Module1 1

Lecture-3

We shall continue our discussion on some more theoretical aspects. In these lecture we
shall be introducing an important concept of “extreme point” of a convex set and how it
can be related to the optimal solution of an LPP. But before that we need to understand
the following concepts.

Definition1.9 Interior point of a set: A point x ∈ S ⊆ Rn is said to be an interior

point of the set S if ∃ ε > 0 such that B(x, ε) ⊆ S, where B(x, ε) is an open ball centred
at x with radius ε.

Definition1.10 Boundary point of a set: A point x ∈ S ⊆ Rn is said to be a

boundary point of the set S if for every ε > 0, B(x, ε) ∩ S 6= φ, B(x, ε) ∩ S c 6= φ and S c
is complement of S in Rn .

The following is the one in which we shall mainly be interested.

Definition1.11 Extreme point of a set: Let S ⊆ Rn be a convex set. A point x ∈ S
is said to be an extreme point of the set S if it is impossible to express x as a convex
combination of two other distinct points in S, i.e. x ∈ S is an extreme point of S if and
only if x1 , x2 ∈ S, 0 < λ < 1 such that x = (1 − λ)x1 + λx2 then x = x1 = x2 .

Example 1.13 S = {x : cT x ≥ z} is a convex set with no extreme

Feasible se point.

x
X
X

x x !

x x
x
•
(184)
y

S = {(x, y) : x2 + y 2 ≤ 1} is a convex set in• which every point on boundary is an extreme

point(infinite extreme points).
The following pictorial representation of S has four extreme points.

Copyright
c Reserved IIT Delhi
 Feasible se

x 2
x4
X
X

x x !

x x
S x3
x
•
x1
(184)
y

x2

xi =, i = 1, 2, 3, 4 extreme points of S.
•

Example 1.14 Graph the convex hull of the points (0, 0),(0, 1), (1, 2), (1, 1), (4, 0).
Which of these points is an interior point of the convex hull? Express it as a convex
combination of extreme points.
Solution:
Let (1, 1) = λ1 (0, 0) + λ2 (4, 0) + λ3 (0, 1) + λ4 (1, 2)
= (4λ2 + λ4 , λ3 + 2λ4 )
 
  λ1  
λ 1 + λ 2 + λ3 + λ4 = 1 1 1 1 1  1
  λ2   

4λ2 + λ4 = 1 ⇒  0 4 0 1   = 1 

 λ3 
λ3 + 2λ4 = 1 0 0 1 2 1
λ4
λ3 = 1 − 2λ4
1 − λ4
⇒ λ2 =
4 Feasible se
1
λ1 = 1 − λ2 − λ3 − λ4 = (5λ4 − 1)
4
x
X
X

(1,2)
x x !
Convex hull

(0,1)
x x
•
(1,1)
x
•
ω

(0,0) (4,0)
(184)
y
(1,1) ε interior

1 3 2
Let λ4 = ⇒ λ3 = , λ2 = , λ1 = 0. Then (1, 1) is a convex combination for extreme
5 5 5 •
point vectors.

Copyright
c Reserved IIT Delhi
 3
Result 1.9 An extreme point of a convex set is a boundary point of a set. Converse need
not be true.
Proof: Let S be a convex set and x ∈ S be its extreme point. To the contrary, let x ∈int
S, Then, ∃ ε > 0 such that B(x, ε) ⊆ S.
Feasible se
We will show that x can be written as convex combination of two distinct points of S.

x
X -
X

x x !

x x
• S ε x

(184)
y

εe εe •
Take y = x + √ , and z = x − √ , e = (1, 1, . . . , 1)T ∈ Rn . Then distance between x
2 n 2 n
and y is r
ε2 ε2 ε2
||y − x|| = + + . . . + (n times)
4n 4n
r 4n
ε2 ε
= = < ε.
4 2
Similarly ||z − x|| < ε
⇒ y, z ∈ B(x, ε) ⊆ S.
Also y 6= z.
Now, x = 12 (y + z) = (1 − λ)y + λz for λ = 12 ∈ (0, 1).
⇒ x is not an extreme point of S, which is not correct.
Thus, x 6∈ intS. But x ∈ S so x has to be a boundary point of S.
Converse is not true as depicted in the following figure.
Feasible se

x
X -
X

x x
–w
1 !

– w x x

–w
2 x
•
S (184)
y

Copyright
c •
Reserved IIT Delhi
 4
w is a boundary point of S but it is not an extreme point of S.

Example 1.15 The extreme points of the polygon convex set S described by S = {2x1 +
x2 + 9 ≥ 0, − x1 + 3x2 + 6 ≥ 0, x1 + x2 ≤ 0, x1 + 2x2 − 3 ≤ 0} are
( 32 , −3
2
), (−3, −3), (−7, 5).
Feasible se

x
X -
– X
–
(-7,5)
– x x !
– (0, 3)
– 2
–
x x
ι ι ι ι ι ι ι ι –ι ι ι ι ι ι ι ι ι
• – x

– (184)
–
– y
(-3,-3) – ( 3, -3)
2 2
–

Perhaps one of the most significant result in linear programming theory is the following
one, also known as separation theorem. We state it but skip the proof. However, we
strongly encourage you to get yourself convinced at least graphically as to what the result
intends to say.
Theorem 1.2 A point y in space Rn either belongs to given closed convex set S or there
exists a hyperplane through y so that the whole of S lies in one se
Feasible open half space produced
by that hyperplane.
x
X
X

x x !
y
º
x x
•w
S x
•
ω

(184)
y

T n
Definition1.12 Supporting Hyperplane: • A hyperplane c x = z (c ∈ R , z ∈ R are given)
is said to be a supporting hyperplane at a boundary point w of a convex set S if
(i) cT w = z, i.e., the hyperplane passes through w.

Copyright
c Reserved IIT Delhi
 5
T T
(ii) c u ≥ z or c u ≤ z, ∀u ∈ S, i.e., the whole of theFeasible
S liessein one half closed space
T
produced by hyperplane c x = z.
x
X
X

x x !

w
x x
T
C x=z
x
•
(184)
Sw
y

Definition1.13 Optimal Hyperplane: Consider • a Linear programming problem (LPP)

T
Max z = c x subject to Ax = b, x ≥ 0. Let S be a feasible set of LPP and z0 be the
optimal objective value. Then, S is a convex set. A hyperplane cT x = z0 is called an
optimal hyperplane to the set S at x0 (x0 is optimal solution of (LPP) with corresponding
objective value z0 ).

Theorem 1.3 The optimal hyperplane of LPP is a supporting hyperplane to the feasible
solution set of LPP.

Remark 1.1 Converse of above result need not hold, i.e., a supporting hyperplane need
not be an optimal hyperplane. This can be observed from immediate above figure.

Theorem 1.4 Every supporting hyperplane of a closed convex set bounded from below
contains atleast one extreme point of the set.

Remark 1.2 As the optimal supporting hyperplane passes through at least one extreme
point, the optimum value of z is also attained at at least one extreme point, i.e., at least
one extreme point will give an optimal solution of LPP.

Theorem 1.5 If the convex set of the feasible solutions of Ax = b, x ≥ 0 is a convex

bounded polyhedron then at least one extreme point gives an optimal solution of the
LPP.
Proof: We know that the convex bounded polyhedral set of the feasible solutions of
Ax = b, x ≥ 0, has finite number of extreme points and this being a convex polyhedron
any point of it can be written as a convex combination of its extreme points.
Let y1 , y2 , . . . , yp be the extreme points of S and let x0 be the optimal solution of LPP.
If we are maximizing the objective functioncT x = f (x), then we have

Copyright
c Reserved IIT Delhi
 6
f (x0 ) ≥ f (x), ∀ x ∈ S. . . . (1)
Now, if x0 is not an extreme point of S then
p p
X X
x0 = λi yi , for some λi ≥ 0, 1 ≤ i ≤ p, λi = 1
i=1 i=1
p p
X X
⇒ f (x0 ) = f ( λi y i ) = λi f (yi ) (f is a linear function)
i=1
i=1 p
 X p
X
≤ max f (yi ) λi = f (yk ) (∵ λi = 1)
1≤i≤p
i=1 i=1
⇒ f (x0 ) ≤ f (yk )
But by (1) we get f (x0 ) = f (yk ).
⇒ Maximum of the objective function is also attained at an extreme point yk of S,
completing the proof.

Theorem 1.6 If an optimal value of the objective function is attained at more than
one extreme point then every convex combination of these extreme points also gives the
optimal value of the objective function.
Proof: Let the objective function f (x) = z = cT x assumes the optimal value z ∗ at
extreme points y1 and y2 , i.e. z ∗ = cT y1 = cT y2 .
Let x0 = (1 − λ)y1 + λy2 . Then, x0 ∈ S asS is a convex set and
cT x0 = (1 − λ)cT y1 + λcT y2 = z ∗
⇒ x0 is also an optimal solution of LPP.

Copyright
c Reserved IIT Delhi