Big M method

Artificial variables

In the previous chapter, we considered LPP involving upper bound constraints (≤ type) only. These
constraints were changed to equalities by introducing slack variables. These slack variables become the
initial basic variables and provide a feasible solution to start with.
Now we consider problems having lower bound constraints (≥ type) and equality constraints. The
inequality of ≤ type can be changed toequality by subtracting a surplus variable. But its coefficient is
-1 and hence it has a negative value , and therefore it gives rise to an infeasible solution. Thus a surplus
variable cannot become a starting basic variable. In order to obtain an initial basic feasible solution we
introduce another variable called artificial variable.
For example, consider the constraint

2x1 + 3x2 + 4x3 ≥ 10

Introducing a surplus variable s, we get

2x1 + 3x2 + 4x3 − s = 10

∴ s = 2x1 + 3x2 + 4x3 − 10

Assigning zero values to the the non-basic variables we get s=-10 (negative value).
Therefore we introduce an artificial variable R and write

2x1 + 3x2 + 4x3 − s + R = 10

∴ R = 10 − 2x1 − 3x2 − 4x3 + s

Now R becomes a starting basic variable with a value 10.

Big M method

The purpose of introducing the artificial variable is just to obtain an initial basic feasible solution.
But already we have equality with a surplus variable. Adding one more variable to only one side of
this equality causes violation of the equality constraint. In order to overcome this difficulty we need to
get rid of these variables and not allow them to appear as optimal basic variables. If at all an artificial
variable becomes a basic variable in the final table, its value must be zero. This will ensure a solution
to the problem. To achieve this, these artificial variables are assigned a large penalty cost coefficient
(-M ) in the objective function to be maximised. However, if there is an artificial variable in the optimal
basis with positive value then it indicates that the given problem has no solution.
Solve the modified LPP by simplex method, until anyone of the three cases may arise.

1. If no artificial variable appears in the basis and the optimality conditions are satisfied, then the
current solution is an optimal basic feasible solution.
2. If at least one artificial variable in the basis at zero level and the optimality condition is satisfied
then the current solution is an optimal basic feasible solution (though degenerated solution).
3. If at least one artificial variable appears in the basis at positive level and the optimality condition is
satisfied, then the original problem has no feasible solution. The solution satisfies the constraints but
does not optimise the objective function, since it contains a very large penalty M.
 Example 1
Maximise Z = 3x1 − x2 subject to the constraints

2x1 + x2 ≥ 2
x1 + 3x2 ≤ 3
x2 ≤ 4
x1 , x2 ≥ 0
Solution: The standard form of the problem is
Maximise Z = 3x1 − x2 + 0s1 + 0s2 + 0s3 − M R subject to the constraints

2x1 + x2 − s1 + R = 2
x1 + 3x2 + s2 = 3
x2 + s 3 = 4
x1 , x2 , s1 , s2 , s3 , R ≥ 0
Here s1 is a surplus variable, s2 and s3 are slack variables and R is an artificial variable with a penalty
cost −M in the objective function. M is a very large number tending to infinity. The starting simplex
table is

Table(1)
z x1 x2 s1 s2 s3 R const.(b)
CB B 3 −1 0 0 0 −M 0 Ratio
−M R 2 1 −1 0 0 1 2 1 −→
0 s2 1 3 0 1 0 0 3 3
0 s3 0 1 0 0 1 0 4 −
Zj − cj (−2M − 3) ↑ (−M + 1) M 0 0 0 −2M

Table(2)
z x1 x2 s1 s2 s3 R const.(b)
CB B 3 −1 0 0 0 −M 0 Ratio
3 x1 1 1/2 −1/2 0 0 1/2 1 −
0 s2 0 5/2 1/2 1 0 −1/2 2 4 −→
0 s3 0 1 0 0 1 0 4 −
Z j − cj 0 5/2 −3/2 ↑ 0 0 3/2 + M 3

Table(3)

z x1 x2 s1 s2 s3 R const.(b)
CB B 3 −1 0 0 0 −M 0 Ratio
3 x1 1 3 0 1 0 0 3 −
0 s1 0 5 1 2 0 −1 4 −
0 s3 0 1 0 0 1 0 4 −
Zj − cj 0 10 0 3 0 M 9 −

All cj ≥ 0 and therefore optimality is reached. The optimal solution is x1 = 3, x2 = 0, Zmax = 9.

 Example 2

Minimise Z = 5x1 + 4x2 + 3x3 subject to the constraints

x1 + x2 + x3 = 100
x1 ≤ 20
x2 ≥ 30
x3 ≤ 40
x1 , x2 , x3 ≥ 0

Solution: Here the first constraint is an equality and hence we introduce an artificial variable only.
In the second and fourth constraint slack varaibles are introduced. In the third constraint we have to
introduce a surplus variable and an artificial variable.

The standard form of the LPP is

Maximise W = −Z = −5x1 − 4x2 − 3x3 + 0s1 + 0s2 + s3 − M R1 − M R2 subject to the constraints

x1 + x2 + x3 + R1 = 100
x1 + s1 = 20
x2 − s2 + R2 = 30
x3 + s3 = 40
x1 , x2 , x3 s1 , s2 , s3 , R1 , R2 ≥ 0

The starting simplex table is

Table(1)

z x1 x2 x3 s1 s2 s3 R1 R2 const.(b)
CB B −5 −4 −3 0 0 0 −M −M 0 Ratio
−M R1 1 1 1 0 0 0 1 0 100 100
0 s1 1 0 0 1 0 0 0 0 20 −
−M R2 0 1 0 0 −1 0 0 1 30 30 −→
0 s3 0 0 1 0 0 1 0 0 40 −
Zj − cj (−M + 5) (−2M + 4) ↑ (−M + 3) 0 M 0 0 0 −130M

Table(2)

z x1 x2 x3 s1 s2 s3 R 1 R2 const.(b)
CB B −5 −4 −3 0 0 0 −M −M 0 Ratio
−M R1 1 0 1 0 1 0 1 −1 70 70 −→
0 s1 1 0 0 1 0 0 0 0 20 −
−4 x2 0 1 0 0 −1 0 0 1 30 −
0 s3 0 0 1 0 0 1 0 0 40 −
Zj − cj (−M + 5) 0 (−M + 3) 0 (−M + 4) ↑ 0 0 2M − 4 −120M − 70M −

Table(3)
 z x1 x2 x3 s 1 s 2 s 3 R1 R2 const.(b)
CB B −5 −4 −3 0 0 0 −M −M 0 Ratio
0 s2 1 0 1 0 1 0 1 −1 70 70
0 s1 1 0 0 1 0 0 0 0 20 −
−4 x2 1 1 1 0 0 0 1 0 100 100
0 s3 0 0 1 0 0 1 0 0 40 40 −→
Zj − cj 1 0 −1 ↑ 0 0 0 −4 + M M −400 −

Table(4)

z x1 x2 x3 s 1 s 2 s 3 R1 R2 const.(b)
CB B −5 −4 −3 0 0 0 −M −M 0 Ratio
0 s2 1 0 0 0 1 −1 1 −1 30 −
0 s1 1 0 0 1 0 0 0 0 20 −
−4 x2 1 1 0 0 0 −1 1 0 60 −
−3 x3 0 0 1 0 0 1 0 0 40 −
Zj − cj 1 0 0 0 0 1 −4 + M M −360 −
The optimal solution is x1 = 0, x2 = 60, x3 = 40, Zmin. = −W = 360.