Outline

Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Lecture I: Convex Sets

Convex Analysis

Zakaria El Allali, Polydisciplinary Faculty of Nador

Master of Mathematical Modeling and Scientific Computing

Z.El Allali Mathematical Preliminaries 1/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

1 Introduction

2 Historical Development of Convex Analysis

3 Convex Sets

4 Convex combinations

5 Convex hull

6 Convex cones

Z.El Allali Mathematical Preliminaries 2/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Introduction to Convex Analysis

Convex analysis is a specialized and essential branch of mathematics that

deals with the properties and structures of convex sets and convex
functions. These two objects form the backbone of many theoretical and
applied disciplines. These fundamental concepts unify geometry, linear
algebra, and calculus, providing a framework for understanding a wide
variety of mathematical phenomena.
Convex analysis forms the theoretical foundation of optimization theory, a
field critical for solving practical problems in numerous domains.
Optimization seeks to maximize or minimize objectives, often under
constraints, and convex analysis provides the tools to guarantee
tractability and global solutions for these problems.

Z.El Allali Mathematical Preliminaries 3/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Introduction to Convex Analysis

The applications of convex analysis are vast, spanning fields like economics,
machine learning, operations research, and signal processing.
In economics: convex sets are used to model feasible production
possibilities and consumer preferences, while convex functions represent
utility and cost functions that are naturally well-behaved.
In machine learning: convex loss functions and constraints allow
algorithms like support vector machines and logistic regression to perform
efficiently. Operations research relies heavily on convex sets and functions
to model feasible regions and costs in optimization problems, such as
those in transportation, logistics, and supply chain management.
In signal processing: convex optimization helps recover signals corrupted
by noise or compress large datasets effectively.

Z.El Allali Mathematical Preliminaries 4/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Historical Development of Convex Analysis

The study of convex sets can be traced back to the ancient Greeks, who
studied shapes like polygons and spheres. Convexity formally entered
mathematics in the 19th century through the work of mathematicians like:
Augustin-Louis Cauchy (1821): Investigated convex shapes in geometry.
Hermann Minkowski (1896): Introduced the Minkowski sum, a
fundamental operation in convex geometry.
Jean-Pierre Serre: Developed foundational concepts of convex functions.
In the 20th century ; convex analysis gained prominence due to its
applications in optimization. Researchers like John von Neumann and
Leonid Kantorovich demonstrated its importance in game theory and linear
programming, respectively.

Z.El Allali Mathematical Preliminaries 5/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Historical Development of Convex Analysis

Year Contributor Contribution

300 BCE Euclid Formalized geometry, implicitly introducing convexity.
3 rd BCE Archimedes Studied volumes of convex shapes (spheres, cylinders)
1637 R. Descartes Cartesian coordinates enabled analytical geometry.
1821 A.L.Cauchy Inequalities and convex surface area formula.
1896 H. Minkowski Minkowski sum and convex geometry contributions.
1927 Hahn-Banach Separation of convex sets in functional analysis.
1928 J. von Neumann Convexity in game theory and economic strategies.
1939 L. Kantorovich Linear programming with convex polyhedra.
1947 G. Dantzig Simplex Method for linear programming.
1969 Rockafellar Formalized convex analysis as a distinct field.
2000 s Various Applications in Al, machine learning, and robotics.

Z.El Allali Mathematical Preliminaries 6/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex Sets

Definition: A set C ⊆ Rn is called convex iffor any x, y ∈ C and λ ∈ [0, 1],

we have λx + (1 − λ)y ∈ C .
Note 1: C is convex ⇐⇒ for any x, y ∈ C , the line segment [x, y ] is in C .
i.e., [x, y ] ⊆ C .
Note 2: The empty set ∅ is a convex set. ($ not! then ∃ . . . →← )
Example: A line in Rn is a set of the form, L = {z + td : t ∈ R}, where
z, d ∈ Rn . Let x = z + t1 d ∈ L and y = z + t2 d ∈ L. Then for any
λ ∈ [0, 1], λx + (1 − λ)y = z + (λt1 + (1 − λ)t2 ) d ∈ L. Therefore, L is a
convex set.

Z.El Allali Mathematical Preliminaries 7/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convexity of hyperplanes and half-spaces

Note 1: For any x, y ∈ Rn , the closed and open line segments [x, y ] and
(x, y ) are convex sets.
Note 2: The entire space Rn is a convex set.
Note 3: Let a ∈ Rn \{0} and b ∈ R. The following sets are convex:
the hyperplane H = x ∈ Rn : a⊤ x = b ;

1

the half-space H = x ∈ Rn : a⊤ x ≤ b ;
−

2

the open half-space x ∈ Rn : a ⊤ x < b .


3

Proof of (2): Let x, y ∈ H − and λ ∈ [0, 1]. We will show that

z = λx + (1 − λ)y ∈ H − . Indeed,
   
a ⊤ z = a ⊤ (λx + (1 − λ)y ) = λ a ⊤ x + (1 − λ) a ⊤ y
≤ λb + (1 − λ)b = b

which implies z ∈ H − .

Z.El Allali Mathematical Preliminaries 8/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convexity of balls

Let c ∈ Rn and r > 0. Let ∥ · ∥ be an arbitrary norm defined on Rn . Then the

open ball B(c, r ) := {x ∈ Rn : ∥x − c∥ < r } and the closed ball
B[c, r ] := {x ∈ Rn : ∥x − c∥ ≤ r } are convex. Proof: We will show the
convexity of the closed ball. Let x, y ∈ B[c, r ] and λ ∈ [0, 1]. Then
∥x − c∥ ≤ r and ∥y − c∥ ≤ r . Let z = λx + (1 − λ)y . We will show that
z ∈ B[c, r ]. Indeed,

∥z − c∥ = ∥λx + (1 − λ)y − c∥ = ∥λ(x − c) + (1 − λ)(y − c)∥

≤ ∥λ(x − c)∥ + ∥(1 − λ)(y − c)∥
= λ∥x − c∥ + (1 − λ)∥y − c∥
≤ λr + (1 − λ)r
= r.

Therefore z ∈ B[c, r ], establishing the result. Note: The above result is true
for any norm defined on Rn .

Z.El Allali Mathematical Preliminaries 9/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convexity of ellipsoids

An ellipsoid is a set of the form

n o
E = x ∈ Rn : f (x) := x ⊤ Qx + 2b ⊤ x + c ≤ 0 ,

where Q ∈ Rn×n is positive semidefinite, b ∈ Rn , and c ∈ R. Then E is a

convex set.
Proof: Let x, y ∈ E , λ ∈ [0, 1], and z := λx + (1 − λ)y . Then
f (x) ≤ 0, f (y ) ≤ 0 and

z ⊤ Qz = (λx + (1 − λ)y )⊤ Q(λx + (1 − λ)y )

= λ2 x ⊤ Qx + (1 − λ)2 y ⊤ Qy + 2λ(1 − λ)x ⊤ Qy .
 ⊤  
Since x ⊤ Qy = Q 1/2 x Q 1/2 y , by the Cauchy-Schwarz inequality, we have
p 1 ⊤ 
x ⊤ Qy ≤ Q 1/2 x x Qx + y ⊤ Qy .
p
Q 1/2 y = x ⊤ Qx y ⊤ Qy ≤
2

Z.El Allali Mathematical Preliminaries 10/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convexity of ellipsoids

Thus, z ⊤ Qz ≤ λx ⊤ Qx + (1 − λ)y ⊤ Qy . Hence,

f (z) ≤ λx ⊤ Qx + (1 − λ)y ⊤ Qy + 2λb ⊤ x + 2(1 − λ)b ⊤ y + c

   
= λ x ⊤ Qx + 2b ⊤ x + c + (1 − λ) y ⊤ Qy + 2b ⊤ y + c
= λf (x) + (1 − λ)f (y ) ≤ 0,

establishing the desired result that z ∈ E .

Z.El Allali Mathematical Preliminaries 11/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convexity is preserved under the intersection

Lemma : Let Ci ⊆ Rn be a convex set for any i ∈ I , where I is an

arbitrary index set. Then ∩i∈I Ci is convex.
Proof: Let x, y ∈ ∩i∈I Ci and λ ∈ [0, 1]. Then x, y ∈ Ci , ∀i ∈ I . Since Ci is
convex, it follows that λx + (1 − λ)y ∈ Ci , ∀i ∈ I . Therefore,
λx + (1 − λ)y ∈ ∩i∈I Ci . That is, ∩i∈I Ci is convex.
Example (convex polytopes): A set P is called a convex polytope if it has
the form P = {x ∈ Rn : Ax ≤ b}, where A ∈ Rm×n and b ∈ Rm . The
convexity of P follows from the fact that it is an intersection of half-spaces
and half-spaces are convex:
m
\
P= {x ∈ Rn : Ai x ≤ b i } ,
i=1

where Ai is the i th row of A.

Z.El Allali Mathematical Preliminaries 12/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Preservation of convexity

1 Let C1 , · · · , Ck ⊆ Rn be convex sets and let µ1 , · · · , µk ∈ R. Then the

following set is convex:
( k )
X
µ1 C1 + µ2 C2 + · · · + µk Ck := µi xi : xi ∈ Ci , 1 ≤ i ≤ k
i=1

Note: if C ⊆ R is a convex set and b ∈ Rn , then the set

n

C + b := {x + b : x ∈ C } is also convex.
2 Let Ci ⊆ Rki be a convex set for any i = 1, 2, · · · , m. Then the following
Cartesian product is convex:

C1 × C2 × · · · × Cm := {(x1 , x2 , · · · , xm ) : xi ∈ Ci , 1 ≤ i ≤ m}
3 Let M ⊆ Rn be a convex set and let A ∈ Rm×n . Then the image set
A(M) := {Ax : x ∈ M} is convex.
4 Let D ⊆ Rm be a convex set and let A ∈ Rm×n . Then the inverse image
set, A−1 (D) := {x ∈ Rn : Ax ∈ D}, is convex.
Z.El Allali Mathematical Preliminaries 13/ 29
 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex combinations

Definition: Given x1 , x2 , · · · , xk ∈ Rn , a convex combination of these k

vectors is a vector of the form λ1 x1 + λ2 x2 + · · · + λk xk , where λi ∈ R and
λi ≥ 0 for 1 ≤ i ≤ k, satisfying λ1 + λ2 + · · · + λk = 1, i.e.,
λ := (λ1 , λ2 , · · · , λk )⊤ ∈ ∆k .
Note : A convex set can be defined by the property that any convex
combination of two points from the set is also in the set.

Z.El Allali Mathematical Preliminaries 14/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex combinations

Theorem: Let C ⊆ Rn be a convex set  and let x1 , x2 , · · · , xm ∈ C . Then

⊤ m Pm
for any λ = (λ 1 , · · · , λm ) ∈ ∆m := α ∈ R+ : i=1 αi = 1 , we have
Pm
i=1 λi xi ∈ C . That is, a convex combination of any finite number of
points from a convex set is in the set.
Proof: We prove the theorem by induction on m. The case m = 1 is
trivial. Suppose that P m = k holds. Let x1 , x2 , · · · , xk+1 ∈ C and λ ∈ ∆k+1 .
If λk+1 = 1, then k+1 i=1 λi xi = xk+1 ∈ C . If λk+1 < 1, then

v
z }| {
k+1 k k
X X X λi
z := λi xi = λi xi +λk+1 xk+1 = (1 − λk+1 ) xi +λk+1 xk+1 .
i=1 i=1 i=1
1 − λk+1
Pk
i=1 λi
Pk λi
Since i=1 1−λk+1 = 1−λk+1
= 1, we have v ∈ C and hence, z ∈ C .

Z.El Allali Mathematical Preliminaries 15/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex hull

Definition: (convex hull) Let S ⊆ Rn . Then the convex hull of S is the

set comprising all the convex combinations of vectors from S, i.e.,
( k )
X
conv(S) := λi xi | x1 , x2 , · · · , xk ∈ S, λ ∈ ∆k , k ∈ N .
i=1

Note: The convex hull conv (S) is a convex set (Exercise!). In fact, conv
(S) is the ”smallest” convex set containing S.
Lemma: Let S ⊆ Rn . If S ⊆ T and T is convex, then conv(S) ⊆ T .
Proof: Let z ∈ conv(S). Then we have z = ki=1 λi xi , for some
P

x1 , · · · , xk ∈ S ⊆ T and λ = (λ1 , · · · , λk )⊤ ∈ ∆k . That is, z is a convex

combination of elements from T . Since T is convex, by the previous
theorem, we obtain z ∈ T .

Z.El Allali Mathematical Preliminaries 16/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Carathédory Theorem

Let S ⊆ Rn and let x ∈ conv(S). Then ∃x1 , x2 , · · · , xn+1 ∈ S such that

x ∈ conv
P ({x1 , · · · , xn+1 }). That is, ∃λ = (λ1 , λ2 , · · · , λn+1 ) ∈ ∆n+1 such that
x = n+1 i=1 λi xi .
Proof: Let x ∈ conv(S). Then ∃x1 , · · · , xk ∈ S, λ ∈ ∆k s.t. x = ki=1 λi xi with
P
λi > 0∀i. If k ≤ n + 1, the result is proven. If k ≥ n + 2, then
x2 − x1 , · · · , xk −
Px1 are linearly dependent. Therefore, P ∃µ2 , · · · , µk not all
zeros such that ki=2 µi (xi − x1 ) = 0. Let µ1 := − ki=2 µi , we obtain
Pk Pk
i=1 µi xi = 0 and i=1 µi = 0, where ∃i for which µi < 0. Let α ∈ R+ . Then

k
X k
X k
X k
X k
X
x= λi x i = λi x i +α µi x i = (λi + αµi ) x i and (λi + αµi ) = 1.
i=1 i=1 i=1 i=1 i=1

Z.El Allali Mathematical Preliminaries 17/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Carathédory Theorem

The above representation is a convex combination if and only if

λi + αµi ≥ 0, ∀i = 1, · · · , k.

Since λi > 0∀i,n theoabove set of inequalities is satisfied for all α ∈ [0, ε], where
ε = mini:µi <0 −λµi
i
. Taking α = ε, then λj + αµj = 0 for
n o
−λi
j = argmini:µi <0 µi . This means that we have found a representation of x
as a convex combination of k − 1 vectors. This process can be carried on until
a representation of x as a convex combination of no more than n + 1 vectors is
derived.

Z.El Allali Mathematical Preliminaries 18/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex cones

Definition: A set S is called a cone if for any x ∈ S and λ ≥ 0, we have

λx ∈ S.
Lemma: A set S is a convex cone if and only if the following properties
hold: (1) x, y ∈ S ⇒ x + y ∈ S; (2) x ∈ S, λ ≥ 0 ⇒ λx ∈ S.
Proof: (⇒) Let x, y ∈ S. By the convexity, we have 12 x + 1 − 12 y ∈ S.

Since S is a cone, we have 2 × 12 (x + y ) = x + y ∈ S, i.e., property (1)

holds. Property (2) is true because S is a cone. (⇐) By property (2), S is
a cone. Let x, y ∈ S and λ ∈ [0, 1]. Since S is a cone, we have λx ∈ S
and (1 − λ)y ∈ S. By property (1), we further have λx + (1 − λ)y ∈ S,
establishing the convexity.
Example : Consider the convex polytope C = {x ∈ Rn : Ax ≤ 0}, where
A ∈ Rm×n . The set C is clearly a convex set. It is also a cone since

x ∈ C , λ ≥ 0 ⇒ Ax ≤ 0, λ ≥ 0 ⇒ A(λx) ≤ 0 ⇒ λx ∈ C

Z.El Allali Mathematical Preliminaries 19/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Lorentz cone (Ice cream cone)

The Lorentz cone, also called the ice cream cone, is given by
n o
Ln := (x, t)⊤ ∈ Rn+1 : x ∈ Rn , t ∈ R, and ∥x∥ ≤ t

The Lorentz cone is in fact a convex cone. Let (x, t)⊤ , (y , s)⊤ ∈ Ln . Then
∥x∥ ≤ t and ∥y ∥ ≤ s. The triangle inequality implies that

∥x + y ∥ ≤ ∥x∥ + ∥y ∥ ≤ t + s.

That is, (x, t)⊤ + (y , s)⊤ = (x + y , t + s)⊤ ∈ Ln . We have property (1). To

show property (2), take (x, t)⊤ ∈ Ln and λ ≥ 0. Then we obtain
∥λx∥ = λ∥x∥ ≤ λt, so λ(x, t)⊤ = (λx, λt)⊤ ∈ Ln

Z.El Allali Mathematical Preliminaries 20/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Conic combination

Definition: Given x1 , x2 , · · · , xk ∈ Rn , a conic combination of these k

vectors is a vector of the form λ1 x1 + λ2 x2 + · · · + λk xk , where λi ≥ 0 for
all i = 1, 2, · · · , k.
Lemma: Let C be a convex cone, and let x1 , x2 P , · · · , xk ∈ C and
λ1 , λ2 , · · · , λk ≥ 0. Then the conic combination ki=1 λi xi ∈ C .
Proof: Since C P is a convex cone, by property (2), we have λi xi ∈ C , ∀i.
By property (1), ki=1 λi xi ∈ C .
Definition: (conic hull) Let S ⊆ Rn . Then the conic hull of S is the set
comprising all the conic combinations of vectors from S, i.e.,
( k )
X k
cone(S) := λi xi | x1 , x2 , · · · , xk ∈ S, λ ∈ R+ , k ∈ N .
i=1

Note that cone (S) is a convex cone. (Exercise!) In fact, we have

Lemma: Let S ⊆ Rn . If S ⊆ T for some convex cone T , then cone
(S) ⊆ T , i.e., the conic hull of S is the smallest convex cone containing S.
(Exercise!)
Z.El Allali Mathematical Preliminaries 21/ 29
 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Conic representation theorem

Theorem: Let S ⊆ Rn and let x ∈ cone(S). Then ∃k linearly
independent vectors x1 , x2 , · · · , xk ∈ S such that x P
∈ cone ({x1 , · · · , xk });
that is, ∃λ = (λ1 , λ2 , · · · , λk ) ∈ Rk+ such that x = ki=1 λi xi and k ≤ n.
Pm
Proof: Let x ∈ cone(S). Then ∃x1 , · · · , xm ∈ S, λ ∈ Rm + s.t. x = i=1 λi xi
with λi > 0∀i. If x1 , · · · , xm are linearly independent, then k := m ≤ n and the
Pm is proven. Otherwise, ∃µ1 , · · · , µm ∈ R not all zeros such that
result
i=1 µi x i = 0. Let α ∈ R. Then
m
X m
X m
X m
X
x= λi x i = λi x i + α µi x i = (λi + αµi ) x i .
i=1 i=1 i=1 i=1

The above representation is a conic combination if and only if

λi + αµi ≥ 0, ∀i = 1, · · · , m
Since λi > 0 for all i, we can find αe ∈ R s.t. λj + α eµj = 0 for some j and
λi + αeµi ≥ 0 for the others. Thus we obtain a representation of x as a conic
combination of at most m − 1 vectors. Continuing this process, we can obtain
k linearly independent vectors x1 , x2 , · · · , xk ∈ S with k ≤ n such that
x ∈ cone ({x1 , · · · , xk }).
Z.El Allali Mathematical Preliminaries 22/ 29
 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Basic feasible solution (BFS)

Linear systems consisting of linear equalities and nonnegativity constraints
often appear as constraints in standard formulations of linear programming
problems.
Definition: (basic feasible solution) Let P := {x ∈ Rn : Ax = b, x ≥ 0},
where A ∈ Rm×n and b ∈ Rm . Suppose that the rows of A are linearly
independent. Then x̄ ∈ P is a basic feasible solution (BFS) of P if the
columns of A corresponding to the indices of the positive values of x̄ are
linearly independent.
Note) Since the columns of A reside in Rm , it follows that a BFS has at
most m nonzero elements.
Example : Consider the linear system

x1 + x2 + x3 = 6, x2 + x3 = 3, x1 , x2 , x3 ≥ 0.

A BFS of the system is (3, 3, 0). It satisfies all the constraints and the
columns corresponding to the positive elements, (1, 0)⊤ , (1, 1)⊤ are
linearly independent.
Z.El Allali Mathematical Preliminaries 23/ 29
 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Existence of a BFS in P

Theorem: Let P := {x ∈ Rn : Ax = b, x ≥ 0}, where A ∈ Rm×n and

b ∈ Rm . If P ̸= ∅, then it contains at least one BFS.
Proof: Let x ∈ P ̸= ∅. Then Ax = b and x ≥ 0. It follows that
b = x1 a 1 + x2 a 2 + · · · + xn a n , i.e., b ∈ cone ({a 1 , a 2 , · · · , a n }), where ai denotes
the i th column of A. By the conic representation theorem, there exist P indices
i1 < i2 < · · · < ik and k numbers xi1 , xi2 , · · · , xik > 0 such that b = kj=1 xij a ij
and a i1 , a i2 , · ·P
· , a ik are linearly independent.
Denote x̄ := kj=1 xij e ij . Then x ≥ 0 and

k
X k
X
Ax = xij Ae ij = xij a ij = b.
j=1 j=1

Therefore, x̄ ∈ P and satisfies that the columns of A corresponding to the

indices of the positive components of x̄ are linearly independent. That is, P
contains at least one BFS.

Z.El Allali Mathematical Preliminaries 24/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Closure and interior of a convex set

Theorem: Let C ⊆ Rn be a convex set. Then the closure cl(C ) is

convex. Proof: Let x, y ∈ cl(C ) and λ ∈ [0, 1]. Then ∃ sequences {xk },
{y k } ⊆ C such that x k → x and y k → y as k → ∞. By the convexity of
C , λxk + (1 − λ)yk ∈ C for any k. Since
λxk + (1 − λ)yk → λx + (1 − λ)y , we can conclude that
λx + (1 − λ)y ∈ cl(C ), which implies that cl(C ) is convex.
(line segment principle): Let C ⊆ Rn be a convex set, and assume that
int(C ) ̸= ∅. Suppose that x ∈ int(C ), y ∈ cl(C ). Then
(1 − λ)x + λy ∈ int(C ) for any λ ∈ (0, 1).
Theorem: Let C ⊆ Rn be a convex set. Then the interior int(C ) is
convex.
Proof: If int(C ) = ∅, then int(C ) is convex. Let x, y ∈ int(C ) and λ ∈ (0, 1).
Then by the line segment principle, (1 − λ)x + λy ∈ int(C ). We can conclude
that int(C ) is convex.

Z.El Allali Mathematical Preliminaries 25/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Other topological properties

Let C ⊆ Rn be a convex set and int(C ) ̸= ∅. Then we have

cl(int(C )) = cl(C ).
Proof: (⊆) : Since int(C ) ⊆ C , we have cl(int(C )) ⊆ cl(C ). (⊇) : Let
x ∈ cl(C ). We take
y ∈ int(C ). Then by the line segment principle, we have
xk := k1 y + 1 − k1 x ∈ int(C ) for any k ≥ 1. Since xk → x as k → ∞, we
obtain x ∈ cl(int(C )).
int(cl(C )) = int(C ).
Proof: (⊆) : Let x ∈ int(cl(C )). Then ∃ε > 0 s.t. B(x, ε) ⊆ cl(C ). Let
y ∈ int(C ). If y = x, then the result is proved. Otherwise, define
ε
z := x + α(x − y ), where α = 2∥x−y ∥
. Since ∥z − x∥ = ε2 , we have z ∈ cl(C ).
By the line segment principle, we have (1 − λ)y + λz ∈ int(C ) for λ ∈ [0, 1).
1
Taking λ = 1+α ∈ (0, 1), we obtain (1 − λ)y + λz = x ∈ int(C ).

Z.El Allali Mathematical Preliminaries 26/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Convex hull of compact set

Theorem: Let S ⊆ Rn be a compact set. Then conv(S) is compact.
Proof:
(Boundedness) Since S is bounded, ∃M > 0 such that ∥x∥ ≤ M for any
x ∈ S. Let y ∈ conv(S). By the CarathéodoryP theorem it follows that
∃x1 , · · · , xn+1 ∈ S and λ ∈ ∆n+1 s.t. y = n+1
i=1 λi x i . Therefore,
n+1
X n+1
X n+1
X
∥y ∥ = λi xi ≤ λi ∥xi ∥ ≤ M λi = M
i=1 i=1 i=1

(Closedness) Let yk be a sequence in conv(S) and yk → y as k → ∞. We

wish to show that y ∈ conv(S). By the Carathéodory P theorem it follows
that ∃x1k , · · · , xn+1
k
∈ S and λk ∈ ∆n+1 s.t. yk = n+1 k k
i=1 λi xi .

By the
compactness of S and ∆n+1 , the sequence λk , x1k , · · · , xn+1 k
has a
subsequence such that
 
k kj
lim λkj , x1 j , · · · , xn+1 = (λ, x1 , · · · , xn+1 )
j→∞

with λ ∈ ∆n+1 and x1 , · · · , xn+1 ∈ S. Therefore, we have

kj kj
y = limj→∞ y kj = limj→∞ n+1
P Pn+1
i=1 λi x i = i=1 λi x i which means that
Z.El Allali Mathematical Preliminaries 27/ 29
 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

Extreme points

Definition: (extreme points) Let S ⊆ Rn be a convex set. A point

x ∈ S is called an extreme point of S if there do not exist x1 , x2 ∈ S,
x1 ̸= x2 and λ ∈ (0, 1) such that x = λx1 + (1 − λ)x2 . The set of extreme
points of S is denoted by ext(S). That is, an extreme point is a point in
the set S that cannot be represented as a nontrivial convex combination of
two different points in S.
Example: The set of extreme points of a convex polytope consists of all its
vertices. The convex set S = conv {x1 , x2 , x3 , x4 }.
The extreme points set is ext(S) = {x1 , x2 , x3 }.

Z.El Allali Mathematical Preliminaries 28/ 29

 Outline
Introduction
Historical Development of Convex Analysis
Convex Sets
Convex combinations
Convex hull
Convex cones

The Krein-Milman theorem

We will state this theorem without a proof.

Krein-Milman theorem: Let S ⊆ Rn be a compact convex set. Then

S = conv(ext(S)).

That is, a compact convex set is the convex hull of its extreme points.

Z.El Allali Mathematical Preliminaries 29/ 29