b)(c, d) = (ac bd, ad + bc), a, b, c, d R.

Since addition and multiplication of real numbers are commutative operations, it

follows from the definitions just given that the same is true of the new operations:

(a, b) + (c, d) = (c, d) + (a, b), (a, b)(c, d) = (c, d)(a, b), a, b, c, d R.

Notice that multiplication distributes over addition:

(a, b) ((c, d) + (e, f )) = (a, b)(c, d) + (a, b)(e, f ), a, b, c, d, e, f R.

As well, the new operations are associative, properties which are inherited from the
real numbers. For instance, a, b, c, d, e, f R,

(a, b) + ((c, d) + (e, f )) = (a, b) + (c + e, d + f )

= (a + (c + e)), b + (d + f ))
= ((a + c) + e, (b + d) + f )
= (a + c, b + d) + (e, f )
= ((a, b) + (c, d)) + (e, f ).

1
Exercise 1. Prove that

(a, b) ((c, d)(e, f )) = ((a, b)(c, d)) (e, f ), a, b, c, d, e, f R.

This means that we can unambiguously define the sum, u + v + w, (respectively, the
product uvw) of three complex numbers u, v, w as either u + (v + w) or (u + v) + w
(respectively, as u(vw) or (uv)w). In particular, we can define successive powers of
a complex number in a clear way.
We denote the set of complex numbers by C, and usually use the letter z as a generic
complex number. The real numbers can be thought of as forming a subset of the
complex numbers R: this is because pairs of the form (a, 0) have algebraic properties
exactly similar to the real numbers, and we simply identify such pairs with a, i.e.,
we write a in place of (a, 0). (Thus (an isomorphic copy of) R sits inside C: R C.)
In particular, we write 1 for (1, 0). The pair (0, 1) is also singled out for special
mention, and denoted by the letter i. Its square is given by

i2 = (0, 1)(0, 1) = (02 12 , 0 1 + 1 0) = (1, 0) = 1.

Also, by the defining laws of operation,

a + bi = (a, 0)(1, 0) + (b, 0)(0, 1) = (a, 0) + (0, b) = (a, b), a, b R.

This gives the customary expression for a complex number, in which a is its real part,
and b its imaginary part. Such expressions are manipulated according to the usual
operations of real numbers with the proviso that whenever i2 occurs it is replaced
by 1. For example, treating everything as a real number,

(a + bi)(c + di) = ac + a(di) + (bi)c + (bi)(di)

= ac + adi + bci + bdi2
= ac + (ad + bc)i ad
= (ac bd) + (ad + bc)i,

in agreement with the definition of multiplication given at the start of this discussion.
real, its complex conjugate is defined to be z = a bi,
If z = a + bi, where a, b are
and its modulus by |z| = a2 + b2 .

Exercise 2. Suppose z, w C. Prove that

z + w = z + w, zw = z w, |z| = z z.

Remark 1. Under the usual laws of addition and multiplication of matrices, complex
numbers can be viewed as 2 2 matrices of the form
 
a b
, a, b R.
b a

2
For,
       
a b c d a+c b+d a+c b+d
+ = = ,
b a d c b d a + c (b + d) a + c

and
      
a b c d ac bd ad + bc ac bd ad + bc
= = .
b a d c bc ad bd + ac (ad + bc) ac bd

Remark 2. Buoyed by his success of defining complex numbers in the above manner,
Hamilton tried for many years to find a method of defining operations on triplets of
real numbers (a, b, c) so that they could be manipulated as if they were real numbers.
But he didnt succeedin fact, no such operations can be devised. However, on
October 16, 1843, on his way to Dunsink Observatory along the Royal Canal, he
discovered a way of multiplying quadruples of real numbers (a, b, c, d), which he wrote
as a + bi + cj + dk, and called quaternions. Crucially, this was a non-commutative
operation. He scratched the defining rules that i, j, k should obey on Broom Bridge.
The year 2005 was designated the Hamilton Year by the Irish Government. The
Central Bank of Ireland issued a special 10 euro coin in his honour, and An Post
struck a special stamp which carried the rules of the non-commutative algebra of
quaternions:
i2 = j 2 = k 2 = ijk = 1.

Remark 3. Quaternions can also be thought of as 22 matrices of pairs of complex

numbers of the form  
z w
, z, w C.
w z

2 Polynomials
A polynomial is a function p defined on the complex numbers C whose value at
z C is given by a linear combination of powers of z:

p(z) = a0 + a1 z + a2 z 2 + + an z n z C.

The numbers a0 , a1 , . . . , an attached to the various powers of z are independent of

z; they are called the coefficients of p, and can be real or complex numbers. A
polynomial is known when these are specified. The highest power of z present in the
display, by which is meant that the corresponding coefficient is non-zero, is called
the degree of p. In the polynomial p displayed, its degree is n provided that an 6= 0.
Polynomials of degree 0 are constants. Those of degree 1 are called linear polynomi-
als, of degree 2, quadratic polynomials, of degree 3, cubic polynomials, of degree 4,
quartic polynomials, of degree 5, quintic polynomials, and so on. For short, we may
refer to them as lines, quadratics, cubics, etc.

3
A complex number z0 is called a root or zero of a polynomial p if its value at z0 is
zero:
p(z0 ) = 0.
A polynomial all of whose coefficients are real numbers may not have any real roots.
The simplest example is the quadratic x2 +1. This has two complex roots, viz., i, i.
Theorem 1. A polynomial of odd degree whose coefficients are real numbers, has at
least one real root.
Intuitive solution. Consider their graphs in the plane. For instance, the graphs of
ones of degree 1 are straight lines, which consist of sets of points of the plane that
lie above and below the horizontal axis, and contain a point of it. Likewise, graphs
of cubics occupy regions of the plane that lie above and below the horizontal axis.
Since these regions are joined (??), their union must contain a point of the real axis.
Generally, if the degree of
p(x) = an xn + + a0
is odd, for all sufficiently large x > 0, the sign of the output p(x) matches that of
an , while if x is large and negative, p(x) has the same sign as an . Thus, p takes
both positive and negative values, and being a smooth function it must therefore
assume the value zero. In other words, it must have a real root. 

The reality of the coefficients specified in this theorem is essential. For example, the
linear polynomial x i has no real root.
Theorem 2 (Gauss). Every non-constant polynomial has at least one root, which
may be complex. A polynomial of degree n has at most n distinct roots.
This deep result is called the Fundamental Theorem of Algebra. We take it for
granted.
Theorem 3. Suppose p has real coefficients and z is one of its a complex roots.
Then z is also one of its roots.
Proof. Suppose
p(x) = a0 + a1 x + a2 x2 + + an xn x C,
where a0 , a1 , . . . , an are real numbers. Then, by repeated application of Exercise 2,
0 = p(z)
= a0 + a1 z + a2 z 2 + + an z n
= a0 + a1 z + a2 z 2 + + an z n
= a0 + a1 z + a2 z 2 + + an z n
= a0 + a1 z + a2 z 2 + + z n
= p(z).
Thus p(z) = 0 implies p(z) = 0. 
Some further information about polynomials is given in an Appendix.

4
3 Linear and quadratic polynomials
Here, we focus on linear and quadratic polynomials whose coefficients are real num-
bers, and restrict their domain of definition to be the set of real numbers. We
can then consider their graphs and thence examine their properties in a geometric
manner. These are functions of the form
ax + b, ax2 + bx + c, x R,
where a, b, c are real, and a 6= 0.

3.1 Graphs of lines

While the Greeks worked out the geometry of the straight line, they had no away
of representing them in an algebraic manner. It fell to Descartes to describe a way
of doing this: one introduces the cartesian plane R2 of points with two coordinates,
a procedure with which you are familiar. Graphs of functions are then identified as
subsets of this plane whose intersection with every vertical line consists of at most
a single point. As a consequence, a circle is not the graph of a function, nor is the
parabola {(x, y) R2 : y 2 = 4x}.
If c + mx is a linear polynomial, its graph in R2 is the set of points
{(x, y); y = mx + c, x R} = {(x, mx + c) : x R},
which is abbreviated to y = mx + c; m is its slope, and c its y-intercept.
Given two distinct points (x1 , y1 ), (x2 , y2 ) in its graph, so that
y1 = mx1 + c, y2 = mx2 + c,
in which case x2 6= x1 , we can solve these equations for m, c and, doing so, find that
y2 y1 y1 x2 y2 x1
m= , c= .
x2 x1 x2 x1
Substituting these into the equation, and rearranging the resulting expression, we
obtain the familiar formula for the equation of a line through two points which are
not on the same vertical line:
y2 y1
y y1 = (x x1 ).
x2 x1
Equivalently in this case,
(x2 x1 )(y y1 ) = (y2 y1 )(x x1 ).
But this formula works even if x2 = x1 , provided that y2 6= y1 ; or if y2 = y1 and
x2 6= x1 . But we cant have y2 y2 = x2 x1 = 0 unless we wish to regard the entire
plane as a line!
To cover all possibilities, then, the equation of a line in R2 is an expression of the
form
ax + by + c = 0,
where a2 + b2 > 0. This is the graph of a linear polynomial iff ab 6= 0. It is the
graph of a constant polynomial if a = 0, b 6= 0.

5
3.2 Distance formula
Various candidates present themselves that qualify as a distance between a pair of
points P = (x1 , y1 ), Q = (x2 , y2 ). The usual onewhich youll recogniseis given
by p
|P Q| d2 (P, Q) = (x2 x1 )2 + (y2 y1 )2 .
Another onecalled the taxi-cab metricis given by

d1 (P, Q) = |x2 x1 | + |y2 y1 |.

An even simpler one is given by


6 Q,
1, if P =
d0 (P, Q) =
0, if P = Q.

If, for the moment, d is any one of these, L is a straight line, and P0 = (x0 , y0 ) is
any point, is there a point R in L which is nearest to P0 ? If so, is it unique? In
other words, does there exist a point R L such that

d(P0 , R) d(P0 , X), X L,

and, if so, is it unique?

This is an example of whats called an Existence and Uniqueness Problem, and
possibly the first one of its kind that is encountered in second level mathematics. If
d = d2 , the answer to both questions is in the affirmative, as you know: the foot of
the perpendicular from P0 onto L is the point that is nearest to P0 in this metric.
If L = {(x, y) : ax + by + c = 0, a2 + b2 > 0}, then

ac aby0 + b2 x0 bc + a2 y0 abx0
R=( , ),
a2 + b 2 a2 + b 2
and
|ax0 + by0 c|
|P0 R| = .
a2 + b 2
However, by contrast, while the answer to the first question is still in the affirmative
if we measure distance using either d0 or d1 , we lose uniqueness. For instance, if we
use d0 , then the distance between every point in L and P0 is 1, unless P0 L, in
which case the distance between them is 0, and R = P0 .

Exercise 3. Work out a solution to the problem when d1 is used.

Exercise 4. Show that if P, Q, R are three points in the plane, then

d1 (P, Q) d1 (P, R) + d1 (R, Q).

6
3.3 Herons problem
Given two distinct points P, Q on the same side of a line L, is there a point R L
such that
|P R| + |RQ| |P X| + |XQ|, X L?
This is another Existence and Uniqueness Problem that was first considered by
Heron1 who gave a beautiful solution based on the notion of the reflection of a point
in a line. (Recall that P 0 is the reflection of P in L if both points are equidistant
from L, and their mid-point belongs to L.)

Theorem 4 (Heron). Given two distinct points P, Q on the same side of a line L,
there is a unique point R L such that

|P R| + |RQ| |P X| + |XQ|, X L.

Proof. Let P 0 be the reflection of P in L, so that if X L, then |P X| = |P 0 X|.

Join P 0 and Q. The line M passing through these points meets L at the desired
point R. For

|P X| + |XQ| = |P 0 X| + |XQ| (by reflection)

|P 0 Q| (by the triangle inequality)
= |P 0 R| + |RQ| (since R M L)
= |P R| + |RQ| (again by reflection).

This result incorporates Herons Principle of the Shortest Path of Light: If a ray of
light propagates from point A to point B within the same medium, the path-length
followed is the shortest possible.

Remark 4. Snooker players use this fact instinctively!

3.4 Graphs of quadratics

The simplest quadratic is the square function x x2 , whose graph in R2 is given
by the set
{(x, y) : y = x2 : x R}.
This is a subset in the upper half-plane that is symmetric about the vertical axis,
and the origin is the lowest point on it, which is its only turning point. Also, the
square function is convex.i.e., it is a smile. To see this, let P = (x1 , y1 ), Q = (x2 , y2 )
be two distinct points on its graph, so that y1 = x21 , y2 = x22 . The line through P
1
Heron (or Hero) of Alexandria (c. 10 70 AD) was a mathematician and engineer, who is said
to be the greatest experimenter of antiquity. For instance, he is credited with inventing a vending
machine to dispense water, and a windwheel to operate an organ.

7
and Q has equation
y2 y1
y = y1 + (x x1 )
x2 x1
x2 x21
= x21 + 2 (x x1 )
x2 x1
= x21 + (x2 + x1 )(x x1 )
= (x2 + x1 )x x1 x2 .

Thus, the line P Q lies above the arc P

d Q since

x2 (x2 + x1 )x + x1 x2 = (x x1 )(x x2 ) 0,

for all x between x1 , x2 , i.e.,

x2 (x2 + x1 )x x1 x2 ,

for all x between x1 , x2 , as required.

The square function separates the plane into two disjoint regions, viz., the sets of
points above and below its graph. The set above it is {(x, y) : y > x2 }; the set below
is {(x, y) : y > x2 }. The upper set is convex: the line segment joining any two of its
points is wholly contained in it. The lower set is neither convex nor concave.
The graph of any quadratic can be reduced to that of the square function or to its
reflection in the horizontal axis, by a process known as completing the square.
To justify this, suppose
p(x) = ax2 + bx + c
is a quadratic polynomial, so that a 6= 0. Then, if (x, y) is a point in its graph,

y = ax2 + bx + c
b
= a(x2 + x) + c
a
2 b b2 b2
= a(x + x + 2 ) + c a 2
2a 4a 4a
b 2 4ac b2
= a(x + ) + .
2a 4a
Thus
4ac b2 b
y = a(x + )2 ,
4a 2a
Or, changing the coordinate axes by translating the origin, we have Y = aX 2 , where

4ac b2 b
Y =y , X =x+ .
4a 2a
Hence, the graph is convex, i.e., a smile if a > 0, and concave, i.e., a frown, if a < 0.

8
 b
We can also conclude that the graph of p is symmetric about the line x = 2a ,
b 4acb2
and that it has a single turning point at ( 2a , 4a ). Moreover, this is a minimum
point if a > 0, and a maximum point if a < 0, i.e.,
2
4acb

4a
, if a > 0,
p(x) 4acb2
4a , if a < 0,
b
with equality here iff x = 2a . In other words,

4ac b2
min{p(x) : x R} =
4a
if a > 0, and
4ac b2
max{p(x) : x R} =
4a
if a < 0.

Example 1. If a, b, c are real numbers, and a 6= 0, then

p(x) = ax2 + bx + c 0, x R,

iff a > 0, c 0 and b2 4ac. These conditions hold iff p is the square of the modulus
of a linear polynomial.

Proof. Suppose p 0 on R. Then, in particular, c = p(0) 0. Next, for x 6= 0,

b c 1
a+ + 2 = p(x) 2 0,
x x x
and so, letting x , we deduce that a 0. But, a 6= 0. Hence, a > 0, so that

4ac b2 b
= p( ) 0,
4a 2a
whence b2 4ac. Thus the stated conditions are necessary to ensure the nonnega-
tivity of p. Conversely, if they hold, then, by completing the square, we see that

b 2 4ac b2 4ac b2
p(x) = a(x + ) + 0, x R,
2a 4a 4a
from which it also follows that p is the square of the modulus of the linear polynomial

b + 4ac b2 i
ax +
2 a

which has complex coefficients.

9
Example 2. The pairs of real numbers (m, c) such that y = mx + c is a line below
the graph of the square function comprise the set

{(m, c) : m2 + 4c 0}.

Proof. Suppose (m, c) generates such a line. Then

mx + c x2 , x R.

Equivalently, the quadratic polynomial x2 mx c is nonnegative for all real x. By

the previous example, this occurs iff m2 4(c). The result follows. 

4 Exercises
1. Sketch the graphs of the polynomials

3x + 2, 2x 3, 3x2 + 4x 2, 3x2 + 4x 2, (x )(x ), (x )(x ),

where , are arbitrary real numbers.

2. Determine the minimum of each of the quadratics

(x 1)(x 2), (x 3)(x 4), (x 1)(x 3), (x 2)(x 4).

3. Determine the minimum of the quadratic

(x 1)2 + (x 2)2 + (x 3)2 + (x 4)2 .

4. Determine the minimum of the quadratic

a(x )2 + b(x )2 + c(x )2 ,

where a, b, c, , , are arbitrary real numbers, and at last one of a, b, c is

non-zero.

5. Determine the minimum of each of the quartics

(x1)(x2)(x3)(x4), (x1)(x2)+(x1)(x2)(x3)(x4)+(x3)(x4).

6. Let P = (1, 1), Q = (1, 1). Show that there is a point R on the line L,
whose equation is x + y = 1, such that

|P R|2 + |RQ|2 |P X|2 + |XQ|2 , X L.

10
 7. Let P, Q be two points not necessarily on the same side of a line L. Prove that
there is a unique point R L such that

|P R|2 + |RQ|2 |P X|2 + |XQ|2 , X L.

8. Let x, y be a pair of real numbers, prove that

x2 + y 2 + 2 (x + 1)(y + 1),

with equality iff x = y = 1.

9. Show that the set {(x, y) R2 : y |x|} is convex, and describe all the pairs
(m, c) such that the line y = mx + c lies below the graph of y = |x|.

10. Show that the semicircle {(x, y) : 1 x 1, y = 1 x2 } is convex, and
describe all the pairs (m, c) such that the line y = mx + c lies below its graph.

11. For what real numbers a 6= 0, b, c is the quadratic ax2 + bx + c nonnegative on

the half-line [0, )? On the interval [0, 1]?

5 Quartic polynomials
A quartic is a polynomial of degree 4, i.e., a linear combination of the simple mono-
mials 1, x, x2 , x3 , x4 , and is therefore of the form

p(x) = ax4 + bx3 + cx2 + dx + e,

where the coefficients a, b, c, d, e are real or complex numbers, and a 6= 0.

Every quartic is a product of two quadratics. For, counting their multiplicity, such
a polynomial has 4 roots, and if these are denoted by , , , , then

p(x) = a(x )(x )(x )(x ) = a(x2 ( + )x + )(x2 ( + )x + ),

a product of two quadratics, which evidently are not unique. Conversely, its easy
to see that the product of two quadratics is a quartic.
Theorem 5. If a quartic has real coefficients, then it can be expressed as a product
of two quadratics each having real coefficients.
Proof. Suppose
p(x) = ax4 + bx3 + cx2 + dx + e,
where a, b, c, d, e are real, and a 6= 0. The result is immediate if p has only real
roots. If z is a complex root of p, then so is z, by Theorem 3. Since p has at most
4 distinct roots, either they are all complex, and come in pairs, or at most two are
non-real. If they are all complex, then we can write them as , , , , in which case

p(x) = a(x2 (+)x+)(x2 (+)x+ ) = a(x2 (+)x+||2 )(x2 (+)x+||2 ).

11
Since the coefficients of x in each factor are real numbers, the result follows in this
case. If exactly two are non-real, they must be each others complex conjugate, and
so, this time, the roots can be denoted by , , , , where , are real numbers. As
before,
p(x) = a(x2 ( + )x + ||2 )(x2 ( + )x + ),
and, once more, we see that each factor has real coefficients. 

5.1 Graphs of quartics

Here, we consider quartics with real coefficients, and say a few words about their
graphical representation in the cartesian plane.
The simplest quartic is p(x) = x4 = (x2 )2 , whose graph is symmetric about x = 0
and convex; it is a large smile. Whereas the graph of every quadratic is symmetric
about some vertical line, the graph of a quartic need not be symmetric about any
such line. For example, the graph of the quartic p(x) = x4 + x is not symmetric
about any vertical line. For, if for some real (or complex) h, p(x + h) = p(h x)
for all x, then (?) every number is a root of the cubic 4hx3 + (4h3 + 1)x, which is
absurd.
Since x4 is positive for all non-zero x, the sign of the general quartic p(x) = ax4 +
bx3 + cx2 + dx + e, a 6= 0, is the same as that of a, for all large values of x. In
other words, if a > 0, the graph lies in the upper half-plane for all sufficiently large
x; whereas, if a < 0, it lies in the lower half-plane for all sufficiently large x. Thus,
to describe its graph completely we must examine the behaviour for all x in some
symmetric interval. Thus, we need to determine its turning points, and points of
inflexion, if any. These will provide information about the shape of the graph, and
determine where its convex, concave, increasing and decreasing. To do this, we
must use calculus methods. The turning points, are found by finding the first and
second derivative p0 , and solving the equation p0 (x) = 0; the points of inflexion are
found by determining the second derivative, p00 , and solving p00 (x) = 0. Since p0 has
real coefficients there must be at least one turning point, but there need not be any
point of inflexion. For instance, this is the case when p(x) = x4 + 2x2 + 1, which is
symmetric about x = 0, and convex. It has no real root, only one turning point at
x = 0, where it has a local minimum, and no point of inflexion.

5.2 Exercises
1. Let p be an arbitrary quartic. Prove that

p00 (h)x2 p000 (h)x3 p(iv) (h)x4

p(x + h) = p(h) + p0 (h)x + + + ,
2! 3! 4!
where p0 (h), p00 (h), p000 (h), p(iv) (h) stand for the first, second, third and fourth
derivatives of p evaluated at h.

12
 2. Let p be a arbitrary quartic. Use the previous exercise to show that we may
choose h so that the coefficient of x3 in the expansion of p(x + h) is zero.

3. Suppose the roots of p(x) = ax4 + bx3 + cx2 + dx + e, a 6= 0, are in arithmetic

progression. Prove that b3 4abc + 8a2 d = 0.

4. Determine the roots of x4 12x3 + 49x2 78x + 40. Show that its graph is
symmetric about x = 3.

5. Show that
(x 1)(x 2)(x 4)(x 5) 9/4, x.

6. Prove that the quartic p(x) = ax4 + bx3 + cx2 + dx + e, a 6= 0 has no points
of inflexion iff 3b2 < 8ac.

6 Cubic polynomials
We consider cubics of the form

p(x) = ax3 + bx2 + cx + d,

where a 6= 0.
Theorem 6. Suppose a, b, c, d are real numbers. Then p has at least one real root.
Proof. Weve already outlined one proof of this fact based on the fact that ap(x)
is positive for large positive x and negative for large negative x, a statement which
extends to cover the case of any polynomial of odd degree having real coefficients.
Heres another approach which uses Theorem 2 due to Gauss. According to this,
every cubic (with real or complex coefficients) has at least one root, which may be
complex, and at most three distinct roots. So, p has three roots, which may not all
be different. Since its coefficients are real, by Theorem 3, its complex roots come in
pairs. Hence, one of the roots must be real. 

As the next example shows, this is the most we can say, in general.
Example 3. Find the roots of p(x) = x3 1.
Solution. Clearly 1 is a root. Hence x 1 is a factor of p. In other words, there
is a quadratic q such that p(x) = (x 1)q(x). Its straightforward to verify that
q(x) = x2 + x + 1. Now, the discriminant of the latter is b2 4ac = 1 4 = 3,
which is negative, and so, by the usual formula, the roots of q are complex, and
equal to
1 3i
.
2
Thus, p has only one real root. 

13
Its standard to let
1 + 3i
= ,
2
so that
6= 1, 3 = 1, 2 + + 1 = 0.
Well refer to this later on.

6.1 Graphs of cubics

The simples cubic is p(x) = x3 . This has precisely one real root, of multiplicity
three. Note that its first and second derivatives vanish at x = 0 also. While it
has a point of inflexion there, since p takes on both positive and negative values
in every neighbourhood of it, 0 is neither a local maximum nor a local minimum.
Since, p(x) p(x) for all x, p is an odd function, which means that its graph is
anti-symmetric about the y-axis. Also, p0 (x) = 3x2 0, x, which means that
the graph is strictly increasing on (, ). In addition, the graph is convex over
the interval (0, ), and concave over (, 0). Putting these facts together we can
sketch the graph of y = x3 .
Example 4. Discuss the graph of y = x3 + 3x.
Solution. Since x3 + 3x = x(x2 + 3), and x2 + 3 has no real root, this cubic crosses
the x-axis only at 0. Next, p0 (x) = 3(x2 + 1), which signifies that the cubic has no
tangent parallel to the x axis. Thus it has no turning points. But since p0 is every-
where positive, the graph is strictly increasing on (, ). Finally, p00 (x) = 6x,
which vanishes only when x = 0. From the fact that p00 (x) > 0 if x > 0, we may
conclude that the graph is convex on (0, ). Since p00 (x) < 0 if x < 0, the graph is
concave on (, 0). 

Theorem 7. Suppose a cubic has three real roots. Then it has two turning points.
Proof. Suppose
p(x) = (x )(x )(x ) = x3 ( + + )x2 + ( + + )x ,
where , , are real. Then
p0 (x) = 3x2 2( + + )x + ( + + ),
which has real roots iff b2 4ac. Now
b2 4ac = 4( + + )2 12( + + )
= 4[( + + )2 3( + + )]
= 4[2 + 2 + 2 ]
= 2[( )2 + ( )2 + ( )2 ]
0,

14
with equality iff = = . Thus p0 has two real roots, which may be equal. 

Theorem 8. Suppose p is a cubic, not necessarily with real coefficients, and

p0 () = p0 () = 0 = p00 ().

Then
+
= .
2
Proof. Suppose
p(x) = ax3 + bx2 + cx + d, a 6= 0.
Then
p0 (x) = 3ax2 + 2bx + c, p00 (x) = 6ax + 2b.
Hence
b
= .
3a
But also,
p0 (x) = 3a(x )(x ) = 3a(x2 ( + )x + ),
which means that
2b
3a( + ) = 2b, + = = 2.
3a
This means the if a cubic with real coefficients has two turning points, then the
mid-point of their first coordinates coincides with the point of inflexion.

6.2 Roots of cubics

To discuss the roots of a general polynomial with real coefficients, we need to compile
some preliminary facts. We start with an identity which is useful in other situations
as well.

Lemma 1. Suppose x, y, z are real or complex numbers. Then

x3 + y 3 + z 3 3xyz = (x + y + z)(x2 + y 2 + z 2 xy yz zx).

Proof. Simply expand the RHS. 

Taking y = 1, z = 0 we get the more familiar factorisation:

x3 + 1 = (x + 1)(x2 x + 1),

while the choice y = 1, z = 0 tells us that

x3 1 = (x 1)(x2 + x + 1).

15
More generally, the identities

x3 + y 3 = (x + y)(x2 xy + y 2 ), x3 y 3 = (x y)(x2 + xy + y 2 ),

hold for all real or complex x, y.

Recall the definition of

Lemma 2. Suppose x, y, z are real or complex numbers. Then

x2 + y 2 + z 2 xy yz zx = (x + y + 2 z)(x + 2 y + z).

Proof. For,

(x + y + 2 z)(x + 2 y + z)
= x2 + xy( 2 + ) + xz( + 2 ) + 3 y 2 + yz( 2 + 4 ) + 3 z 2
= x2 xy xz + y 2 + yz( 2 + ) + z 2
= x2 + y 2 + z 2 xy yz zx.

Combining these factorisations we deduce that, for all x, y, z,

x3 + y 3 + z 3 3xyz = (x + y + z)(x + y + 2 z)(x + 2 y + z).

As an immediate consequence, we have the following statement.

Corollary 1. The roots of the cubic polynomial

p(x) = x3 3abx + a3 + b3

are given by
(a + b), (a + 2 b), ( 2 a + b).

We can exploit this to determine the roots of cubic polynomials of the form x3 +qx+r,
as long as we can find a, b so that

3ab = q, a3 + b3 = r. (1)

Before discussing the general case, we look at a simple example.

Example 5. Determine the roots of p(x) = x4 + 3x2 3x + 4.

Solution. The first step is the eliminate the x2 term. We can do this by shifting the
x-axis. Noting that

p(x) = (x + 1)3 6x + 3 = (x + 1)3 6(x + 1) + 9,

its enough to find the roots of X 3 6X + 10. So, choose a, b so that

2 = ab, a3 + b3 = 9.

16
Plainly, the pair a = 1, b = 2 works. Hence,

X 3 6X +10 = (X +3)(X + +2 2 )(X + 2 +2) = (X +3)(X 1+ 2 )(X 1+).

In other words,
p(x) = (x + 4)(x + 2 )(x + )
and 4, , 2 are the roots of p.
Return to (1). For these relations to hold, a3 , b3 must be the roots of the quadratic

q3
z 2 rz ,
27
and so can be determined by the usual formula,
q
4q 3
r r2 + 27
3 3
a ,b = .
2
So, a, b can be determined by extracting cube roots of possibly complex numbers.
We could, for instance, choose a to be one of the cube roots of
q
3
r + r2 + 4q 27
,
2
something we gloss over.
Turning to the determination of the roots of a general cubic p(x) = ax3 + bx2 +
cx + d, a 6= 0, since the roots dont depend on the sign of a, we can suppose, for
simplicity, that a = 1. The first step to perform is to eliminate the term involving
x2 by shifting x. In fact,

1 1 b3
x3 + bx2 + cx + d = (x + b)2 b2 x + cx + d
3 3 27
1 1 1 b3 1 1
= (x + b)2 + (c b2 )(x + b) (c b2 )( b) + d
3 3 3 27 3 3
3
1 1 2b
= X 3 + (c b2 )X + d bc +
3 3 27
= X 3 + qX + r,

where
1 1 1 2b3
X = x + b, q = c b2 , r = d bc + .
3 3 3 27
Next we select a, b so that

3ab = q, a3 + b3 = r;

this entails solving the quadratic equation z 2 rz q 3 /27 = 0. The final step is to
use the factorisation in the above Corollary.

17
6.3 Exercises
1. Determine the roots of the cubic x3 (a2 + ab + b2 )x + ab(a + b).

2. Suppose x1 6= x2 and yi = x31 , y2 = x32 . Write down the equation of the line
through the points (x1 , y1 ), (x2 , y2 ) and prove that it cuts the graph of y = x3
at three points, two of which may coincide.

3. Prove that the cubic y = x3 is convex on (0, ), and concave on (, 0).

7 Appendix
This contains additional information about polynomials and some IMO-type prob-
lems.

7.1 Some more facts

1. The collection of polys is closed under addition, multiplication and composi-
tion. In other words, if

p(x) = a0 + a1 x + + an xn , q(x) = b0 + bx + + bm xm ,

are polys of degree n, m then p + q, pq, p q, q p are polys and

deg(p + q) max(deg p, deg q), deg(pq) = m + n, deg(p q) = deg(q p) = mn.

2. The following are special polynomials. Let a be a fixed complex number and
n N:
n1 n  
n n
X
nk+1 k
X n nk k
x a , x a , x a ,
k=0 k=0
k
where

n!
  
n (nk)!k!
, if 0 k n,
=
k 0, if k < 0 or k > n
are the Binomial coefficients.
Note that
n1 n  
n n
X
nk+1 k n
X n
x a = (x a) x a , (x + a) = xnk ak ,
k=0 k=0
k

3. a is root (zero) of a poly p if p(a) = 0. If this is so, then we can factor p: there
is a poly q, with deg q = deg p 1, such that

p(x) = (x a)q(x).

18
 Note especially that if a is rational and all the coeffs of p are rational, then
the coeffs of q are also rational.
If b is another root of p, then q(b) = 0, and so we can factor q: q(x) =
(x b)r(x), deg r = deg q 1. Thus p(x) = (x a)(x b)r(x), and so on: if
x1 , x2 , . . . , xn are roots of p, then
n
Y
p(x) = c(x x1 )(x x2 ) (x xn ) = c (x xi ).
i=1

4. Suppose p/q, with p, q Z, (p, q) = 1, q 6= 0, is a rational root of a poly

p(x) = an xn + + a0 whose coeffs are integers. Then p|a0 , q|an .
This is another theorem due to Gauss. It reduces the search for rational roots
of polys with integer coeffs to an examination of a finite number of possibilities
which arise by factorising the integers a0 , an .
5. How are the roots of a poly related to its coeffs? If ax2 + bx + c is a quadratic,
so that a 6= 0, and its roots are x1 , x2 , then

ax2 + bx + c = a(x x1 )(x x2 ) = a(x2 (x1 + x2 )x + x1 x2 ,

whence
b c
x1 + x2 = , x1 x2 = .
a a
3 2
If ax + bx + cx + d is a cubic, so that a 6= 0, and its roots are x1 , x2 , x3 , then

ax3 + bx2 + cx + d = a(x x1 )(x x2 )(x x3 )

= a(x3 (x1 + x2 + x3 )x + (x1 x2 + x2 x3 + x3 x1 )x x1 x2 x3 ),

whence
b c d
x1 + x2 + x3 = , x1 x2 + x2 x3 + x3 x1 = , x 1 x2 x3 = .
a a a
Similar relations hold for higher degree polynomials.
6. The poly xn 1 has roots k , k = 0, 1, 2, . . . , n where
2 2 i2
= cos( ) + i sin( ) = e n .
n n
Thus
n1
Y
n
x 1= (x k ),
k=0

so that, if x is a real number, then

n1
n 2
Y 2k
|x 1| = (x2 2x cos( ) + 1).
k=0
n

19
 The complex number , which depends on n, is never equal to 1, unless n = 1;
it is called an nth root of unity. Its powers 1, , 2 , . . . , n1 are also roots
of xn 1, and form a cyclic group under multiplication. They are also the
vertices of a regular n-gon inscribed in the unit circle. The perimeter of this
n-gon is given by
n
X
| k k1 | = n|1 | = 2n sin( ),
k=1
n

which tends to 2 as n .

7.2 Some identities

If X X
p(x) = ai xi , q(x) = bj x j
i0 j0

are two polys of degrees m, n, their product is also a poly of degree m + n, and
X X
p(x)q(x) = ai x i bj x j
i0 j0
X
= ai bj xi+j
i,j0
X X
= ai bj xi+j
k0 i+j=k
X
= ck xk ,
k0

where
X
ck = ai bj = a0 bk + a1 bk1 + + ak b0 , k = 0, 1, . . . , m + n.
i+j=k

For instance, if p(x) = (1 + x)m , q(x) = ax2 + bx + c, then

m  
m 2
X m i
(1 + x) (ax + bx + c) ( x ) (b0 + b1 x + b2 x2 )
i=0
i
m+2
X X m 
k
= x bj
k=0 i+j=k,j2
i
m+2
X X  m  m+2
X
k
= x bj = ck xk ,
k=0 0j2
k j k=0

where    
m m
c0 = b0 = c, c1 = b0 + b1 = mc + b,
1 0

20
and, for k 2,
           
m m m m m m
ck = b0 + b1 + b2 = c+ b+ a.
k k1 k2 k k1 k2
Again,
m   n  
X m i
X n
( x )( xj ) = (1 + x)m (1 + x)n
i=0
i j=0
j
= (1 + x)m+n
m+n
X m + n 
= xk ,
k=0
k
and comparing coeffs of powers of x,
  X mn
m+n
= , k = 0, 1, 2 . . . , m + n.
k i+j=k
i j

In particular, if k = m = n
  n  2
2n X n
= .
n k=0
k
2 n n n
Since (1 x ) = (1 + x) (1 x) , we see in the same way that
n   n   n   2n X n   
k n n iX j n j n
X X X
2k j k
(1) x = x (1) x = x (1) ,
k=0
k i=0
i j=0
j k=0 i+j=k
i j

whence
X n    
0, if k is odd,
j n
(1) = k/2 n


i j (1) k/2
, if k is even.
i+j=k

In particular,
2m+1
X   
2(2m + 1) i 2(2m + 1)
(1) = 0, m = 0, 1, 2, . . .
i=0
i 2m + 1 i
and
2m      
X 4m i 4m m 4m
(1) = (1) , m = 0, 1, 2, . . .
i=0
i 2m i m

7.3 Interpolation
As weve mentioned, if we know the roots of a poly and their multiplicities, we can
determine the poly to within a constant multiple. What if we know the values that
a poly takes on a prescribed set? Can we determine it? In geometric terms, can we
always find a poly of smallest degree to pass through a finite set of points in the
(x, y)-plane? Since a poly is a function, the x-coordinates of the points better be
distinct, but the y-coordinates dont have to be.

21
Example 6. Suppose x1 , x2 , x3 are three distinct real numbers and y1 , y2 , y3 are any
given set of real numbers, determine the equation of the poly that passes through the
points
(x1 , y1 ), (x2 , y2 ), (x3 , y3 ).

Let the desired poly be p(x) = ax2 + bx + c. Then p(xi ) = yi , i = 1, 2, 3, i.e.,

ax21 + bx1 + c = y1 ,
ax22 + bx2 + c = y2 ,
ax23 + bx3 + c = y3 ,

three equations for three unknowns a, b, c, which can be solved by the process of
elimination. 

Another approach is to solve three similar, but easier problems by letting two of the
ys be 0 and the other 1. Then, say, p1 (x2 ) = p1 (x3 ) = 0, p1 (x1 ) = 1. The first two
conditions say that x2 , x3 are roots of the quadratic p1 . So

p1 (x) = a(x x2 )(x x3 ).

The third condition now fixes a and so p1 :

1 (x x2 )(x x3 )
a= , p1 (x) = .
(x1 x2 )(x1 x3 ) (x1 x2 )(x1 x3 )
Cycling the points x1 , x2 , x3 we determine p2 , p3 as
(x x1 )(x x3 ) (x x1 )(x x2 )
p2 (x) = , p3 (x) = .
(x2 x1 )(x2 x3 ) (x3 x1 )(x3 x2 )
Then
p(x) = y1 p1 (x) + y2 p2 (x) + y3 p3 (x)
solves the problem.
This can be extended to deal with the general problem of finding a poly of degree n
that passes through n + 1 points (xi , yi ), i = 1, 2, . . . , n with distinct x-coordinates.

7.4 Some Olympiad style problems

1. The value of the polynomial x2 +x+41 is a prime number for x = 0, 1, 2, . . . , 39.
The value of the polynomial x2 81x + 1681 is a prime number for x =
0, 1, 2, . . . , 80. These examples might suggest that there is a poly p, with
integer coeffs, such that p(n) is a prime for every integer n 0. This is false.
No such poly exists. [Hint: If such a poly f exists, p = f (0) is a prime and
p|f (mp), m = 0, 1, . . ., i.e., Hence f (mp) = p, m = 0, 1, 2, . . ., which is absurd.]

22
2. For which n N is x2 + x + 1 a factor of x2n + xn + 1? [Hint: Say x2n + xn + 1 =
(x2 + x + 1)q(x), for some poly q.]

3. (IMO 1973) Find the minimum value of a2 + b2 , where a, b are real numbers
for which the equation

x4 + ax3 + bx2 + ax + 1 = 0

has at least one real root.

4. USAMO75. Let p be a poly of degree n and suppose that

k
p(k) = , k = 0, 1, 2, . . . , n.
k+1
Determine p(n + 1). [Hint: Consider the poly q(x) = (x + 1)p(x) x.]

5. IRMO89. Let a be a positive real number, and let

q q
3 3
b = a + a + 1 + a a2 + 1.
2

Prove that b is a positive integer if, and only if, a is a positive integer of the
form 21 n(n2 + 3), for some positive integer n.

6. IRMO91. Find all polynomials

f (x) = a0 + a1 x + + an xn

satisfying the equation

f (x2 ) = (f (x))2
for all real numbers x.

7. IRMO91. Find all polynomials

f (x) = xn + a1 xn1 + + an

with the following properties:

(a) all the coefficients a1 , a2 , . . . , an belong to the set {1, 1}

(b) all the roots of the equation

f (x) = 0

are real.

23
 8. Let a, b, c and d be real numbers with a 6= 0. Prove that if all the roots of the
cubic equation
az 3 + bz 2 + cz + d = 0
lie to the left of the imaginary axis in the complex plane, then

ab > 0, bc ad > 0, ad > 0.

9. IRMO93. The real numbers , satisfy the equations

3 32 + 5 17 = 0,

3 3 2 + 5 + 11 = 0.
Find + .

10. IRMO93. Let a0 , a1 , . . . , an1 be real numbers, where n 1, and let the
polynomial
f (x) = xn + an1 xn1 + . . . + a0
be such that |f (0)| = f (1) and each root of f is real and satisfies 0 < < 1.
Prove that the product of the roots does not exceed 1/2n .

11. IRMO93. Let a1 , a2 . . . an , b1 , b2 . . . bn be 2n real numbers, where a1 , a2 . . . an

are distinct, and suppose that there exists a real number such that the
product
(ai + b1 )(ai + b2 ) . . . (ai + bn )
has the value for i = 1, 2, . . . , n. Prove that there exists a real number
such that the product

(a1 + bj )(a2 + bj ) . . . (an + bj )

has the value for j = 1, 2, . . . , n.

12. IRMO94. Determine, with proof, all real polynomials f satisfying the equation

f (x2 ) = f (x)f (x 1),

for all real numbers x.

24
13. IRMO95. Suppose that a, b and c are complex numbers, and that all three
roots z of the equation
x3 + ax2 + bx + c = 0
satisfy |z| = 1 (where | | denotes absolute value). Prove that all three roots w
of the equation
x3 + |a|x2 + |b|x + |c| = 0
also satisfy |w| = 1.

14. IRMO97. Find all polynomials p satisfying the equation

(x 16)p(2x) = 16(x 1)p(x)

for all x.

15. Prove that x4 + x3 + x2 + x + 1 is a factor of x44 + x33 + x22 + x11 + 1.

16. The coeffs of the cubic ax3 + bx2 + cx + d are integers, ad is odd andbcis even.
Prove that at least one root is irrational.

17. USAMO77. Suppose a, b are two roots of x4 + x3 1. Prove that ab is a root

of x6 + x4 + x3 x2 1.

18. Prove that if x3 + px2 + qx + r has three real roots, then p2 3q.

19. IMO93. Let n > 1. Show that the poly xn + 5xn1 + 3 cannot be written as a
product of two non-constant polys with integer coeffs.

20. The coeffs of p(x) = xn + + 1 are 0. Assume p has real roots. Prove that

p(2) 3n .

8 Some more facts about cubics

1. Suppose f (x) = ax3 + bx2 + cx + d is a cubic poly with a, b, c, d real numbers
and a 6= 0. For x 6= 0,
f (x) b c d
3
=a+ + 2 + 3
x x x x
and, apart from the first, the terms are small for large |x|. Precisely,
f (x)
lim = a.
|x| x3

25
 This means that for large |x|, f (x)/x3 has the same sign as a. Hence, supposing
a > 0, f (x) > 0 if x > 0 is large enough, and f (x) < 0 if x < 0 and |x| is
large enough. (If a < 0, then xf (x) < 0 if |x| is large enough.) Hence, by
the Intermediate-value theorem, the graph of y = f (x) crosses the real axis at
some point. Conclusion: f has at least one real roots.
Note that this is false without the assumption that the coeffs a, b, c, d are real.
For instance, x3 3ix2 3x i = (x i)3 has no real roots.
Notice too that the same reasoning applies to confirm that any poly of odd
degree that has real coeffs has at least one real root.

2. The previous statement tells us that a cubic with real coeffs has at least one
real root; the other roots may be complex numbers, and if they are, they are
each others complex conjugate, once more from the reality of the coeffs.
Is there a formula for finding the roots? To answer this, we begin by reducing
the general cubic to normal form. First divide across by a, obtaining a cubic
which has the same roots as f . Renaming the coeffs, if necessary, we may as
well suppose that a = 1, thereby getting a monic poly of the form f (x) =
x3 + bx2 + cx + d. Now we seek another cubic which has no term involving x2 ,
using a procedure similar to completing the square in a quadratic. Consider

f (x + t) = (x + t)3 + b(x + t)2 + c(x + t) + d

= x3 + (3t + b)x2 + (3t2 + 2bt + c)x + t3 + bt2 + ct + d
f 00 (t)
= x3 + x + f 0 (t)x + f (t).
2

Now choose t so that f 00 (t) = 3t + b = 0 and let q = f 0 (t), r = f (t). Then

f (x + t) = x3 + qx + r.

This is the normal form.

(As a general principle, it is often a very good idea to convert a cubic to
normal form, just as completing the square in a quadratic is often a good
idea.) Observe, too, that x0 is a root of f iff x0 t is a root of the reduced
form of f .

3. From now on we deal with a cubic in its the normal form x3 + qx + r. We can
proceed in two directions to solve this. Using the method previously suggested
we can let q = 3uv, r = u3 + v 3 and then use the factorization

x3 + qx + r = x3 + u3 + v 3 3uvx = (x + u + v)(x2 (u + v)x + u2 + v 2 uv),

to confirm that (u+v) is one root and the other two are roots of the quadratic
x2 (u + v)x + u2 + v 2 uv. Thus, were down to solving the pair of equations

26
 q = 3uv, r = u3 + v 3 for u, v. Its easy to see that u3 , v 3 must be the roots of
the quadratic equation.
q3
z 2 rz = 0.
27
Hence, q
4q 3
r r2 + 27
3 3
u ,v = .
2
So, let u be a cube root of one of these and let v = q/(3u). Then (u + v)
3
is a root,
q possibly complex, of x + qx + r. Having selected one cube root u of
3
(r + r2 + 4q27
)/2, say, and setting v = q/3u, the remaining roots can then
be found by solving x2 q(u + v)x + u2 + v 2 uv = 0, or by figuring out the
4q 3
other cube roots of (r + r2 + 27
)/2.
Notice that u3 is real iff 27r2 + 4q 3 0. So, if this condition is satisfied,
then (u + v) is a real rootpossibly the only real root, as the following
example shows. (In fact, its not too hard to see that if 27r2 + 4q 3 > 0
holds, then (u + v) is the only real root of the cubic. What may happen if
27r2 + 4q 3 = 0?)
Example 7. Solve x3 + 3x + 1 = 0.

Solution. For this, uv = 1, and u3 , v 3 are roots of z 2 z 1 = 0. So,

3 3 1 5
u ,v = .
2
q q
3 1+ 5
Say u = 2
. Then (?) v = 3 12 5 , and so
s s
3 1+ 5 3 1 5

2 2
is a real root of x3 + 3x + 1. The remaining roots are given by
p
(u + v) (u + v)2 4(u2 + v 2 uv) (u + v) 3i(u v)
= ,
2 2
and are complex numbers.
4. The last remark raises an interesting question: what conditions on the coeffs
q, r guarantee that all three roots of x3 +qx+r are real? To see whats involved,
let a, b, c be the real roots of this cubic. Then a + b + c = 0, ab + bc + ca =
q, abc = r. Hence r2 = a2 b2 c2 . Since the geometric mean of positive numbers
doesnt exceed their arithmetic mean, we see that

3 a2 + b 2 + c 2 (a + b + c)2 2(ab + bc + ca) 2q
r2 = = .
3 3 3
27
 Hence q 0 and
27r2 + 8q 3 0
are necessary conditions for the cubic to have three real roots.
(This is an algebraic approach. What does a geometric approach suggest?)

5. Are these conditions sufficient In other words, if they hold, will x3 +qx+r have
three real roots? (Answer: not necessarily; consider 2x3 3x + 1.) Suppose
q 0 to begin with, and see what effect this has on the shape of the graph
of y = x3 + qx + r. (It will help you to visualise p the shape of such a cubic.)
The graph has two turning points given by x = q/3. The corresponding
y values are r
2q q
+ r.
3 3
The smallest of these is r
2q q
+ r.
3 3
The graph tells us that the cubic has three real roots if this is 0. This is so
if
27r2 4q 3 .
In other words, the cubic has three real roots if 27r2 + 4q 3 0. (This implies
that q < 0.)
This, then, is a sufficient condition, which implies the necessary condition
obtained above, but the two conditions are different! Can the gap be closed
between these two conditions? Is it the case, perhaps, that 27r2 + 4q 3 0 is
true if the roots are real? To see that this is indeed the case, we revisit our
attempt to derive a necessary condition assuming the roots are real. This time

27r2 + 4q 3 = 27a2 b2 c2 + 4(ab + bc + ca)3

= 27a2 b2 (a + b)2 4(a2 + ab + b2 )3
= 3a4 b2 + 26a3 b3 + 3a2 b4 4a6 12a5 b 4b6 12b5 a
= (a b)2 (a + 2b)2 (b + 2a)2
0.

(Query: how did we factorize this expression? By the way, this also points up
that the AM-GM inequality can sometimes be a blunt instrument.)
So, to sum up: the cubic x3 + qx + r has three real roots iff 27r2 + 4q 3 0. It
has one real root and two complex roots iff 27r2 + 4q 3 > 0.

6. Heres an alternative method of solving x3 + qx + r = 0 that relies on the

trigonometric identity:

4 cos3 = cos(3) + 3 cos .

28
 To utilize this, replace x by cos() and multiply through by 4 to get

3 cos(3) + (32 + 4q) cos + 4r = 0.

Now choose so that 32 + 4q = 0. Then

4r
cos 3 = .
3
Now solve this for ! This is the strategy. When does it work? For the last
equation to be solvable for real , we require to be real and 1 4r 3

4r
1. Hence, we require q 0 and 1 3 1 to hold, i.e., we require
27r2 + 4q 3 0, the same condition we encountered before that is necessary
and sufficient for the roots to be real. (NB. The method can also be used even
if these conditions arent fulfilled, by allowing to be complex and using the
fact that cos z = (eiz + eiz )/2, which is valid for complex z. In particular,
cos(i) = (e + e )/2 = cosh , the hyperbolic equivalent of cosine.)

9 Outline solutions to some problems on polys

1. (IMO 1973) Find the minimum value of a2 + b2 , where a, b are real numbers
for which the equation

x4 + ax3 + bx2 + ax + 1 = 0

has at least one real root.

Solution. Denote by the set of points (x, y) in the plane for which there is
a real number w such that

w4 + xw3 + yw2 + xw + 1 = 0.

Suppose (x, y) . Letting

1 2 1
t=w+ , t = w2 + 2 + 2 4,
w w
we see that there is a real number t, with |t| 2, such that

t2 + xt + y 2 = 0.

The reality of t requires that

x2 4(y 2) 0, (x, y) .

Thus,
{(x, y) : 4(y 2) x2 }.

29
 Moreover, since t2 4 is a requirement,

(x, y) 2t2 {(x, y) : tx + y + 2 0}

= {(x, y) : 2x + y + 2 0} {(x, y) : 2x + y + 2 0}.

Its easily seen that the lines 2x + y + 2 = 0, 2x + y + 2 = 0 are tangents

to the parabola x2 = 4(y 2) at the points (4, 6), (4, 6), respectively, and
intersect at (0, 2). Hence

{(x, y) : 2x + y + 2 0} {(x, y) : 2x + y + 2 0} {(x, y) : 4(y 2) x2 },

and so

{(x, y) : 2x + y + 2 0} {(x, y) : 2x + y + 2 0}.

Plainly, equality holds here. What we seek is the square of the distance from
to the origin. Its clear from geometric considerations that the distance is
given by the distance from the origin to one of the tangent lines, i.e., it is
|0 + 0 + 2| 2
p = .
(2)2 + 1 5

Hence the required minimum value is 4/5.

2. USAMO75. Let p be a poly of degree n and suppose that

k
p(k) = , k = 0, 1, 2, . . . , n.
k+1
Determine p(n + 1).

Consider the poly q(x) = (x + 1)p(x) x. Then (?)

n
Y
q(x) = a (x k)
k=0

for some a. But q(1) = 1. Hence a can be determined, whence

(1)n+1 (n + 1)
p(n + 1) = .
n+2

3. IRMO89. Let a be a positive real number, and let

q q
3 3
b = a + a + 1 + a a2 + 1.
2

30
 Prove that b is a positive integer if, and only if, a is a positive integer of the
form 21 n(n2 + 3), for some positive integer n.
By direct computation it can be seen that b is a root of the cubic x3 + 3x 2a.
Hence, if b is a positive integer, then so is 2a = b3 + 3b. But the latter is even.
Hence, a is a positive integer. Conversely, if a is of the form 21 n(n2 + 3), for
some positive integer n, then n is also a root of the same cubic. But this cubic
has only one real root. Thus b = n.

4. IRMO95. Suppose that a, b and c are complex numbers, and that all three
roots z of the equation
x3 + ax2 + bx + c = 0
satisfy |z| = 1 (where | | denotes absolute value). Prove that all three roots w
of the equation
x3 + |a|x2 + |b|x + |c| = 0
also satisfy |w| = 1.

Denote the roots of x3 + ax2 + bx + c = 0 by p, q, r. Then

a = p + q + r, b = pq + qr + rp, c = pqr.

Hence |c| = 1, and

1 1 1
|b| = |(pqr)( + + )| = |r + p + p| = |a|.
r p q
Thus

x3 + |a|x2 + |b|x + |c| = x3 + |a|(x2 + x) + 1 = (x + 1)(x2 + (|a| 1)x + 1);

and (?) since ||a| 1| 2, the roots of the quadratic x2 + (|a| 1)x + 1 are
complex numbers of unit modulus, the result follows.

5. Prove that x4 + x3 + x2 + x + 1 is a factor of x44 + x33 + x22 + x11 + 1.

Since x5 1 = (x 1)(x4 + x3 + x2 + x + 1) = 0 we see that x5n = 1 for every

integer n. So x44 + x33 + x22 + x11 + 1 = x8.5 x4 + x6.5 x3 + x4.5 x2 + x2.5 x + 1 =
x4 + x3 + x2 + x + 1 = 0. Thus every root of x4 + x3 + x2 + x + 1 is a root of
x44 + x33 + x22 + x11 + 1, and the Remainder Theorem does the rest.

6. The coeffs of the cubic ax3 + bx2 + cx + d are integers, ad is odd and bc is even.
Prove that at least one root is irrational.
Suppose all roots are rational, and apply the rational roots theorem. According
to this, if p/q is a rational root with p, q in their lowest form, then p divides

31
 d and q divides a. But ad is odd. Hence both of a, d are odd and so p, q are
odd. Next note that

ap3 + bp2 q + cpq 2 + dq 3 = 0.

The first and last terms are odd, so their sum is even, and since bc is even
one of b, c is even. Hence one of bp2 q, cpq 2 is even. But their sum is even.
Hence both are even. Hence b, c are even. Finally, if the roots are rational we
can denote them by p1 /q1 , p2 /q2 , p3 /q3 , where, for k = 1, 2, 3, pk , qk are odd
integers, with pk , qk having no common factors. Then
p1 p2 p3 b
+ + = ,
q1 q2 q3 a

whence we derive (?) a contradiction. Thus at least one root is irrational.

7. Suppose m, n 1 and

f (x) = xn + + a1 x + a0 , g(x) = bm xm + + b1 x + b0 , bm 6= 0,

are two polynomials with integer coefficients and share a common real root.
Prove that the common root is an integer. This is false! For instance x2
2, x(x2 2) share a common real root, which is irrational.
To see whats involved, let n = 3, m = 2 and let x denote the common root.
Then
0 = x 3 + a2 x 2 + a1 x + a0 = b 2 x 2 + b 1 x + b 0 .
Then
0 = b2 x3 + a2 b2 x2 + a1 b2 x + a0 b2 = b2 x3 + b1 x2 + b0 x,
whence

0 = (a2 b2 b1 )x2 + (a1 b2 b0 )x + a0 b2 = b2 x2 + b1 x + b0 .

Now eliminate the x2 term and conclude that

[b2 (a1 b2 b0 ) b1 (a2 b2 b1 )]x + [a0 b22 b0 (a2 b1 b1 )] = 0.

It follows from this that if

[b2 (a1 b2 b0 ) b1 (a2 b2 b1 )] = 0,

then
[a0 b22 b0 (a2 b1 b1 )] = 0.
Otherwise, x is rational, and so a rational root of f . By the rational roots
theorem, it must then be an integer.g

32