Chapitre 1

General information on mathematical

programming

Sommaire
2.1 Linear algebra reminders . . . . . . . . . . . . . . . . . . . . . . . 14

2.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.1.2 Full rank, basic solutions . . . . . . . . . . . . . . . . . . . . . . . 15

2.2 Feasible bases of a linear program . . . . . . . . . . . . . . . . . 17

2.2.1 Putting a linear program in canonical form with respect to a basis 18

2.3 Fundamental theorem of linear programming . . . . . . . . . . 21

1.1 A Mathematical Program

D ⊂ Rn , x ∈ Rn , x = (x1 , x2 , ..., xn )

f :D→R

Définition 1 A mathematical program is a problem which consists in finding x∗ ,x∗ ∈ D(if

2
Z. Benméziane 1.1. A Mathematical Program

it exists) such that:

 



 f (x∗ ) ≥ f (x) 


 f (x∗ ) ≤ f (x)
 
(P ) ∀x ∈ D or (P 0 ) ∀x ∈ D

 


 

 maximisation problem  minimisation problem

 



 M ax f (x) 


 M in f (x)
 
Notation 1 (P ) x∈D or (P 0 ) x∈D

 


 

 maximisation problem  minimisation problem

A point x ∈ D is called a feasable solution of (P ).

A point x∗ ∈ D,such that f (x∗ ) is maximum (or minimum) is called an optimum solution of

Let (P ) be a mathematical program, we can have:

1. D = ∅. (P ) has no feasable solution.





 max(x2 + 2)

Exemple 1 (P ) x ≤ −2




 x≥3

2. D 6= ∅.(P ) has no optimal solution.


 max(x)

Exemple 2 (P )
 x≥3


3. D 6= ∅. (P ) has an optimal solution.


 max(x)

Exemple 3 (P )
 x≤0


3
Z. Benméziane 1.1. A Mathematical Program

Remarque 1 Let (P ) be the following mathematical program:


 M in f (x)

(P )
 x∈D


0
To solve
 (P ) equivale to solve (P ) defined as follows:
 M ax g(x)

0
(P ) such that: g(x) = −f (x)
 x∈D


Définition 2 A mathematical program (P ) is linear if :

1. The function f is a linear function of the decision variables x1 , x2 , ..., xn :

f (x1 , x2 , ..., xn ) = c1 x1 + c2 x2 + ... + cn xn

The function f is called the objective function of (P ).

2. The feasable space D is defined by a linear system of equalities (=) or inequalities (≤

or ≥)

a11 x1 +a12 x2 +... +a1n xn ≤ b1







 a21 x1 +a22 x2 +... +a2n xn ≥ b2


+...







 am x1 +am2 x2 +... +amn xn = bm


These equations and inequations are called the constraints of the linear program (P ).

A general linear program in the decision variables x1 , x2 , ..., xn is therefore of the following

form:

M aximise or M inimise Z= c1 x1 + c2 x2 + ... + cn xn

subject to: ai1 x1 +ai2 x2 +... +ain xn ≤ bi , i = 1, .., m

4
Z. Benméziane 1.1. A Mathematical Program

1.1.1 Linear program example

A firm produces two products A and B using raw materials I,II and III.

The quantities (in units) of materials I,II and III are 8,7 and 3 respectively.

To make a unit of A, we use 2 units of I and 1 unit of II, and to make a unit of B, we use 1

unit of I, 2 units of II and 1 unit of III.

The sale of a unit of A brings a profit of 4 and that of a unit of B brings a profit of 5.

The problem is to maximise the profit given constraints.

The decision variables of the problem are:

x1 :quantity of product A to be produced

x2 :quantity of product B to be produced

The objective function:

maxZ = 4x1 + 5x2

The contraints:

x1 ≥ 0

x2 ≥ 0

2x1 + x2 ≤ 8

x1 + 2x2 ≤ 7

x2 ≤ 3

So we have the following linear program:

5
Z. Benméziane 1.1. A Mathematical Program

2x1 + x2 ≤ 8







 x1 + 2x2 ≤ 7

(P 1)
x2 ≤ 3







 4x1 + 5x2 = Z(max), x1 , x2 ≥ 0


1.1.2 Notations

We consider that all the operations of addition, difference and product of matrices are known.

1. If J is a finite set, |J| represents its cardinality.

2. A is a m by n matrix

Aji :is the entry in the i-th row and j-th column of the matrix A.

3. The matrix with all its elements equal to 0 is denoted: 0.

4. Let A and B two m by n matrices:

. A ≤ B if and only if Aji ≤ Bij , for each i and j, i = 1, .., m, j = 1, .., n

. A = B if and only if Aji = Bij , for each i and j, i = 1, .., m, j = 1, .., n

5. At : is the transpose of the matrix A.

6. Un :is the n by n matrix identity defined by:


 1 if i = j

j
(Un )i =
 0 else


7. A−1 : is the inverse matrix of the matrix A.

8. Let A be a m by n matrix.

I ⊂ {1, 2, .., m}, J ⊂ {1, 2, .., n}

i ∈ {1, 2, .., m}, j ∈ {1, 2, .., n}

6
Z. Benméziane 1.1. A Mathematical Program

(a) Ai : represents the ith row of A

AI : represents the |I| ∗ n matrix in which the rows are the Ai , i ∈ I.

(b) Aj : represents the j th column of A

AJ : represents the m ∗ |J| matrix in which the columns are the Aj , j ∈ J.

(c) AJI : represents the |I| ∗ |J| matrix with the elements are Aji , i ∈ I and j ∈ J.

Exemple 4 Given :
 
3
     
 
2 1 1 0 0 8 2
   
     
A= , b = , x =
 
1 2 0 1 0 7 0
  
     
 
0 1 0 0 1 3 0
 
 
1
   
c= 4, 5, 0, 0, 0 , y = 1, 2, 0 ,

J = {1, 2, 5}, I = {1, 2}

explicite:

AJ , AI , AJI , xJ , cJ , AJ xJ , y.A, cJ .xJ .

Remarque 2 The matrix form of the linear program (P (1)) is as follows:





 C.x = Z(max)

(P1 ) A.x ≤b




 x≥0
   
 2 1   8 
 
  x 1  
where A =  ,x =  , b =

 1 2     7 
 
  x2  
0 1 3

c = (4, 5)

7
Z. Benméziane 1.1. A Mathematical Program

1.1.3 Different forms of a linear program:

A standard form:

A linear program is said to be in standard form if:

. it is a maximization program,

. there are only equalities (no inequalities) and

. all decision variables are restricted to be non negative.

A linear program (P ) in standard form has the following matrix form:


 C.x

 = Z(max)


(P ) A.x =b




 x≥0

A canonical form

A linear program is in canonical form if it is of the form:

 



 C.x = Z(max) 


 C.x = Z(min)
 
 A.x ≤b  A.x ≥b

 


 x≥0 
 x≥0

canonical form of maximization canonical form of minimisation

Définition 3 Two linear programs (P ) and (P 0 ) are equivalent if for every feasable solution

of one, we can associate a feasble solution of the other such that the objective functions’s

values are equal.





 C.x = Z(max)

Remarque 3 1. every linear program A.x ≤b




 x≥0

8
Z. Benméziane 1.1. A Mathematical Program

in canonical form of maximisation can be written in standard form:

indeed, If we have an inequality constraint:

ai1 x1 + ... + ain xn ≤ bi

then we can transform it into an equality constraint by adding a slack variable, say

xn+1 , restricted to be non negative:

ai1 x1 + ... + ain xn + xn+1 = bi , and xn+1 ≥ 0.





 C.x = Z(max)

2. every linear program (P ) A.x =b




 x≥0

in standard form can be written in canonical form of maximisation:

indeed,
 Ai x = bi the ith constraint of (P ), can be transformed in two inequalities
 A i x ≤ bi


 −Ai x ≤ −bi


Théorème 1 Every linear program can be transformed into an equivalent linear program

written in canonical form of maximisation.

Preuve.

1. If the objective function is to minimize z, z = c1 x1 + c2 x2 + ... + cn xn then we can

maximize z 0 = −z = −c1 x1 − c2 x2 − ... − cn xn .

2. The constraints:

Let us consider the ith constraint:

(a) Ai x ≥ bi ⇐⇒ −Ai x ≤ −bi


 A i x ≤ bi

(b) Ai x = bi ⇐⇒
 −Ai x ≤ −bi


3. The variables:

9
Z. Benméziane 1.2. Duality

(a) if xi ≤ 0, then we replace every occurence of xi , by −x0i , where x0i is a non negative

decision variable.

(b) if xi is unrestricted in sign, then we can introduce two new decision variables x0i

and x00i restricted to be non negative and replace every occurrence of xi by x0i − x00i .

1.2 Duality

1.2.1 Find a dual from its canonical form

Définition 4 Let (P ) a linear program in canonical form of maximisation





 A.x ≤ b, x ≥ 0

(P ) C.x = Z(max)






The dual of (P ) is the linear program (D) defined by:


 y.A ≥ c, y ≥ 0

(D)
 y.b = W (min)


developped forms of (P ) and (D) :

 
 A11 x1 +.....+ An1 xn

 ≤ b1 , xi ≥ 0 

 A11 y1 +.....+ A1m ym ≥ c1 , yi ≥ 0

 

 
A12 x1 +.....+ An2 xn ≤ b2 , ∀i = 1, .., n A21 y1 +.....+ A2m ym ≥ c2 , ∀i = 1, .., m

 


 


 

(P ) .. .. .. .. (D) .. .. .. ..
 . . . .  . . . .

 

A12 x1 +.....+ A12 x1  An y +.....+ Anm ym ≥ cn
 



 ≤ bm 

 1 1

 


 cx 
1 1 +.....+ cn xn = Z(max)  by
1 1 +.....+ bm ym = W (min)

Exemple 5 the dual of the following linear program (P1 ):

10
Z. Benméziane 1.2. Duality





 

≤8



 2x1 +x2 



 

 
≤7
 



 x1 +2x2 



 
(P1 ) x2 ≤3 is (D1 )

 


 




 4x1 +5x2 = Z(max) 




 


 x ≥0 x ≥0 

1 2 







Remarque 4 As every linear program is equivalent to a linear program in canonical form

of maximisation, then evry linear program has a dual. Especially, the dual of the dual is the

primal.

P rimal(dual) Dual(P rimal)

max min

ith constraint ≤ ith decision variable ≥ 0

ith constraint ≥ ith decision variable ≤ 0

ith constraint = ith decision variable is unrestricted in sign

ith decision variable ≥ 0 ith constraint ≥

ith decision variable ≤ 0 ith constraint ≤ 0

ith decision variable is unrestricted in sign ith constraint =

Exemple 6 (exercise 7 Tutorial sheet 1)

Give the dual of (P ) :

11
Z. Benméziane 1.2. Duality

x1 +x2 ≤ 8







−1x1 +2x2 = 3

 



 

 
 x1 −x2 ≥ −2
 

(P ) (D)
5x1 +x2 ≥ 1, x1 ≥ 0

 


 


 

3x1 −2x2 = −1, x2 ≤ 0







 5x1

+3x2 = Z(max)

1.2.2 Interaction between a linear program and its dual:




 A.x ≤ b, x ≥ 0 

  y.A ≥ c, y ≥ 0

Théorème 2 Let (P ) C.x = Z(max) and let (D) its dual.


  y.b

= W (min)



If x is a feasible solution of (P ) and y is a feasible solution of (D) then: cx ≤ yb.

Preuve. Let x a feasible solution of (P ) and y is a feasible solution of (D) then:

A.x ≤ b, x ≥ 0 and y.A ≥ c, y ≥ 0

so, A.x ≤ b ⇐⇒ y.A.x ≤ y.b, ....(1) because y ≥ 0

and y.A ≥ c ⇐⇒ y.A.x ≥ c.x....(2) because x ≥ 0

from (1) and (2), we have:

c.x ≤ y.A.x ≤ y.b ,

then: cx ≤ yb




 A.x ≤ b, x ≥ 0 

  y.A ≥ c, y ≥ 0

Corollaire 1 Let (P ) C.x = Z(max) and let (D) . its dual.


  y.b

= W (min)



If x is a feasible solution of (P ) and y is a feasible solution of (D) and cx = yb, then x is

an optimal solution of (P ) and y is an optimal solution of (D).

12
Z. Benméziane 1.2. Duality

Preuve. by the contrary.

suppose x and y are feasble solutions of (P ) and (D) respectively, such that cx = yb and

suppose x is not an optimal solution of (P ), so there exists a feasble solution x∗ of (P ) such

that cx∗ > cx, , but we have cx = yb, so cx∗ > yb, but this contradicts theorem 17.

13