The Dual Simplex Method

The Dual Simplex method is the mirror image of Regular Simplex Method. It is particularly useful for solving LPPs for
which negative value(s) appear in the (requirement vector) B column of the tableau, e.g. during sensitivity testing for
cases where changes are made in the requirement vector B. The new XB = S-1 B, may contain negative elements.

The DUAL Simplex Method: The regular simplex method moves the initial feasible but nonoptimal solution to an
optimal solution while maintaining feasibility through an iterative procedure. On the other hand, the dual simplex
method moves the initial optimal infeasible solution to a feasible solution while maintaining optimality through an
iterative procedure.

Step 1: Rewrite the linear programming problem by expressing all the constraints in <= form and transforming them
into equations through slack variables.
Step 2: Exhibit the above problem in the form of a Simplex tableau. If the optimality condition is satisfied and one or
more basic variables have negative values, the dual simplex method is applicable.
Step 3 Feasibility condition: The basic variable with the most negative value becomes the departing variable (D.V.).
Call this row in which this value appears the work row. If more than one candidate for D.V. exists, choose arbitrarily.
Step 4: Optimality Condition: Form ratios by dividing all but the last element of the last row of (cj – zj) values
(minimization problem) or the (zj – cj ) values (maximization problem) by the corresponding negative coefficients of the
work row. The nonbasic variable with the smallest absolute ratio becomes the entering variable (E.V.). Designate this
element in the work row at the pivot element and the corresponding column the work column. If more than one
candidate for E.V. exists, choose one arbitrarily. If no element in the work row is negative, this problem has no
feasible solution.
Step 5: Use elementary row operations to convert the pivot element to 1 and then to reduce all the other elements in the
work column to zero.
Step 6: Repeat step 3 through 5 until there are no negative values for the basic variables.

Ex: Solve the following LPP,

Minimize, Z = 4 x1 + 3 x2 + 2 x3 + 5 x4
s.t. x1 + 2 x2 + 3 x2 + x4 >= 5
2 x1 – x2 + 5 x3 – x4 >= 1
2 x1 + x2 + x3 + 3 x4 >= 10
with all variables non-negatives.

Ans: Expressing all the constraints in the form <= and adding the stack variables x5, x6 and x7, the problem becomes,

Minimize, Z = 4 x1 + 3 x2 + 2 x3 + 5 x4 + 0 x5 + 0 x6 + 0 x7
s.t. - 1 x1 - 2 x2 - 3 x2 - 1 x4 + 1 x5 + 0 x6 + 0 x7 = - 5
- 2 x1 + 1 x2 - 5 x3 + 1 x4 + 0 x5 + 1 x6 + 0 x7 = - 1
- 2 x1 - 1 x2 - 1 x3 - 3 x4 + 0 x5 + 0 x6 + 1 x7 = - 10
with all variables non-negatives.
Now we form Tableau 1
Tableau 1
x1 x2 x3 x4 x5 x6 x7
B
4 3 2 5 0 0 0
x5 0 -1 -2 -3 -1 1 0 0 -5
x6 0 -2 1 -5 1 0 1 0 -1
x7 0 -2 -1 -1 - 3* 0 0 1 -10
(cj – zj) 4 3 2 5 0 0 0
xj work row coefficients -2 -1 -1 -3 0 0 1
Absolute ratios by
2 3 2 5/3
negative coefficients only

Since all the (cj – zj) values are non-negative, the above solution is optimal. However, it is infeasible because it has non
positive values for the basic variables x 5, x6 and x7. Since x7 has the most non-positive vale, it becomes the departing
variable (D.V.)

1
Since x4 has the smallest absolute ratio, it becomes the entering variable (E.V.). Thus the element -3, marked by
asterisk, becomes the pivot element. Using elementary row operation, we get Tableau 2

Tableau 2
x1 x2 x3 x4 x5 x6 x7
B
4 3 2 5 0 0 0
x5 0 -1/3 -5/3 -8/3 0 1 0 -1/3 -5/3
x6 0 - 8/3 2/3 -16/3* 0 0 1 1/3 -13/3
x4 5 2/3 1/3 1/3 1 0 0 -1/3 10/3
(cj – zj) 2/3 4/3 1/3 1 0 0 5/3
xj work row coefficients - 8/3 2/3 -16/3 0 0 1 1/3
-50/3
Absolute ratios by
1/4 1/16
negative coefficients only

Since x6 has the most non-negative value, it becomes the departing variable. Since x 3 has the smallest absolute ratio, it
becomes the entering variable (E.V.). Thus the element -16/3, marked by the asterisk, becomes the pivot element. Using
elementary row operations, we obtain tableau 3.
Tableau 3
x1 x2 x3 x4 x5 x6 x7
B
4 3 2 5 0 0 0
x5 0 1 2 0 0 1 -1/2 -1/2 ½
x3 2 ½ -1/8 1 0 0 -3/16 -1/16 13/16
x4 5 ½ 9/24 0 1 0 1/16 15/48 147/48
(cj – zj) 1/2 33/24 0 0 0 1/16 81/48 -813/48

Since all the variables have non-negative values, the above optimal solution is feasible. The optimal and feasible
solution is x1* = 0, x2* = 0, x3* = 13/16, x4*= 147/48; with z* = 813/48.

2
 The inside story of the Simplex Method

Illustration: Maximize , Z = 3x1 + 5 x2

s.t. x1 <= 4
2x2 <= 12
3x1 + 2 x2 <= 18
With x1, x2 >= 0

Ans: Initial Simplex Tableau

x1 x2 x3 x4 x5
3 5 0 0 0
x3 0 1 0 1 0 0 4
x4 0 0 2 0 1 0 12
x5 0 3 2 0 0 1 18
(zj – cj) -3 -5 0 0 0 0

Final Simplex Tableau

x1 x2 x3 x4 x5
3 5 0 0 0
x3 0 0 0 1 1/3 -1/3 2
x4 0 0 1 0 ½ 0 6
x5 0 1 0 0 -1/3 1/3 2
(zj – cj) -3 -5 0 0 0 0

Matrix A*
Matrix S*

The Initial tableau is: [ A : I : b ]

The Final tableau is: [ A* : S* : b* ]

Where, A* = S* A

b* = S* b

This can be explained as:

b* = S* b =
1 1/3 -1/3 4 = 2
0 ½ 0 12 6
0 -1/3 1/3 18 2

A* = S* A =
1 1/3 -1/3 1 0 = 0 0
0 ½ 0 0 2 0 1
0 -1/3 1/3 3 2 1 0

3
 Sensitivity Analysis

This analysis explains how sensitive the optimal solution is to changes in the coefficients of the LP model. This process
is also known as the post-optimality analysis. In general, we are interested in finding the effects of the following
changes on the optimal LP solution: -

1) Changes in profit or cost/unit (coefficients) of the objective function.

2) Changes in the availability of resources or capacities of production/service centers or limits on demands
(requirements vector or RHS of constraints B).
3) Changes in resource requirements/units of products or activities (technological coefficients of variables) in
constraints.
4) Addition of a new product or activity (variable).
5) Addition of a new constraint.
Optimize, Z = CT X
Subject to, A X=B
With, X >= 0
Where, X is the column vector of unknowns; CT is the row vector of the corresponding costs (cost vector); A is the
coefficient matrix of the constraints (matrix of technological coefficients and B is the column vector of the right hand
sides of the constraints (requirement vector).

Illustration of Sensitivity Analysis:

Solve the following LPP:-
Maximize, z = 20 x1 + 10 x2,
Subject to, x1 + 2 x2 <= 40
3 x1 + 2 x2 <= 60
With, x1 and x2 nonnegative.

A. Modification of the Cost Vector CT: Using the final simplex tableau of the above LPP, again solve the followings
LPPs:-
(i) Coefficient of non-basic variables: The objective function is Maximize, z = 20 x 1 + 15 x2, and the constraints
remain the same as above,
(ii) Coefficient of basic variables: The objective function is Maximize, z = 10 x 1 + 10 x2, and the constraints
remain the same as above.
Ans to A.:
Initial Simplex tableau of the original LPP is:-
x1 x2 x3 x4
20 10 0 0
x3 0 1 2 1 0 40
x1 0 3 2 0 1 20
(zj – cj) -20 -10 0 0 0

Final Simplex tableau of the original LPP is:-

x1 x2 x3 x4
20 10 0 0
x3 0 0 4/3 1 -1/3 20
x1 20 1 2/3 0 1/3 20
(zj – cj) 0 10/3 0 20/3 400

(i) Modification of the cost vector CT: Here the coefficient of the nonbasic variable has changed. The new value
of the cost coefficient corresponding to the nonbasic variable x 2 is 15 instead of 10. The corresponding Simplex
tableau is
x1 x2 x3 x4
20 15 0 0
x3 0 0 4/3* 1 -1/3 20
x1 20 1 2/3 0 1/3 20
(zj – cj) 0 ( - 5/3) 0 20/3 400

4
 Since (z2– c2) < 0, the new solution is not optimal. The regular Simplex method is used to re-optimize the problem,
starting with x2 as the entering variable.
x1 x2 x3 x4
20 15 0 0
x2 15 0 1 3/4 -1/4 15
x1 20 1 0 -1/2 1/2 10
(zj – cj) 0 0 5/4 25/4 425
The optimal solution is x1 *= 10 ; x2 * = 15 , with Z* = 425.

(ii) Modification of the cost vector C T: Here the coefficient of the basic variable has changed. The new value of
the cost coefficient corresponding to the basic variable x 1 is 10 instead of 20. The corresponding Simplex tableau
becomes
x1 x2 x3 x4
10 10 0 0
x3 0 0 4/3* 1 -1/3 20
x1 10 1 2/3 0 1/3 20
(zj – cj) 0 ( - 10/3) 0 10/3 200

Since (z2– c2) < 0, the new solution is not optimal. The regular Simplex method is used to re-optimize the problem,
first by entering x2.
x1 x2 x3 x4
10 10 0 0
x2 10 0 1 3/4 -1/4 15
x1 10 1 0 -1/2 1/2 10
(zj – cj) 0 0 5/2 5/2 250

The optimal solution is x1 *= 10; x2 * = 15, with Z* = 250.

B. Modification of the requirements vector B: Using the final simplex tableau of the above LPP, again solve the
followings LPPs:-
The RHs of the second constraint be changed from 60 to 130 i.e. the LPP becomes: -
Maximize, z = 20 x1 + 10 x2,
Subject to, x1 + 2 x2 <= 40
3 x1 + 2 x2 <= 130
With, x1 and x2 nonnegative.

Final Simplex tableau of the original LPP is:-

x1 x2 x3 x4
20 10 0 0
x3 0 0 4/3 1 -1/3 20
x1 20 1 2/3 0 1/3 20
(zj – cj) 0 10/3 0 20/3 400

Here S* means the

x3 x4
1 -1/3
0 1/3

The new basic variable vector will be XB = S* B

x3 = 1 -1/3 40 = - 10/3
x1 0 1/3 130 130/3

This means that the new Simplex Tableau will be: -

5
 New Simplex tableau
x1 x2 x3 x4
20 10 0 0
x3 0 0 4/3 1 -1/3 -10/3
x1 20 1 2/3 0 1/3 130/3
(zj – cj) 0 10/3 0 20/3 866-2/3

Since x3 value in the new Simplex tableau is negative, the new solution is not feasible. The dual simplex method is
used to clear the infeasibility staring with the new simplex tableau.

New Simplex tableau

x1 x2 x3 x4
20 10 0 0
x3 0 0 4/3 1 -1/3* -10/3
x1 20 1 2/3 0 1/3 130/3
(zj – cj) 0 10/3 0 20/3 866-2/3
x3 work row coefficients 0 4/3 1 -1/3
Absolute ratios by negative
20
coefficients only

The next Simplex tableau is

Next and Final tableau
x1 x2 x3 x4
20 10 0 0
x4 0 0 -4 -3 1 10
x1 20 1 2 1 0 40
(zj – cj) 0 30 20 0 800

The optimal and feasible solution is x1 *= 40; x2 * = 0, with Z* = 800.

C. Modification of the matrix of coefficient A: The problem becomes more complicated, when the technological
coefficients of the basic variables are considered. This is because the matrix under the starting solution changes.
In this case it is easier to solve the new problem than resort the sensitivity analysis approach.

Therefore our analysis is limited to the case of the coefficients of nonbasic variables only.

Let the technological coefficients of x be changed from (2 , 2 )T to (2 , 1 )T .

Maximize, z = 20 x1 + 10 x2,
Subject to, x1 + 2 x2 <= 40
3 x1 + 1 x2 <= 60
With, x1 and x2 nonnegative.

Then the new technological coefficients of x2 in the optimal Simplex Tableau of the original problem are given by,

XB = S* P2
= 1 -1/3 2 = 5/3
0 1/3 1 1/3

This means that the new Simplex Tableau will be: -

x1 x2 x3 x4
20 10 0 0
x3 0 0 5/3 1 -1/3 20
x1 20 1 1/3 0 1/3 20
(zj – cj) 0 -10/3 0 20/3 400

6
 Since here (z2 – c2) < 0, the new solution is not optimal. Again regular Simplex method is resorted to first by
entering x2. The new tableau is: -
Final tableau
x1 x2 x3 x4
20 10 0 0
x2 10 0 1 3/5 -1/5 12
x1 20 1 0 -1/5 2/5 16
(zj – cj) 0 0 2 6 440

The optimal and feasible solution is x1 *= 16; x2 * = 12, with Z* = 440.

D. Addition of a variable: Let a new variable xk be added to the original problem. This is accompanied by the
addition to “A” of the column Pk = (3 , 1)T and to CT of a component ck = 30. Thus the new problem becomes:-

Maximize, z = 20 x1 + 10 x2 + 30 xk,
Subject to, x1 + 2 x2 + 3 xk <= 40
3 x1 + 2 x2 + 1 xk <= 60
With, x1, x2 and xk nonnegative

Then the technological coefficients of xk in the optimal Simplex Tableau are given by,

xk = S* Pk
= 1 -1/3 3 = 8/3
0 1/3 1 1/3

This means that the new Simplex Tableau will be: -

x1 x2 xk x3 x4
20 10 30 0 0
x3 0 0 5/3 8/3 1 -1/3 20
x1 20 1 1/3 1/3 0 1/3 20
(zj – cj) 0 10/3 -70/3 0 20/3 400

Since here (zk – ck) < 0, the new solution is not optimal. Again regular Simplex method is resorted to first by
entering xk. The new tableau is: -
Final Simplex Tableau
x1 x2 xk x3 x4
20 10 30 0 0
Xk 30 0 1/2 1 3/8 -1/8 15/2
x1 20 1 1/2 0 -1/8 3/8 35/2
(zj – cj) 0 15 0 70/8 30/8 575

The optimal and feasible solution is x1 *= 17.5; x2 * = 0, with Z* = 575.

E. Addition of a constraint: If a new constraint added to the system is not active i.e. it is satisfied by the optimal
values of the variables of the original LPP, it is called a secondary or redundant constraint and the optimality of the
problem remains unchanged.

On the other hand, if the new constraint is active i.e. it is not satisfied by the optimal values of the variables of the
original LPP, the current optimal solution becomes infeasible. Let us consider the case of the addition of an
active constraint e.g., 2 x1 + 3 x2 >= 50 to the original problem. Thus the new problem becomes:-

Maximize, z = 20 x1 + 10 x2
Subject to, x1 + 2 x2 <= 40
3 x1 + 2 x2 <= 60
2 x1 + 3 x2 >= 50
With, x1, x2 nonnegative

7
The current optimal solution (x1*= 20, x2*= 0) does not satisfy the above new constraint, 2 x1 + 3 x2 >= 50, as
[2 x 20 + 3 x 0] = 40, i.e. the new constraint is active and hence the current optimal solution becomes infeasible.
Therefore, we add the new constraint to the current optimal tableau. The new slack variable is x5.

We first transform the new constraint in <= form, so that we can resort to Dual Simplex Method.

-2 x1 - 3 x2 <= -50
The new Simplex tableau is: -
New simplex tableau
x1 x2 x3 x4 x5
20 10 0 0 0
x3 0 0 4/3 1 -1/3 0 20
x1 20 1 2/3 0 1/3 0 20
x5 0 -2 -3 0 0 1 -50
(zj – cj) 0 10/3 0 20/3 0 400

By using row operations, the coefficient of x1 in the new constraint is made zero. The modified tableau becomes: -

Modified Simplex Tableau

x1 x2 x3 x4 x5
20 10 0 0 0
x3 0 0 4/3 1 -1/3 0 20
x1 20 1 2/3 0 1/3 0 20
x5 0 0 -5/3 0 2/3 1 -10
(zj – cj) 0 10/3 0 20/3 0 400

The Dual Simplex method is used to overcome the infeasibility by departing the variable x 5. The new Simplex
Tableau is: -
New & final Simplex Tableau
x1 x2 x3 x4 x5
20 10 0 0 0
x3 0 0 0 1 1/5 4/5 12
x1 20 1 0 0 3/5 2/5 16
x2 10 0 1 0 -2/5 -3/5 6
(zj – cj) 0 0 0 8 2 380

The above tableau gives the optimal and feasible solution as x1 *= 16; x2 * = 6, with Z* = 380.