Artificial Variables

Artificial Variables

V O Thomas

Department of Mathematics, Faculty of Science

V O Thomas Department of Mathematics, Faculty of Science 1

 Artificial Variables

Artificial Variables

If the constraints are equations or (≥) type, getting an initial basic

feasible solution of the LPP often becomes difficult. In order to obtain
an initial basic feasible starting solution, we first put the LPP in the
standard form with the help of slack or surplus variables. Then a
non-negative variable is added to the left side of each constraint
equation that lacks the starting basic variables. The variables so
added are called artificial variables. These variables have no physical
meaning with regard to the given LPP and hence they should be
forced to go out of the basis or exist in the basis at zero level when
the optimum solution is obtained.

V O Thomas Department of Mathematics, Faculty of Science 2

 Artificial Variables

There are two methods for solving LPP with artificial variables:
1 Two-phase method
2 Big-M method or Charnes’ M-method or Method of Penalties
We shall discuss here only the second method.

Big-M Method

Different steps involved in Big-M method are given below.

Step 1
Write the LPP in the standard form and check whether it admits a
starting solution or not.
(a) If there is a starting feasible solution go to Step 3.
(b) If there does not exist a starting feasible solution go to Step 2.

V O Thomas Department of Mathematics, Faculty of Science 3

 Artificial Variables

Step 2
Add artificial variables to the LHS of each equation that has no
obvious starting basic variables. Assign a very high penalty (say M)
to these variables in the objective function.

Step 3
Apply simplex method to the modified LPP. The following cases may
arise in the last iteration.
(a) At least one artificial variable is present in the basis with zero
value and the optimality condition is reached, then the current
solution is an optimum basis feasible solution.

V O Thomas Department of Mathematics, Faculty of Science 4

 Artificial Variables

(b) If no artificial variable appears in the basis and the optimality

condition is reached, then the current solution is optimal basic
feasible solution.

(c) At least one artificial variable is present in the basis with a

positive value, then the LPP has no optimum basic feasible solution.

Example
Use Penalty ( Big-M ) method to M aximize z = 6x1 + 4x2 subject to
2x1 + 3x2 ≤ 30 , 3x1 + 2x2 ≤ 24 , x1 + x2 ≥ 3 and x1 , x2 ≥ 0. Is the
solution unique? If not give two different solutions

V O Thomas Department of Mathematics, Faculty of Science 5

 Artificial Variables

Solution
First we shall write the LPP in the standard form:

M ax : z = 6x1 + 4x2 + 0x3 + 0x4 + 0x5

Subject to

2x1 + 3x2 + x3 = 30

3x1 + 2x2 + x4 = 24

x1 + x2 − x5 = 3

and
x1 , x2 , x3 , x4 , x5 ≥ 0

V O Thomas Department of Mathematics, Faculty of Science 6

 Artificial Variables

This LPP has no starting solution which is basic feasible. So we

introduce an artificial variable A ≥ 0 in the last constraint. We shall
assign a large penalty M in the objective function. Then the LPP
becomes

M ax : z = 6x1 + 4x2 + 0x3 + 0x4 + 0x5 − M A

subject to

2x1 + 3x2 + x3 = 30

3x1 + 2x2 + x4 = 24

x1 + x2 + x5 + A = 3

and
x1 , x2 , x3 , x4 , x5 , A ≥ 0

V O Thomas Department of Mathematics, Faculty of Science 7

 Artificial Variables

6 4 0 0 0 −M
cB B.V. xB P1 P2 P3 P4 P5 P6 Min. Ratio
30
0 x3 30 2 3 1 0 0 0 2 = 15
24
0 x4 24 3 2 0 1 0 0 3 =8
3
−M A 3 1 1 0 0 −1 1 1 =3→
z = −3M −M − 6 ↑ −M − 4 0 0 M 0 ← ∆j
24
0 x3 24 0 1 1 0 2 − 2 = 12
15
0 x4 15 0 −1 0 1 3 − 3 =5
6 x1 3 1 1 0 0 −1 − −
z = 18 0 2 0 0 −6 ↑ − ← ∆j
5 −2 42
0 x3 14 0
3 1 3 0 − 5 = 8.4 →
−1 1
0 x5 5 0 3 0 3 1 − −
2 1 24
6 x1 8 1 3 0 3 0 − 2 = 12
z = 48 0 0↑ 0 2 0 − ← ∆j

V O Thomas Department of Mathematics, Faculty of Science 8

 Artificial Variables

All ∆j ≥ 0 and hence the solution is optimum. The optimum basic

feasible solution is
x1 = 8 , x2 = 0 and M ax z = 48.

The net evaluation corresponding to the non-basic variable x2 is zero.

This is an indication that the current solution is not unique and an
alternative solution exists. Now bringing P2 into the basis and
removing P3 from the basis we get.

6 4 0 0 0
cB B.V. xB P1 P2 P3 P4 P5 Min. Ratio
42 3 −2
4 x2 5 0 1 5 5 0
39 1 1
0 x5 5 0 0 5 5 1
12 −2 3
6 x1 5 1 0 5 5 0
z = 48 0 0 0 2 0

V O Thomas Department of Mathematics, Faculty of Science 9

 Artificial Variables

All ∆j ≥ 0 and hence the solution is optimum. The optimum solution

is
x1 = 12 42
5 , x2 = 5 and M ax z = 48.
! !
12
8 5
Let X1 = and X2 = 42
, then
0 5

X = λX1 + (1 − λ)X2

is also a solution.
! !
12
8 5
X=λ + (1 − λ) 42
0 5
! !
8λ (1 − λ) 12
5
= +
0 (1 − λ) 42
5

V O Thomas Department of Mathematics, Faculty of Science 10

 Artificial Variables

!
12
+ 8λ − 12
5 5 λ
= 42 42
0+ 5 − 5 λ
!
12
5 + 285 λ
= 42 42
5 − 5 λ
! !
12
1 5 + 28
5 · 21 26
5
For λ = 2 , we get X3 = 42 42 1 = 21
5 − 5 · 2 5

! !
12 28 1 14
1 5 + 5 · 14 5
For λ = 14 , we get X4 = 42 42 1
= 39
5 − 5 · 14 5

V O Thomas Department of Mathematics, Faculty of Science 11

 Artificial Variables

Example

Use penalty method to M ax z = 3x1 + 2x2 subject to

2x1 + x2 ≤ 2 , 3x1 + 4x2 ≥ 12 , x1 , x2 ≥ 0

Solution

We shall rewrite the equation in standard form and then add artificial
variable to the second constraint.

M ax z = 3x1 + 2x2 + 0x3 + 0x4 − M A

subject to

2x1 + x2 + x3 =2

3x1 + 4x2 − x4 + A = 12
and
x1 , x2 , x3 , x4 ≥ 0.
V O Thomas Department of Mathematics, Faculty of Science 12
 Artificial Variables

3 2 0 0 −M
cB B.V. xB P1 P2 P3 P4 P5 Min. Ratio
2
0 x3 2 2
1 1 0 0 1 =2
12
−M A 12 3 4 0 −1 1 4 =3
z = −12M −3M − 3 −4M − 2 ↑ 0 M 0
2 x2 2 2 1 1 0 0
−M A 4 −5 0 −4 −1 1
z = −4M + 4 5M + 1 0 4M + 2 M 0

Since all ∆j ≥ 0, optimality has been reached. Since artificial variable

Ais appearing in the basis at positive level, the given LPP has no
feasible solution.

V O Thomas Department of Mathematics, Faculty of Science 13

 Artificial Variables

Example
Use Big-M method to M aximize z = 8x2 Subject to 2x1 + 3x2 ≤ 6
and x1 , x2 unrestricted.

Solution
If we write the unrestricted variables x1 , x2 in the form x1 = x01 − x001
andx2 = x02 − x00 2 , we can rewrite the LPP as:
M ax z = 8(x02 − x00 2 )
subject to
x01 − x001 − x02 + x002 ≥ 0,
2x01 − 2x001 + 3x02 − 3x002 ≤ 6,
and
x01 , x02 , x001 , x002 ≥ 0

V O Thomas Department of Mathematics, Faculty of Science 14

 Artificial Variables

∴ The LPP in the standard form is

M ax z = 0x01 − 0x001 + 8x02 − 8x002 + 0x3 + 0x4 − M A

subject to

x01 − x001 − x02 + x002 − x3 +A=0

2x01 − 2x001 + 3x02 − 3x002 + x4 =6

and
x01 , x001 , x02 , x002 , x3 , x4 , A ≥ 0

V O Thomas Department of Mathematics, Faculty of Science 15

 Artificial Variables

0 0 8 −8 0 0 −M
cB B.V. xB P1 P2 P3 P4 P5 P6 P7 Min. Ratio
−M A 0
1 −1 −1 1 −1 0 1 0→
6
0 x4 6 2 −2 3 −3 0 1 0 2 =3
z=0 −M ↑ M M − 8 −M + 8 M 0 0 ← ∆j
0 x002 0 1 −1 −1 1 −1 0 − −
6
0 x4 6 0 0
5 −5 2 1 − 5 →
z=0 0 0 −8 ↑ 8 0 0
−3
0 x01 6
5 1 −1 0 0 5
1
5 -
0 6 2 1
8 x2 5 0 0 1 −1 5 5 -
z = 485 0 0 0 3 16
5
5M−16
5 −

V O Thomas Department of Mathematics, Faculty of Science 16

 Artificial Variables

Since all ∆j ≥ 0, optimality has been reached. The optimum solution

is
x01 = 65 , x02 = 65 and M ax z = 485 .
Reverting to the original variables
x1 = x01 − x001 = 56 − 0 = 65 and x2 = x02 − x002 = 65 − 0 = 65
Thus we have
x1 = 56 , x2 = 65 and M ax z = 48 5 .

V O Thomas Department of Mathematics, Faculty of Science 17

 Artificial Variables

Exercise
1 Solve the following LPP by Big-M method:

(a) M inimize : z = 2x1 + x2

s.t.
3x1 + x2 = 3 , 4x1 + 3x2 ≥ 6 , x1 , x2 ≥ 0.

(b) M aximize : z = 2x1 + x2 + 3x3

s.t.
x1 + x2 + 2x3 ≤ 5 , 2x1 + 3x2 + 4x3 = 12 , x1 , x2 , x3 ≥ 0.

(c) M inimize : z = 5x1 − 6x2 − 7x3

s.t.
x1 + x2 + 2x3 ≤ 5 , 5x1 − 6x2 + 10x3 ≥ 0 ,
x1 + x2 + x3 = 5 , x1 , x2 , x3 ≥ 0.

V O Thomas Department of Mathematics, Faculty of Science 18

 Artificial Variables

V O Thomas Department of Mathematics, Faculty of Science 19