Chapitre 4

Simplex method initialization

Let us consider the linear program (P ) written in standard form:


 Ax

=b
(P )
 Cx = Z(max), x ≥ 0


We suppose that b ≥ 0. The aim of this chapter is to write (P ) in canonical form with

respect to a feasible basis if such a basis exists. The linear program (P ) is associated with

the following auxiliary program:


 Ax +Um .V = b

(P A) Pm
i=1 vi = Ψ(min) , vi ≥ 0, i = 1, .., m


The vi variables are called artificial variables.

34
Z. Benméziane 4. Simplex method initialization

Exemple
 13 . 



 2x1 −3x2 = −4 


 −2x1 +3x2 +v1 =4
 
(P ) x1 +2x2 = 17 y (P A) x1 +2x2 +v2 = 17

 


 

 2x +5x = Z(max) x , x ≥ 0  v1 + v2 = Ψ(min)v1 , v2 ≥ 0, x1
1 2 1 2

1. The ith constraint of (P A) is written as follows:

Ai x + vi = bi .

when all the artificial variables vi , i = 1, .., m are equal to zero then the solution x of

(P A) is also solution of (P ).

Pm Pm Pm
2. i=1 vi = i=1 bi − i=1 Ai x = e.b − e.A.x, e = (1, 1, .., 1)
| {z }
m times


Ax +Um .V =b


(P A0) (P A0 )
 −eAx

= Ψ(min) − e.b , xi ≥ 0, i = 1, .., n, vi ≥ 0, i = 1, .., m

is written in canonical form withe respect to the feasible basis J = {n + 1, n + 2, ..., n +

m}. we can then apply the algorithm simplex to the linear program (P A0 ), and the

algorithm leads to a basis optimal solution, because the objective function f is lower

bounded by 0( m
P
i=1 vi ≥ 0).

Let (P
d A) be the linear program written in canonical form with respect to an optimal

basis:

35
Z. Benméziane 4. Simplex method initialization

 Ax
 b +B.V
b = bb
(P
d A)
cx +dv = Ψ(min) − Ψ
b , xi ≥ 0, i = 1, .., n, vi ≥ 0, i = 1, .., m

 b b

Ψ:
b the optimal fuction objective value.

Théorème 6 Let (x, v) be an optimal solution of (P A). A necessary and sufficient condition

so that (P ) has a feasible solution is that vi ≥ 0, ∀i = 1, .., m.

1. If no artificial variable is in the optimal basis of (P A), then by removing all the artificial

variables (which are all out of basis), we obtain the linear program (P ) written in

canonical form with respect to a feasible basis.

2. Suppose that there is an artificial variable vs which is basis variable of the linear

program (P
d A) , then vs = 0; and A
bs x + vs = 0. If there exists an index j such that

bj 6= 0, then by applying the pivot algorithm around A

A bj , the artificial variable vs will
s s

leave the basis and the variable xj will enter in the basis.

If no pivoting operation is possible (which means that there is no an index j such that
bj 6= 0), then the sth equation is redundant, so we delete it.
As

3. After having examined all the artificial variables which may be found in the final basis,

having eliminated them by pivoting if this is possible or having deleted the corresponding

36
Z. Benméziane 4. Simplex method initialization

equation is such pivoting is impossible we arrive at a basis which can be used to initialize

the simplex algorithm to solve (P ). These operations constitute the first phase of the

simplex algorithm.

The simplex algorithm is the second pahse.

Exemple 14

x1 +x2 +x3 −x4 =2







 −x1

+x2 +2x3 −2x4 =1
(P )
6x1 −3x3 +3x4 =3







 3x1 −2x2 −x3 +2x4 = Z(max) xi ≥ 0, i = 1, .., 4



x1 +x2 +x3 −x4 +v1 =2







 −x1 +x2 +2x3 −2x4

+v2 =1
(P A0 )
6x1 −3x3 +3x4 +v3 = 3, vi ≥ 0, i = 1, .., 3







v1 +v2 +v3 = Ψ(min), xi ≥ 0, i = 1, .., 4



↓

x1 +x2 +x3 −x4 +v1 =2







 −x1

+x2 +2x3 −2x4 +v2 =1
(P A0 )
6x1 −3x3 +3x4 +v3 = 3, vi ≥ 0, i = 1, .., 3







 −6x1 −2x2

= Ψ(min) − 6, xi ≥ 0, i = 1, ..,

37
Z. Benméziane 4. Simplex method initialization

1. Suppose that in the initial linear program (P ), one variable, say x1 , appears only in the

line r of (P ), then if Asr and b have the same sign, it is not necessary to introduce an

artificial variale with respect to this equation, endeed we can consider x1 as the basis

variable corresponding to the rth equation of (P ).

In that case, to go from (P A) to (P A0 ), we only subtract from the objective function

of (P A) the equations for which we have introduced artificial variables.

2. At each iteration of the simplex algoritm applied to (P A), if one artificial variable

leaves the basis and is pelaced by an initial variable, then we can completly ignore this

artificial variable and delete the corresponding column.


x1 +2x2 +x3 =2







 x1 +x2 +5x3

= 12
Exemple 15 (P )
x1 +2x2 +6x3 +x4 = 13







 x1 +x2 +x3 +x4 = Z(min) xi ≥ 0, i = 1, .., 4


38