Scribd
Upload
5 views

Lecture 9

PDF

Uploaded by

ahmedEm
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
5 views

Lecture 9

PDF

Uploaded by

ahmedEm
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

CS 435 : Linear Optimization Fall 2008

Lecture 9: Convex Hull of Extreme Points


Lecturer: Sundar Vishwanathan
Computer Science & Engineering Indian Institute of Technology, Bombay

In this lecture, we complete the proof of the theorem on extreme points mentioned in the previous lecture
and begin the last part of understanding the object {x : Ax ≤ b}.
Proof: (Continuing Part 2.) Here we prove that every extreme point of {x : Ax ≤ b} can be expressed
as an intersection of n linearly independent hyperplanes. We show that if a point in the set cannot be
expressed as an intersection of n linearly independent hyperplanes then we can express it as a convex
combination of two other points in the set.
Let x0 be such a point. We split Ax0 ≤ b into two parts
0 0
A x0 = b (1)
00 00
A x0 < b (2)

By the assumption on x0 , A0 has rank strictly less than n. We wish to express x0 as the convex
combination of two other points y and z. Where do we find this y and z ? At this point, we recommend
visualising examples in two and three dimensions. It will not take long to conjecture that y and z must
also satisfy A0 x = b0 . For instance, if the object is a cube, and x0 is a point on one face of the cube, it
is easy to see that both y and z must also be on that face.
0 0
We need to use the fact that A has rank less than n. One fallout of this is that the solutions to A x = b0
is a subspace of dimension at least one (at least a line) shifted by a vector, and x0 lies in this shifted
subspace. Again it is natural to consider y and z on this line passing through x0 , very close to, and on
either side of x0 . This is what we will show below.
00 00
Since A x0 is strictly less than b , we can draw a small enough sphere around x0 such that every point
00 00
x within the sphere satifies A x0 < b . This means that there is an  such that for which all unit vectors
v, v,
00 00
A (x0 + v) < b (3)
0 0
Since A does not have n linearly independent vectors, we can find an x0 6= ~0 such that A x0 = 0. Then,
0 0 0
A (x0 + x0 ) = A (x0 − x0 ) = b Also,
00 00
A (x0 + x0 ) < b (4)
00 00
0
A (x0 − x ) < b (5)

So x0 + x0 and x0 − x0 are points in {x : Ax ≤ b} and x0 = (1/2)(x0 + x0 ) + (1/2)(x0 − x0 ). Thus
we have expressed x0 as a convex combination of two points in the set. 

1 Convex hull of the extreme points


Definition 1 A convex hull of points p1 , p2 , . . . pn is the set of all points which can be written as convex
combination of p1 , p2 , . . . pn .

One can make a definition even when the set of points is infinite, but we will only deal with finite sets
in this course.
The last step in understanding {x : Ax ≤ b} is the following theorem.

1
2

Theorem 1 Let p1 , p2 , . . . pn be extreme points of x : Ax ≤ b. Then every point in x : Ax ≤ b can be


expressed as a convex combination of the points p1 , p2 , . . . pn .

Take any x0 such that Ax0 ≤ b. We need to show that this can be written as:

x0 = Σi λi pi where Σλi = 1 , 0 ≤ λi ≤ 1. (6)

Again, it is instructive to look at examples especially in two dimensions and see what can be done.
Our proof will be by induction on the dimension of the object {x : Ax ≤ b}. Base Case: Exercise: Do
this when the dimension is 0 and 1.
For the rest of this lecture, we will use an example in two dimensions to illustrate the technique of the
proof. So consider a set {x : Ax ≤ b} as shown in the figure.
Let p be any point inside the set. Now take the extreme point pi , join it to p, and we extend this line
till it touches one of the bounding segments (pk pk+1 in this case) at a point, say q.
Since q lies on the segment joining the extreme points pk and pk+1 it can be expressed as a convex
combination of pk and pk+1 . Therefore,

q = λ1 pk + λ2 pk+1 where λ1 + λ2 = 1. (7)

Also p lies on the segment joining q and pi . So p can expressed as a convex combination of q and pi .
Therefore,
p = λ3 pi + λ4 q where λ3 + λ4 = 1. (8)
Combining the above equations we get

p = λ3 pi + λ4 λ1 pk + λ4 λ2 pk+1 (9)

Now λ3 + λ4 λ1 + λ4 λ2 = 1. Why?
Thus we have expressed p as a convex combination of the extreme points. Where did we use induction ?
In expressing q as a convex combination of pk and pk+1 . Once we have q, our focus was the line segment
between pk and pk+1 .
In the next lecture we use the same technique and complete the proof. This needs some clear thinking
and the ability to be able to express your intuition in the language we have developed.

You might also like