Duality Theory

Duality in Linear Programming

Every Linear programming problem (called the primal) is
associated with another linear programming problem (called
the dual). Either of the problem can be considered as
personal with the other one its dual.
Properties of Duality
The Primal and Dual problems in L.P. bear distinct
properties relating one another. These properties are cited
in terms of relationship between the primal and dual
problems. They are:
(1)

The Dual of the Dual is Primal.

2

(2)The number of variables in one is equal to the number of

constraints in the other and vice versa.

(3) If, for one of the problems, a finite optimal solution exists,
then the other also has a finite optimal solution and optional
values of the two objective functions are equal.
(4) If one of the problems has no bounded solution, then the
other has no feasible solution.
(5) If one of the problems has no feasible solution, then the
other may have an unbounded solution or may have no
feasible solution.
(6) If there is an equality in the Primal, the Dual variable
corresponding to the Equality, constraint will be unrestricted
in sign and vice versa.
3

(7) The real variable of the primal correspond to the

slacks/surpluses of the Dual and vice versa.
To Convert Primal to Dual
Given the primal problem in standard form, the forming steps
are followed to obtain its dual.
1. The maximize problem in primal is minimize problem in its
equivalent dual and vice versa.
2. The R.H.S. value of the constraints of the primal are
objective functions co-efficient in its dual.
3. Each constraints in primal will correspond to a decision
variable of the dual problem.
4

4. The constraints matrix of the dual is the matrix transpose of

the constrain matrix of primal.
5. All decision variable must be non-negative.
6. The objective function co-efficient of the primal will be the
R.H.S. values of the constraints of the dual problem.
7. For a Max. problem of primal, the constraints are
converted to constraint and for Min. problem of the
primal the constraint are converted to constraint in
dual.
8. If there is an equality in the primal, the dual variable
corresponding to the equality constraint will be unrestricted
in sign and vice versa.
5

To Correct the given Problem to Standard Form

1 If the objective function must be in maximization form, the
constraints must be in the form type and if the objective
function must be in minimization form, the constraint must be
in the form .
2. All decision variables must be non-negative.
Example: To convert constraints of to type.
Example: 911 X1 + 912 X2 b1
Multiply by ve sing to to type.
911 X2 912 X2 b1
6

Example: To convert constraints of type to type

Example: 911 X1 + 912 X2 b1
Multiply by 1 on both sides we get
911 X1 912 X2 b1
Non-Negative Variable: An variable X is said to be
unrestricted in sign, when it can take +ve or ve negative
value. It is represent as X = X+ X.
Write the dual for the following
(1) Max.
Z = X1 X2 + 3 X3
Subject to: X1 + X2 + X3 10
2X1
X3 2
2X12X2 + 3X3 0
X1, X2, X3 0

W1
W2
W3
7

Let W1, W2 and W3 be the variable of the dual corresponding

to each constraint.
Dual objective Min. W0 = 10 W1 + 2 W2 + 0 W3
Subject to:

W1 + 2 W2 + 2 W3 1
W1

2 W3 1

W1 W2

+3 W3 3

W1, W2, W3 0

(2)Max.

Z = 2 X1 +3 X2 4 X3

Subject to: X1 + 2X2 + X3 6

2X + X2 + X3 4
4X2 + 2 X3
8
X1, X2, X3 0

LetW1, W2 and W3 be the variable of the dual corresponding

tp each constraint.
Converting the given problem to standard form
Min.
Z = 2 X1 +3 X2 4 X3
Subject to: X1 + 2X2 + X3 6
2X X2 X3 4
4X2 + 2 X3
8

W1
W2
W3

Dual
Obj. Max.
Subject to:

Y = 6 W1 4 W2 + 8 W3
W1

2 W2 2

2 W1 W2 + 4 W3 3
W1 W2 + 2 W3 4
W1, W2, W3 0

10

(3)Max.

Z = 2 X1 +4 X2

Subject to: 3X1 + 2 X2 14

4 X1 + 12 X2 21
X1
=6
Dual
Min.
Subject to:

W0 = 14 Y1 + 21 Y2 + 6 Y3
3 Y1 + 4Y2 + Y3 2
2 Y1 + 12Y2
Y1, Y2 0

Y3 is unrestricted in sign.
11

(6) Max.
Z = 5 X1 + 6 X2
Subject to: X1 + 2 X2 = 5
X1 + 5 X2 3
4 X1 + 7 X2 8
X2 0, X1 is unrestricted.
Solution: Let
Max.
Z = 5 X1 + 6 X2
Subject to: X1 + 2 X2 = 5 Y1
+X1 5 X2 -3 Y2
4 X1 + 7 X2 8 Y3
Dual
Minimize.
Y0 = 5 Y1 3 Y2 + 8 Y3
Subject to:
Y1 + Y2 + 4 Y3 =5
2 Y1 5 Y2 +7 Y3 6
Y1. is unrestricted
Y2, Y3 0

12

One of the key uses of duality theorem is the interpretation and

implementation of sensitivity analysis
Post optimality Analysis:
Large linear programming problems usually will have hundreds
of decision variables thousands of functional constraints.
Many variations of the basic model may be of interest for
considering different scenarios.
After finding the optimal solution for one version of a linear
programming model, it is required to find solution for a slightly
different model (often many times).
One approach is to re-apply the simplex method from scratch
for each new version of the model.
Re-optimize by deducing changes in the model.
Revised tableau and optimal solutions for the prior model are13

then used as the initial tableau and the initial basic solution for
solving new model.
If this solution is feasible for the new model, then the simplex
method is applied in the usual way starting from the initial
basic feasible solution.
If the solution is not feasible, duel simplex method can be
applied for this initial basic solution.

Advantage:
1. Re-optimization technique will be closer to the optimal
solution than Initial basic feasible solution.
2. The number of iteration will reduce or may require only one
optimality test and no interation.
14

Example: Consider a problem where there are three plants

and two products need to be manufactured. Product 1
requires some of the production capacity in plant 1 and 3, but
not plant 2. Product 2 needs only plants 2 and 3. OR team
came out with the result that the production time available per
week in hours form plants 1, 2 and 3 are 4, 12 and 18 hrs
respectively. The profit per batch from the two products are
$3000 and $5000 respectively. Formulate the mathematical
model and find the optimal batch for each product and the
optimal profit.
Production time per batch, hrs
1

Production Time
available per
week, Hrs

12

18

Profit

3000

5000

Plant

Product

15

Max Z = 3X1 + 5X2

X1 4

Subject to:

2X2 12
3X1 +2X2 18
X1, X2 0

Basic
Z

Equ
(0)

Z
1

X1
-3

X2
-5

S1
0

S2
0

S1
S2
S3

(1)
(2)
(3)

0
0
0

1
0
3

0
2
2

1
0
0

0
1
0

S3 RHS
0
0

0
0
1

4
12
18
16

Basic

Equ

X1

X2

S1

S2

S3 RHS

(0)

-3

5/2

30

S1

(1)

X2

(2)

1/2

S3

(3)

-1

Basic

Equ

X1

X2

S1

S2

S3

RHS

(0)

3/2

36

S1

(1)

1/3 -1/3

X2

(2)

X1

(3)

1/3

1/3

2
17

X1* = 2, X2* = 6 Z* = 36
Graphical Method: Considering the three constraints we
have:
X1 4 -------------------------(1)
2X2 12 ---------------------(2)
3X1 +2X2 18--------------(3)
X1 = 4;
X2 = 6;
When X1 = 0 , X2 = 9
When X2 = 0, X1 = 6

The points are (0,9) and (5,0)

18

Objective:3X1+5X2 = 15

X1=0, X2 = 3

10

X1=0, X1 = 5
8

Optimal Point
2 X2

2
4

0,0

X1
1

19

The shadow Prices for the three resources are

Y*1 = 0 Shadow price for resource 1
Y2* = 3/2 shadow price for resource 2
Y3* = 1 Shadow price for resource 3
With just two decision variables, these numbers can be
verified by checking graphically that individually increasing
any RHS (bi) by 1 would increase the optimal value Z by Yi*
Example:
Let us consider changing the resource 2 by a value 1.
i.e. 2X2 12 is changed to 2X2 13

20

3
New Optimal Point (5/3, 13/2)

10

Old Optimal Point (2,6)

2 X2

2
4

0,0

X1
1

21

Original Equation and Resource

Changed Equation Resource

2X2 12

2X2 13

Graphical solution

2X2 = 12

2X2 = 13

X2 =6

X2 = 6.5

The optimal point from the

The optimal point is

Previous graph is (2,6)

shifted to X1=5/3 and

The optimal solution Z* = 36

X2=13/2
Z * = 37.5

The change is optimal value is 37.5-36 = 1.5.

22

If the profit from each batch is $ 1500 per week, then it has
bee seen that whether the management need to consider
increasing the working resource time from 12 to 13 hrs?
Hence we have proved that there is an increase in3/2 as
seen in the table Y2* co-efficient.

23

Dual Simplex Algorithm.

1. Obtain the canonical form of the constraint by introducing
slacks; if constraints exists, multiply the constraints by 1
and introduce slacks. (dont introduce surplus variables into
the constraint system).
2. Transmit the problem data into the standard dual simplex
tableau.
3. First the leaving basic variable is found by selecting the
most ve value of the R.H.S. of the (solution column) basic
variable. If all the basic variables are non-negative, the
process ends and the feasible solution is reached. Else
step 4.
4. With respect to deporting row, computer rations

24

Z
1=

Provided aij is negative (Break tie if

the denominator are zero or
positive; the problem has
a1j
no feasible solution.

5. Choose the variable with minimum 1 to enter basis. The

entering variable is the one with the smallest ratio if the
problem is minimization or smallest ratio of the problem is
minimization.
6. Identify Pivot as the element at the intersection of the
departing row and entering column.
7. Iterate and proceed to next iteration as follows:
(a) New basic is introduced in the place of the old
basic (which is the leaving variable).
(b) New pivot row = old pivot row pivot element.

25

(c) All other rows will be

old
New element =

Pivot New
elements

Pivot
Column
Element

Row
element

8. Go to step 3
Examples: Solve by dual simplex method
Minimize:.

Z = 2 X1 + X2

Subject to:

3 X1 + X2 3
4 X1 +3 X2 6
X1 +2 X2 3

X1, X2 0
26

Min.
Z = 2 X1 + X2
Subject to: 3 X1 X2 + S1 = 3
4 X1 3 X2 + S2 = 6
+ 4 X1 +2 X2 + S3 =+8
X1, X2, X3 0
Basic

X1

X2

S1

S2

S3

Solution

S1

S2

S3

+1

+2

+3

Ratio

--

--

--

-27

S2 is leaving variable with most ve value.

Note: Since this a minimization problem and all the objective
equation co-efficient are 0, the starting basic solution S1 =
3, S2 = 6, S3 = 3 is optional but infeasible. This problem is
typical of the type that can be handled by the dual simplex
method.
Entering variable will be X2 because it has the minimum ratio.
Iteration 1
Basic

Z
S1
X2
S3
Ratio

X1

X2

S1

S2

S3

Solution

2/3
3
4
+1
2/5

0
0
1
0

0
1
0
0
--

0
0
0
1
--

--

1
2

4
--

28

2/5

1/5

12/5

X1

3/5

+1/5

3/5

X2

4/5

3/5

6/5

S3

The above table is optional and feasible.

X1 = 3/5 X2 = 6/5 Z = 12/5

29

(2) Max.
Z = 2 X1 X3
Subject to:
X1 + X2 X3 5
X1 2 X2 + 4 X3 8
X1, X2, X3 0

30

X2

5/2

X3

3/2

X2 = 1

X3 = 9

X=9

31