The University of Jordan

Department of Mathematics
Simplex Method : Background for Integer Programming
Prepared and presented by :
Nadjiba MANSOURI Soraya ABDELAZIZ
Ilhem KADRI Mohamed BENRABIA
Samia MAZOUZ Said ATLAOUI
Ahmed BOUCHENAK Ibtissem ATTAFI

Supervised and edited by : Dr. Baha ALZALG

Integer and Combinatorial Optimization (0301973)

Spring 2018
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 1 / 12
 Sommaire

1 Introduction

2 Example of the Simplex Method

3 Example of the Dual Simplex Method

Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 2 / 12

 Introduction

In mathematical optimization, the Simplex Method is a popular

algorithm for linear programming.
In the simplex method we move from one solution point to an
adjacent point.

Why do we care ?
The first step for solving an integer programming problem is to solve
the linear programming relaxation by using the simplex method.
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 3 / 12
 Introduction

Relaxation

Let P be a mixed integer linear program :



 min c T x + d T y + e T z
s.t. Ax + By + Cz ≤ b


 x, y , z > 0 (P)
y ∈ Z.


The linear programming relaxation is :


 min c T x + d T y + e T z
s.t. Ax + By + Cz ≤ b (P 0 )
x, y > 0.


Problem (P 0 ) is called linear relaxation of (P).

Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 4 / 12

 Example of the Simplex Method

Example (the simplex method)

When solving a (mixed) integer program, the simplex method is used
to solve the relaxed problem.
Consider the linear programming problem :

max z = 12x1 + 16x2

s.t. 10x1 + 20x2 ≤ 120
8x1 + 8x2 ≤ 80
x1 , x2 ≥ 0.

The problem in the standard form is :

max 12x1 + 16x2 + 0e1 + 0e2

s.t. 10x1 + 20x2 + e1 = 120
8x1 + 8x2 + e2 = 80
x1 , x2 , e1 , e2 ≥ 0.
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 5 / 12
 Example of the Simplex Method

Iteration 1
cj 12 16 0 0
CBi basic variable x1 x2 e1 e2 Solution Ration
120
0 e1 10 20 1 0 120 20
=6
80
0 e2 8 8 0 1 80 8
= 10
zj 0 0 0 0 0
cj − zj 12 16 0 0
Here,
2
X
zj = (CBi )(aij ).
i=1

(0 × 10) + (0 × 8) = 0 + 0 = 0.
The optimality condition is :
For max LP : All cj − zj ≤ 0. For min LP : All cj − zj ≥ 0.
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 6 / 12
 Example of the Simplex Method

Iteration 2

cj 12 16 0 0
CBi basic variable x1 x2 e1 e2 Solution Ration
1 1
16 x2 2
1 20
0 6 12
−2
0 e2 4 0 5
1 32 8
4
zj 8 16 5
0 96
−4
cj − zj 4 0 5
0

Corr. key column value × Corr. key row value

New value = Old value − Key element
.

8×10
For instance, 4 = 8 − 20
.

Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 7 / 12

 Example of the Simplex Method

Iteration 3
cj 12 16 0 0
CBi basic variable x1 x2 e1 e2 Solution Ration
1 −1
16 x2 0 1 10 8
2
−1 1
12 x1 1 0 10 4
8
2
zj 12 16 5
1 128
−2
cj − zj 0 0 5
-1

We have that
cj − zj ≤ 0.
So, the optimal solution is
x ? = (x1 , x2 ) = (8, 2),
and hence
zopt = 128.
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 8 / 12
 Example of the Dual Simplex Method

Example (the dual simplex method)

When solving a (mixed) integer program, the dual simplex method is
also used to solve the relaxed problem.
Consider the linear programming problem :
max z = −2x1 − x3
s.t. −x1 − x2 + x3 ≤ −5
−x1 + 2x2 − 4x3 ≤ −8
x1 , x2 , x3 ≥ 0.
The problem in the standard form is :
max z = −2x1 − x3
s.t. −x1 − x2 + x3 + s1 = −5
−x1 + 2x2 − 4x3 + s2 = −8
x1 , x2 , x3 , s1 , s2 ≥ 0.

Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 9 / 12

 Example of the Dual Simplex Method

Iteration 1
Note that, as a condition for using this method, at least one of the
components of b must be negative.
This is the outcome of the first iteration.
cj -2 0 -1 0 0
CBi basic variable x1 x2 x3 s1 s2 Solution
0 s1 -1 -1 1 1 0 -5
0 s2 -1 2 -4 0 1 -8
zj 0 0 0 0 0 0
cj − zj -2 0 -1 0 0
1
Ratio 2 - 4
- -
 
c −zj
Here, Ratio = Keyj row . The key column is corresponding to the
minimum positive value of the “Ratio Raw”, and hence the entering
variable is x3 as min(Ratio) = min 2, 14 = 41 .


Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 10 / 12
 Example of the Dual Simplex Method

Iteration 2

The key row is corresponding to the most negative value of “Solution

Column”, and hence the leaving variable is s2 .
The second iteration yields
cj -2 0 -1 0 0
CBi basic variable x1 x2 x3 s1 s2 Solution
−5 −1 1
0 s1 4 2
0 1 4
-7
1 −1 −1
-1 x3 4 2
1 0 4
2
−1 1 −1
zj 4 2
-1 0 4
-2
−7 −1 1
cj − z j 4 2
0 0 4
7
Ratio 5
1 − - -

Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 11 / 12

 Example of the Dual Simplex Method

Iteration 3
We repeat the same work in each iteration and we stop when we see
the values of cj − zj are nonnegative and the values of solution are
positive.
The third iteration produces

cj -2 0 -1 0 0
CBi basic variable x1 x2 x3 s1 s2 Solution
5 −1
0 x2 2
1 0 -2 4
14
−1
-1 x3 -1 0 1 -1 2
9
1
zj 1 0 -1 1 2
-9
−1
cj − zj -3 0 0 -1 2

The above tableau is optimal ! We obtain x1 = 0, x2 = 14, x3 = 9

and zopt = −9.
Integer Optimization (University of Jordan) THE SIMPLEX METHOD 29-04-2018 12 / 12