School of Economics

ECON6004: Mathematical Economics

Lecture 1.8: Convex Sets in Euclidean Spaces

Instructor: William Schworm Semester 1, 2017

1. Introduction

Convex sets have numerous important properties and many applications in Economics. We will use convex sets as an interesting
environment through out the course.
We begin by defining convex combinations of vectors and then define convex sets.

2. Convex Combinations

Let x ∈ Rn and y ∈ Rn and suppose there exists a scalar s with 0 ≤ s ≤ 1 such that z = sx + (1 − s)y. Then, z is a convex combination
of x and y.
Recall that z ∈ Rn is a linear combination of x, y ∈ Rn if there exist scalars sx and sy such that z = sx x + sy y. By restricting the scalars
to be nonnegative and sum to one, we obtain a convex combination.
In this case, note that z can also be expressed as
z = y + s(x − y) (1)

PDF version of lecture-01.8.tex produced on 28 February 2017.

Lecture 1.8: Convex Sets in Euclidean Spaces Page 2
so that
s = 0 =⇒ z = y
(2)
s = 1 =⇒ z = x

and for any 0 < s < 1, z is contained in the line segment connecting x and y.

Figure 1: Convex Combination.

y
y+
s(x
−y
)
x

s( x
−y
)
x−y

−y

Pn Pn
More generally, let xi ∈ Rn and si ∈ R with 0 ≤ si ≤ 1 for i = 1, . . . , n with i=1 si = 1. Then, z = i=1 si x
i is a convex combination
 Lecture 1.8: Convex Sets in Euclidean Spaces Page 3

of x1 , x2 , . . . , xn .
A Euclidean space has the property that all linear combinations of a set of vectors are in the space. Since convex combinations are linear
combinations with further restrictions on the coefficients, any convex combination of any set of points in a Euclidean space will also be
in the space.

3. Convex Sets

Definition: A set C in Rn is convex if, for all pairs of points x ∈ C and y ∈ C, all convex combinations of x and y are contained in C.

The following examples display some simple convex sets.

Figure 2: Convex Sets.

(a) (b) (c)

x x
y
y

The following examples display some sets that are not convex due to “holes” in the interior of the set or the shape of the boundary of
the set.
 Lecture 1.8: Convex Sets in Euclidean Spaces Page 4
Figure 3: Non-Convex Sets.
(a) (b) (c)
y c
x y

x
c
c

,→ Examples:

(i) Let S = (x, y) ∈ R2 | x2 + y 2 ≤ 1 . One can see from the shape that the line between any pair of points is contained in the

set. Also, a simple proof of convexity can be given based on the triangle inequality.
(ii) Let S = (x, y) ∈ R2 | x2 + y 2 ≥ 1 . This is not convex since it has a hole in the middle. A proof that this set is not convex


is provided by an example of two points in the set with a convex combination not in the set. For example, (0, 1) and (0, −1)
are contained in S and (0, 0) is a convex combination of these two points. But (0, 0) is not contained in S.
(iii) Let T be a triangle defined by T = (x, y) ∈ R2 | y ≥ 0, x + y ≤ 1, y − x ≤ 1 and define a translated triangle by S =


{ (1, 0) + T }. Note that both of these triangles are convex. Now define the set R = S ∪ T and note that this set is not
convex since (0, 1) ∈ R and (1, 1) ∈ R but none of the convex combinations of these points are in R. Therefore, the union of
two convex sets is not necessarily convex.

♦ ♦ ♦

4. Operations that Preserve Convexity

 Lecture 1.8: Convex Sets in Euclidean Spaces Page 5
In this section, we describe operations on sets that preserve convexity.
The first theorem states that a convex set remains convex if either all boundary points are excluded or if all boundary points are included.

Theorem 1. If C is a convex set in Rn , then the interior of C and the closure of C are convex sets.

The next theorem states that intersections, sums and Cartesian products of convex sets are convex.

Theorem 2. Suppose C and D are convex sets. Then, C ∩ D, C + D, and C × D are convex sets.

A few examples will show that unions and complements do not preserve convexity.

5. Convexity and Separating Hyperplanes

There are numerous connections between convex sets, hyperplanes, and half-spaces that are extremely useful in a variety of contexts.
We begin with the idea of two sets being “separated” by a hyperplane. Let C and D sets in Rn . A hyperplane separates C from D if
C is contained in one of its (closed) half spaces and D in its other (closed) half-space.
The following theorem describes the circumstances in which a separating hyperplane exists.

Theorem 3. Let C and D be sets in Rn . If C and D are convex with disjoint interiors, then there exists a hyperplane that separates
them.
 Lecture 1.8: Convex Sets in Euclidean Spaces Page 6
,→ Examples:

(i) Let C = [0, 1] and D = [1, 2] and note that the interiors of these intervals are disjoint: (0, 1) ∩ (1, 2) = ∅. Therefore, the theorem
implies there is a separating hyperplane. A one-dimensional hyperplane is a point and it is easy to see that the “hyperplane”
given by the point 1 separates C and D.
(ii) Let C = (x1 , x2 ) ∈ R2 | x21 + x22 ≤ 1 and let D = (0, 2) + C. The sets C and D are convex with disjoint interiors so the

theorem implies that there is a separating hyperplane. The separating hyperplane is unique and can be represented with the
general form by selecting β = (0, 1) and α = 1 so that we obtain

H(1, (0, 1)) = { (x1 , x2 ) | x2 = 1, x1 ∈ R } = { (x1 , 1) | x1 ∈ R } . (3)

This is a horizontal line that intersects (0, 1).

♦ ♦ ♦

The following figures show examples of the existence and non-existence of separating hyperplanes.
Lecture 1.8: Convex Sets in Euclidean Spaces Page 7
Figure 4: Convex Sets with Separating Hyperplane.
(a) (b)

T S

In the first example in Figure 4 (a), the sets S and T have a non-empty intersection but the intersection only contains a boundary point.
Therefore, the sets have disjoint interiors and are convex. The theorem implies there exists a separating hyperplane and one is displayed
in the figure. The separating hyperplane is unique in this case.
In the second example in Figure 4 (b), the sets S and T have an empty intersection since S does not contain its boundary points.
Therefore, the sets have disjoint interiors and are convex. The theorem implies there exists a separating hyperplane and one is displayed
in the figure. There are many separating hyperplanes in this case which can be characterized by two extreme separating hyperplanes:
One with maximal slope and one with minimal slope. These can be defined explicitly if you have an explicit definition of S and T .
In the first example in Figure 5 (a), the sets S and T are disjoint and T is convex but S is not convex. The theorem is uninformative in
this case but we can see that there are no separating hyperplanes. Any hyperplane that has T in one of its half-spaces will have at least
some points of S in the same half-space.
Lecture 1.8: Convex Sets in Euclidean Spaces Page 8
Figure 5: Sets with No Separating Hyperplane.

(a) (b)

T
S

T S

In the second example in Figure 5 (b), the sets S and T are disjoint and S is convex. But T is not convex and its position relative to
S ensures there is no separating hyperplane. Any hyperplane with S in one half-space will also have at least some elements of T in the
same half-space.
The next theorem shows the close connection between closed convex sets and their containing half-spaces.

Theorem 4. A closed convex set C is equal to the intersection of all the closed half-spaces containing C.