Linear Programming

S.Subrahmanyam

LP is a technique for solving a special set of constrained optimization problems

where the objective function is linear and there are one or more linear constraints.

Applications:

1. Managers in multi product firms have to decide the combination of output rates
of several products that would maximize their profit subject to a limited number
of machine hours and worker days.
2. Dairy farmers used the technique to determine the least cost combination of
feeds that would meet minimum nutrition requirements for their animals.
3. Farmers use the technique to determine the allocation of land to different crops
subject to such constraints as fixed land area, differing soil characteristics and
limited water availability

Linearity assumption:

All the relationships in the problem must be linear. This assumption implies that
there are constant returns to scale in production, that the price of output levels, and
that input prices are constant regardless of the amount of input purchase. In other
words production cost per unit of output is constant. Since price is also constant
profit per unit is constant.

Key Concepts
• LP is a technique for solving constrained optimization problem where both
the objective function and the constraints are linear.
• The linearity assumption means that there are constant output and input
prices, constant returns to scale in production, and constant cost and profit
per unit of output.

1
Constrained Optimization

Structuring the problem:

Consider a firm that produces two products, A and B, which require processing
on three different machines. The number of hours available on each machine is
limited. Assume that profit per each unit of output is constant. The problem facing
the firm is to determine the quantities of A and B (QA, QB) that maximize profit, but
not requiring more than the limited number of machine hours available. QA and QB
are the decision variables in the problem. The objective function is:

π = a QA + b QB (1)

Where Π represents total profit and a and b are the profit per unit of A ad B

Solving for QB yields

QB = π/b - a / b QA (2)

Equation 2 describes a straight line where the vertical intercept is determined by the
ratio of total profit to profit per unit of B. The slope of the line is negative of the ratio
a/b or the relative profitability of the two products.

Suppose that Rs. 3 is the profit per unit of A and Rs. 1 is the profit per unit of B.
The profit or objective function is:

π = 3 QA + 1 QB

Which can be written as

2
 QB = π - 3. QA (3)

Equation 3 can be considered as isoprofit equation. For example, π = 90, some

combinations that yield this profit are shown in the table given below. The table also
gives points for profit levels for Rs.120 and Rs.60. These three iso-profit lines are
shown in the following diagram.

Table: Points for iso-profit curves

π = 90 π = 60 π = 120
0 90 0 60 0 120
10 60 10 30 10 90
20 30 15 15 20 60
30 0 20 0 30 30
40 0

140

120

100

80 Series1
Series2
60 Series3

40

20

0
0 10 20 30 40 50

3
 For producing A and B the required processing on three machines, say X, Y, Z.
The hours of time per unit of output that are required on each machine and the
hours of total time available on each machine are given below.

Resource requirements and availability

Machine Hours required per unit Total time available

Product A Product B

X 1 3 90

Y 2 2 80

Z 2 0 60

On machine X each unit of product A required one hour and each unit of B requires
three hours. Thus for any output rate, QA , the number of machine X hours required
1QA. Similarly, for any output rate QB,the number of hours required on machine X is
3QB. Thus the total hours required on machine X is 1QA + 3QB . This total should
not be more than 90. This is the machine X constraint which can be written as:
QA + 3QB ≤ 90
The constraint on machine Y can be written as:
2QA + 2QB ≤ 80
and the constraint on machine Z can be written as:
2QA ≤ 60

In addition, a set of non negativity requirements is added to ensure that the solution
makes economic sense. Therefore, two non-negativity constraints are added:

QA ≥ 0
QB ≥ 0

4
Now the LP problem is complete. It can be stated as:
Maximize π = 3QA + 1QB (objective function)

Subject to 1QA + 3QB ≤ 90 (machine X constraint)

2QA + 2QB ≤ 80 (machine Y constraint)
2QA ≤ 60 (machine Z constraint)

In general, the number of decision variables will be no greater than the resource
constraints.

The feasible region

The next step is to determine the set of output rates QA and QB that can be
produced without violating the constraints. The five constraints are graphically
shown below:

QA +3QB = 90 QA QB
QB = 30 – 1/3QA 0 30
90 0

2QA + 2QB = 80 QA QB
QB = 40 - QA 0 40
40 0

2QA = 60
QA = 30

5
 60

50

40
Series1
30 Series2
Series3
20

10

0
0 20 40 60 80 100

In general, the feasible region consists of all values of the decision variables that
satisfy all the constraints simultaneously. The LP problem is to identify the one
combination (QA, QB ) within the feasible region that yields maximum profit.

Key Concepts

• All LP problems consist of an objective function, one or more resource

constraints, and a non negativity constraint for each decision variable.
• The feasible region consists of all combinations of values for the decision
variables that satisfy all the constraints simultaneously.
• The goal of LP is to find that point in the feasible region that optimizes the
objective function.

Graphic Solution
We have to find the highest iso-profit line that meets the constraints. In other words, at
least one point on the iso-profit line has to be in common with the feasible region. For
finding this point we super-impose the graph of iso-profit lines on the graph of feasible
region. It is found that QA = 30 and QB = 10 provides the solution. Then the total profit is
100 (30 x 3 + 10 x 1). A higher profit line would have no point in common with the
feasible region. Similarly, a lower profit line has points in common with the feasible
region but the profit not maximum.

6
 We can calculate the machine time used for this level of output. Machines Y and
Z are fully utilized whereas 30 hours of machine X are unused. In the language of LP, it
is said that the constraints for Y and Z are binding and X is said to be a non-binding
constraint. The optimal point lies on the binding constraints. If the constraint is binding,
the resource has opportunity cost. If one or more of these hours are reallocated to some
other use, total profit would be reduced.

Table 2: Hours of Machine Time Available and Utilized

Machine Hours of time utilized Available Slack hours

A B Total
Time
X 30 30 60 90 30
Y 60 20 80 80 -
Z 60 - 60 60 -

The solution to LP always occurs at the corner of the feasible region. Hence we
have to evaluate only the corner points. This reduces the number of computations to a
great extent. The slope of the objective function is critical in determining which corner
point is optimal.

Corner Point Quantity of A Quantity of B profit

O 0 0 0
E 0 30 30
F 15 25 70
G 30 10 100
H 30 0 90

Constrained Cost Minimization

Structuring the problem
Consider a milk producer who wants to feed the cows adequately but at a minimum
cost. Suppose he has to feed 40 units of protein, 60 units of calcium and 60 units of
carbohydrates. He has to determine how much of the two feeds he has to use. One ton
of A contains 1 unit of P, 3 units of C ad 1 units of CH. One ton of B contains 1 unit of P,
1 unit of C and 6 units of CH. Let the price of feed A be Rs. 100 per ton and price of
feed B be Rs. 200 per ton. The problem is to find the quantities of the two feeds
purchased so that the feed cost is minimum. The objective function to be minimized is

7
 C = 100 XA + 200 XB

subject to the constraint to the minimum nutritional requirements or constraints.

1 XA + 1 XB ≥ 40 (protein constraint)
3 XA + 1 XB ≥ 60 (calcium constraint)
1 XA + 6 XB ≥ 60 (carbohydrates constraint)
XA ≥ 0
XB ≥ 0

Summary of the data for cost minimization problem.

Feed
A B
Price per ton 100 200

Nutrients per ton

A B Minimum Units Required

Protein 1 1 40
Calcium 3 1 60
Carbohydrates 1 6 60

We find the intercepts for the three constraints.

XA XB
Protein 40 40
Calcium 20 60
Carbohydrates 60 10

The objective function is

C = 100 XA + 200 XB
If C = 1000

8
 XA XB
0 5
10 0

If C = 2000

XA XB
0 10
20 0

The constraints are plotted in a graph. The feasible region extends upward and
to the right of the constraints.

Sensitivity Analysis

Decision makers are often faced with the problems of what – if variety. A
manager might be asked how the process should be changed if the price of labour
increased by 10 percent. That is, how should the mix of labour and capital is altered to
adjust for the higher price of labour? We change the parameters of the problem, solve it
again, and compare the original and new solutions. This is the essence of the sensitivity
analysis. The coefficients of the objective function, the coefficients of the constraints and
/or the resource amounts on the right hand side of the constraints can be changed. Also,
a decision variable or a constraint could be added to or deleted from the problem.

Key Concepts

• The algebraic solution to LP requires that the coordinates of each corner point of
the feasible solution be determined and then the objective function evaluated for
each set of coordinates.
• Sensitivity analysis involves solving LP problem, changing one or more
components of the objective function or the constraints, solving the new problem,
and then comparing the solution with the earlier one.

Special Problems in Linear Programming

9
 We encounter three problems viz., multiple solutions, redundant constraints and
no feasible solution.

Multiple Solutions

If the objective function has the same slope as one of the constraints, the result
will be an infinite number of optimal combinations of the decision variables. This
situation does not pose a problem, and the solution is exactly as outlined as earlier.

Redundant Solutions

Sometimes, one or more constraints are unnecessary or redundant. This does

not present a problem in solving LP.

No Feasible Solution
A serious problem arises when the set of constraints is such that there are no
values of the decision variables that simultaneously satisfy all the constraints.
Consider the following constraints.

2X + 3Y ≥ 120
X + 2Y ≥ 40
As there are no points that satisfy both the conditions, there is no feasible solution.
In large LP problems with many constraints, there is no way of knowing that there is
no solution until the computer programme is run. In such cases, obtaining a feasible
solution means that one or more constraints must be modified. For example,
additional hours of machine time may have to be acquired in order for any
production to take place.

The Dual Problem

When a resource constraint is binding, an opportunity cost is associated with that

resource. Consider a profit – maximization problem. Time available on machines Y
and Z are binding. If one hour of time on either of these machines had to be given

10
up, profit would fall. But the constraint for hours of machine X is not binding. This
means that there is a zero opportunity cost for this resource.

In terms of planning for expansion of facilities, it would be useful to have an

estimate of the value or opportunity cost of time on machines Y and Z. For example,
if the firm could hire one more hour of machine, should it hire more hours of Y or Z?
Suppose it was known that the opportunity cost of one hour of time on machines Y
and Z are Rs.20 and 10 respectively. This means that an additional hour on machine
Y would add Rs.20 to profit in contrast to adding only Rs.10 to profit for an additional
profit on machine Z. The duel LP is a way to determine the opportunity cost of the
resource associated with each constraint.

For every LP problem (called the primal problem), there an associated problem,
referred to as the duel problem. If the objective of the primal problem is
maximization of an objective function, the objective of the duel problem is the
minimization of an associated objective function and vice versa. For example, if the
objective of the primal problem is to maximize production subject to a set of machine
hour constraints, the objective of the duel problem would be to minimize the cost of
the resources (i.e., the machine hours) subject to a constraint on production. The
solution to the duel problem would allocate the same number of machine-hours to
each product as did the primal problem.

Structuring the Duel Problem

Consider the profit maximization problem discussed earlier. This will be referred
to as the primal problem:

Maximize: π = 3 QA + 1 QB

Subject to: 1 QA + 3 QB ≤ 90

2 QA + 2 QB ≤ 80

2 QA + 0 QB ≤ 60

11
 The duel problem is constructed by using the columns of the coefficients of the
primal problem. Define the three new variables Cx, Cy and Cz that represent the
opportunity cost of hours on machines X, Y, and Z, respectively. The objective
function for the duel problem is

Minimize: C = 90 Cx + 80 Cy + 60 Cz

That is, the objective is to minimize the opportunity cost of the firm’s scarce
resources (hours of machine time).

The coefficients of first constraint in the duel problem are the coefficients of QA
from the first column of the primal problem. That is

1 Cx + 2 Cy + 3 Cz ≥ 3

It means that the value of machine hours used to produce one unit of product A
must be at least as great as the profit on that unit of output. Thus one unit of
output requires one hour of time on machine X, two hours on Y, and two hours
on Z, and those hours are value at Cx, Cy,and Cz, respectively.

The second constraint is based on the coefficients in the second column

in the primal problem. Hence

3 Cx + 2 Cy + 0 Cz ≥ 1

Note that output B does not require processing on machine Z, so that

there is no cost associated with that machine. As there are three decision
variables but only two resource constraints, there will be at most two decision
variables with nonzero values.

Solving the Duel Problem

12
 Because there are three variables, the graphic solution would require three-
dimensional analysis and would be extremely difficult. An algebraic approach is
more straightforward. Since it is known that the constraint X on the primal is not
binding, and therefore the opportunity cost of machine X hours is known to be zero
(i.e, Cx =0). Thus the problem reduces to

Minimize: C = 80 Cy + 60 Cz

Subject to 2 Cy + 2 Cz = 3

2 Cy = 1

Cy = 0.5 and Cx = 1, represent the opportunity cost per hour of time on

machines Y and Z, respectively. That is a one hour reduction in time available on
machine Y would reduce profit by Rs.0.5 and on machine Z by Rs. 1. The total
cost is Rs. 100 (80 x 0.5 + 60 x 1).The minimum cost is same as the maximum
rate of profit for the primal problem. That is, at the point of optimal resource
allocation, profit equals the value or the opportunity cost of the resources being
used to generate that profit.

The opportunity cost of any resource is its value to the firm. The term
shadow price is often used in the context of duel programming problems to
describe this opportunity cost. That is, the Ci in the problem are the shadow
prices of the machine hours. As the firm makes plans to add more machine
capacity, it can compare with the shadow price to the market or acquisition price
of additional hours of machine time. If the shadow price exceeds the acquisition
price, profit can be increased by acquiring more hours. Thus, the dual problem
provides direction to management when making decisions about expanding
productive capacity.

13