Duality Problem

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

C h a p t e r 5

Duality in Linear Programming


“A boat can’t have two captains.”
– Akira Mori

PREVIEW
The chapter deals with how to find the marginal value (also known as shadow price) of each resource.
This value reflects an additional cost to be paid to obtain one additional unit of the resource to get the
optimal value of objective function under resource constraints.

LEARNING OBJECTIVES

After studying this chapter, you should be able to


z appreciate the significance of the duality concept.
z formulate the dual LP problem and understand the relationship between primal and dual LP problems.
z understand the concept of shadow prices.

CHAPTER OUTLINE
5.1 Introduction • Conceptual Questions
5.2 Formulation of Dual Linear Programming • Self Practice Problems B
Problem • Hints and Answers
• Self Practice Problems A ‰ Chapter Summary
• Hints and Answers ‰ Chapter Concepts Quiz
5.3 Standard Results on Duality ‰ Case Study
5.4 Managerial Significance of Duality ‰ Appendix: Theorems of Duality
5.5 Advantages of Duality
146 Operations Research: Theory and Applications

5.1 INTRODUCTION
The term ‘dual’, in general, implies two or double. The concept of duality is very useful in mathematics,
physics, statistics, engineering and managerial decision-making. For example, in a two-person game theory,
one competitor’s problem is the dual of the opponent’s problem.
In linear programming, duality implies that each linear programming problem can be analyzed in two
different ways but would have equivalent solutions. Any LP problem (either maximization and minimization)
can be stated in another equivalent form based on the same data. The new LP problem is called dual linear
programming problem or in short dual. In general, it is immaterial which of the two problems is called primal
or dual, since the dual of the dual is primal.
Dual LP problem
For example, consider the problem of production planning. The production manager attempts to
provides useful determine quantities for each product to be produced with an objective to optimize the use of available
economic resources so that profit is maximum. But through a dual LP problem approach, he may develop a production
information about plan that optimizes resource utilization so that marginal opportunity cost of each unit of a resource is equal
worth of resources
to be used.
to its marginal return (also called shadow price). The shadow price indicates an additional price to be
paid to obtain one additional unit of the resources in order to maximize profit under the resource constraints.
If a resource is not completely used, i.e. there is slack, then its marginal return is zero.
The shadow price is also defined as the rate of change in the optimal objective function value with
respect to the unit change in the availability of a resource. To be more precise for any constraint, we have

Change in optimal objective function value


Shadow price =
Unit change in the availability of resource
The interpretation of rate of change (increase or decrease) in the value of objective function depends on
whether LP problem is of maximization or minimization type. The shadow price for a less than or equal to (≤) type
constraint will always be positive. This is because increasing the right-hand side resource value cannot
decrease the value of objective function. Similarly, the shadow price for a greater than or equal to (≥) type
constraint will always be negative because increasing the right-hand side resonance value cannot increase the
value of the objective function.
There is no need to solve both LP problems separately. Solving one LP problem is equivalent to solving
the other simultaneously. Thus, if the optimal solution to one is known, the optimal solution of the other
can also be read from the cj – zj row. In some cases, considerable computing time can be reduced by solving
the dual LP problem.

5.2 FORMULATION OF DUAL LINEAR PROGRAMMING PROBLEM

There are two important forms of primal and dual LP problems, namely the symmetrical (canonical) form
and the standard form.

Shadow price 5.2.1 Symmetrical Form


represents increase
in the objective
function value due
Suppose the primal LP problem is given in the form
to one-unit increase
in the right hand
Maximize Zx = c1 x1 + c2 x2 + . . . + cn xn
side (resource) of subject to the constraints
any constraint.
a11 x1 + a12 x2 + . . . + a1n xn ≤ b1
a21 x1 + a22 x2 + . . . + a2n xn ≤ b2
.. ..
. .
am1 x1 + am2 x2 + . . . + amn xn ≤ bm
and x1, x2 . . ., xn ≥0
Duality in Linear Programming 147

Then the corresponding dual LP problem is written as:


Minimize Zy = b1 y1 + b2 y2 + . . . + bm ym
subject to the constraints
a11 y1 + a21 y2 + . . . + am1 ym ≤ c1
a12 y1 + a22 y2 + . . . + am2 ym ≤ c2
. .
. .
. .
a1n y1 + a2n y2 + . . . + amn ym ≤ cn
and y1, y2, . . ., ym ≥ 0

In general, the primal-dual relationship between a pair of LP problems can be expressed as follows:
Dual variables
represent the
Primal Dual potential value of
n m resources
Max Z x = ∑ cj xj Min Z y = ∑ bi y i
j =1 i =1

subject to the constraints subject to the constraints

n m
∑ aij xj ≤ bi ; i = 1, 2 , . . ., m
j =1
∑ aji yi ≥ cj ; j = 1, 2, . . ., n
i =1
aij = aji
and xj ≥ 0; j = 1, 2, . . ., n and yi ≥ 0; i = 1, 2, . . ., m

A summary of the general relationships between primal and dual LP problems is given in Table 5.1.

If Primal Then Dual

(i) Objective is to maximize (i) Objective is to minimize


(ii) jth primal variable, xj (ii) jth dual constraint
(iii) ith primal constraint (iii) ith dual variable, yi
(iv) Primal variable xj unrestricted in sign (iv) Dual constraint j is = type Table 5.1
(v) Primal constraint i is = type (v) Dual variable yi is unrestricted in sign Primal-Dual
(vi) Primal constraints ≤ type (vi) Dual constraints ≥ type Relationship

5.2.2 Economic Interpretation of Dual Variables


In the maximization primal LP model, we may define each parameter as follows:
Z = return cj = profit (or return) per unit of variable (activity) xj
xj = number units of variable j aij = units of resource, i consumed (required)
bi = maximum units of resource, i available per unit of variable j
The new variables introduced in the dual problem are Zy and yi (dual variables). When both the primal
and the dual solutions are optimal, the value of objective function satisfies the strict equality, i.e. Zx = Zy.
The interpretation associated with the dual variables yi (i = 1, 2, . . ., m) is discussed below. Rewrite the
primal LP problem as follows:

Primal LP Problem
n n
Maximize (return) Zx = ∑ cj xj = Σ (Profit per unit of variable xj) (Units of variable xj)
j =1 j =1
subject to the constraints
n
∑ a ij xj ≤ bi
j =1
n
or ∑ (Units of resource i, consumed per unit of variable, xj) (Units of variable xj)
j =1
≤ Units of resource, i available
and xj ≥ 0, for all j i = 1, 2, ..., m
148 Operations Research: Theory and Applications

Dual LP Problem
m m
Minimize (cost) Zy = ∑ bi y i = Σ (Units of resource, i) (Cost per unit of resource, i )
i =1 i =1

subject to the constraints


m
∑ aji y i ≥ cj
i =1
m
or Σ (Units of a resource, i consumed per unit of variable yi ) (Cost per unit of resource, i)
i =1 ≥ Profit per unit for each variable, xj
and yi ≥ 0, for all i ..., n
From these expressions of parameters of both primal and dual problems, it is clear that for the unit of
measurement to be consistent, the dual variable (yi) must be expressed in terms of return (or worth) per
unit of resource i. This is called dual price (simplex multiplier or shadow price) of resource i. In other
words, optimal value of a dual variable associated with a particular primal constraint indicates the marginal
change (increase, if positive or decrease, if negative) in the optimal value of the primal objective function.
For example, if y2 = 5, then this indicates that for every additional unit (up to a certain limit) of resource 2
(resource associated with constraint 2 in the primal), the objective function value will increase by 5 units.
The value y2 = 5 is also called the marginal (or shadow or implicit) price of resource 2.
Similarly, for feasible solutions of both primal and dual LP problems, objective function value satisfy
the inequality Zx ≤ Zy. This inequality is interpreted as: Profit ≤ Worth of resources. Thus, so long as the
total profit (return) from all activities is less than the worth of the resources, the feasible solution of both
primal and dual is not optimal. The optimality (maximum profit or return) is reached only when the resources
have been completely utilized. This is only possible if the worth of the resources (i.e. input) is equal to
profit (i.e. output).

5.2.3 Economic Interpretation of Dual Constraints


As stated earlier, the dual constraints are expressed as:
m
∑ a ji yi − c j ≥ 0
i =1
Since coefficients aji represents the amount of resource bi consumed by per unit of activity xj, and the dual
variable yi represents shadow price per unit of resource bi, the quantity Σ aji yi (= zj) should be total shadow
price of all resources required to produce one unit of activity xj.
For maximization LP problem, if cj – zj > 0 value corresponds to any non-basic variable, then the value of
objective function can be increased. This implies that the value of variable, xj can be increased from zero to a
positive level provided its unit profit (cj) is more than its shadow price, i.e.
m
cj ≥ ∑ a ji yi
i =1
Profit per unit of activity xj ≥ Shadow price of resources used per unit of activity, xj.

5.2.4 Rules for Constructing the Dual from Primal


The rules for constructing the dual from the primal and vice-versa using the symmetrical form of LP problem are:
1. A dual variable is defined corresponds to each constraint in the primal LP problem and vice versa. Thus,
for a primal LP problem with m constraints and n variables, there exists a dual LP problem with m
variables and n constraints and vice-versa.
2. The right-hand side constants b1, b2, . . ., bm of the primal LP problem becomes the coefficients of the
dual variables y1, y2, . . ., ym in the dual objective function Zy. Also the coefficients c1, c2, . . ., cn of
the primal variables x1, x2, . . ., xn in the objective function become the right-hand side constants in
the dual LP problem.
Duality in Linear Programming 149

3. For a maximization primal LP problem with all ≤ (less than or equal to) type constraints, there exists
a minimization dual LP problem with all ≥ (greater than or equal to) type constraints and vice versa.
Thus, the inequality sign is reversed in all the constraints except the non-negativity conditions.
4. The matrix of the coefficients of variables in the constraints of dual is the transpose of the matrix of
coefficients of variables in the constraints of primal and vice versa.
5. If the objective function of a primal LP problem is to be maximized, the objective function of the dual
is to be minimized and vice versa.
6. If the ith primal constraint is = (equality) type, then the ith dual variables is unrestricted in sign and
vice versa.
The primal-dual relationships may also be memorized by using the following table:

Dual Variables Primal Variables Maximize Zx


x1 x2 . . . xj . . . xn
y1 a 11 a 12 . . . a 1j . . . a 1n ≤ b1
y2 a 21 a 22 . . . a 2j . . . a 2n ≤ b2
. . .
.. .. ..
ym a m1 a m2 . . . a mj . . . a mn ≤ bm

Minimize Zy ≥ c1 ≥ c2 . . . ≥ cj . . . ≥ cn ↑ Dual objective function coefficients

↑ jth dual constraint


The primal constraints should be read across the rows, and the dual constraints should be read across the
columns.
Example 5.1 Write the dual to the following LP problem.
Maximize Z = x1 – x2 + 3x3
subject to the constraints
(i) x1 + x2 + x3 ≤ 10, (ii) 2x1 – x2 – x3 ≤ 2, (iii) 2x1 – 2x2 – 3x3 ≤ 6
and x1, x2, x3 ≥ 0

Solution In the given LP problem there are m = 3 constraints and n = 3 variables. Thus, there must be
m = 3 dual variables and n = 3 constraints. Further, the coefficients of the primal variables, c1 = 1,
c2 = –1, c3 = 3 become right-hand side constants of the dual. The right-hand side constants b1 = 10,
b2 = 2, b3 = 6 become the coefficients in the dual objective function. Finally, the dual must have a minimizing
objective function with all ≥ type constraints. If y1, y2 and y3 are dual variables corresponding to three primal
constraints in the given order, the resultant dual is
Minimize Zy = 10y1 + 2y2 + 6y3
subject to the constraints
(i) y1 + 2y2 + 2y3 ≥ 1, (iii) y1 – y2 – 2y3 ≥ –1, (iii) y1 – y2 – 3y3 ≥ 3
and y1, y2, y3 ≥ 0

Example 5.2 Write the dual of the following LP problem.


Minimize Zx = 3x1 – 2x2 + 4x3
subject to the constraints
(i) 3x1 + 5x2 + 4x3 ≥ 7, (ii) 6x1 + x2 + 3x3 ≥ 4, (iii) 7x1 – 2x2 – x3 ≤ 10
(iv) x1 – 2x2 + 5x3 ≥ 3, (v) 4x1 + 7x2 – 2x3 ≥ 2
and x1, x2, x3 ≥ 0

Solution Since the objective function of the given LP problem is of minimization, the direction of each
inequality has to be changed to ≥ type by multiplying both sides by – 1. The standard primal LP problem
so obtained is:
150 Operations Research: Theory and Applications

Minimize Zx = 3x1 – 2x2 + 4x3


subject to the constraints
(i) 3x1 + 5x2 + 4x3 ≥ 7, (ii) 6x1 + x2 + 3x3 ≥ 4, (iii) –7x1 + 2x2 + x3 ≥ – 10
(iv) x1 – 2x2 + 5x3 ≥ 3, (v) 4x1 + 7x2 – 2x3 ≥ 2
and x1, x2, x3 ≥ 0
If y1, y2, y3, y4 and y5 are dual variables corresponding to the five primal constraints in the given order,
the dual of this primal LP problem is stated as:
Maximize Zy = 7y1 + 4y2 – 10y3 + 3y4 + 2y5
subject to the constraints
(i) 3y1 + 6y2 – 7y3 + y4 + 4y5 ≤ 3, (ii) 5y1 + y2 + 2y3 – 2y4 + 7y5 ≤ – 2
(iii) 4y1 + 3y2 + y3 + 5y4 – 2y5 ≤ 4
and y1, y2, y3, y4, y5 ≥ 0

Example 5.3 Obtain the dual LP problem of the following primal LP problem:
Minimize Z = x1 + 2x2
subject to the constraints
(i) 2x1 + 4x2 ≤ 160, (ii) x1 – x2 = 30, (iii) x1 ≥ 10
and x1, x2 ≥ 0

Solution Since the objective function of the primal LP problem is of minimization, change all ≤ type
constraints to ≥ type constraints by multiplying the constraint on both sides by –1. Also write = type
constraint equivalent to two constraints of the type ≥ and ≤. Then the given primal LP problem can be
rewritten as:
Minimize Zx = x1 + 2x2
subject to the constraint
(i) – 2x1 – 4x2 ≥ –160, (ii) x1 – x2 ≥ 30
(iii) x1 – x2 ≤ 30 or – x1 + x2 ≥ – 30, (iv) x1 ≥ 10
and x1, x2 ≥ 0
Let y1, y2, y3 and y4 be the dual variables corresponding to the four constraints in the given order.
The dual of the given primal LP problem can then be formulated as follows:
Maximize Zy = – 160y1 + 30y2 – 30y3 + 10y4
subject to the constraints
(i) – 2y1 + y2 – y3 + y4 ≤ 1, (ii) – 4y1 – y2 + y3 ≤ 2
and y1, y2, y3, y4 ≥ 0

Let y = y2 – y3 (y2, y3 ≥ 0). The above dual problem then reduces to the form
Maximize Zy = – 160y1 + 30y + 10y4
subject to the constraints
(i) – 2y1 + y + y4 ≤ 1, (ii) – 4y1 – y ≤ 2
and y1, y4 ≥ 0; y unrestricted in sign

Remark Since second constraint in the primal LP problem is equality, therefore as per rule 6 corresponding
second dual variable y (= y2 – y3) should be unrestricted in sign.
Example 5.4 Obtain the dual LP problem of the following primal LP problem:
Minimize Zx = x1 – 3x2 – 2x3
Duality in Linear Programming 151

subject to the constraints


(i) 3x1 – x2 + 2x3 ≤ 7, (ii) 2x1 – 4x2 ≥ 12, (iii) – 4x1 + 3x2 + 8x3 = 10
and x1, x2≥ 0; x3 unrestricted in sign.

Solution Let y1, y2 and y3 be the dual variables corresponding to three primal constraints in the given
order. As the given LP problem is of minimization, all constraints can be converted to ≥ type by multiplying
both sides by –1, i.e., –3x1 + x2 – 2x3 ≥ –7. Since the third constraint of the primal is an equation, the third
dual variable y3 will be unrestricted in sign. The dual of the given LP primal can be formulated as follows:

Maximize Zy = –7y1 + 12y2 + 10y3


subject to the constraints
(i) – 3y1 + 2y2 – 4y3 ≤ 1, (ii) y1 – 4y2 + 3y3 ≤ – 3, (iii) – 2y1 + 8y3 ≤ – 2
and y1, y2 ≥ 0; y3 unrestricted in sign.

Example 5.5 Obtain the dual of the following primal LP problem


Maximize Zx = x1 – 2x2 + 3x3
subject to the constraints
(i) – 2x1 + x2 + 3x3 = 2, (ii) 2x1 + 3x2 + 4x3 = 1
and x1, x2, x3 ≥ 0

Solution Since both the primal constraints are of the equality type, the corresponding dual variables y1
and y2, will be unrestricted in sign. Following the rules of duality formulation, the dual of the given primal
LP problem is
Minimize Zy = 2y1 + y2
subject to the constraints
(i) – 2y1 + 2y2 ≥ 1, (ii) y1 + 3y2 ≥ – 2, (iii) 3y1 + 4y2 ≥ 3
and y1, y2 unrestricted in sign.

Example 5.6 Write the dual of the following primal LP problem


Maximize Z = 3x1 + x2 + 2x3 – x4
subject to the constraints
(i) 2x1 – x2 + 3x3 + x4 = 1, (ii) x1 + x2 – x3 + x4 = 3
and x1, x2 ≥ 0 and x3, x4 unrestricted in sign.

Solution Here we may apply the following rules of forming a dual of given primal LP problem.
(i) The x3 and x4 variables in the primal are unrestricted in sign, third and fourth constraints in the
dual shall be equalities.
(ii) The given primal LP problem is of maximization; the first two constraints in the dual LP problem
will therefore be ≥ type constraints.
(iii) Since both the constraints in the primal are equalities, the corresponding dual variables y1 and y2
will be unrestricted in sign.
If y1 and y2 are dual variables corresponding to the two primal constraints in the given order, the dual
of the given primal can be written as:
Minimize Zy = y1 + 3y2
subject to the constraints
(i) 2y1 + y2 ≥ 3, (ii) – y1 + y2 ≥ 1, (iii) 3y1 – y2 = 2
(iv) y1 + y2 = – 1
and y1, y2 unrestricted in sign.
152 Operations Research: Theory and Applications

SELF PRACTICE PROBLEMS A

Write the dual of the following primal LP problems 6. Min Zx = x1 + x2 + x3


1. Max Zx = 2x1 + 5x2 + 6x3 subject to (i) x1 – 3x2 + 4x3 = 5, (ii) x1 – 2x2 ≤ 3
subject to (i) 5x1 + 6x2 – x3 ≤ 3 (ii) 2x2 – 3x3 ≥ 4
(ii) – 2x1 + 3x2 + 4x3 ≤ 4 . and x1, x2 ≥ 0, x3 is unrestricted.
(iii) x1 – 5x2 + 3x3 ≤ 1 [Meerut, MSc (Maths), 2005]
(iv) – 3x1 – 3x2 + 7x3 ≤ 6 7. Min Zx = 8x1 + 3x2
and x1, x2, x3 ≥ 0 subject to (i) x1 – 6x2 ≥ 2, (ii) 5x1 + 7x2 = – 4
[Sambalpur MSc (Maths), 1996] and x1, x2 ≥ 0.
2. Min Zx = 7x1 + 3x2 + 8x3 8. Max Zx = 8x1 + 8x2 + 8x3 + 12x4
subject to (i) 8x1 + 2x2 + x3 ≥ 3 subject to (i) 30x1 + 20x2 + 25x3 + 40x4 ≤ 800
(ii) 3x1 + 6x2 + 4x3 ≥ 4 (ii) 25x1 + 10x2 + 7x3 + 15x4 ≤ 250
(iii) 4x1 + x2 + 5x3 ≥ 1 (iii) 4x1 – x2 = 0
(iv) x1 + 5x2 + 2x3 ≥ 7 (iv) x3 ≥ 5
and x1, x2, x3 ≥ 0 and x1, x2, x3, x4 ≥ 0
3. Max Zx = 2x1 + 3x2 + x3 9. Min Zx = 18x1 + 10x2 + 11x3
subject to (i) 4x1 + 3x2 + x3 = 6, (ii) x1 + 2x2 + 5x3 = 4 subject to (i) 4x1 + 6x2 + 5x3 ≥ 480
and x1, x2, x3 ≥ 0 (ii) 12x1 + 10x2 + 10x3 ≥ 1,200
4. Max Zx = 3x1 + x2 + 3x3 – x4 (iii) 10x1 + 15x2 + 7x3 ≤ 1,500
subject to (i) 2x1 – x2 + 3x3 + x4 = 1 (iv) x3 ≥ 50
(ii) x1 + x2 – 3x3 + x4 = 3 (v) x1 – x2 ≤ 0
and x1, x2, x3, x4 ≥ 0 and x1, x2, x3 ≥ 0
5. Min Zx = 2x1 + 3x2 + 4x3 10. Min Zx = 2x1 – x2 + 3x3
subject to (i) 2x1 + 3x2 + 5x3 ≥ 2 subject to (i) x1 + 2x2 + x3 ≥ 12
(ii) 3x1 + 3x2 + 7x3 = 3 (ii) x2 – 2x3 ≥ –6
(iii) x1 + 4x2 + 6x3 ≤ 5 (iii) 6 ≤ x1 + 2x2 + 4x3 ≥ 24
and x1, x2, ≥ 0, x3 is unrestricted and x1, x2 ≥ 0, x3 unrestricted.

HINTS AND ANSWERS

1. Min Zy = 3y1 + 4y2 + y3 + 6y4 5. Max Zy = 2y1 + 3y2 – 5y3


subject to (i) 5y1 – 2y2 + y3 – 3y4 ≥ 2 subject to 2y1 + 3y2 – y3 ≤ 2
(ii) 6y1 + y2 – 5y3 – 3y4 ≥ 5 3y1 + 3y2 – 4y3 ≤ 3
(iii) – y1 + 4y2 + 3y3 + 7y4 ≥ 6
5y1 + 7y2 – 6y3 = 4
and y1, y2, y3, y4 ≥ 0
and y1, y3 ≥ 0 and y2 unrestricted.
2. Max Zy = 3y1 + 4y2 + y3 + 7y4
6. Max Zy = – 5y1 – 3y2 + 4y3
subject to (i) 8y1 + 3y2 + 4y3 + 5y4 ≤ 7
(ii) 2y1 + 6y2 + 5y3 + 5y4 ≤ 3 subject to (i) – y1 – y2 ≤ 1, (ii) –3y1 + 2y2 + 2y3 ≤ 1
(iii) y1 + 4y2 + 5y3 + 2y4 ≤ 8 4y1 – y3 ≤ 1
and y1, y2, y3, y4 ≥ 0 and y2, y3 ≥ 0 and y1 is unrestricted.
3. Min Zy = 6y1 + 4y2 7. Max Zy = 2y1 – 4y2
subject to (i) 4y1 + 2y2 ≥ 2, (ii) 3y1 + 2y2 ≥ 3 subject to (i) y1 + 5y2 ≤ 8, (ii) – 6y1 + 7y2 ≤ 3
(iii) y1 + 5y2 ≥ 1 and y1 ≥ 0 and y2 is unrestricted.
and y1, y2 unrestricted in sign.
8. Min Zy = 800y1 + 250y2 + y3 + 5y4
4. Min Zy = y1 + 3y2
subject to 30y1 + 25y2 + 4y3 ≥ 8
subject to (i) 2y1 + y2 ≥ 3
20y1 + 10y2 – y3 ≥ 8
(ii) – y1 + y2 ≥ 1
(iii) 3y1 – y2 ≥ 3 25y1 + 7y2 + y4 ≥ 8
(iv) y1 + y2 ≥ –1 40y1 + 15y2 ≥ 12
and y1, y2 unrestricted in sign. y1, y2 ≥ 0 and y3, y4 are unrestricted.
Duality in Linear Programming 153

5.3 STANDARD RESULTS ON DUALITY


See appendix for detail proof of the following standard results:
1. The dual of the dual LP problem is again the primal problem.
2. If either the primal or the dual problem has an unbounded objective function value, the other problem
has no feasible solution.
3. If either the primal or dual problem has a finite optimal solution, the other one also possesses the same,
and the optimal value of the objective functions of the two problems are equal, i.e. Max Zx = Min Zy.
This analytical result is known as the fundamental primal-dual relationship. These results are
summarized as follows.

Dual Problem (Max) Primal Problem (Min)

Feasible Infeasible
Feasible Max Zy = Min Zx Max Zy → + ∞
Infeasible Min Zx → – ∞ Unbounded or infeasible

4. Complementary slackness property of primal-dual relationship states that for a positive basic variable
in the primal, the corresponding dual variable will be equal to zero. Alternatively, for a non-basic
variable in the primal (which is zero), the corresponding dual variable will be basic and positive.

5.3.1 Principle of Complementary Slackness


The principle of complementary slackness establishes the relationship between the optimal value of the main
variables in one problem with their counterpart slack or surplus variables in other problem. Thus, this
principle can also be helpful in obtaining the primal LP problem solution when only the dual solution is
known. To illustrate this concept, let us consider the following example.
1. If at the optimal solution of a primal problem, a primal constraint has a positive value of a slack variable,
the corresponding resource is not completely used and must have zero opportunity cost (shadow
price). This means that having more of this resource will not improve the value of the objective
function. But if the value of slack variable is zero in that constraint, the entire resource is being used
and must have a positive opportunity cost, i.e. additional resource will improve the value of objective
function by allowing more production.
Since resources are represented by slack variables in the primal and by main variables in the dual,
therefore, the principle of complementary slackness states that for every resource, the following
condition must hold:
Primal slack variable × Dual main variable = 0
2. In cases where the resources are not completely used, the opportunity cost of such resources exceeds
the unit profit from that variable. But, if few units of the variable are obtained, then the opportunity
cost of resources used must equal the profit from each unit of the variable, so that the surplus cost
is zero. A primal variable in primal LP problem represents the quantity to be obtained, and the surplus
variables in dual represent surplus costs. In this case the principle of complementary slackness states
that for every unit of the variable, the following condition must hold.
Primal main variable × Dual surplus variable = 0

5.4 MANAGERIAL SIGNIFICANCE OF DUALITY


The study of dual LP problem provides information about the value of the resources. Thus, a thorough
economic analysis is required to decide whether additional units of any resource are needed. If yes, then
how much cost per unit is required to be paid for any resource.
The significance of the study of dual is summarized as follows:
(i) The right-hand side of the constraint in primal LP problem represents the amount of a resource
available and the associated dual variable value is interpretated as the maximum amount likely to
be paid for an additional unit of this resource.
154 Operations Research: Theory and Applications

(ii) The maximum amount that should be paid for one additional unit of a resource is called its shadow
price (also called simplex multiplier).
(iii) The total shadow price (or marginal value) of the resources equals the optimal value of objective
function.
(iv) The value of the ith dual variable represents the rate at which the primal objective function value
will increase by increasing the right-hand side (resource value) of constraint i, assuming that all
other data remain unchanged.

Remark When the cost of a resource is sunk (cost that is incurred no matter irrespective to the value the dual
variables). When the cost of a resource vary with the value of decision variables the shadow price represents
the extra cost likely to be paid for one unit of the resource.

Example 5.7 The optimal solution simplex table for the primal LP problem
Maximize Z = 3x1 + 4x2 + x3
subject to the constraints
(i) x1 + 2x2 + 3x3 ≤ 90 (time for operation 1), (ii) 2x1 + 2x2 + 3x3 ≤ 60 (time for operation 2)
(iii) 3x1 + x2 + 2x3 ≤ 80 (time for operation 3)
and x1, x2, x3 ≥ 0
is given below:
cj → 3 4 1 0 0 0

Basic Variables Basic Basic Variables x1 x2 x3 s1 s2 s3


Coefficient Variables Value
cB B xB (= b)
4 x2 40 0 1 10/6 4/6 – 1/3 0
3 x1 10 1 0 – 1/3 – 1/3 2/3 0
0 s3 10 0 0 8/6 8/6 – 10/6 1

Z = 190 cj – zj 0 0 – 28/6 – 10/6 – 2/3 0

(a) Find the solution, maximum profit, idle capacity and the loss of the total contribution of every unit
reduced from the right-hand side of the constraints.
(b) Write the dual of the given problem and give the initial simplex table.
Solution Considering this LP problem in the context of profit maximization production problem with
constraints on time input resource.
The primal LP problem is concerned with maximizing the profit contribution from three products say
A, B and C, while the dual will be concerned with evaluating the time used in the three operations for
producing the three products.
The production capacity of the three operations is a resource to the firm. This LP problem can be solved
along the following lines.
(a) The optimal solution of the primal LP problem can be read from the given table as:
x1 = 10, x2 = 40, x3 = 0 and Max (profit) Zx = Rs 190
The value of slack variable, s3 = 10 units in the optimal solution is representing idle capacity in operation 3.
Interpretation of dual variables The absolute values of cj – zj in the optimal solution simplex table,
under the slack variables columns represent the values of dual variables, i.e. marginal value (shadow price)
or worth of one hour in each operation 1 and 2. The optimal solution of dual LP problem is:
y1 = 10/6, y2 = 2/3, y3 = 0 and Mix Zy = 190
Thus optimal value of the dual variable y1 = Rs 10/6 or Rs 1.66 represents the per unit price (worth or
marginal price) of the first resource (i.e. time available in operation 1). This means that marginal contribution
of operation 1 to the total profit is Rs 1.66. In other words, if one productive unit is removed from the right-
hand side of the first constraint, the total contribution would reduce by Rs 10/6, or Rs 1.66.
Duality in Linear Programming 155

However, if one productive unit is removed up to the extent of idle capacity from the right-hand side
of the third constraint (time available in operation 1), the total contribution would not be affected.
(b) Let y1, y2 and y3 be the dual variables representing per unit price (worth) for time available in the
operations 1, 2 and 3, respectively. Then the dual of the given problem is stated below:

Minimize Z = 90y1 + 60y2 + 80y3


subject to the constraints
(i) y1 + 2y2 + 3y3 ≥ 3, (ii) 2y1 + 3y2 + 3y3 ≥ 4, (iii) 3y1 + 3y2 + 2y3 ≥ 1
and y1, y2, y3 ≥ 0

Interpretation of dual constraints Suppose company wants to continue production of product C. To


see economic worth of C, examine c3 – z3 value under column-x3. Producing C will be economical only if c3 > z3.
This is possible either by increasing the profit per unit c3 or decreasing the cost of the used resources z3 (= 3y1
+ y2 + 2y3). An increase in the unit profit may not be possible because of competitive market. A decrease in z3 The shadow price
is possible by reducing the unit usage of time in operation 3. Let t1, t2 and t3 be the unit times of three operations is represented by
to be reduced. Then to determine t1, t2 and t3 such that the new cost, z3 of three operations falls below the unit the cj – zj value
under slack variable
profit, i.e. columns in the
optimal simplex
3 (1 – t1) y1 + (1 – t2)y2 + 2 (1 – t3) y3 < 1 table.
For given values of y1 = 5/3, y2 = 2/3 and y3 = 0, we get
3 (1 – t1) (5/3) + (1 – t2) (2/3) < 1 or 5t1 + 2t2 > 16
Thus, any values of t1 and t2 that satisfy 5t1 + 2t2 > 16 should make the product profitable. However, this may
not be possible because this requires a reduction in the times of operation 1 and 2.
Example 5.8 XYZ Company has three departments – Assembly, Painting and Packing. The company
can make three types of almirahs. An almirah of type I requires one hour of assembly, 40 minutes of painting
and 20 minutes of packing time, respectively. Similarly, an almirah of type II needs 80 minutes, 20 minutes
and one hour, respectively. The almirah of type III requires 40 minutes each of assembly, painting and
packing time. The total time available at assembly, painting and packing departments is 600 hours, 400 hours
and 800 hours, respectively. Determine the number of each type of almirahs that should be produced in
order to maximize the profit. The unit profit for types I, II and III is Rs 40, 80 and 60, respectively.
Suppose that the manager of this XYZ company is thinking of renting the production capacities of the
three departments to another almirah manufacturer – ABC company. ABC company is interested in
minimizing the rental charges. On the other hand, the XYZ company would like to know the worth of
production hours to them, in each of the departments, in order to determine the rental rates. (a) Formulate
this problem as an LP problem and solve it to determine the number of each type of almirahs that should
be produced by the XYZ company in order to maximize its profit. (b) Formulate the dual of the primal LP
problem and interpret your results. [Delhi Univ., MBA, 2002, 2005]

Mathematical formulation The production data given in the problem may be summarized as below:

Types of Almirah Number of Hours Required (per Unit) Profit per Unit
Assembly Painting Packing of Almirah (Rs)

I 1 2/3 1/3 40
II 4/3 1/3 1 80
III 2/3 2/3 2/3 60

Total availability (hrs) 600 400 800

Let x1, x2 and x3 be the number of units of the three types of almirahs to be produced, respectively.
The given problem can then be represented as an LP model:
156 Operations Research: Theory and Applications

Maximize Z = 40x1 + 80x2 + 60x3


subject to the constraints
(i) x1 + (4/3)x2 + (2/3)x3 ≤ 600, (ii) (2/3) x1 + (1/3) x2 + (2/3) x3 ≤ 400
(iii) (1/3) x1 + x2 + (2/3) x3 ≤ 800
and x1, x2, x3 ≥ 0
Let y1, y2 and y3 be the per hour rental price for assembly, painting and packing departments,
respectively to be paid by ABC company. Since company ABC desires to minimize total rental changes, then
objective function becomes:

Min (total rental charges) Zy = 600y1 + 400y2 + 800y3


The given data show that an almirah of type I requires 1 assembly hour, 2/3 painting hour and 1/3
packing hour. The cost (i.e. rent) of time used for making almirah I is: y1 + (2/3) y2 + (1/3) y3. If ABC Company
used that time to make almirah I, it would earn Rs c1 = 40 in contribution to profit, and so it will not rent
out the time unless: y1 + (2/3) y2 + (1/3) y3 ≥ 40. Similarly, the company will work for almirahs II and III.
Since the total rent of all the departments should be greater than or equal to the profit from one unit
of the almirah, the dual objective function along with the constraints that determines for ABC, the value
of the productive resources, can be written as:
Minimize (total rent) Zy = 600y1 + 400y2 + 800y3
subject to the constraints
(i) y1 + (2/3) y2 + (1/3) y3 ≥ 40, (ii) (4/3) y1 + (1/3) y2 + y3 ≥ 80
(iii) (2/3) y1 + (2/3) y2 + (2/3) y3 ≥ 60
and y1, y2, y3 ≥ 0
This dual objective function (minimum total rent acceptable by the ABC company) is equal to the primal
objective function (maximum profit that could be earned by XYZ company) from its resources.

Example 5.9 A firm manufactures two products A and B on machines I and II as shown below:

Machine Product Available Hours


A B
I 30 20 300
II 5 10 110

Profit per unit (Rs) 6 8

The total time available is 300 hours and 110 hours on machines I and II, respectively. Products A and
B contribute a profit of Rs 6 and Rs 8 per unit, respectively. Determine the optimum product mix. Write the
dual of this LP problem and give its economic interpretation.

Mathematical formulation The primal and the dual LP problems of the given problem are
Primal problem Dual problem
x1 and x2 = number of units of A and B to y1 and y2 = cost of one hour on machines I and II,
x1 and x2 = be produced, respectively respectively
Max Zx = 6x1 + 8x2 Min Zy = 300y1 + 110y2
subject to the constraints subject to the constraints
30x1 + 20x2 ≤ 300 30y1 + 15y2 ≥ 6
5x1 + 10x2 ≤ 110 20y1 + 10y2 ≥ 8
and x1, x2 ≥ 110 and y1, y2 ≥ 0
Duality in Linear Programming 157

Solution of the primal problem The optimal solution of the primal problem is given in Table 5.2.

cj → 6 8 0 0

Basic Variables Basic Basic Variables x1 x2 s 1p s 2p


Coefficient Variables Value
cB B xB (= b)

6 x1 4 1 0 1/20 – 1/10
Table 5.2
8 x2 9 0 1 – 1/10 3/20
Optimal Solution
Z = 96 cj – zj 0 0 – 1/10 – 6/10 of Primal Problem

Table 5.2 indicates that the optimal solution is to produce: x1 = 4 units of product A; x2 = 9 units
of product B and Zx = total maximum profit, Rs 96.
Solution of the dual problem The optimal solution of the dual problem can be obtained by applying
the Big-M method. The optimal solution is shown in Table 5.3.
bi → 300 110 0 0

Basic Variables Basic Basic Variables y1 y2 s 1d s 2d


Coefficient Variables Value
cB B yB
300 y1 1/10 1 0 – 1/20 1/40
Table 5.3
110 y2 6/10 0 1 1/10 – 3/20
Optimal Solution
Z = 96 bi – zj 0 0 4 9 of Dual Problem

The optimal solution as given in Table 5.3 is: y1 = Rs 1/10 per hour on machine I; y2 = Rs 6/10 per
hour on machine II and Zy = total minimum cost, Rs 96.
The values of y1 = Rs 1/10 and y2 = Rs 3/5, indicate the worth of one hour of machine time of machines
I and II, respectively. However, since the machines hours can be increased beyond a specific limit, this result
holds true only for a specific range of hours available on machines I and II.
Comparison of the solutions For interpreting the optimal solution of the primal (or dual), its solution
values can be read directly from the optimal simplex table of the dual (or primal). The method can be
summarized in the following steps.
1. The slack variables in the primal, correspond to the dual basic variables in the optimal solution and
vice versa. For example, s1p is the slack variable of the first primal constraint. It corresponds to the
first dual variable y1. Likewise s2p corresponds to the second dual variable y2. Moreover, s1d and s2d
are the dual surplus variables and they correspond to x1 and x2, respectively of the primal problem.
The correspondence between primal and dual variables is summarized in Table 5.4.

Primal Dual

Main variables
RS x 1 s1 d UV Surplus variables
Tx 2 s2 d W
R|S s1p y1 UV Main variables
Slack variables
|T s
2p y2 W Table 5.4

2. The value in the cj – zj row under columns of the slack/surplus variables, ignoring the negative sign,
directly gives the optimal values of the dual/primal basic variables. For example, c3 – z3 = 1/10 and
c4 – z4 = 6/10 values in Table 5.2 under primal slack s1p and s2p correspond to solution values of dual
variables y1 and y2 in Table 5.3 and vice versa. The primal-dual relationship for these problems which
are feasible can be summarized in Table 5.5.
158 Operations Research: Theory and Applications

Primal Dual

y Values of the basic variables y (cj – zj) values in the non-basic


surplus variable columns
y (cj – zj) values in the non-basic y Value of the basic variables
slack variable columns

3. The optimal value of the objective function is the same for primal and dual LP problems.
4. The dual variable yi represents the worth (dual price or shadow price) of one unit of resource i. For
example, for y1 the worth of time on machine I, equals Rs 1/10 and for y2, the worth of time on machine
II, equals Rs 6/10, as shown in Table 5.3.
Example 5.10 A company wishes to get at least 160 million ‘audience exposures’ the number of times one
of the advertisements is seen or heard by a person. Because of the nature of the product the company wants at
least 60 million month and at least 80 million of the exposures to involve persons between 18 and 40 years of
age. The relevant information pertaining to the two advertising media under consideration—magazine and
television is given below:
Magazine Television
• Cost per advertisement (Rs. thousand) 40 200
• Audience per advertisement (million) 4 40
• Audience per advertisement with monthly income
over Rs. 10,000 (million) 3 10
• Audience (per advertisement) in the age group
18–40 (million) 8 10
The company wishes to determine the number of advertisements to be released each in magazine and
television so as to keep the advertisement expenditure to the minimum. Formulate this problem as a LP problem.
What will be the minimum expenditure and its allocation among the two media? Write ‘dual’ of this problem.
Solve the ‘dual’ problem to find answer to the problem.
Solution The primal and dual LP problems of the given problem are:
Primal Problem Dual Problem
x1, x2 = number of advertisements in magazine y1, y2, y3 = shadow price (or worth) of one unit of
and television, respectively. advertisement over audience characte-
ristics, respectively.
Minimize Zx = 40x1 + 200x2 Maximize Zy = 160y1 + 60y2 + 80y3
subject to 4x1 + 40x2 ≥ 160 subject to 4y1 + 3y2 + 8y3 ≤ 40
3x1 + 10x2 ≥ 60 40y1 + 10y2 + 10y3 ≤ 200
8x1 + 10x2 ≥ 80 and y1, y2, y3 ≥ 0.
and x1, x2 ≥ 0.
Example 5.11 XYZ manufacturing company operates a three-shift system at one of its plants. In a certain
section of the plant, the number of operators required on each of the three shifts is as follows:
Shift Number of Operators

Day (6 a.m. to 2 p.m.) 50


Afternoon (2 p.m. to 10 p.m.) 24
Night (10 p.m. to 6 p.m.) 10

The company pays its operators at the basic rate of Rs. 10 per hour for those working on the day shift. For
the afternoon and night shifts, the rates are one and a half times the basic rate and twice the basic rate,
respectively. In agreement with each operator at the commencement of his employment, he is allocated to one
of three schemes A, B or C. These are as follows:
A : Work (on average) one night shift, one afternoon shift, and two day shifts in every four shifts.
B : Work (one average) equal number of day and afternoon shifts.
C : Work day shifts only.
Duality in Linear Programming 159

In schemes A and B, it is necessary to work strictly alternating sequences of specified shifts, as long as the
correct proportion of shifts is worked in the long run.
(a) Formulate a linear programming model to obtain the required number of operators at minimum cost.
(b) By solving the dual of the problem, determine how many operators must be employed under each of the
three schemes. Does this result in over-provision of operators on any one of the three shifts?

Solution The primal and dual LP problems of the given problem are:

Primal Problem Dual Problem


x1, x2 and x3 = number of operators employed under y1, y2 and y3 = shadow price (or worth) per unit
scheme A, B and C, respectively. of resources—operators in three
1 1 1  shifts, respectively.
Minimize Zx = 20 × x1 + 15  x1 + x2  Maximize Zy = 10y1 + 24x2 + 50y3
4 4 2 
1 1  subject to the constraints
+ 10  x1 + x2 + x3  (1/4)y1 + (1/4)y2 + (1/2)y3 ≤ 55/4
 2 2 
55 25 (1/2)y2 + (1/2)y3 ≤ 25/2
= x1 + x2 + 10 x3 y3 ≤ 10
4 2
subject to the constraints and y1, y2, y3 ≥ 0.
(1/4) x1 ≥ 10,
(1/4) x1 + (1/2) x2 ≥ 24
(1/2) x1 + (1/2) x2 + x3 ≥ 50
and x1, x2, x3 ≥ 0.

5.5 ADVANTAGES OF DUALITY


1. It is advantageous to solve the dual of a primal that has a less number of constraints because the
number of constraints usually equals the number of iterations required to solve the problem.
2. This avoids the necessity for adding surplus or artificial variables and solves the problem quickly (the
technique is known as the primal-dual method ). In economics, duality is useful in the formulation of
the input and output systems. It is also useful in physics, engineering, mathematics, etc.
3. The dual variables provide an important economic interpretation of the final solution of an LP problem.
4. It is quite useful when investigating changes in the parameters of an LP problem (the technique is
known as the sensitivity analysis).
5. Duality is used to solve an LP problem by the simplex method in which the initial solution is infeasible
(the technique is known as the dual simplex method ).

CONCEPTUAL QUESTIONS

1. Define the dual of a linear programming problem. State the 7. What is a shadow price? How does the concept relate to the
functional properties of duality. dual of an LP problem?
2. Explain the primal-dual relationship. 8. How can the concept of duality be useful in managerial decision-
3. Briefly discuss ‘duality’ in linear programming. making?
4. What is the principle of duality in linear programming? Explain its 9. State and prove the relationship between the feasible solutions
advantages. of an LP problem and its dual.
5. What is duality? What is the significance of dual variables in an 10. Prove that the necessary and sufficient condition for any LP
LP model? problem and its dual, in order to have optimal solutions is that
6. State the general rules for formulating a dual LP problem from both have feasible solutions.
its primal.

SELF PRACTICE PROBLEMS B

Write the dual of the following primal LP problems 2. Min Zx = 7x1 + 3x2 + 8x3
1. Max Zx = 2x1 + 5x2 + 6x3 subject to (i) 8x1 + 2x2 + 3x3 ≥ 3 (ii) 3x1 + 6x2 + 4x3 ≥ 4
subject to (i) 5x1 + 6x2 – x3 ≤ 3 (ii) – 2x1 + x2 + 4x3 ≤ 4 (iii) 4x1 + 3x2 + 5x3 ≥ 1 (iv) x1 + 5x2 + 2x3 ≥ 7
(iii) x1 – 5x2 + 3x3 ≤ 1 (iv) – 3x1 – 3x2 + 7x3 ≤ 6 and x1, x2, x3 ≥ 0.
and x1, x2, x3 ≥ 0.
160 Operations Research: Theory and Applications

3. Max Zx = 2x1 + 3x2 + x3 II. The first one provides 7, 5 and 2 units of the three vitamins per
subject to (i) 4x1 + 3x2 + x3 = 6 (ii) x1 + 2x2 + 5x3 = 4 gram, respectively and the second one provides 2, 4 and 8 units of
and x1, x2, x3 ≥ 0. the same three vitamins per gram of the foodstuff, respectively.
The first foodstuff costs Rs. 3 per gram and the second Rs. 2 per
4. Max Zx = 3x1 + x2 + 3x3 – x4
gram. The problem is how many grams of each foodstuff should
subject to (i) 2x1 – x2 + 3x3 + x4 = 1 (ii) x1 + x2 – x3 + x4 = 3 the housewife buy everyday to keep her food bill as low as possible?
and x1, x2, x3, x4 ≥ 0. (a) Formulate this problem as an LP model.
5. Min Zx = 2x1 + 3x2 + 4x3 (b) Write and then solve the dual problem.
subject to (i) 2x1 + 3x2 + 5x3 ≥ 2 (ii) 3x1 + 3x2 + 7x3 = 3 12. A large distributing company buys coffee seeds from four different
(iii) x1 + 4x2 + 6x3 ≤ 5 plantations. On these plantations, the seeds are available only in a
and x1, x2 ≥ 0, x3 is unrestricted. blend of two types A and B. The company wants to market a blend
consisting of 30 per cent of type A and 70 per cent of type B. The
6. Min Zx = x1 + x2 + x3 percentage of each type used by each plantation and the selling
subject to (i) x1 – 3x2 + 4x3 = 5 (ii) x1 – 2x2 ≤ 3 prices per 10 kg of the blends of each plantation are as follows:
(iii) 2x2 – x3 ≥ 4
and x1, x2 ≥ 0, x3 is unrestricted. Type Plantation Desired
[Meerut Univ., MSc (Maths), 2004] 1 2 3 4
7. One unit of product A contributes Rs. 7 and requires 3 units of raw
material and 2 hours of labour. One unit of product B contributes A 40% 20% 60% 80% 30%
Rs. 5 and requires one unit of raw material and one hour of labour. B 60% 80% 40% 20% 70%
Availability of raw material at present is 48 units and there are 40 Selling price
hours of labour. per 5 kg Rs. 3 Rs. 2 Rs. 1.20 Rs. 1.50
(a) Formulate this problem as a linear programming problem.
(b) Write its dual. What quantity of coffee seeds should the company buy from each
(c) Solve the dual by the simplex method and find the optimal plantation so that the total mixture will contain the desired
product mix and the shadow prices of the raw material and percentages of A and B and at the same time keep the purchasing
labour. cost at a minimum? Also write the dual of the problem.
8. A company makes three products: X, Y and Z out of three raw 13. The XYZ Plastic Company has just received a government contract
materials A, B and C. The raw material requirements are given to produce three different plastic valves. These valves must be
below: highly heat and pressure resistant and the company has developed
a three-stage production process that will provide the valves with
Raw Materials Number of Units of Raw Material Required the necessary properties involving work in three different chambers.
to Produce One Unit of Product Chamber 1 provides the necessary pressure resistance and can
process valves for 1,200 minutes each week. Chamber 2 provides
X Y Z heat resistance and can process valves for 900 minutes per week.
Chamber 3 tests the valves and can work 1,300 minutes per
A 1 2 1 week. The three valve types and the time in minutes required in
B 2 1 4 each chamber are:
C 2 5 1
Time Required in
The unit profit contribution of the products: X, Y and Z is Rs. 40, 25 Valve Type
and 50, respectively. The number of units of raw material available Chamber 1 Chamber 2 Chamber 3
are 36, 60 and 45, respectively. A 5 7 4
(a) Determine the product mix that will maximize the total profit. B 3 2 10
(b) Using the final simplex table, write the solution to the dual
C 2 4 5
problem and give its economic interpretation.
9. Three food products are available at costs of Rs. 10, Rs. 36 and Rs. The government will buy all the valves that can be produced and
24 per unit, respectively. They contain 1,000, 4,000 and 2,000 the company will receive the following profit margins on each valve:
calories per unit, respectively and 200, 900 and 500 protein units per A, Rs. 15; B, Rs. 13.50; and C, Rs. 10.
unit, respectively. It is required to find the minimum-cost diet How many valves of each type should the company produce
containing at least 20,000 calories and 3,000 units of protein. each week in order to maximize profits? Write the dual of the given
Formulate and solve the given problem as an LP problem. Write the LP problem and give its economic interpretation?
dual and use it to check the optimal solution of the given problem. 14. A medical scientist claims to have found a cure for the common
10. A company produces three products: P, Q and R from three raw cold that consists of three drugs called K, S and H. His results
materials A, B and C. One unit of product P requires 2 units of A and indicate that the minimum daily adult dosage for effective treatment
3 units of B. A unit of product Q requires 2 units of B and 5 units of C is 10 mg of drug K, 6 mg of drug S and 8 mg of drug H. Two
and one unit of product R requires 3 units of A, 2 units of B and 4 units substances are readily available for preparing pills for distribution
of C. The company has 8 units of material A, 10 units of material B to cold sufferers. Both substances contain all three of the required
and 15 units of material C available to it. Profits per unit of products P, drugs. Each unit of substance A contains 6 mg, 1 mg and 2 mg of
Q and R are Rs. 3, Rs. 5 and Rs. 4, respectively. drugs K, S and H, respectively and each unit of substance B
(a) Formulate this problem as an LP problem. contains 2 mg, 3 mg and 2 mg of the same drugs. Substance A
(b) How many units of each product should be produced to costs Rs. 3 per unit and substance B costs Rs. 5 per unit.
maximize profit? (a) Find the least-cost combination of the two substances that
(c) Write the dual of this problem. will yield a pill designed to contain the minimum daily
11. A diet conscious housewife wishes to ensure certain minimum recommended adult dosage.
intake of vitamins A, B and C for the family. The minimum daily (b) Suppose that the costs of the two substances are
(quantity) needs of the vitamins A, B and C for the family are interchanged so that substance A costs Rs. 5 per unit and
respectively 30, 20 and 16 units. For the supply of these minimum substance B costs Rs. 3 per unit. Find the new optimal solution.
vitamin requirements, the housewife relies on two fresh food I and [Delhi Univ., MBA (HCA), 2008]
Duality in Linear Programming 161

15. The XYZ company has the option of producing two products during 18. A firm produces three articles X, Y, Z at a total cost of Rs. 4 , Rs.
the period of slack activity. For the next period, production has 3, and Rs. 6 per item respectively. Total number of X and Z item
been scheduled so that the milling machine is free for 10 hours and produced should be at least 2 and number of Y and Z together be
skilled labour will have 8 hours of time available. at least 5. The firm wants to minimize the cost. Formulate this
problem as an LP problem. Write its dual. Solve the dual by the
Product Machine Time Skilled Labour Profit simplex method. Can you point out the solution of the primal
per Unit Time per Unit Contribution problem? If yes, what is it?
per Unit (Rs.) 19. A firm produces three types of biscuits A, B and C. It packs them
in assortments of two sizes I and II. The size I contains 20 biscuits
A 4 2 5 of type A, 50 of type B and 10 of type C. The size II contains 10
B 2 2 3 biscuits of type A, 80 of type B and 60 of type C. A buyer intends to
buy at least 120 biscuits of type A, 740 of type B and 240 of type C.
Solve the primal and dual LP problems and bring out the fact that Determine the least number of packets he should buy. Write the
the optimum solution of one can be obtained from the other. Also dual LP problem and interpret your answer.
explain in the context of the example, what you understand by
20. Consider the following product mix problem: Let x1 denote
shadow prices (or dual prices or marginal value) of resource.
number of units of Product 1 to be produced daily and x2 the
[Jammu Univ. MBA 2006]
number of units of Product 2 to be produced daily. The production
16. A company produces three products P, Q and R whose prices per
of Product 1 requires one hour of processing time in department
unit are 3, 5 and 4 respectively. On unit of product P requires 2
D1 . Production of 1 unit of Product 2 requires 2 hours of
units of m1 and 3 units of m2. A unit of product Q requires 2 units of
processing time in department D1 and one hour in department
m2 and 5 units of m3 and one unit of product R requires 3 units of
D2. The number of hours available in department D1 are 32
m1, 2 units of m2 and 4 units of m3. The company has 8 units of
hours and in department D2, 8 hours. The contribution of one
material m1, 10 units of material m2 and 15 units of material m3
unit of Product 1 is Rs. 200 and of Product 2 is Rs. 300.
available to it.
The solution to this LP model is given below:
(a) Formulate the problem as on LP model.
(b) How many units of each product should be produced to cj → 200 300 0 0
maximize revenue?
(c) Write the dual problem. Basic Solution x1 x2 s1 s2
Variables Values
17. A person consumes two types of food A and B everyday to obtain 8 cB xB
units of proteins, 12 units of carbohydrates and 9 units of fats which
is his daily minimum requirements. 1 kg of food A contains 2, 6 and 1 200 x1 32 1 2 1 0
units of protein, carbohydrates and fats, respectively. 1 kg of food B 0 s2 8 0 1 0 1
contains 1, 1 and 3 units of proteins, carbohydrates and fats
Z = 6400 cj – zj –100 –200 0 6400
respectively. Food A costs Rs. 8.50 per kg, while B costs Rs. 4 per
kg. Determine how many kg of each food should he buy daily to Given the dual of the primal model. Obtain the optimum solution
minimize his cost of food and still meet the minimum requirements. to the dual LP model from the above table. Interpret the dual
Formulate on LP problem mathematically. Write its dual and variables.
solve the dual by the simplex method. [Gujarat Univ. MBA, 2006]

HINTS AND ANSWERS

1. Min Zy = 3y1 + 4y2 + y3 + 6y4 5. Max Zy = 2y1 + 3y2 – 5y3


subject to (i) 5y1 – 2y2 + 5y3 – 3y4 ≥ 2 subject to (i) 2y1 + 3y2 – y3 ≤ 2 (ii) 3y1 + 3y2 – 4y3 ≤ 3
(ii) 6y1 + y2 – 5y3 – 3y4 ≥ 5 (iii) 5y1 + 7y2 – 6y3 = 4
(iii) – y1 + 4y2 + 3y3 + 7y4 ≥ 6 and y1, y3 ≥ 0 and y2 unrestricted.
and y1, y2, y3, y4 ≥ 0. 6. Max Zy = – 5y1 – 3y2 + 4y3
subject to (i) – y1 – y2 ≤ 1 (ii) 3y1 + 2y2 + 2y3 ≤ 1
2. Max Zy = 3y1 + 4y2 + y3 + 7y4
(iii) – 4y1 – y3 ≤ 1
subject to (i) 8y1 + 3y2 + 4y3 + 5y4 ≤ 7
and y2, y3 ≥ 0 and y1 is unrestricted.
(ii) 2y1 + 6y2 + 5y3 + 5y4 ≤ 3
7. Primal
(iii) y1 + 4y2 + 5y3 + 2y4 ≤ 8
x1 and x2 = number of units of products A
and y1, y2, y3, y4 ≥ 0.
x1 and x2 = and B, respectively to be produced.
3. Min Zy = 6y1 + 4y2
Max Zx = 7x1 + 5x2
subject to (i) 4y1 + 2y2 ≥ 2 (ii) 3y1 + 2y2 ≥ 3
subject to (i) 3x1 + x2 ≤ 48 (ii) 2x1 + x2 ≤ 40
(iii) y1 + 5y2 ≥ 1
and x1, x2 ≥ 40
and y1, y2 unrestricted in sign.
4. Min Zy = y1 + 3y2 Ans. x1 = 0, x2 = 40, Max Zx = Rs. 200
Dual
subject to (i) 2y1 + y2 ≥ 3 (ii) – y1 + y2 ≥ 1
y1 and y2 = worth of one unit of raw material
(iii) 3y1 – y2 ≥ 3 (iv) y1 + y2 ≥ – 1
y1, and y2 = and labour, respectively.
and y1, y2, unrestricted in sign.
162 Operations Research: Theory and Applications

Min Zy = 48y1 + 40y2 and x1, x2, x3, x4 ≥ 0.


subject to (i) 3y1 + 2y2 ≥ 7 (ii) y1 + 2y2 ≥ 5 Dual
and y1, y2 ≥ 0 y1 and y2 = worth of coffee seeds A and B, respectively
Max Zy = 0.3y1 + 0.7y2
Ans. y1 = 0, y2 = 5 and Min Zy = Rs. 200
8. Primal subject to (i) 0.4y1 + 0.6y2 ≤ 3 (ii) 0.2y1 + 0.8y2 ≤ 2
x1, x2 and x3 = units of the products X, Y and (iii) 0.6y1 + 0.4y2 ≤ 1.2 (iv) 0.8y1 + 0.2y2 ≤ 1.5
x1, x2, and x3 = Z, respectively to be produced. and y1, y2 ≥ 0
Max Zx = 40x1 + 25x2 + 50x3 13. Primal
subject to (i) x1 + 5x2 + 4x3 ≤ 36 (ii) 2x1 + 5x2 + 4x3 ≤ 60 x1, x2 and x3 = number of valves of the types A, B and C,
respectively to be produced.
(iii) 2x1 + 5x2 + 4x3 ≤ 45
Max Zx = 15x1 + 13.5x2 + 10x3
and x1, x2, x3 ≥ 0.
subject to (i) 5x1 + 13x2 + 2x3 ≤ 1,200
Ans. x1 = 20, x2 = 0, x3 = 5; Max Zx = 1,050
Dual (ii) 7x1 + 12x2 + 4x3 ≤ 900
y1, y2 and y3 = worth (or shadow price) per unit (iii) 4x1 + 10x2 + 5x3 ≤ 1,300
y1, y2, and y3 = of raw materials A, B and C, respectively. and x1, x2, x3 ≥ 0.
Min Zx = 36y1 + 60y2 + 45y3 Ans. x1 = 22,400/217, x2 = 5,500/62, x3 = 0 and
subject to (i) y1 + 2y2 + 2y3 ≥ 40 (ii) y1 + 5y2 + 5y3 ≥ 25 Ans. Max Zx = Rs. 58,125/31
(iii) y1 + 4y2 + 5y3 ≥ 50 Dual
and x1, x2, x3 ≥ 0. y1, y2 and y3 = work of Chambers 1, 2 and 3 production capacity,
y1, y2 and y3 = respectively.
Ans. y1 = 0, y2 = 0, y3 = 10; Min Zy = 1,050
10. Primal Min Zy = 1,200y1 + 900y2 + 1,300y3
x1, x2 and x3 = units of the products P, Q and R to be produced, subject to (i) 5y1 + 7y2 + 14y3 ≥ 15
respectively. (ii) 3y1 + 2y2 + 10y3 ≥ 13.5
Max Z = 3x1 + 5x2 + 4x3 (iii) 2y1 + 4y2 + 15y3 ≥ 10
subject to (i) 2x1 + 3x3 ≤ 8 (ii) 5x1 + 2x2 + 2x3 ≤ 10 and y1, y2, y3 ≥ 0.
(iii) 5x2 + 4x3 ≤ 15 Ans. y1 = 0, y2 = – 41/31, y3 = – 645/62.
and x1, x2, x3 ≥ 10.
Dual 15. Primal
Min Z = 8y1 + 10y2 + 15y3 x1, x2 = number of units of product A and B respectively to be
produced
subject to (i) 2y1 + 5y2 ≥ 3 (ii) 2y2 + 5y3 ≥ 5
Max Zx = 5x1 + 3x2
(iii) 3y1 + 2y2 + 4y3 ≥ 4
subject to (i) 4x1 + 2x2 ≤ 10 (ii) 2x1 + 2x2 ≤ 8
and y1, y2, y3 ≥ 0.
11. Primal and x1, x2 ≥ 0
x1 and x2 = units of fresh food I and II, to be bought, respectively Ans. x1 = 1, x2 = 3 and Max Zx = 14
Min Z = 3x1 + 2x2 Dual
subject to (i) 7x1 + 2x2 ≥ 30 (ii) 5x1 + 4x2 ≥ 20 y1, y2, y3 = worth (or shadow price) per unit of resource operators
(iii) 2x1 + 8x2 ≥ 16 in three shifts respectively.
and x1, x2 ≥ 10. Min Zy = 10y1 + 8y2
Dual subject to (i) 4y1 + 2y2 ≥ 5 (ii) 2y1 + 2y2 ≥ 3
y1, y2 and y3 = worth per unit of vitamins A, B and C, respectively and y1, y2 ≥ 0.
to the body.
Ans. y1 = 1, y2 = 1/2 and Min Zy = 14.
Max Z = 30y1 + 20y2 + 16y3
16. Primal
subject to (i) 7y1 + 5y2 + 2y3 ≤ 3 (ii) 2y1 + 4y2 + 8y3 ≤ 2
x1, x2, x3 = number of units of product P, Q and R to be produced,
and y1, y2, y3 ≥ 0.
respectively.
(b) The optimal solution to the dual problem is: y1 = 20/52,
Max Zx = 3x1 + 5x2 + 4x3
y2 = 0, y3 = 8/52 and Max Z = Rs. 14.
subject to (i) 2x1 + 3x3 ≤ 8 (ii) 5x1 + 2x2 + 2x3 ≤ 10
12. Primal (iii) 5x2 + 4x3 ≤ 15
x1, x2, x3 and x4 = quantities of coffee seeds, respectively which and x1, x2, x3 ≥ 0.
the company buys from each plantation. Ans. x1 = 44/41, x2 = 59/41, x3 = 80/41
Min = 3x1 + 2x2 + 1.2x3 + 1.5x4 Dual
subject to (i) 0.4x1 + 0.2x2 + 0.6x3 + 0.8x4 ≥ 0.3 y1, y2, y3 = worth per unit of material m 1 , m 2 and m 3
(ii) 0.6x1 + 0.8x2 + 0.4x3 + 0.2x4 ≥ 0.7 respectively.
Duality in Linear Programming 163

Min Zy = 8y1 + 10y2 + 15y3 19. Primal


subject to (i) 2y1 + 3y2 ≥ 3 (ii) 2y2 + 5y3 ≥ 5 x1, x2 = number of assortments of size I and II, respectively.
(iii) 3y1 + 2y2 + 4y3 ≥ 4 Max Zx = x1 + x2
subject to (i) 20x1 + 10x2 ≥ 120 (ii) 50x1 + 80x2 ≥ 740
and y1, y2, y3 ≥ 0.
(iii) 10x1 + 60x2 ≥ 240
17. Primal
and x1, x2 ≥ 0
x1, x2 = number of units of food A and B to be consumed,
Ans. x1 = 2, x2 = 8 and Min Zx = 10
respectively.
Dual
Max Zx = 8.50x1 + 4x2
y1, y2, y3 = worth per unit of biscuits of type A, B and C,
subject to (i) 2x1 + x2 ≥ 8 (ii) 6x1 + x2 ≥ 12
respectively.
(iii) x1 + 3x2 ≥ 9 Min Zy = 120y1 + 740y2 + 240
and x1, x2 ≥ 0. subject to (i) 20y1 + 50y2 + 10y3 ≤ 1
Ans. x1 = 1, x2 = 6 and Min Zx = 65/2; (ii) 10y1 + 80y2 + 60y3 ≤ 1
Dual and y1, y2, y3 ≥ 0
y1, y2, y3 = worth per unit of proteins, carbohydrates and fats 20. Primal
respectively. Max Zx = 200x1 + 300x2
Min Zy = 8y1 + 12y2 + 9y3 subject to (i) x1 + 2x2 ≤ 32 (ii) 0x1 + x2 ≤ 8
subject to (i) 2y1 + 6y2 + y3 ≤ 8.50 (ii) y1 + y2 + 3y3 ≤ 4 and x1, x2 ≥ 0.
and y1, y2, y3 ≥ 0 Dual
Ans. y1 = 31/8, y2 = 1/8, y3 = 0 and Max Zy = 65/2; Min Zy = 32y1 + 8y2
subject to (i) y1 + 0y2 ≥ 200 (ii) 2y1 + y2 ≥ 300
18. Primal
and y1, y2 ≥ 0
x1, x2, x3 = number of units of articles X, Y and Z respectively.
The optimum solution to the dual LP problem is: y1 = 200, y2 = 0s1
Max Zx = 4x1 + 3x2 + 6x3 = 0s2 = 100, Z* = Rs. 600
subject to (i) x1 + x3 ≥ 2 (ii) x2 + x3 ≥ 5 Note that, y1 = marginal increase in profit for an addition 1 hour
and x1, x2, x3 ≥ 0. of capacity. Since y1 = Rs. 200, profit becomes Rs. 6400 + 200
= Rs. 6600. Also y2 = marginal increase in profit given an addition
Ans. x1 = 0, x2 = 3, x3 = 2 with Min Z = 21
hour of capacity in D2. Since y2 = 0, there would be no change in
Dual the profit.
y1, y2 = worth per unit of resources to be used, respectively In primal LP problem solution, s2 = 8 indicates that there is an
excess capacity of 8 hours in D2. Thus, in dual solution we have y2 = 0.
Min Zy = 2y1 + 5y2
Dual constraint (i) is written as: y1 + 0y2 ≥ 200 or
subject to (i) y1 ≤ 4 (ii) y2 ≤ 3
 total value of hours in department D1 and D 2 
(iii) y1 + y2 ≤ 6  
 required to produce 1 unit of Product 1 
and y1, y2 ≤ 0  unit profit received from manufacturing 
* ≥  
Ans. y1 = 3, y2 = 3 with Max Z = 21 and selling 1 unit of Product 1 

CHAPTER SUMMARY
In this chapter we discussed the relationship between an LP problem and its dual. The method of formulating a dual LP problem
of the given LP problem was explained with several examples. The main focus, while solving a dual LP problem, was to find for
each resource its best marginal value, also known as dual or shadow price. This value reflects the maximum additional price to
be paid in order to obtain one additional unit of any resource in order to maximize profit (or minimize cost) under resource
constraints.

CHAPTER CONCEPTS QUIZ


True or False
1. If the objective function of the primal LP problem is maximized, 3. The dual of dual is a primal problem.
then the objective function of the dual is to be minimized. 4. The value in the cj – zj row under columns of slack/surplus
2. A dual variable is not defined for each constraint in the primal variables ignoring negative signs do not give the direct optimal
LP problem. values of the dual/primal basic variables.
164 Operations Research: Theory and Applications

5. The shadow price indicates how much increase in a right hand (a) marginal opportunity cost of a resource equals its marginal
side coefficient will change the optimality range of an objective return
function coefficient. (b) marginal opportunity cost of a resource is less than its
6. The right hand side coefficient of some constraint in a primal LP marginal return
problem appears in the corresponding dual as a coefficient in the (c) both (a) and (b)
objective function. (d) none of the above
7. The change in the objective function value due to per unit 25. Shadow price indicates how much one unit change in the
decrease in the right hand side of a constraint is given by the resource value will change the
negative of the shadow price for that constraint. (a) optimality range of an objective function
8. A variable which is in the solution of a linear programming (b) optimal value of the objective function
problem is called basic variable. (c) value of the basic variable in the optimal solution
9. The cj – zj values in the non-basic slack variable columns is (d) none of the above
equal to the value of the basic variable. 26. Principle of complementary slackness states that
10. The number of constraints in any LP problem usually equals the (a) primal slack × dual main = 0
number of iterations required to solve the problem. (b) primal main × dual surplus = 0
(c) both (a) and (b)
(d) none of the above
Fill in the Blanks
27. If dual has an unbounded solution, primal has
11. The optimal value of the objective function is same for the — (a) no feasible solution
————— and —————— problem. (b) unbounded solution
12. The —————— variables provide the important economic (c) feasible solution
interpretation of the final solution of the LP problem. (d) none of the above
13. Duality is used to solve an LP problem by —————— method 28. If at the optimality a primal constraint has a positive value of a
in which the initial solution is infeasible. slack variable, then
14. If either of the primal or dual LP problem has an ————— (a) dual variable corresponding to that constraint has zero
objective function, then the other problem has —————— value
solution. (b) corresponding resource is not completely used up
15. If the ith primal constraint is of —————— type, then the ith (c) corresponding resource have zero opportunity cost
dual variable will be ————— in sign and vice versa. (d) both (b) and (c) but not (a)
16. —————— property states that for a positive basic variable 29. The shadow price for a resources is the
in the primal, the corresponding dual variable will be ———— (a) price that is paid for the purchase of a resource
—— to zero. (b) saving by eliminating one of the excess quantities of
17. The maximum amount that should be paid for one additional unit resource
of a resource is called its ———. (c) increase in the objective function value by providing one
18. The —————— variables in the primal correspond to the dual additional unit of resource
—————— variables in the optimal solution. (d) none of the above
19. Linear programming models are composed of three parts: —— 30. The value of dual variable
————, ——————, and ———. (a) represents worth of each additional unit of resource
20. Associated with any linear programming problem is another (b) can be obtained by examining the zj row of primal optimal
linear problem called the ——————, which can be used to simplex table
evaluate the resources in the original problem. If the original (c) can be obtained by examining cj – zj row of primal optimal
linear program, called the —————— is a maximization simplex table
problem, then this associated problem is a —————— problem. (d) all of the above
31. If value of primal slack variable in the optimal simplex table is zero,
Multiple Choice then
(a) LP problem has multiple solutions
21. The dual of the primal maximization LP problem having m (b) dual variable value associated with that constraint is zero
constraints and n non-negative variables should (c) resource measured by that constraint is not fully utilized
(a) have n constraints and m non-negative variables (d) all of the above
(b) be a minimization LP problem 32. The shadow price of a greater than or equal to (≥) constraint in a
(c) both (a) and (b) maximization LP problem can be
(d) none of the above (a) zero (b) positive
22. For any primal problem and its dual, (c) negative (d) all of the above
(a) optimal value of objective functions is same 33. The imputed cost of used resources per unit of an activity is
(b) primal will have an optimal solution if and only if dual does (a) less than or equal to the profit per unit of the activity
too (b) greater than or equal to the profit per unit of the activity
(c) both primal and dual cannot be infeasible (c) strictly equal to the profit per unit of the activity.
(d) all of the above (d) neither (a) nor (b)
23. The right-hand side constant of a constraint in a primal problem 34. Dual constraints for the maximization LP problem can be written
appears in the corresponding dual as as
(a) a coefficient in the objective function (a) Σ aij yi ≥ cj (b) Σ aji yi ≥ cj
(b) a right-hand side constant of a constraint
(c) Σ aij yj ≤ cj (d) Σ aji yi ≤ cj
(c) an input-out coefficient
(d) none of the above 35. If a primal LP problem has a finite solution then the dual LP problem
24. Dual LP problem approach attempts to optimize resource should have
allocation by ensuring that (a) finite solution (b) infeasible solution
(c) unbounded solution (d) none of the above
Duality in Linear Programming 165

Answers to Quiz
1. T 2. F 3. T 4. F 5. F 6. T 7. T 8. T 9. T 10. T
11. Primal, dual 12. Dual 13. Simplex 14. Unbounded; infeasible 15. Equality; unrestricted
16. Complementary slackness; equal 17. Shadow price 18. Slack; basic
19. Decision variables; constraints; objective function 20. Dual; primal; minimization
21. (c) 22. (b) 23. (a) 24. (a) 25. (b) 26. (c) 27. (a) 28. (d) 29. (c) 30. (a) 31. (c)
32. (d) 33. (a) 34. (b) 35. (a)

CASE STUDY
Case 5.1: Bhawani Manufacturer
Bhawani Manufacturer has discontinued production of a certain unprofitable product line and this has created considerable
excess production capacity. The management is considering to devote this excess capacity to produce one or more of three
products 1, 2 and 3. The available excess capacity on the machines which might limit output, is summarized in the following
table:

Machine type Available Excess Capacity


(Machine Hours per Week)

Milling machine 250


Lathe 150
Grinder 50

The number of machine-hours for each unit of the respective product is given below.

Capacity Requirement (Machine-hours per Unit)


Machine Type Product 1 Product 2 Product 3

Milling machine 8 2 3
Lathe 4 3 0
Grinder 2 0 1

The per unit contribution would be Rs. 20, Rs. 6 and Rs. 8, respectively for products 1, 2 and 3.
Suggest to the management of the company a production plan to produce and sell its three products so as to maximize
total revenue. Also, evaluate shadow price of each resource.

Case 5.2: Raman Traders


The procurement manager of Raman Traders that manufactures special gasoline additives must determine the proper
amounts of each raw material to be purchased for the production of one of its products. Three raw materials are available.
Each litre of the finished product must have at least a combustion point of 222°F. In addition, the gamma content (which
cause hydrocarbon pollution) cannot exceed 6 per cent of volume. The zeta content (which is goods for cleaning the internal
moving parts of engines) must be at least 12 per cent by volume. Each raw material contains three elements in varying
amounts, as shown in the given table.
Raw material A costs Rs. 9.00 per litre, whereas raw materials B and C cost Rs. 6.00 and Rs. 7.50 per litre, respectively.

Raw Material
A B C
Combustion point, °F 200 180 280
Gamma content, % 4 3 10
Zeta content, % 20 10 8

The procurement manager wishes to minimize the cost of raw materials per litre of product.
You as management trainee are expected to suggest method to find the optimal proportions of each raw material to be
used in a litre of the finished product. Also write the dual of the given problem and give its economic interpretation.
166 Operations Research: Theory and Applications

APPENDIX: THEOREMS ON DUALITY


Theorem 5.1: The dual of the dual is the primal.
Proof: Consider the primal problem in canonical form:
Minimize Zx = cx
subject to Ax ≥ b; x ≥ 0 (1)
where A is an m × n matrix, bT ∈ Em and c, xT ∈ En.
Applying the transformation rules, the dual of this problem is:
Maximize Zy = bTy
subject to A Ty ≤ c T ; y ≥ 0 (2)
This dual problem can also be written as:
Minimize Zy* = (– b)Ty (3)
subject to (– AT)y ≥ (– cT); y ≥ 0, where Z *y = – Zy
If we consider LP problem (3) as primal, then its dual can be constructed by considering x as the
dual variable. Thus, we have
Maximize Z * = (– cT)Tx = (– c)x
x
subject to (– AT)Tx ≤ ( – bT)T or (– A)x ≤ (– b)
and x ≥0 (4)
But LP problem (4) is identical to the given primal LP problem (1). This completes the proof of the
theorem.

Theorem 5.2: Let x* be any feasible solution to the primal LP problem


Maximize Zx = cx
subject to Ax ≤ b; x ≥ 0
and y* be any feasible solution to the dual LP problem
Maximize Zy = bTy
subject to A Ty ≥ c T; y ≥ 0
of the above primal problem. Then prove that cx* ≤ bTy*, i.e. Zx ≤ Zy

Proof: Since x* and y* are the feasible solutions to the primal and dual LP problems, respectively,
therefore from the constraints in primal and dual, we have
Ax* ≤ b; x* ≥ 0 (5)
T T
A y* ≥ c ; y* ≥ 0 (6)
T T T
From inequality (6), we have c ≤ A y* or c ≤ A (*y)
Multiplying both sides by x*, we get
cx* ≤ (Ay*T)x* = y*T(Ax*) = y*Tb = bTy* [since y*Tb = bTy*]
This completes the proof of the theorem.

Theorem 5.3: If x* is the feasible solution to the primal LP problem,


Minimize Zx = cx
subject to Ax ≤ b; x ≥ 0
and y* is the feasible solution to the dual problem
Minimize Zy = bTy
subject to A Ty ≥ c T ; x ≥ 0
of the above primal problem, such that, cx* = bTy, then x* and y* are the optimal solutions to their
respective problems.
Duality in Linear Programming 167

Proof: Let x *0 be any other feasible solution to the primal. Then from Theorem 5.2 we have
c x *0 ≤ bTy or c x *0 ≤ cx* [since cx* = bTy*]. Hence x* is an optimal solution to the given maximization
primal LP problem.
Similarly, for any other feasible solution y *0 to the dual LP problem, we have cx* ≤ bT y *0 or bTy*
≤ bT y *0 [since cx* = bTy*]. Hence, y* is also an optimal solution to the given minimization dual problem.

Theorem 5.4: If ith constraint in the primal is an equality, then the ith dual variable is unrestricted in sign.
Proof: Consider the primal LP problem in its standard form as
Minimize Zx = cx
subject to Ax = b; x ≥ 0
This problem can also be written in equivalent form as:
Maximize Zx = cx
subject to Ax ≤ b and (– A) x ≤ – b; x ≥ 0
The dual of this problem can now be written as:
Minimize Zy = y1bT + y2 (– bT ) = ( y1 – y2) bT
subject to ATy1 + (– AT ) y2 ≥ c; y1, y2 ≥ 0
Let y = y1 – y2. Then we can rewrite these equations as
Minimize Zy = ybT
subject to ATy ≥ c; y unrestricted in sign.
Theorem 5.5 (Unboundedness Theorem): If either the primal or the dual LP problem has an
unbounded objective function value, then the other problem has no feasible solution.
Proof: Let the given primal LP problem have an unbounded solution. Then for any value of the
objective function, say + ∞, there exists a feasible solution say x yielding this solution, i.e. cx → ∞.
From Theorem 5.2, for each feasible solution y* of the dual, there exists a feasible solution x* to
the primal such that cx* ≤ bTy*. That is, bTy* → + ∞. As b is a constant number and y* has to satisfy
the constraint ATy* ≤ cT. Hence the dual objective function Zy = bTy* must be finite. This contradicts
the result bTy* → ∞. Hence the dual LP problem has no feasible solution.
A similar argument can be used to show that when the dual LP problem has an unbounded solution,
the primal LP problem has no solution.
Theorem 5.6 (Duality Theorem): If either the primal or the dual problem has a finite optimal solution,
then the other one also possess the same, and the optimal values of the objective functions of the two
problems are equal, Max Zx = Min Zy.
Proof: Let the following LP problem represent a primal problem:
Maximize Zx = cx
subject to Ax ≤ b or Ax + Is = b; x ≥ 0
where I is the identify matrix of order m and s is the slack variable.
Let xB = B– 1b be an optimal basic feasible solution to this primal problem and cB the cost vector
of the basic variables. Then according to the optimality condition, we have, cj – zj ≤ 0, for any vector
aj of A but not in B. Therefore,
cj – zj = cj – cByj = cj – cB (B– 1 aj) ≤ 0, for all j
–1
This is equivalent to c ≤ cBB A, when written in matrix notation.
Let y* = cB B– 1. Then c ≤ cB B– 1A becomes c ≤ y*A or cT ≤ y* AT. Hence y* is a feasible solution
to the dual of the given primal because it satisfies the dual constraints. The corresponding dual objective
function is given by
Zy = y*bT = cB B– 1 bT = cB xB = Zx
Hence, it has been shown that xB and cB B– 1 are the feasible solution to the primal and dual LP problems,
respectively for which Max Zx = Min Zy.
168 Operations Research: Theory and Applications

Theorem 5.7 (Complementary Slackness Theorem): If x* and y* be feasible solutions to the primal
and dual LP problems, respectively, then a necessary and sufficient condition for x* and y* to be optimal
solutions to their respective problems is,
yi · xn + i = 0, i = 1, 2, . . ., m
and xj · ym + j = 0, j = 1, 2, . . ., n
where xn + i is the ith slack variable in the primal LP problem and ym + j the jth surplus variable for the
dual LP problem.
Proof: Let x* be the feasible solution to the primal LP problem,
n
Maximize Zx = ∑ c j x j
j =1
n
subject to ∑ a ij x j + x n + i = bi ; xj ≥ 0 (7)
j =1
and y* be the feasible solution to the dual of the above primal.
n
Minimize Zy = ∑ bi y i (8)
j =1
n
subject to ∑ a ij y i − y m + j = cj ; y i ≥ 0
j =1
Multiplying constraints in (7) by yi, i = 1, 2, . . ., m, and then adding we get
m m
x j ∑ a ij y i + x n + i y i = ∑ y i bi ; j = 1, 2, ..., n (9)
i =1 i =1
Subtracting (9) from the objective function of (7), we get
RS c m
− ∑ a ij y i
UV x m
– x n + i y i = Z x – ∑ y i bi (10)
T j
i =1 W j
i =1
Substituting the value of Zy and ym + j from (8) to (10) we get,
n m
− ∑ x j y m + j − ∑ y i x n + i = Z x – Zy (11)
j =1 i =1
If x* and y* are optimal solutions to the primal and dual problems, then we have cx* = bTy* (or Zx
= Zy ) Thus, from (11), we have
n m
∑ xj y m + j = ∑ y i x n + i = 0
j =1 i =1
m
Now it follows that, for each xj > 0, j (= 1, 2, . . ., m) implies ym + j = 0 and ∑ a ij y i = cj, i.e. jth constraint
i =1
m
in the dual is an equation. Also xj = 0 implies ym + i = 0 and ∑ a ij y i > cj.
i =1
n
Similarly, for each yi > 0, i (= 1, 2, . . ., m) implies xn + i = 0 and ∑ a ij x i = bi, i.e. ith constraint in
j =1
n
the primal is an equation. Also yi = 0 implies xn + i > 0 and ∑ a ij x j < bi. This completes the proof of
j =1
the theorem.

You might also like