MB0032 or Complete

Unit 1- Introduction To OR
Introduction
Optimization is the act of obtaining the best result under any given circumstance. In
various practical problems we may have to take many technical or managerial decisions at
several stages. The ultimate goal of all such decisions is to either maximize the desired
benefit or minimize the effort required. We make decisions in our every day life without
even noticing them. Decision-making is one of the main activity of a manager or
executive. In simple situations decisions are taken simply by common sense, sound
judgment and expertise without using any mathematics. But here the decisions we are
concerned with are rather complex and heavily loaded with responsibility. Examples of
such decision are finding the appropriate product mix when there are large numbers of
products with different profit contributions and productional requirement or planning
public transportation network in a town having its own layout of factories, apartments,
blocks etc. Certainly in such situations also decision may be arrived at intuitively from
experience and common sense, yet they are more judicious if backed up by mathematical
reasoning. The search of a decision may also be done by trial and error but such a search
may be cumbersome and costly. Preparative calculations may avoid long and costly
research. Doing preparative calculations is the purpose of Operations research. Operations
research does mathematical scoring of consequences of a decision with the aim of
optimizing the use of time, efforts and resources and avoiding blunders.
Learning Objective 1
Know about Significant Features in Operations Research
Historical Background
During Second World War in United Kingdom a team of scientists from different
disciplines studied the strategic and tactical problems associated with air and land defence
of the country. Their objective was to determine the most effective utilization of limited
military resources to win the battle and the technique they developed was named as
Operations research. After the war, Operations research techniques rapidly developed in
the fields of industrial, academic and government organizations.
The application of Operations research methods helps in making decisions in such
complicated situations. Evidently the main objective of Operations research is to
provide a scientific basis to the decision-makers for solving the problems
involving the interaction of various components of organization, by employing a
team of scientists from different disciplines, all working together for finding a
solution which is the best in the interest of the organization as a whole. The
solution thus obtained is known as optimal decision.
Definition of Operations Research: Churchman, Ackoff and Aruoff have defined
Operations research as the application of scientific methods, techniques and tools
to operation of a system with optimum solutions to the problems. Here Optimum
implies the one, which is best of all possible alternatives. Another definition is that,
Operations research is the use of scientific methods to provide criteria for decisions
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
regarding man, machine, systems involving repetitive operations. This definition is
more general and comprehensive and seems to be more exhaustive than the previous
definition.
Scope of Operations Research (OR)
In general, whenever there is any problem simple or complicated, the OR techniques may
be applied to find the best solution. In this section we shall try to find the scope of OR by
seeing its application in various fields of everyday life.
i) In Defence Operations: In modern warfare the defence operations are carried out
by a number of independent components namely Air Force, Army and Navy. The
activities in each of these components can be further divided in four sub-components
viz.: administration, intelligence, operations and training, and supply. The application
of modern warfare techniques in each of the components of military organizations
requires expertise knowledge in respective fields. Further more, each component
works to drive maximum gains from its operations and there is always a possibility
that strategy beneficial to one component may have an adverse effect on the other.
Thus in defence operations there is a necessity to co-ordinate the activities of various
components which gives maximum benefit to the organization as a whole, having
maximum use of the individual components. The final strategy is formulated by a
team of scientists drawn from various disciplines who study the strategies of different
components and after appropriate analysis of the various courses of actions, the best
course of action, known as optimum strategy, is chosen.
ii) In Industry: The system of modern industries are so complex that the optimum
point of operation in its various components cannot be intuitively judged by an
individual. The business environment is always changing and any decision useful at
one time may not be so good some time later. There is always a need to check the
validity of decisions continually, against the situations. The industrial revolution with
increased division of labour and introduction of management responsibilities has
made each component an independent unit having their own goals. For example:
Production department minimize cost of production but maximizes output. Marketing
department maximizes output but minimizes cost of unit sales. Finance department
tries to optimize capital investment and personnel department appoints good people at
minimum cost. Thus each department plan their own objectives and all these
objectives of various department or components come to conflict with each other and
may not conform to the overall objectives of the organization. The application of OR
techniques helps in overcoming this difficulty by integrating the diversified activities
of various components so as to serve the interest of the organization as a whole
efficiently.
OR methods in industry can be applied in the fields of production, inventory controls
and marketing, purchasing, transportation and competitive strategies etc.
iii) Planning: In modern times it has become necessary for every government to have
careful planning, for economic development of the country. OR techniques can be
fruitfully applied to maximize the per capita income, with minimum sacrifice and
time. A government can thus use OR for framing future economic and social policies.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
iv)Agriculture: With increase in population there is a need to increase agriculture
output. But this cannot be done arbitrarily. There are a number of restrictions under
which agricultural production is to be studied. Therefore there is a need to determine a
course of action, which serves the best under the given restrictions. The problem can
be solved by the application of OR techniques.
v) In Hospitals: The OR methods can be used to solve waiting problems in out-
patient department of big hospitals. The administrative problems of hospital
organization can also be solved by OR techniques.
vi) In Transport: Different OR methods can be applied to regulate the arrival of
trains and processing times, minimize the passengers waiting time and reduce
congestion, formulate suitable transportation policy, reducing the costs and time of
trans-shipment.
vii) Research and Development: Control of R and D projects, product introduction
planning etc. and many more applications.
Important features of OR are:
i. It is System oriented: OR studies the problem from over all point of view of
organizations or situations since optimum result of one part of the system may not be
optimum for some other part.
ii. It imbibes Inter disciplinary team approach. Since no single individual can have a
thorough knowledge of all fast developing scientific know-how, personalities from
different scientific and managerial cadre form a team to solve the problem.
iii. It makes use of Scientific methods to solve problems.
iv. OR increases the effectiveness of a management Decision making ability.
v. It makes use of computer to solve large and complex problems.
vi. It gives Quantitative solution.
vii. It considers the human factors also.
Phases of Operations Research
The scientific method in OR study generally involves the following three phases:
i) Judgment Phase: This phase consists of
a) Determination of the operation.
b) Establishment of the objectives and values related to the operation.
c) Determination of the suitable measures of effectiveness and
d) Formulation of the problems relative to the objectives.
ii) Research Phase: This phase utilizes
a) Operations and data collection for a better understanding of the problems.
b) Formulation of hypothesis and model.
c) Observation and experimentation to test the hypothesis on the basis of
additional data.
d) Analysis of the available information and verification of the hypothesis using
pre-established measure of effectiveness.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
e) Prediction of various results and consideration of alternative methods.
iii) Action Phase: It consists of making recommendations for the decision process by
those who first posed the problem for consideration or by anyone in a position to make a
decision, influencing the operation in which the problem is occurred.
Types of operation Research Models
A model is a representation of the reality. It is an idealized representation or abstraction
of a real life system. The objective of the model is to identify significant factors and their
interrelationship. A model is helpful in decision making as it provides a simplified
description of complexities and uncertainties of a problem in logical structure. Major
advantage of models is it does not interfere with real system.
A broad classification of OR models is
a) Physical Model b) Mathematical or symbolic model c) Models by nature of
Environment and d) Models by the extent of generality.
a. Physical Modes include all form of diagrams, graphs and charts. They are designed
to deal with specific problems. They bring out significant factors and inter-
relationship in pictorial firm so as to facilitate analysis. There are two types i) Iconic
models and ii) Analog models.
Iconic model is an image of an object or system, represented on a small scale. These
models can simulate the actual performance of a product.
Analog models are small physical systems that has similar characteristics and work
l ike an objects it represents Eg: Toy
b. Mathematical Model or symbolic models employ a set of mathematical symbols
to represent the decision variable of the system. The variables are related by
mathematical system Eg: Allocation, sequencing, replacement models etc.
c. By nature of Environment
We have i) Deterministic model in which every thing is defined and the results are
certain. Eg: Eo2 model ii) Probabilistic Models in which the input and output
variables follow a probability distribution Eg: Games Theory.
d. By the extent of Generality: The two models belonging to this class are i) General
models can be applied in general and does not pertain to one problem only. Eg: Linear
programming ii) Specific Model is applicable under specific condition only. Eg: Sales
response curve or equation as a function of advertising is applicable in the marketing
function alone.
Understand the Methodology of Operations Research
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Methodology of Operations Research
The basic dominant characteristic feature of operations research is that it employs
mathematical representations or model to analyze problems. This distinctive approach
represents an adaptation of the scientific methodology used by the physical sciences. The
scientific method translates a real given problem into a mathematical representation
which is solved and retransformed into the original context. The OR approach to problem
solving consists of the following steps:
Definition of the problem. 1.
Construction of the model. 2.
Solution of the model. 3.
Validation of the model. 4.
Implementation of the final result. 5.
Definition of the problem
The first and the most important requirement is that the root problem should be identified
and understood. The problem should be identified properly, this indicates three major
aspects: (1) a description of the goal or the objective of the study, (2) an identification of
the decision alternative to the system, and (3) a recognition of the limitations, restrictions
and requirements of the system.
Construction of the model
Depending on the definition of the problem, the operations research team should decide
on the most suitable model for representing the system. Such a model should specify
quantitative expressions for the objective and the constraints of the problem in terms of
its decision variables. A model gives a perspective picture of the whole problem and
helps tackling it in a well-organized manner. If the resulting model fits into one of the
common mathematical models, a convenient solution may be obtained by using
mathematical techniques. If the mathematical relationships of the model are too complex
to allow analytic solutions, a simulation model may be more appropriate. There are
various types of models which can be constructed under different conditions.
Solution of the model
Once an appropriate model has been formulated, the next stage in the analysis calls for its
solution and the interpretation of the solution in the context of the given problem. A
solution to a model implies determination of a specific set of decision variables that
would yield an Optimum solution. An Optimum solution is one which maximize or
minimize the performance of any measure in a model subject to the conditions and
constraints imposed on the model.
Validation the model
A model is a good representative of a system, then the Optimal solution must improve the
systems performance. A common method for testing the validity of a model is to
compare its performance with some past data available for the actual system. The model
will be valid if under similar conditions of inputs, it can reproduce the past performance
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
of the system. The problem here is that there is no assurance that future performance will
continue to duplicate past behaviour. Also, since the model is based on careful
examination of past data, the comparison should always reveal favorable results. In some
instances this problem may be overcome by using data from trial runs of the system. It
must be noted that such a validation method is not appropriate for non-existent systems,
since data will not be available for comparison.
Implementation of the final result
The optimal solution obtained from a model should be applied practice to improve the
performance of the system and the validity of the solution should be verified under
changing conditions. It involves the translation of these results into detailed operating
instructions issued in an understandable form to the individuals who will administer and
operate the recommended system. The interaction between the operations research team
and the operating personnel will reach its peak in this phase.
Techniques or Tools of OR
Linear Programming: 1.
It is used to find a solution for optimizing a given objective. Objectives may be to
maximize profit or minimize cost. Both objective function and constraints must be
capable of being expressed as linear expression of decision variables. Its various
uses will be seen in Chapter-2.
Inventory control Methods: 2.
The production, purchasing and material managers are always confronted with
question of when to buy, how much to buy and how much to keep in stock.
Inventory models aims at optimizing inventory levels.
Goal Programming 3.
Single objective function is taken in the linear programming and all other factors
are considered as constraints, but in actual practice there may be number of
important objective functions. Goal programming has several objective functions,
each having a target value and programme models are developed to minimize
deviation from these targets.
The tools, namely, queuing model, sequence model, transportation and assignment 4.
model, network analysis are discussed in detail in later chapters.
Learn about Structure of Mathematical Model
The Structure of Mathematical Model
Many industrial and business situations are concerned with planning activities. In each
case of planning, there are limited sources, such as men, machines, material and capital at
the disposal of the planner. One has to make decision regarding these resources in order
to either maximize production, or minimize the cost of production or maximize the profit
etc. These problems are referred to as the problems of constrained optimization. Linear
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
programming is a technique for determining an optimal schedule of interdependent
activities, for the given resources. Programming thus means planning and refers to the
process of decision-making regarding particular plan of action amongst several available
alternatives.
Any business activity of production activity to be formulated as a mathematical model
can best be discussed through its constituents; they are:
- Decision Variables,
- Objective function,
- Constraints.
Decision variables and parameters
The decision variables are the unknowns to be determined from the solution of the model.
The parameters represent the controlled variables of the system.
Learn about the Significance of Objective function
Objective functions
This defines the measure of effectiveness of the system as a mathematical function of its
decision variables. The optimal solution to the model is obtained when the corresponding
values of the decision variable yield the best value of the objective function while
satisfying all constraints. Thus the objective function acts as an indicator for the
achievement of the optimal solution.
While formulating a problem the desire of the decision-maker is expressed as a function
of n decision variables. This function is essentially a linear programming problem (i.e.,
each of its item will have only one variable raise to power one). Some of the Objective
functions in practice are:
- Maximization of contribution or profit
- Minimization of cost
- Maximization of production rate or minimization of production time
- Minimization of labour turnover
- Minimization of overtime
- Maximization of resource utilization
- Minimization of risk to environment or factory etc.
Constraints
To account for the physical limitations of the system, the model must include constraints,
which limit the decision variables to their feasible range or permissible values. These are
expressed in the form of constraining mathematical functions.
For example, in chemical industries, restrictions come from the government about
throwing gases in the environment. Restrictions from sales department about the
marketability of some products are also treated as constraints. A linear programming
problem then has a set of constraints in practice.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The mathematical models in OR may be viewed generally as determining the values of
the decision variables x
J
, J = 1, 2, 3, n, which will optimize Z = f (x
1
, x
2
, - x
n
).
Subject to the constraints:
g
i
(x
1
, x
2
x
n
) ~ b
i
, i = 1, 2, - m
The function f is called the objective function, where g
i
~ b
i
, represent the i
th
constraint for i = 1, 2, 3 - m where b
i
is a known constant. The
constraints x
j
2 0 are called the non-negativity condition, which restrict the variables to
zero or positive values only.
Diet Problem
Formulate the mathematical model for the following:
Vitamin A and Vitamin B are found in food 1 and food 2. One unit of food 1
contains 5 units of vitamin A and 2 units of vitamin B. One unit of food 2 contains 6
units of vitamin A and 3 units of vitamin B. The minimum daily requirement of a
person is 60 units of vitamin A and 80 units of Vitamin B. The cost per one unit of
food 1 is Rs. 5/- and one unit of food 2 is Rs. 6/-. Assume that any excess units of
vitamins are not harmful. Find the minimum cost of the mixture (of food1 and food2)
which meets the daily minimum requirements of vitamins.
Mathematical Model of the Diet Problem: Suppose x1 = the number of units of food1 in
the mixture, x2 = the number of units of food2 in the mixture.
Now we formulate the constraint related to vitamin-A. Since each unit of food 1
contains 5 units of vitamin A, we have that x1 units of food 1 contains 5x1 units of
vitamin A. Since each unit of food 2 contains 6 units of vitamin A, we have that x2
units of food 2 contains 6x2 units of vitamin A. Therefore the mixture contains 5x1 +
6x2 units of vitamin - A. Since the minimum requirement of vitamin A is 60 units, we
have that
Now we formulate the constraint related to vitamin B. Since each unit of food 1
contains 2 units of vitamin B we have that x1 units of food 1 contains 2x1 units of
vitamin - B. Since each unit of food 2 contains 3 units of vitamin B, we have that x2
units of food 2 contains 3x2 units of vitamin B. Therefore the mixture contains 2x1 +
3x2 units of vitamin B. Since the minimum requirement of vitamin B is 80 units, we
have that
Next we formulate the cost function. Given that the cost of one unit of food 1 is Rs. 5/-
and one unit of food 2 is Rs. 6/-. Therefore x1 units of food1 costs Rs. 5x1, and x2
units of food 2 costs Rs. 6x2. Therefore the cost of the mixture is given by Cost = 5x1 +
6x2. If we write z for the cost function, then we have z = 5x1 + 6x2. Since cost is to be
minimized, we write min z = 5x1 + 6x2.
Since the number of units (x1 or x2) are always non-negative we have that
Therefore the mathematical model is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Limitations of OR
The limitations are more related to the problems of model building, time and money
factors.
i) Magnitude of computation: Modern problem involve large number of variables and
hence to find interrelationship, among makes it difficult.
ii) Non quantitative factors and Human emotional factor cannot be taken into
account.
iii) There is a wise gap between the managers and the operation researches
iv) Time and Money factors when the basic data is subjected to frequent changes
then incorporation of them into OR models is a costly affair.
v) Implementation of decisions involves human relations and behaviour.
Summary
The OR approach needs to be equally developed in various agricultural problems on a
regional or international basis. With the explosion of population and consequent shortage
of food, every country faces the problem of optimum allocation of land in various crops
in accordance with climate conditions and available facilities. The problem of optimal
distribution of water from a resource like a reservoir for irrigation purposes is faced by
each developing country, and a good amount of scientific work can be done in this
direction.
Unit -2-Linear Programming
Introduction
One of the most important problems in management decision is to allocate limited and
scarce resource among competing agencies in the best possible manner. Resources may
represent man, money, machine, time, technology on space. The task of the management
is to derive the best possible output (or set of outputs) under given restraints on resources.
The output may be measured in the form of profits, costs, social welfare, effectiveness,
etc. In many situations the output (or the set of outputs) can be expressed as a linear
relationship among a number of variables. The amount of available resources can also be
expressed as a linear relationship among some system variables. The management
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
problem may be to optimize (maximize or minimize) the out-put or the objective function
subject to the set of constraints An optimization problem in which both the objective
function and the constraints are represented by linear forms is a problem in linear
programming.
Learn Formulating LPP and Observing the Feasible Region
Requirements of L.P.P.
i. Decisions variables and their relationship
ii. Well defined objective function
iii. Existence of alternative courses of action
iv. Non-negative conditions on decision variables.
Basic assumptions of L.P.P
1. Liniarity: Both objective function and constraints must be expressed as linear
inequalities.
2. Deterministic: All coefficient of decision variables in the objective and constraints
expressions
should be known and finite.
3. Additivity: The value of objective function for the given values of decision variables
and the
total sum of resources used, must be equal to sum of the contributions earned from each
decision variable and the sum of resources used by decision variables respectively.
4. Divisibility: The solution of decision variables and resources can be any non-negative
values
including fractions.
Linear Programming
The Linear Programming Problem (LPP) is a class of mathematical programming in
which the functions representing the objectives and the constraints are linear. Here, by
optimization, we mean either to maximize or minimize the objective functions. The
general linear programming model is usually defined as follows:
Maximize or Minimize
Z = c
1
x
1
+ c
2
x
2
+ - - + c
n
x
n
subject to the constraints,
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Where c
j
, b
i
and a
ij
(i = 1, 2, 3, .. m, j = 1, 2, 3 - n) are constants determined from
the technology of the problem and x
j
(j = 1, 2, 3 - n) are the decision variables. Here ~
is either (less than), 2 (greater than) or = (equal). Note that, in terms of the above
formulation the coefficient c
j
, a
ej
, b
j
are interpreted physically as follows. If b
i
is the available amount of resources i, where a
ij
is the amount of resource i, that must be allocated to each unit of activity j, the worth
per unit of activity is equal to c
j
.
Canonical forms:
The general Linear Programming Problem (LPP) defined above can always be put in the
following form which is called as the canonical form:
Maximise Z = c
1
x
1
+c
2
x
2
+ + c
n
x
n
Subject to
The characteristics of this form are:
all decision variables are non-negative. 1.
all constraints are of 2.
type.
the objective function is of the maximization type. 3.
Any LPP can be put in the cannonical form by the use of five elementary transformations:
1. The minimization of a function is mathematically equivalent to the maximization
of the negative expression of this function. That is, Minimize Z = c
1
x
1
+ c
2
x
2
+ .
+ c
n
x
n
is equivalent to
Maximize Z = c
1
x
1
c
2
x
2
c
n
x
n
.
Any inequality in one direction may be changed to an inequality in the 1.
opposite direction by multiplying both sides of the inequality by 1.
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
3. An equation can be replaced by two inequalities in opposite direction. For
example, 2x
1
+3x
2
= 5 can be written as
4. An inequality constraint with its left hand side in the absolute form can be
changed into two regular inequalities. For example: | 2x
1
+3x
2
| 5 is equivalent to
2x
1
+3x
2
5 and 2x
1
+3x
2
5 or 2x
1
3x
2
5.
5. The variable which is unconstrained in sign (i.e., 0, 0 or zero) is
equivalent to the difference between 2 non-negative variables. For example, if x is
unconstrained in sign then x
Examples Of A Linear Programming Problem:
Example 1: A firm engaged in producing 2 models, viz., Model A and Model B,
performs only 3 operations painting, assembly and testing. The relevant data are as
follows:
Unit Sale Price Hours required for each unit
Assembly Painting Testing
Model A Rs. 50.00
Model B Rs. 80.00
1.0
1.5
0.2
0.2
0.0
0.1
Total number of hours available each week are as under assembly 600, painting 100,
testing 30. The firm wishes to determine the weekly product-mix so as to maximize
revenue.
Solution: Let us first write the notations as under:
Since the objective (goal) of the firm is to maximize its revenue, the model can be stated
as follows:
The objective function, Z = 50x
1
+ 80x
2
is to be maximized subject to the constraints
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Example 2: A milk distributor supplier milk in bottles to houses in three areas A, B, C in
a city. His delivery charges per bottle is 30 paise in area A, 40 paise in area B and 50
paise in area C. He has to spend on an average, 1 minute to supply one bottle in area A, 2
minutes per bottle in area B and 3 minutes per bottle in area C. He can spare only 2 hours
30 minutes for this milk distribution but not more than one hour 30 minutes for area A
and B together. The maximum number of bottles he can deliver is 120. Find the number
of bottles that he has to supply in each area so as to earn the maximum. Construct a
mathematical model.
Solution: The decision variables of the model can be defined as follows:
x
1
: Number of bottles of milk which the distributor supplies in Area A.
x
2
: Number of bottles of milk which the distributor supplies in Area B.
x
3
: Number of bottles of milk which the distributor supplies in Area C.
The objective :
Maximize Z = in rupees.
constraints:
1. Maximum number of milk bottles is 120, that is x
1
+x
2
+x
3
120.
Since he requires one minute per bottle in area A, 2 minutes per bottle in area B 1.
and 3 minutes per bottle in area C and he cannot spend more than 150 minutes for
the work,
1.x
1
+ 2.x
2
+ 3.x
3
150.
Further, since he cannot spend more than 90 minutes for areas A and B. 1.x
1
+2.x
2
2.
90.
Non-negativity x
1
3.
2 0, x
2
2 0.
The problem can now be stated in the standard L.P. form is
Maximize Z = 0.3x
1
+ 0.4x
2
+ 0.5x
3
Subject to
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Example 3: An oil company has two units A and B which produce three different grades
of oil super fine, medium and low grade oil. The company has to supply 12, 8, 24 barrels
of super fine, medium and low grade oils respectively per week. It costs the company Rs.
1,000 and Rs. 800 per day to run the units A and B respectively. On a day Unit A
produces 6, 2 and 4 barrels and the unit B produces 2, 2 and 12 barrels of super fine,
medium and low grade oil per day. The manager has to decide on how many days per
week should each unit be operated in order to meet the requirement at minimum cost.
Formulate the LPP model.
Solution: The given data can be presented in summary as follows:
Product Capacity Requirements
Super fine
Medium
Low grade
Cost
Unit A Unit B
12
8
24
6
2
4
Rs. 1,000
2
2
12
Rs. 800
Let x
1
and x
2
be the number of days the units A and B be operated per week respectively.
Then the objective of the manager is to,
Minimize the cost function
Graphical Analysis
Linear programming with 2 decision variables can be analysed graphically. The graphical
analysis of a L.P.P. is illustrated with the help of the following example:
Maximize Z = 700 x
1
+500 x
2
Let the horizontal axis represent x
1
and the vertical axis x
2
. First we draw the line
4x
1
+ 3x
2
= 210. (by replacing the inequality symbols by the equality) which meets the x
1
-
axis at the point A (52.50, 0) (put x
2
= 0 and solve for x
1
in 4x
1
+ 3x
2
= 210) and the x
2
axis at the point B(0, 70) (put x

1
= 0 in 4x
1
+ 3x
2
= 210 and solve for x
2
).
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Any point on the line 4x
1
+3x
2
= 210 or inside the shaded portion will satisfy the
restriction of the inequality, 4x
1
+3x
2
210. Similarly the line 2x
1
+x
2
= 90 meets the x
1
-axis at the point C(45, 0) and the x
2
axis at the point D(0, 90).

Combining we can sketch the area as follows:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The 3 constraints including non-negativity are satisfied simultaneously in the shaded
region OCEB. This region is called feasible region.
Some Basic Definitions
Definition: Any non-negative value of (x
1
, x
2
) (i.e.: x
1
2 0, x
2
2 0) is a feasible solution of the LPP if it satisfies all the constraints. The collection of all
feasible solutions is known as the feasible region.
Definition: A set X is convex if for any points x
1
, x
2
in X, the line segment joining these
points is also in X.
. By convention, a set containing only a single
point is also a convex set.
Definition: A linear inequality in two variables is known as a half plane. The
corresponding equality or the line is known as the boundary of the half- plane.
Definition: A convex polygon is a convex set formed by the inter-section of finite
number of closed half-planes.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Note: The objective function is maximized or minimized at one of the extreme points
which is the Optimum solution. Extreme points are referred to as vertices or corner points
of the convex regions.
Definition: A redundant constraint is a constraint which does not affect the feasible
region.
Definition: A basic solution of a system of m equations and n variables (m < n) is a
solution where at least n-m variables are zero.
Definition: A basic feasible solution of a system of m equations and n variables (m < n)
is a solution where m variables are non-negative (2 0) and n-m variables are zero.
Definition: Any feasible solution that optimizes the objective function is called an
optimal feasible
solution.
Example: Find all basic solutions for the system x
1
+ 2x
2
+ x
3
= 4, 2x
1
+ x
2
+ 5x
3
= 5.
Solution: Here A = , X = and b = .
i) If x
1
= 0, then the basis matrix is B = . In this case 2x
2
+ x
3
= 4, x
2
+ 5x
3
= 5.
If we solve this, then x
2
= and x
3
= . Therefore x
2
= , x
3
= is a basic feasible
solution.
ii) If x
2
= 0, then the basis matrix is B = . In this case, x
1
+ x
3
= 4, 2x
1
+ 5x
3
=
5.If we solve this, then x
1
= 5 and x
3
= -1. Therefore x
1
= 5, x
3
= -1 is a basic solution.
(Note that this solution is not feasible, because x
3
= -1 < 0).
iii) If x
3
= 0, then the basis matrix is B = . In this case, x
1
+ 2x
2
= 4.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
2x
1
+ x
2
= 5. If we solve this, then x
1
= 2, and x
2
= 1. Therefore x
1
= 2, x
2
= 1 is a
basic feasible solution.
Therefore (i) (x
2
, x
3
) = (5/3, 2/3), (ii) (x
1
, x
3
) = (5, -1), and
(iii) (x
1
, x
2
) = (2, 1) are only the collection of all basic solutions.
Learn about the Graphical Method of Solving the Linear Programming Problem
Graphical Methods To Solve The Linear Programming Problems
A LPP with 2 decision variables x
1
and x
2
can be solved easily by graphical method. We
consider the x
1
x
2
plane where we plot the solution space, which is the space enclosed
by the constraints. Usually the solution space is a convex set which is bounded by a
polygon; since a linear function attains extreme (maximum or minimum) values only on
boundary of the region, it is sufficient to consider the vertices of the polygon and find the
value of the objective function in these vertices. By comparing the vertices of the
objective function at these vertices, we obtain the optimal solution of the problem.
The method of solving a LPP on the basis of the above analysis is known as the graphical
method. The working rule for the method is as follows:
Working Rule:
Step I: Write down the equations by replacing the inequality symbols by the equality
symbol in the given constraints.
Step II: Plot the straight lines represented by the equations obtained in step I.
Step III: Identify the convex polygon region relevant to the problem. We must decide on
which side of the line, the half-plane is located.
Step IV: Determine the vertices of the polygon and find the values of the given objective
function Z at each of these vertices. Identify the greatest and least of these values. These
are respectively the maximum and minimum value of Z.
Step V: Identify the values of (x
1
, x
2
) which correspond to the desired extreme value of Z.
This is an optimal solution of the problem.
Example 4: We can solve the L.P.P. discussed in Example I.
Maximize Z = 50x
1
+ 80x
2
Subject to the constraints
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Let the horizontal axis represent x
1
and the vertical axis x
2
. Plot the constraint lines and
mark the feasibility region as has been shown in the figure.
Feasible region of the two dimensional LPP
Any point on the thick line or inside the shaded portion will satisfy all the restrictions of
the problem. Then ABCDE is the feasibility region carried out by the constraints
operating on the objective function. This depicts the limits within which the values of the
decision variables are permissible. The inter-section points C and D can be solved by the
linear equations x
2
= 30; x
1
+ 1.5 x
2
= 600, and 0.2x
1
+ 0.2x
2
= 100 and x
1
+ 1.5x
2
= 600
i.e. C (150, 300) and D (300, 180).
After doing this, the next step is to maximise revenues subject to the above shaded area.
We work out the revenues at different corner points as tabulated below:
At point Feasible solution of the product-mix Corresponding revenue Total revenue
x 1 x2 From x1 From x2
A
B
C
D
E
0
0
150
300
500
0
300
300
180
0
0
0
7500
15000
25000
0
2400
24000
14,400
0
0
24000
31500
29400
25,000
From the above table we find that revenue is maximum at Rs. 31,500 when 150 units of
x
1
and 300 units of x
2
are produced.
Example 5: For conducting a practical examination, the chemistry department of a
college requires 10, 12 and 7 units of three chemicals X, Y, Z respectively. The chemicals
are available in two types of boxes: Box A, Box B. Box A contains 3, 2 and 1 units of X,
Y, Z respectively and costs Rs. 300. Box B contains 1, 2 and 2 units of X, Y, Z
respectively and costs Rs. 200. Find how many boxes of each type should be bought by
the department so that the total cost is minimum.
Solution: First, we summarize the given data in the following table:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Units
Units in Box A Units in Box B Units required
X
Y
Z
3
2
1
1
2
2
10
12
7
Cost Rs. 300 Rs. 200
Let x
1
be the number of boxes of A-type to be bought and x
2
be the number of boxes of B-
type. Then the total cost is,
Z = 300x
1
+ 200x
2
.
From the details tabulated in the table, we find that x
1
and x
2
are subject to the following
constraints:
Now, we consider the lines L
1
: 3x
1
+ x
2
= 10, L
2
: 2x
1
+ 2x
2
= 12 L
3
: x
1
+ 2x
2
= 7. These
lines are shown in fig.
We note that for the co-ordinates (x
1
, x
2
) of a point satisfy the inequalities. The convex
region bounded by these lines and the co-ordinate axes is an unbounded region, this is
shaded in fig. We check that a point (x
1
, x
2
) that lies inside or on the boundary lines of
this region satisfies the conditions x
1
2 0, x
2
2 0 and the constraints.
We find that the vertices for the region of interest here are P, Q, R, S. Where P is the
point at which L meets the x
2
axis, Q is the point of inter-section of L
1
and L
2
, R is the
point of inter-section of L
2
and L
3
and S is the point at which L
3
meets the x
1
axis. We
find that P(0, 10), Q(2, 4), R(5, 1) and S(7, 0).:
At P (0, 10), Z = 300 0 + 200 10 = 2000
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
At Q (2, 4), Z = 300 2 + 200 4 = 1400
At R (5, 1), Z = 300 5 + 200 1 = 1700
At S (7, 0), Z = 300 7 + 200 0 = 2100
Evidently, Z is minimum at the vertices Q (2, 4) for which x
1
= 2, x
2
= 4. Thus the cost is
minimum if 2 boxes of type A and 4 boxes of type B are bought. The minimum cost is
Rs. 1400.
Examples 6 on mixed constraints LP problem: By using graphical method, find the
maximum and minimum values of the function Z = x 3y where x and y are non-negative
and are subject to the following conditions:
Solution: First, we write the constraints (conditions) to be satisfied by x, y in the
following standard (less than or equal) form:
Now, consider the equations:
3x 4y = 19, 2x y = 9, 2x + y = 15, x + y = 3 which represents straight lines in the
xy plane. Let us denote them by L
1
, L
2
, L
3
and L
4
respectively. These are shown in fig.:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
From the figure, we note that the lines L
1
, L
2
, L
3
and L
4
form a quadrilateral ABCD that
lies in the first quadrant of the xy plane. We readily see that the region bounded by this
quadrilateral is convex. As such, the points (x, y) that lie within or on the boundary lines
of this quadrilateral satisfy the inequalities x 2 0, y 2 0 and the constraints. The co-
ordinates of the vertices A, B, C, D of the quadrilateral are obtained by solving equations
taken two of them at a time, we find that A (1, 4), B (5, 1), C (6, 3), D (4, 7)
we get the solution
Z
at A(1, 4)
= 1 34 = 11
Z
at B(5, 1)
= 5 31 = 2
Z
at C(6, 3)
= 6 33 = 3
Z
at D(4, 7)
= 4 37 = 17
Evidently, Z is maximum at the vertex B and minimum at the vertex D. The maximum
value of Z is Z
at B(5, 1)
= 2, which corresponds to x = 5, y = 1, and the minimum values of Z
is 17 at D(4, 7) which corresponds to x = 4, y = 7.
Examples 7: Use the graphical method to solve the following LP problem:
Maximize Z = 7x
1
+3x
2
Solution: Rewriting the given constraints as follows:
Note: The equation is called intercept form of the straight line. Here a and b are
the distance from orgin to the intersection points on the co-ordinate axes.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Graph each constraint by first treating it as a linear equation. Then use the inequality
condition of each constraint to make the feasible region as shown in fig.:
The co-ordinates of the extreme points of the feasible region are
and . The value of the objective function at each of these extreme points is
as follows:
Extreme point Co-ordinates (x1, x2) Objective function value
Z= 7x1 + 3x2
A
B
C
7
22
9/2
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The maximum value of the objective function Z= 22 occurs at the extreme points .
Hence the optimal solution to the given LP problem is and Max. Z = 22.
In linear programming problems may have:
i) a unique optimal solution or
ii) many number of optimal solutions or
iii) an unbounded solution or
iv) no solutions.
Example 8: Maximize Z = 100x
1
+ 40x
2
Subject to
Solutions: The given constraints can be rewritten as
The values of (x
1
x
2
) at the points are 0(0, 0), A(200, 0) B(125, 187.5) and C(0, 250). The
feasible region is OABC. The values of Z at the points are
Z
at O(00)
= 0
Z
at A(200, 0)
= 20000
Z
at B(125, 187.5)
= 20000
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Z
at C(0, 250)
= 10,000
Thus the maximum value of Z occurs at 2 vertices at A and B. Any point on the line
joining A and B will also give the same maximum value of Z. Therefore, there are infinite
number of feasible solutions which yield the same maximum value of Z.
Suppose a linear programming problem has an unbounded feasible solution space.
If the set of all values of the objective function at different feasible solutions is not
bounded above (respectively, bounded below), and if the problem is a maximization
(respectively, minimization) problem, then we say that the given problem has an
unbounded solution.
In the following, we present an example with unbounded solution.
Example 9 for Unbounded Solution:
Maximize Z = 2x
1
+3x
2
Subject to
The intersection point A of the straight lines x
1
x
2
= 2 and x
1
+x
2
= 4 is A(3, 1). Here the
solution space is unbounded. The vertices of the feasible region are A(3, 1) and B (0, 4).
Value of objective at these vertices are
Z
atA(31)
= 23+31 = 9
Z
at B(0, 4)
= 20+43 = 12.
But there are points in the convex region for which Z will have much higher values. For
example E (10, 9) lies in the shaded region and the value of Z there at 47. In fact, the
maximum values of Z occurs at infinity. Thus the problem has an unbounded solutions.
Example 10 for Inconsistent:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Maximize Z = 4x
1
+3x
2
Subject to
There being no point (x
1
, x
2
) common to both the shaded regions, the LPP cannot be
solved. Hence the solution does not exist, since the constraints are inconsistent.
Example 11 for redundant Constraint:
A company making cold drinks has 2 bottling plants located at towns T
1
and T
2
. Each
plant produces three drinks A, B and C and their production capacity per day is shown
below:
Cold drinks Plant at
T1 T2
A
B
C
6000
1000
3000
2000
2500
3000
The marketing department of the company forecasts a demand of 80,000 bottles of A,
22,000 botles of B and 40,000 bottles of C during the month of June. The operating costs
per day of plants at T
1
and T
2
are Rs. 6,000 and Rs. 4,000 respectively. Find the number
of days for which each plant must be run in June so as to minimize the operating costs
while meeting the market demand.
Solution: Let the plants at T
1
and T
2
be run for x
1
and x
2
days.
Then the objective is to minimize the operation costs.
Minimum of Z = 6000 x
1
+ 4000 x
2
.
Constraints on the demand for the 3 cold drinks are
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Thus the LPP is to minimize the objective function subject to the constraints (i), (ii) and
(iii). The solution space is unbounded. The constraint (iii) is dominated by the constraints
(i) and (ii) and hence does not affect the solution space. Such a constraint 3000 x
1
+ 3000
x
2
2 40000 is called the redundant constraint.
The values of the convex region A,
B,
C are A (22, 0), B (12, 4) and C (0, 40). The values of the objective function Z at the
vertices are
Z
at A
= 132000
Z
at B
= 88,000
Z
at
C
= 1,60,000
Thus the minimum value of Z is Rs. 80,000 and it occurs at B. Hence the optimal solution
to the problem is x
1
= 12 days, x
2
= 4 days.
Example 12: Final the maximum and minimum value of Z = 2x + 3y
Subject to
Solution: Any point (x, y) satisfies the conditions x 2 0, y 2 0 lies in the first quadrant
only.
The desired point (x, y) lies with in the feasible convex region ABCDE.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Its vertices are A (3, 3) B (10, 3) C (20, 10), D (18, 12) and B (12, 12). The values of Z at
the five vertices are
Z
at
A (3, 3)
= 2 3 + 3 3 =15
Z
at B (20, 3)
= 49
Z
at C (20, 10)
= 70
Z
at D (18, 12)
= 72
Z
zt E (12,12)
= 60
Since the maximum value of Z is 72 which occurs at the vertix D (18, 12). Therefore the
solution of the LPP is x = 18, y = 12 and the minimum value of z is 15 at x = 3, y = 3.
Summary
In LPP we first identify the decision variables which are some economic or physical
quantities whose values are of interest to the management. The problems must have a
well-defined objective function expressed in terms of the decision variable. The objective
function may have to be maximized when it expresses the profit or contribution. In case
the objective function indicates a cost, it has to be minimized. The decision variables
interact with each other through some constraints. These constraints occur due to limited
resources, stipulation on quality, technical, legal or variety of other reasons. The objective
function and the constraints are linear functions of the decision variables. A LPP with two
decision variables can be solved graphically. Any non-negative solution which satisfies
all the constraints is known as a feasible solution of the problem. The collection of all
feasible solutions is known as a feasible region. The feasible region of a LPP is a convex
set. The value of the decision variables which maximise or minimize the objectives
function is located on the extreme point of the convex set formed by the feasible
solutions. Sometimes the problem may be infeasible indicating that no feasible solution
of the problem exists.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
.
Unit - 3- Simplex Method
Introduction
The simplex method provides an efficient technique which can be applied for solving
LPP of any magnitude involving two or more decision variables. In this method the
objective function used to control the development and evaluation of each feasible
solution to the problem.
The simplex algorithm is an iterative procedure for finding the optimal solution to a
linear programming problem. In the earlier methods, if a feasible solution to the problem
exists, it is located at a corner point of the feasible region determined by the constraints of
the system. The simplex method, according to its iterative search, selects this optimal
solution from among the set of feasible solutions to the problem. The efficiency of this
algorithm is, because it considers only those feasible solutions which are provided by the
corner points, and that too not all of them. We consider a minimum number of feasible
solutions to obtain an optimal solution.
Objective 1
Learn to write the Standard form of LPP from the Given Hypothesis
The Standard Form of LPP
The characteristics of the standard form are:
All constraints are equations except for the non-negativity condition which remain 1.
inequalities only.
The righthand side element of each constraint equation is non-negative. 2.
All variables are non-negative. 3.
The objective function is of the maximization or minimization type. 4.
The inequality constraints can be changed to equations by adding or substracting the
lefthand side of each such constraints by a non-negative variable. The non-negative
variable that has to be added to a constraint inequality of the form to change it to an
equation is called a slack variable. The non-negative variable that has to be substracted
from a constraint inequality of the form 2 to change it to an equation is called a surplus
variable. The right hand side of a constraint equation can be made positive by
multiplying both sides of the resulting equation by (-1) wherever necessary. The
remaining characteristics are achieved by using the elementary transformations
introduced with the canonical form.
The Standard Form of the LPP
Any standard form of the L.P.P. is given by
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Maximize or Minimize
Subject to:
Fundamental Theorem of L.P.P.
Given a set of m simultaneous linear equations in n unknowns/variables, n m, AX = b,
with r(A) = m. If there is a feasible solution X 0, then there exists a basic feasible
solution.
Solution Of The Linear Programming Program Simplex Method
Consider a LPP given in the standard form,
To optimize z = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
Subject to
a
11
x
1
+ a
12
x
2
+ + a
n
x
n
1 S
1
= b
1
a
21
x
1
+ a
22
x
2
+ -+ a
2n
x
n
1 S
2
= b
2
.
.
a
m1
x
1
+ a
m2
x
2
+ + a
mn
x
n
1 S
m
= b
m
x
1
, x
2
, x
n
, S
1
, S
2
, S
m
2 0.
To each of the constraint equations add a new variable called an artificial variable on
the left hand side of every equation which does not contain a slack variable. Then every
constraint equation contains either a slack variable or an artificial variable.
The introduction of slack and surplus variables do not alter either the constraints or the
objective function. So such variables can be incorporated in the objective function with
zero coefficients. However, the artificial variables do change the constraints, since these
are added only to one side i.e., to the left hand side of the equations. The new constraint
equations so obtained is equivalent to the original equations if and only if all artificial
variables have value zero. To guarantee such assignments in the optimal solutions,
artificial variables are incorporated into the objective function with very large positive
coefficient M in the minimization program and very large negative coefficient
M in the maximization program. These coefficients represent the penalty incurred in
making an unit assignment to the artificial variable.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Thus the standard form of LPP can be given as follows :
Optimize Z = C
T
X
Subject to AX = B,
and
Where X is a column vector with decision, slack, surplus and artificial variables, C is the
vector corresponding to the costs, A is the coefficient matrix of the constraint equations
and B is the column vector of the righthand side of the constraint equations.
Example 1: Consider the LPP
Minimize Z = 4x
1
+ x
2
Subject to 3x
1
+ x
2
= 3
Rewriting in the standard form,
Minimize Z = 4x
1
+ x
2
+ 0.S
1
+ 0.S
2
+ M (A
1
+ A
2
)
Subject to 3x
1
+ x
2
+ A
1
= 3
When S
2
is slack variable, S
1
is a surplus variable and A
1
& A
2
an artificial variables.
Representing this program in matrixes, we have
Minimize Z = (4 1 0 0 M M)
Subject to
=
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
and
2 0
Initial basic feasible solution of a LPP
Consider a system of m equations in n unknowns x
1
, x
2
- x
n
,
a
11
x
1
+ a
12
x
2
+ - + a
1n
x
n
= b
1
a
21
x
1
+ a
22
x
2
+ - + a
2n
x
n
= b
2
a
m1
x
1
+ a
m2
x
2
+ - + a
mn
x
n
= b
n
To solve this system of equations, we first assign any of n m variables with value zero.
These variables which have assigned value zero initially are called the non-basic
variables, the remaining variables are called basic variables. Then the system can be
solved to obtain the values of the basic variables. If one or more values of the basic
variables are also zero valued, then solution of the system is said to degenerate. If all
basic variable, have non-zero values, then the solution is called a non-degenerate
solution.
A basic solution is said to be feasible, if it satisfies all constraints.
Example 2: Consider the system of equations
2x1 + x2 x3 = 2
3x1 + 2x2 + x3 = 3
Since there are 3 variables and two equations, assign 3 2 = 1 variable, the value zero
initially.
Case (i): Let x3 = 0 i.e., x3 be a non-basic variable, then equation becomes
2x1 + x2 = 2
3x1 + 2x2 = 3
Solving, we get
x1 = 1, x2 = 0.
. The solution degenerates, but is feasible.
Case (ii) : Let x2 be a non-basic variable i.e., x2 = 0, then solution is
x1 = 1 and x3 = 0
Here also the solution degenerates but feasible.
Case (iii) : Let x1 be non-basic i.e., x1= 0
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Solution is x2 = , x3 = .
The solution non-degenerates, but is not feasible.
Consider a LPP given in the standard form
The initial solution of such a problem denoted by X0, is obtained by treating all decision
and surplus variables as non-basic variables i.e., they have assigned value zero, all
slack and, artificial variables as basic variables and have assigned values which are on
R.H.S. of the corresponding constraint equations.
To Solve problem by Simplex Method
Introduce stack variables (Sis) for type of constraint. 1.
Introduce surplus variables (Sis) and Artificial Variables (Ai) for 2 type of 2.
constraint.
Introduce only Artificial variable for = type of constraint. 3.
Cost (Cj) of slack and surplus variables will be zero and that of Artificial variable 4.
will be M
Find Zj Cj for each variable.
Slack and Artificial variables will form Basic variable for the first simplex table. 5.
Surplus variable will never become Basic Variable for the first simplex table.
Zj = sum of [cost of variable x its coefficients in the constraints Profit or cost 6.
coefficient of the variable].
Select the most negative value of Zj Cj. That column is called key column. The 7.
variable corresponding to the column will become Basic variable for the next
table.
Divide the quantities by the corresponding values of the key column to get ratios 8.
select the minimum ratio. This becomes the key row. The Basic variable
corresponding to this row will be replaced by the variable found in step 6.
The element that lies both on key column and key row is called Pivotal element. 9.
Ratios with negative and O value are not considered for determining key row. 10.
Once an artificial variable is removed as basic variable, its column will be deleted 11.
from next iteration.
For maximisation problems decision variables coefficient will be same as in the 12.
objective function. For minimization problems decision variables coefficients will
have opposite signs as compared to objective function.
Values of artificial variables will always is M for both maximization and 13.
minimization problems.
The process is continued till all Zj Cj 14.
2 0.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Learn Application of Simplex Algorithm to Solve System of Equations
The Simplex Algorithm
To test for optimality of the current basic feasible solution of the LPP, we use the
following algorithm called simplex algorithm. Let us also assume that there are no
artificial variable existing in the program.
Steps
Locate the most negative number in the last (bottom) row of the simplex table, 1.
excluding that of last column and call the column in which this number appears
as the work column.
Form ratios by dividing each positive number in the work column, excluding that 2.
of the last row into the element in the same row and last column. Designate that
element in the work column that yields the smallest ratio as the pivot element. If
more than one element yields the same smallest ratio choose arbitrarily one of
them. If no element in the work column is non negative the program has no
solution.
Use elementary row operations to convert the pivot element to unity (1) and then 3.
reduce all other elements in the work column to zero.
Replace the x -variable in the pivot row and first column by x-variable in the first 4.
row pivot column. The variable which is to be replaced is called the outgoing
variable and the variable that replaces is called the incoming variable. This new
first column is the current set of basic variables.
Repeat steps 1 through 4 until there are no negative numbers in the last row 5.
excluding the last column.
The optimal solution is obtained by assigning to each variable in the first column 6.
that value in the corresponding row and last column. All other variables are
considered as non-basic and have assigned value zero. The associated optimal
value of the objective function is the number in the last row and last column for a
maximization program but the negative of this number for a minimization
problem.
Examples:
3) Maximize z = x1+ 9x2 + x3
Rewriting in the standard form
Maximize z = x1 + 9x2 + x3 + 0.S1 + 0.S2
Subject to the conditions
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Where S1 and S2 are the slack variables.
The initial basic solution is S1 = 9, S2 = 15
. X0 = , C0 =
The initial simplex table is given below :
S1 outgoing variable, x2 incoming variable.
Since there are three Zj Cj which are negative, the solution is not optimal.
We choose the most negative of these i.e. 9, the corresponding column vector x2
enters the basis replacing S1, since ratio is minimum. We use elementary row
operations to reduce the pivot element to 1 and other elements of work column to
zero.
First Iteration The variable x1 becomes a basic variable replacing S1. The
following table is obtained.
Since all elements of the last row are non-negative the optimal solution is obtained.
The maximum value of the objective function Z is which is achieved for x2 = S2 =
6 which are the basic variables. All other variables are non-basic.
4) Use Simplex method to solve the LPP
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Rewriting in the standard form
Maximize Z = 2x1 + 4x2 + x3 + x4 + 0.S1 + 0.S2 + 0.S3
The initial basic solution is S1 = 4, S2 = 3, S3 = 3
.X0 = = C0 =
The initial table is given by
S1 is the outgoing variable, x2 is the incoming variable to the basic set.
The first iteration gives the following table :
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
x
3
enters the new basic set replacing S
3
, the second iteration gives the following table
:
x
1
enters the new basic set replacing S
2
, the third iteration gives the following table:
Since all elements of the last row are non-negative, the optimal solution is Z =
which is achieved for x
2
= 1, x
1
= 1, x
3
= and x
4
= 0.
A manufacturing firm has discontinued production of a certain unprofitable 1.
product line. This created considerable excess production capacity. Management
is considering to devote this excess capacity to one or more of three products:
call them product 1, 2 and 3. The available capacity on the machines which
might limit output are given below :
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The number of machine-hours required for each unit of the respective product is given
below :
Productivity (in Machine hours/Unit)
Machine Type Product 1 Product 2 Product 3
Milling Machine 8 2 3
Lathe 4 3 0
Grinder 2 1
The unit profit would be Rs. 20, Rs. 6 and Rs. 8 for products 1, 2 and 3. Find how
much of each product the firm should produce in order to maximize profit ?
Let x
1
, x
2
, x
3
units of products 1, 2 and 3 are produced in a week.
Then total profit from these units is
Z = 20 x
1
+ 6 x
2
+ 8 x
3
To produce these units the management requires
8x
1
+ 2x
2
+ 3x
3
machine hours of Milling Machine
4x
1
+ 3x
2
+ 0 x
3
machine hours of Lathe
and 2x
1
+ x
3
machine hours of Grinder
Since time available for these three machines are 250, 150 and 50 hours respectively,
we have
Thus the problem is to
Maximize Z = 20x
1
+ 6x
2
+ 8x
3
Maximize Z = 20x
1
+ 6x
2
+ 8x
3
+ 0S
1
+ 0S
2
+ 0S
3
Subject to 8x
1
+ 2x
2
+ 3x
3
+ S
1
= 250
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The initial basic solution is
X
0
= =
The initial simplex table is given by
x
1
enters the basic set of variables replacing the variable s
3
. The first iteration gives
the following table:
x
2
enters the basic set of variables replacing the variable s
2
. The second iteration gives
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
x
3
enters the basic set of variables replacing the variable x
1
. The third iteration yields the
following table:
Since all z
j
c
j
2 0 in the last row, the optimum solution is 700.
i.e., the maximum profit is Rs. 700/- which is achieved by producing 50 units of
product 2 and 50 units of product 3.
Know about the Big-M Method
Penalty Cost Method Or Big-M Method
Consider a L.P.P. when atleast one of the constraints is of the type 2 or = . While
expressing in the standard form, add a non negative variable to each of such constraints.
These variables are called artificial variables. Their addition causes violation of the
corresponding constraints, since they are added to only one side of an equation, the new
system is equivalent to the old system of constraints if and only if the artificial variables
are zero. To guarantee such assignments in the optimal solution, artificial variables are
incorporated into the objective function with large positive coefficients in a minimization
program or very large negative coefficients in a maximization program. These
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
coefficients are denoted by M.
Whenever artificial variables are part of the initial solution X
0
, the last row of simplex
table will contain the penalty cost M. The following modifications are made in the
simplex method to minimize the error of incorporating the penalty cost in the objective
function. This method is called Big M-method or Penalty cost method.
The last row of the simplex table is decomposed into two rows, the first of which 1.
involves those terms not containing M, while the second involves those
containing M.
The Step 1 of the simplex method is applied to the last row created in the above 1.
modification and followed by steps 2, 3 and 4 until this row contains no negative
elements. Then step 1 of simplex algorithm is applied to those elements next to
the last row that are positioned over zero in the last row.
Whenever an artificial variable ceases to be basic, it is removed from the first 2.
column of the table as a result of step 4, it is also deleted from the top row of the
table as is the entire column under it.
The last row is removed from the table whenever it contains all zeroes. 3.
If non zero artificial variables are present in the final basic set, then the program 4.
has no solution. In contrast, zero valued artificial variables in the final solution
may exist when one or more of the original constraint equations are redundant.
Examples:
6) Use Penalty Cost method to
Maximize z = 2x
1
+ 3x
2
Rewriting in the standard form, we have
Maximize z = 2x
1
+ 3x
2
+ 0S
1
+ 0S
2
M A
1
Subject to x
1
+ 2x
2
+ S
1
= 2
6x
1
+ 4x
2
S
2
+ A
1
= 24,
x
1
, x
2
, S
1
, S
2
, A
1
2 0.
The initial simplex table is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Since all elements of the last two rows are non negative, the procedure is complete. But
existence of non zero artificial variable in the basic solution indicates that the problem
has no solution.
Learn about the Two-Phase Method and Dual-Primal Problem
Two Phase Method
The drawback of the penalty cost method is the possible computational error that could
result from assigning a very large value to the constant M. To overcome this difficulty, a
new method is considered, where the use of M is eliminated by solving the problem in
two phases. They are
Phase I: Formulate the new problem by eliminating the original objective function by the
sum of the artificial variables for a minimization problem and the negative of the sum of
the artificial variables for a maximization problem. The resulting objective function is
optimized by the simplex method with the constraints of the original problem. If the
problem has a feasible solution, the optimal value of the new objective function is zero
(which indicates that all artificial variables are zero). Then we proceed to phase II.
Otherwise, if the optimal value of the new objective function is non zero, the problem has
no solution and the method terminates.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Phase II : Use the optimum solution of the phase I as the starting solution of the original
problem. Then the objective function is taken without the artificial variables and is solved
by simplex method.
Examples:
7) Use the two phase method to
Maximise z = 3x
1
x
2
Maximize z = 3x
1
x
2
+ 0S
1
MA
1
+ 0.S
2
+ 0.S
3
Phase I :
Consider the new objective,
Maximize Z* = A
1
Subject to 2x
1
+ x
2
S
1
+ A
1
= 2
Solving by Simplex method, the initial simplex table is given by
x
1
enters the basic set replacing A
1
.
The first iteration gives the following table:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Phase I is complete, since there are no negative elements in the last row.
The Optimal solution of the new objective is Z* = 0.
Phase II:
Consider the original objective function,
Maximize z = 3x
1
x
2
+ 0S
1
+ 0S
2
+ 0S
3
Subject to x
1
+ =1
with the initial solution x
1
= 1, S
2
= 1, S
3
= 4, the corresponding simplex table is
Proceeding to the next iteration, we get the following table:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Since all elements of the last row are non negative, the current solution is optimal.
The maximum value of the objective function
Z = 6 which is attained for x
1
= 2, x
2
= 0.
8) Maximize z = 3x
1
+ 2x
2
,
Maximize z = 3x
1
+ 2x
2
+ 0S
1
+ 0.S
2
MA
1
Subject to 2x
1
+ x
2
+ S
1
= 2
3x
1
+ 4x
2
S
2
+ A
1
= 2
x
1
, x
2
, S
1
, S
2
, A
1
2 0.
Solving by two phase method.
Phase I :
Consider the new objective function
Maximize z* = A
1
Subject to 2x
1
+ x
2
+ S
1
=2
3x
1
+ 4x
2
S
2
+ A
1
= 12,
x
1
, x
2
, S
1
, S
2
, A
1
2 0.
The initial Simplex table is given by
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Since all elements of the last row an non negative, the procedure is complete.
But the existence of non zero artificial variable in the basic set indicates that the problem
has no solution.
Minimization Examples
Example 10
Minimize = Z = 3x
1
+ 8x
2
Subject to
Solution
In a standard form
Minimize Z = 3x
1
+ 8x
2
+ MA
1
+ OS
1
+ MA
2
+ OS
2
Subject to
Simplex Table 1
C
j
- 3 - 8 0 0 - M - M
C.B B.V x1 x2 S1 S2 A1 A3 Qty Ratio
- M A
1
1 1 0 0 1 0 200 200
- M A
2
1
P.E
0 - 1 0 0 0 80 80 + K R
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
0
S
2
0 1 0 1 0 1 60 O
Z
j
C
j
-2M +3 K C -M + 8 M 0 0 0
Simplex Table 2
C
j
- 3 - 8 0 0 - M
C.B B.V x1 x2 S1 S2 A1 Qty Ratio Transformation
- M A
1
0 1 1 0 1 120 120 R
1
1
= R
1
-R
2
1
- 3 x
1
1 0 -1 0 0 80 O R
2
1
= R
2
0 S
2
0 1
P.E
0 1 0 60 60 + K R R
3
1
= R
3
Z
j
C
j
0 -M + 8 K C -M + 3 0 0
Simplex Table 3
C
j
- 3 - 8 0 0
C.B B.V x1 x2 S1 S2 A1 Qty Ratio Transformation
- M A
1
0 0 1
P.E
- 1 1 60 60 R
1
1
= R
1
- 3 x
1
1 0 - 1 0 0 80 -ve R
2
1
= R
2
8 x
2
0 1 0 1 0 60 O R
3
1
= R
3
R
1
1
Z
j
C
j
0 0 - M + 3 KC M 8 0 0
Simplex Table 4
C
j
- 3 - 8 0 0
C.B B.V x1 x2 S1 S2 Qty Ratio Transformation
0 S
1
0 0 1 - 1 60 - ve R
1
1
= R
1
- 3 x
1
1 0 0 - 1 140 -ve R
2
1
= R
2
+ R
1
1
8 x
2
0 1 0 1
P.E
60 80 + K R R
3
1
= R
1
1
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Z
j
C
j
0 0 0 - 5
Simplex Table 5
C
j
- 3 - 8 0 0
C.B B.V x1 x2 S1 S2 Qty Ratio Transformation
0 S
1
0 1 1 0 120 R
1
1
= R
1
+ R
3
1
- 3 x
1
1 1 0 0 200 R
2
1
= R
2
+ R
3
1
0 S
2
0 1 0 1 60 R
3
1
= R
3
Z
j
C
j
0 4 0 0
Since all Z
j
C
j
2 0, the optimum solution is x
1
= 200 x
2
= 0
Min Z = 60
Summary
In this unit we solved the L.P.P by simplex method. The constraints for which slack,
surplus and artificial variables to be introduced and the method of solving L.P.P is
explained with examples.
Unit - 4-Duality in L.P.P
Introduction
Every linear programming problem has associated with it, another linear programming
problem involving the same data and closely related optimal solutions. Such two
problems are said to be duals of each other. While one of these is called the primal, the
other the dual.
Learning Objectives:
After studying this unit, you should be able to understand the following
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Understand the duality concept +
Write the dual problem +
Understand its economical interpretation +
Apply sensitivity analysis +
Understand the Duality Concept and Learn Writing Dual Problem
The Importance Of Duality Concept Is Due To Two Main Reasons
i) If the primal contains a large number of constraints and a smaller number of
variables, the labour of computation can be considerably reduced by converting it
into the dual problem and then solving it
ii) The interpretation of the dual variable from the lost or economic point of view
proves extremely useful in making future decisions in the activities being
programmed.
Formulation of dual problem
Consider the following L.P.P
Maximize Z = c
1
x
1
+c
2
x
2
+ . . .+ c
n
x
n
To construct a dual problem, we adopt the following guide lines:
i) The maximization problem in the primal becomes a minimization problem in the
dual and vice versa
ii) () type of constraints in the primal becomes () type of constraints in the dual
and vice versa.
iii) The coefficients c
1,
c
2,
. . .,c
n
in
the objective function of the primal become b
1
, b
2
,,b
m
in the objective function of
the dual.
iv) The constants b
1
, b
2
,,b
m
in the constraints of the primal becomes c
1,
c
2,
. . .,c
n
in
the constraints of the dual
v) If the primal has n variables and m constraints the dual will have m variables and
n constraints
vi) The variables in both the primal and dual are non-negative
Then the dual problem will be
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Minimize W = b
1
y
1
+ b
2
y
2
+ . . . +b
m
y
m
subject to the constraints
The standard form of L.P.P for maximization must contain type of constraints and for
minimization problem it must contain type of constraints. Then it is easy to form dual
L.P.P
Example 1: Write dual of
max Z = 4x
1
+ 5x
2
Solution: The given problem is in its standard form:
. Its dual is
Mini W = 15y
1
+ 10 y
2
+ 20 y
3
Example 2: Write the dual of
min Z = 10x
1
+ 12x
2
Solution: The given problem is in its standard form. Therefore its dual problem is
Max W = 10y
1
+ 20 y
2
+ 15 y
3
+ 12y
4
Example 3: Write the dual of
max Z = 100x
1
+ 200x
2
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Solution: First 3x
1
10x
2
15 is converted to type as -3x
1
+ 10x
2
-15
Therefore its dual is
Mini Z = -15y
1
+ 20 y
2
Example 4: When the constraints contain = Sign.
Write the dual of Max Z = 40x
1
+ 30x
2
Secondly x
1
+ x
2
= 9 is written as
Therefore Given problem is
max Z = 40x
1
+ 30x
2
Therefore its dual is
Mini W = -15y
1
+ 10 y
2
+ 9y
3
1
9y
3
11
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Then the dual becomes
Max W = 15y
1
-5 y
2
+ y
3
y
3
is unrestricted in sign
Example 5:
Write the dual of
max Z = 12x
1
+ 15x
2
x
2
is unrestricted in sign
Solution:
Let x
2
=x
2
1
x
2
11
0
. Its standard form is
max Z = 12x
1
+ 15x
2
1
-15x
2
11
Therefore dual is
Constraints No.2 and No.4 gives
3y
1
y
2
= 15
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The dual problem is
Max W = 10y
1
15 y
2
Subject to
Example 6: Write the dual of the following L. P. P
minimize Z = 3x
1
2x
2
+ 4x
3
Solution: Since the problem is of minimization all constraints should be of type. We
multiply the third constraint throughout by -1 so that -7x
1
+ 2x
2
+ x
3
-10
Let y
1
, y
2
, y
3
, y
4
and y
5
be the dual variables associated with the above five constraints.
Then the dual problem is given by:
Maximize W = 7y
1
+ 4 y
2
10 y
3
+ 3 y
4
+ 2y
5
Economic interpretation of duality
The linear programming problem can be thought of as a resource allocation model in
which the objective is to maximize revenue or profit subject to limited resources. Looking
at the problem from this point of view, the associated dual problem offers interesting
economic interpretations of the L.P resource allocation model.
We consider here a representation of the general primal and dual problems in which the
primal takes the role of a resource allocation model.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
From the above resource allocation model, the primal problem has n economic activities
and m resources. The coefficient c
j
in the primal represents the profit per unit of activity j.
Resource i, whose maximum availability is b
i
, is consumed at the rate a
ij
units per unit of
activity j.
Understand Economic Interpretation of Dual Problem
Economic interpretation of Dual variables:
For any pair of feasible primal and dual solutions,
(Objective value in the maximization problem) (Objective value in the minimization
problem)
At the optimum, the relationship holds as a strict equation.
Note: Here the sense of optimization is very important.
Hence clearly for any two primal and dual feasible solutions, the values of the objective
functions, when finite, must satisfy the following inequality.
The strict equality, z = w, holds when both the primal and dual solutions are optimal.
Consider the optimal condition z = w first given that the primal problem represents a
resource allocation model, we can think of z as representing profit in Rupees. Because b
i
represents the number of units available of resource i, the equation z = w can be
expressed as profit (Rs) = _ (units of resource i) x (profit per unit of resource i)
This means that the dual variables y
i
, represent the worth per unit of resource i [variables
y
i
are also called as dual prices, shadow prices and simplex multipliers].
With the same logic, the inequality z < w associated with any two feasible primal and
dual solutions is interpreted as (profit) < (worth of resources)
This relationship implies that as long as the total return from all the activities is less than
the worth of the resources, the corresponding primal and dual solutions are not optimal.
Optimality is reached only when the resources have been exploited completely, which can
happen only when the input equals the output (profit).
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Economically the system is said to remain unstable (non optimal) when the input (worth
of the resources) exceeds the output (return). Stability occurs only when the two
quantities are equal.
Learn Applying Sensitivity Analysis in Dual Problem
Sensitivity analysis
In linear programming, all model parameters are assumed to be constant; but in real life
situations, the decision environment is always dynamic. Therefore, it is important for the
management to know how profit would be affected by an increase or decrease in the
resource level, by a change in the technological process, and by a change in the cost of
raw materials .Such an investigation is known as Sensitivity analysis or post optimality
analysis. The results of sensitivity analysis establish upper and lower bounds for input
parameter values within which they can vary without causing major changes in the
current optimal solution.
For e.g.: luminous lamps produces three types of lamps A, B And C. These lamps are
processed on three machines X, Y and Z. the full technology and input restrictions are
given in the following table
Product Machine Profit Per Unit
M1 M2 M3
A 10 7 2 12
B 2 3 4 3
C 1 2 1 1
Available time 100 77 80
The Linear Programming model for the above problem can be written as
Maximize Z = 12 x1 + 3 x2 + x3
subject to
Solving the above problem by the simplex method
x1 = , x2 =
x3 = 0
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
and the optimal value of the objective function is
First we find out whether a previously determined optimal solution remains optimal if the
contribution rate is changed. An increase in Cj of a variable would mean that resources
from other products should be diverted to this more profitable product. The reverse is
true for a minimization problem
Changes in Cj of a non-basic
variable:
A non basic variable can be brought into the basis only if its contribution rate becomes
attractive. Hence, we need to determine the upper limit of the profit contribution (Cj) of
each non basic variable. The reverse is true for a minimization problem.
Consider the final simplex table of the above problem
Cj 12 3 1 0 0 0
CB Basic variables
B
x1 x2 x3 X4 x5 x6 Solution
values
b (=xB )
12 x1 1 0

0
3 x2 0 1
0
0 x6 0 0

1
zj 12 3 0
zj
cj
0 0 0
From the above final simplex table, we note that profit contribution for product C is Rs 1,
which is not greater than its zj . Thus to bring x3 into the basis, its profit contribution rate
cj must exceed Rs to make zj
cj value negative or zero. (i.e. zj -
cj
0)
Specifically
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
In this case c3 = 1 and z3 c3 = then
C3*
1
C3*
+1 =
x3 can be introduced into the basis if its contribution rate c3 increase upto atleast Rs ,
if it increases beyond that then the current solution will no longer be optimal.
Change in Cj of a Basic variable
Let us consider the case of product A (x1 column) and divide each zj cj entry in the index
row by the corresponding co-efficients in the x1 row as shown below
- Minimum
Referring to the final simplex table, we observe that corresponding to the non basic
variables x3 and x5, y13 , y15 <0
Hence, minimum = minimum (11, 3) = 3
Corresponding to the non- basic variable x4 , y14 >0.
Hence, minimum = 5
Thus the optimal solution is insensitive so long as the changed profit coefficient c1
*
varies between Rs 7 and Rs. 15
Change in available resources
Now we investigate whether a previous optimal solution remains feasible if the available
resources change. For long term planning it is important to know the bounds within
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
which each available resource (e.g. machine hours) can vary without causing violent
changes in the current optimal solution. To illustrate divide each quantity in the XB
column by the corresponding coefficient in the X4 column of table
X
B
X
4
10
The least positive ratio indicates to how the number of hours of machine M1 can be
decreased. The least negative ratio (10) indicates to how much the number of hours of
machine M1 can be increased
Calculating the range:
Lower limit = 100
=
Upper limit = 100 (10) = 110
Hence the range of hours for M1 is
to 110, By the same way the range of hours for machine M2 and M3 can be calculated
The management of a company rarely restricts its interest to the numerical values of an
optimal solution. Actually it is interested in knowing the impact of changes in the input
parameter values on the optimal solution. Such a process is known as sensitivity
analysis.
Summary
For every linear programming problem there exists a Dual liner programming problems.
They help us to reduce the amount of calculation involved in original L.P.P. They also
help us to interpret the economic variables more effectively.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Unit 5 -Transportation Problem
Introduction
Here we study an important class of linear programs called the transportation
model. This model studies the minimization of the cost of transporting a
commodity from a number of sources to several destinations. The supply at each
source and the demand at each destination are known.
The transportation problem involves m sources, each of which has available a
i
(i = 1, 2,
..,m) units of homogeneous product and n destinations, each of which requires b
j
(j = 1,
2., n) units of products. Here a
i
and b
j
are positive integers. The cost c
ij
of transporting
one unit of the product from the i
th
source to the j
th
destination is given for each i and j.
The objective is to develop an integral transportation schedule that meets all demands
from the inventory at a minimum total transportation cost.
It is assumed that the total supply and the total demand are equal.
i.e. (1)
The condition (1) is guaranteed by creating either a fictitious destination with a
demand equal to the surplus if total demand is less than the total supply or a
(dummy) source with a supply equal to the shortage if total demand exceeds
total supply. The cost of transportation from the fictitious destination to all
sources and from all destinations to the fictitious sources are assumed to be
zero so that total cost of transportation will remain the same.
Learn Formulating Transportation Problem
Formulation Of Transportation Problem
The standard mathematical model for the transportation problem is as follows.
Let x
ij
be number of units of the homogenous product to be transported from source i to
the destination j
Then objective is to
Minimize z =
Subject to
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(2)
with all x
ij
2 0 and integrals
Theorem: A necessary and sufficient condition for the existence of a feasible solution to
the transportation problem (2) is that
The Transportation Algorithm (MODI Method)
The first approximation to (2) is always integral and therefore always a feasible
solution. Rather than determining a first approximation by a direct application of
the simplex method it is more efficient to work with the table given below called
the transportation table. The transportation algorithm is the simplex method
specialized to the format of table it involves:
i) finding an integral basic feasible solution
ii) testing the solution for optimality
iii) improving the solution, when it is not optimal
iv) repeating steps (ii) and (iii) until the optimal solution is obtained.
The solution to T.P is obtained in two stages. In the first stage we find Basic feasible
solution by any one of the following methods a) North-west corner rale b) Matrix Minima
Method or least cost method c) Vogels approximation method. In the second stage we
test the B.Fs for its optimality either by MODI method or by stepping stone method.
D1 D2 Dn Supply ui
S
1
x
11
x
12
x
1n
a
1
u
1
S
2
x
21
x
22
x
2n
a
2
u
2
S
3
x
31
x
32
x
3n
a
3
u
3
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
S
m
x
m1
x
m2
x
mn
a
m
u
m
Demand b
1
b
2
b
n a
i
= b
i
v
j
v
1
V
2
v
m
Transportation Table
Learn Finding Basic Feasible Solution
The Initial Basic Feasible Solution
Let us consider a T.P involving m-origins and n-destinations. Since the sum of
origin capacities equals the sum of destination requirements, a feasible solution
always exists. Any feasible solution satisfying m + n 1 of the m + n constraints
is a redundant one and hence can be deleted. This also means that a feasible
solution to a T.P can have at the most only m + n 1 strictly positive
components, otherwise the solution will degenerate.
It is always possible to assign an initial feasible solution to a T.P. in such a manner that
the rim requirements are satisfied. This can be achieved either by inspection or by
following some simple rules. We begin by imagining that the transportation table is blank
i.e. initially all x
ij
= 0. The simplest procedures for initial allocation discussed in the
following section.
North West Corner Rule
Step1: The first assignment is made in the cell occupying the upper left hand (north west)
corner of the transportation table. The maximum feasible amount is allocated there, that is
x
11
= min (a
1
,b
1
)
So that either the capacity of origin O
1
is used up or the requirement at destination D
1
is
satisfied or both. This value of x
11
is entered in the upper left hand corner (small square)
of cell (1, 1) in the transportation table
Step 2:
If b
1
> a
1
the capacity of origin O, is exhausted but the requirement at destination D
1
is
still not satisfied , so that at least one more other variable in the first column will have to
take on a positive value. Move down vertically to the second row and make the second
allocation of magnitude
x
21
= min (a
2
, b
1
x
21
) in the cell (2,1). This either exhausts the capacity of origin O
2
or
satisfies the remaining demand at destination D
1
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
If a
1
> b
1
the requirement at destination D
1
is satisfied but the capacity of origin O
1
is not
completely exhausted. Move to the right horizontally to the second column and make the
second allocation of magnitude x
12
= min
(a
1
x
11
, b
2
) in the cell (1, 2) . This either exhausts the remaining capacity of origin O
1
or
satisfies the demand at destination D
2
.
If b
1
= a
1
, the origin capacity of O
1
is completely exhausted as well as the requirement at
destination is completely satisfied. There is a tie for second allocation, An arbitrary tie
breaking choice is made. Make the second allocation of magnitude x
12
= min (a
1
a
1
, b
2
)
= 0 in the cell (1, 2) or x
21
= min (a
2
, b
1
b
2
) = 0 in the cell (2, 1).
Step 3: Start from the new north west corner of the transportation table satisfying
destination requirements and exhausting the origin capacities one at a time, move down
towards the lower right corner of the transportation table until all the rim requirements are
satisfied.
Example 1: Determine an initial basic feasible solution to the following transportation
problem using the north west corner rule:
Where O
i
and D
j
represent the i
th
origin and the j
th
destination respectively.
Solution:
The transportation table of the given T.P. has 12 cells.
Following north west corner rule, the
First allocation is made in the cell
(1,1), the magnitude being
x
11
= min (14, 6) = 6
The second allocation is made
in the 6 10 15 4 cell (1, 2) and the
magnitude of allocation is given by
x
12
= min (14 6, 10) = 8
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The third allocation is made in the cell (2, 2), the magnitude being x
22
= min (16, 10 =
2. The magnitude of fourth allocation, in the cell
(2, 3) is given by x
23
= min (16 2, 15) = 14. The fifth allocation is made in the cell (3, 3),
the magnitude being x
33
= min (5, 15 14) =1. The sixth allocation in the cell (3,4) is
given by x
34
= min (5 1, 4) = 4.
Now all the rim requirements have been satisfied and hence an initial feasible
solution to the T.P. has been obtained. The solution is displayed as
Clearly, this feasible solution is non-degenerate basic feasible solution ; for the
allocated cells do not form a loop. The transportation cost according to the above
loop is given by.
Z = x
11
c
11
+ x
12
c
12
+ x
22
c
22
+ x
23
c
23
+ x
33
c
33
+ x
34
c
34
= + + + +
= 128
Learn Various Methods for Finding Initial Basic Feasible Solution
Matrix Minimum Method
Step 1: Determine the smallest cost in the cost matrix of the transportation table. Let it be
c
ij
, Allocate x
ij
= min ( a
i
, b
j
) in the cell ( i, j )
Step 2: If x
ij
= a
i
cross off the i
th
row of the transportation table and decrease b
j
by a
i
go to
step 3.
if x
ij
= b
j
cross off the i
th
column of the transportation table and decrease a
i
by b
j
go to step
3.
if x
ij
= a
i
= b
j
cross off either the i
th
row or the i
th
column but not both.
Step 3: Repeat steps 1 and 2 for the resulting reduced transportation table until all the rim
requirements are satisfied whenever the minimum cost is not unique make an arbitrary
choice among the minima.
Example 2: Obtain an initial basic feasible solution to the following T.P. using
the matrix minima method.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Where 0
i
and D
i
denote i
th
origin and j
th
destination respectively.
Solution: The transportation table of the given T.P. has 12 cells. Following the matrix
minima method.
The first allocation is made in the cells (3,1) the magnitude being x
21
= 4. This satisfies
the requirement at destination D
1
and thus we cross off the first column from the table.
The second allocation is made in the cell ( 2, 4 ) magnitude x
24
= min( 6, 8 ) =6. Cross off
the fourth column of the table. This yields the table (i)
(i) (ii)
There is again a tie for the third allocation. We choose arbitrarily the cell
(1, 2) and allocate x
12
= min (6 , 6) = 6 there. Cross off either the second column of the
first row .We choose to cross off the first row of the table. The next allocation of
magnitude x
22
= 0 is made in the cell ( 3, 2) cross off the second column getting table( ii)
We choose arbitrarily again to make the next allocation in cell (2,3) of magnitude x
23
=
min (2,8) = 2 cross off the second row this gives table (iii). The last allocation of
magnitude x
23
= min ( 6 , 6) = 6 is made in the cell (3,3).
Now all the rim requirements have been satisfied and hence an initial feasible
solution has been determined. This solution is shown in table ( iv)
Since the cells do not form a loop, the solution is basic one. Moreover the solution is
degenerate also. The transportation cost according to the above route is given by
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Z = + + + +
= 28
Vogels Approximation Method
The Vogels Approximation Method takes into account not only the least cost c
ij
but also
the cost that just exceeds c
ij
.The steps of the method are given below.
Step 1: For each row of the transportation table identify the smallest and the nest to
smallest costs. Determine the difference between them for each row. Display them
alongside the transportation table by enclosing them in parenthesis against the respective
rows. Similarity compute the differences for each column.
Step 2: Identify the row or column with the largest difference among all the rows and
columns. If a tie occurs, use any arbitrary tie breaking choice, let the greatest difference
correspond to the i
th
row and let C
ij
be the smallest cost in the i
th
row. Allocate the
maximum feasible amount x
ij
= min (a
i
, b
j
) in the (i,j)
th
cell and cross off the i
th
row or the
j
th
column in the usual manner.
Step 3: Recompute the column and row differences for the reduced transportation table
and go to step 2. Repeat the procedure until all the rim requirements are satisfied.
Remarks:
1. A row or column difference indicates the minimum unit penalty incurred by
failing to make an allocation to the smallest lost cell in that row or column.
It is clear that VAM determines an initial basic feasible solution which is very 1.
close to the optimum solution, that is, the number of iterations required to reach
the optimal solution is small in this case
Example 3:
Obtain an initial basic feasible solution to the following T.P using the Vogels
approximation method:
Ware houses Stores Availability
I II III IV
A 5 1 3 3 34
B 3 3 5 4 15
C 6 4 4 3 12
D 4 -1 4 2 19
Requirement 21 25 17 17 80
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Solution: The transportation table of the given T.P has 16 cells. The differences between
the smallest and next to smallest costs in each row and each column are computed and
displayed inside the parenthesis against the respective columns and rows. The largest of
these differences is (3) and is associated with the fourth row of the transportation table.
The minimum cost in the fourth row is C
42
= 1. Accordingly we allocate x
42
= min (19,
25) = 19 in the cell (4, 2). This exhausts the availability at ware house D. Cross off the
fourth row. The row and column differences are now computed for the resulting reduced
transportation table (ii).
the largest of this is (2) and is associated with the first row as well as the first and second
column. We select arbitrarily the first row whose minimum cost is C
12
= 1. Thus the
second allocation of magnitude x
12
= min (34, 6) = 6 is made in the cell (1, 2) Cross off
the second column from the table. Continuing in this way, the subsequent reduced
transportation tables and the differences for the surviving rows and columns are shown in
fig (iii)
Eventually, the basic feasible solution is obtained as shown in table (iv) below:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The transportation cost according to the above route is given by:
Z = 6 x 5 + 6 x 1 + 17 x 3 + 5 x 3 + 15 x 3 + 12 x 3 + 19 x ( 1) = 164.
Moving Towards Optimality
After evaluating an initial basic feasible solution to a Transportation problem, the
next question is how to get the optimum solution. The basic techniques are:
Determining the net evaluations for the non basic variables (empty cells) 1.
Determining the entering variable 2.
Determining the leaving variable 3.
Computing a better basic feasible solution 4.
Repeating steps (1) to (4) until an optimum solution has been obtained. 5.
Improving the solution
Definition: A loop is the sequence of cells in the transportation table such that:
i) Each pair of consecutive cells lie either in the same row or same column
ii) No three consecutive cells lies in the same row or same column
iii) The first and the last cells of the sequence lies in the same row or column
iv) No cell appears more than once in the sequence
Consider the non-basic variable corresponding to the most negative of the
quantities.
c
ij
u
i
v
j
. Calculated in the test for optimality; it is made the incoming variable.
Construct a loop consisting exclusively of this incoming variable (cell) and current
basic variables (cells). Then allocate to the incoming cell as many units as possible
such that, after appropriate adjustments have been made to the other cells in the
loop, the supply and demand constraints are not violated, all allocations remain non-
negative and one of the old basic variables has been reduced to zero. (where upon it
ceases to be basic)
Modified Distribution Method / Modi Method / U V Method. 1.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Step 1: Under this method we construct penalties for rows and columns by subtracting
the least value of row / column from the next least value.
Step 2: We select the highest penalty constructed for both row and column. Enter that
row / column and select the minimum cost and allocate min (a
i
, b
j
)
Step 3: Delete the row or column or both if the rim availability / requirements is met.
Step 4: We repeat steps 1 to 2 to till all allocations are over.
Step 5: For allocation all form equation u
i
+ v
j
= c
j
set one of the dual variable u
i
/ v
j
to
zero and solve for others.
Step 6: Use these value to find ^
ij
= c
ij
u
i
v
j
of all ^
ij
2, then it is the optimal solution.
Step 7: If any ^
ij
0, select the most negative cell and form loop. Starting point of the loop is +ve and
alternatively the other corners of the loop are ve and +ve. Examine the quantities
allocated at ve places. Select the minimum. Add it at +ve places and subtract from ve
place.
Step 8: Form new table and repeat steps 5 to 7 till ^
ij
2 0
Balanced T.P
Example 4:
Solve the following transportation problem with cost coefficients demands and supplies
as given in the following table:
Since total demand = 31 = Total supply, the problem is balanced.
The initial basic feasible solution is obtained by Vogels approximation
method. The following table given the initial solution:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The optimum allocations are x13 = 14, x21 = 6, x22 = 5, x23 = 1, x32 = 5.
The minimum transportation cost is 144+64+59+18+52 = 143.
For allocated cells.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Unbalanced T.P
Example 5
A car company is faced with an allocation problem resulting from rental agreement that
allow cars to be returned to locations other than those which they were originally rented.
At the present time there are two cars with 15 and 13 simplex cars respectively and 4
locations requiring 9, 6, 7 and 9 cars respectively. The unit transportation costs (in
dollars) between the locations an given below:
Obtain a minimum cost schedule.
Solution:
Since the supply and requirements are not equal it is called an unbalanced T.P. in
general if then it is called an unbalanced T.P. we introduced either a dummy
row or a dummy column with cost zero an quantity respectively. Applying
VAM Method we find B.F.S as Destinations
Testing for optimality
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
For non-allocated cells determine cij uI vi. Since all then quantities are non-
negative, the current solution is optimal. The minimum transportation cost is = 6
17+321+630+914+419+3.0 = 470.
Which achieved by transporting x12 = 6 cars from sources 1 to dentition 2, x13 = 3,
x14 = 6 cars from sources 1 to destinations 3 and 4 respectively; x21 = 9 and x33 =
4 cars from sources 2 to destinations 1 and 3 respectively.
Learn Handling and Solving Degeneracy in the Transportation Problem
Degeneracy in Transportation Problem
It was shown that a basic solution to an m-origin, n destination Transportation
problem can have at most m+n-1 positive basic variables (non-zero) and so that
otherwise the basic solution degenerates. It follows that whenever the number of
basic cells is less than m + n 1, the transportation problem is a degenerate
one. The degeneracy can develop in two ways:
Case 1: The degeneracy develops while determining an initial assignment via any one of
the initial assignment methods discussed earlier.
To resolve degeneracy, we augment the positive variables by as many zero-
valued variables as is necessary to complete the required m + n 1 basic
variables. These zero-valued variables are selected in such a manner that the
resulting m + n 1 variables constitute a basic solution. The selected zero
valued variables are designated by allocating an extremely small positive value
to each one of them. The cells containing these extremely small allocations are
then treated like any other basic cells. The s are kept in the transportation table
until temporary degeneracy is removed or until the optimum solution is attained,
whichever occurs first.
At that point, we set each = 0.
Case 2: The degeneracy develops at some iteration stage. This happens when the
selection of the entering variable results in the simultaneous drive to zero of two or more
current (pre-iteration) basic variables.
To resolve degeneracy, the positive variables are augmented by as many zero-
valued variables as is necessary to complete m+n-1 basic variables. These zero-
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
valued variables are selected from among those current basic variables which
are simultaneously driven to zero. The rest of the procedure is exactly the same
as discussed in case 1 above.
Note: The extremely small value is infinitesimally small and it never affects the value it
is added to or subtracted from.
We introduce in unallocated minimum cost cell such that we cannot form a loop
Example: Obtain an optimum basic feasible solution to the following degenerate T.P.
Solution: Following the North West Corner rule, an initial assignment is made as shown
in table 1. Since the basic cells do not form a loop, the solution is basic. However, since
the number of basic cells is 4 which is less than 5. (= m + n 1) the basic solution
degenerates.
Table -1
(Initial assignment)
In order to complete the basis and thereby remove degeneracy, we require only one more
positive basic variable. We select the variable x
23
and allocate a negligibly small positive
quantity in the cells (2, 3) as shown in table 2.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Table -2
(Augmented solution)
Notice that even after the inclusion of cell (2, 3) in the basis the basic cells do not form a
loop. i.e. the augmented solution remains basic. The net evaluations can now be
computed and the solution is tested for optimality
Starting table:
Since all the net evaluations for the non-basic variables are not non- positive, the initial
solution is not optimum. The non basic cell (1, 3) must enter the basis. The exit criterion
removes the basic cell (2,3) from the basis max .
First iteration:
Introduce the cell (1, 3) into the basis and drop the cell (2,3) from it. Determine
the current net evaluations. Since all of them are not non-positive the current
solution can be improved.
Second Iteration:
Introduce the cell (3,1) and drop the cell (1,1) from the basis. Since some of the
current net evaluations are still positive, the current solution can further be
improved.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Third iteration:
Introduce the cell (2,3) and drop the cell (2,1) from the basis. Since all the current net
evaluations are non positive, the current solution is an optimum one
The transportation cost according to the above route is given by
= 33
Summary
The transportation problem is a special type of linear programming problem in which the
objective is to transport a homogeneous product manufactured at several plants ( origins)
to a number of different destinations at a minimum total cost. In this chapter, you learned
several different techniques for computing an initial basic feasible solution to a
transportation problem such as north west corner rule, matrix minimum method and vogel
approximation method. Further we studied the degeneracy in transportation problem with
examples on obtaining an optimum basic feasible solution.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Unit - 6-Assignment Problem
Introduction
The assignment problem is a special case of the transportation problem, where
the objective is to minimize the cost or time of completing a number of jobs by a
number of persons and Maximize efficiently Revenue, sales etc In other words,
when the problem involves the allocation of n different facilities to n different
tasks, it is often termed as an assignment problem. This model is mostly used for
planning. The assignment model is also useful in solving problems such as,
assignment of machines to jobs, assignment of salesman to sales territories,
traveling salesman problem etc. It may be noted that with n facilities and n jobs,
there are n! possible assignments. One way of finding an optimal assignment is
to write all the n! possible arrangement ,evaluate their total cost and select the
assignment with minimum cost. But because of many computational procedures
this method is not possible. In this unit we study an efficient method for solving
assignment problems.
There are n jobs for a factory and factory has n machines to process the jobs. A job
i(=1,,n) , when processed by machine j(=1,,n) is assumed to incur
a cost C
ij.
The assignment is to be made in such a way that each job can associate with one
and only one machine Determine an assignment of jobs to machines so as to minimize the
overall cost.
Learn Solving a Routing Problem
Mathematical Formulation Of The Problem
Let x
ij
be a variable defined by
Then, since only one job is to be assigned to each machine we have
x
ij
= 1 and x
ij
= 1
Also the total assignment cost is given by
z = x
ij
c
ij
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Thus the assignment problem takes the following mathematical form
Determine x
ij
0 (i , j =1,, n)
So as to minimize
z = x
ij
c
ij
x
ij
= 1 j =1, 2,, n
and x
ij
= 1 i = 1, 2,, n
with x
ij
= 0 or 1
Note:
In an assignment problem if we add (or subtract) a real number to (from) every element of
a row or column of the cost matrix, then an assignment which is optimum for the
modified matrix is also optimum for the original one.
Hungarian Method: Algorithm
Step 1: Prepare Row ruled Matrix by selecting the minimum values for each row and
subtract it from other elements of the row
Step 2: Prepare column reduced Matrix by subtracting minimum value of the column
from the other values of that column
Step 3: First row-wise assign a zero by if there is only one zero in the row and cross (X)
other zeros in that column.
Step 4: Now assign column wise if there is only one zero in that column and cross other
zeros in that row.
Repeat Step 3 and 4 till all zeros are either assigned or crossed. If the number of
assignments made is equal to number of rows present, then it is the optimal solution
otherwise proceed as follows.
Step 5: Mark the row which is not assigned. Look for crossed zero in that row. Mark
the column containing the crossed zero. Look for assigned zero in that column. Mark the
row containing assigned zero. Repeat this process till all makings are over.
Step 6: Draw straight line through unmarked rows and marked column. The number of
straight line drawn will be equal to number of assignments made.
Step 7: Examine the uncovered elements. Select the minimum.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
a. Subtract it from uncovered elements.
b. Add it at the point of intersection of lines.
c. Leave the rest as it is.
Prepare a New Table.
Step 8: Repeat Steps 5 to 7 till number of allocations = Number of rows.
Example 1: Find the optimum assignment so as to minimize the cost.
Example:
Consider the problem of assigning five jobs to five persons. The assignment
costs are given as follows.
Determine the optimum assignment schedule.
Solution:
Solution: Applying Hungarian Method
Row reduced Matrix
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Column reduced Matrix: Will be same as each columns minimum value is zero. We
start assigning the jobs
Since the number of assignments = 5. It is the optimum solution.
Optimum solution
A to 5 1
B to 1 0
C to 4 2
D to 3 1
E to 2 5
9 Total cost
Routing Problem
Network scheduling is a technique for planning and scheduling of large projects.
It has successfully been applied in transportation and communication problems
.A typical network problem consists of finding route from one node(origin) to
another (destination) between which alternative paths are available at various
stages of the journey. The problem is to select the route that yields minimum
cost. A number of different constraints may be placed on acceptable routes e.g.:
not returning to the node already passed through or passing through every node
once and only once. Problems of such type are called as routing problems.
Although a wide variety of problems other than routing may be developed in
connection with the construction and utilization of networks. Here we consider
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
the special type of routing problem that occurs frequently in O .R.- the travelling
salesman problem.
Unbalanced A.P
If the number of rows = number of column then it is an unbalanced A.P. We
introduce either a dummy row or dummy column and proceed as usual.
Example 2:
Operations Machines
M1 M2 M3 M4
01
02
03
10 15 12 11
9 10 9 12
15 16 16 17
Solution: Introducing a dummy row and applying Hungarian method, we have
Row reduced Matrix and assigning
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Hungarian Method leads to Multiple solution. Selecting (0
3
, M
2
) arbitrarily
The optimum solution is
Infeasible Assignments
Certain jobs cannot be assigned to certain Machines or operators. We insert a high value
O at these cells and Apply Hungarian Method.
Example 3:
Solve the following A.P
Solution: Introducing O at dashes places and applying Hungarian Method. We have
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Row-reduced Matrix
.Optimum assignment is
Maximization In A.P
Hungarian Method cannot be applied to maximization problems. Therefore we reduce it
to maximization problem by subtracting every value form the highest value of the Matrix
and then proceed as usual.
Example 4: Solve the following A.P Efficiency Matrix
Solution: Since it is a Maximization problem we subtract every value from the maximum
value of 90. We have.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Row reduced Matrix
. Optimum Solution
Know about Traveling Salesman Problem
Traveling Salesman Problem
Suppose the salesman has to visit n cities .He wishes to start from a particular city, visit
each city once, and then return to his starting point. The objective is to select the
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
sequence in which the cities are visited in such a way that his total traveling time is
minimized. .Clearly starting from a given city, the salesman will have total of (n-1)!
different sequences .Further, since the salesman has to visit all the n cities; the optimal
solution remains independent of selection of the starting point.
The problem can be represented as a network where the nodes and arcs
represent the cities and the distance between them respectively. Assume that in
a five city problem, a round trip of the salesman is given by the following arcs.
(3,1) ,(1,2) ,(2,4) ,(4,5) ,(5,3)
These arcs taken in order are the first, second, third, fourth and fifth directed arcs for the
trip. Generally the k
th
directed arc represents the k
th
leg of the trip. i.e. on leg
k
The salesman travels from city i to city j
( i , j = 1, 2,, n; )
To formulate the problem whose solution will yield the minimum traveling time ,let
the variables x
ijk
be defined as
Where i, j and k are integers that vary between 1 and n.
Following are the constraints of the problem.
(a) Only one directed arc may be assigned to a specific k ,
Thus
x
ijk
= 1
k =1, 2, 3n
(b) Only one other city may be reached from a specific city i, thus
x
ijk
=1, i = 1, 2, n
(c) Only one other city can initiate a direct arc to a specified city j,thus
x
ijk
=1, j =1, 2, .., n
(d) Given the k
th
directed arc ends at some specific city j,the (k+1)
th
directed arc
must start at the same city j;thus
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
x
ijk
= x
jr
( k +1) for all j and k.
These constraints ensure that the round trip will consist of connected arcs. The
objective function is to minimize
z = d
ij
x
ijk
Where d
ij
is the distance from city i to city j.
Summary
In this unit, we focused on a special type of transportation problem. where the objective
was to allocate n different facilities to n different tasks .Although an assignment problem
can be formulated as a linear programming problem, it is solved by special method know
as Hungarian method. If the number of persons is the same as the number of jobs, the
assignment problem is said to be balanced. Lastly a brief introduction was given about the
traveling salesman problem.
Unit - 7 -Integer Programming Problem
Introduction
The Integer Programming Problem I P P is a special case of L P P where all or
some variables are constrained to assume non-negative integer values. This
type of problem has lot of applications in business and industry where quite often
discrete nature of the variables is involved in many decision making situations.
Eg. In manufacturing the production is frequently scheduled in terms of batches,
lots or runs; In distribution, a shipment must involve a discrete number of trucks
or aircrafts or freight cars
Learning Framing a I.P.P. and Soling a I.P.P. through Gomorys Method
All And Mixed I P P
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
An integer programming problem can be described as follows:
Determine the value of unknowns x
1
, x
2
, , x
n
so as to optimize z = c
1
x
1
+c
2
x
2
+ . . .+ c
n
x
n
subject to the constraints
a
i1
x
1
+ a
i2
x
2
+ . . . + a
in
x
n
=b
i ,
i = 1,2,,m
and x
j
0 j = 1, 2, ,n
where x
j
being an integral value for j = 1, 2, , k n.
If all the variables are constrained to take only integral value i.e. k = n, it is called
an all(or pure) integer programming problem. In case only some of the variables
are restricted to take integral value and rest (n k) variables are free to take any
non negative values, then the problem is known as mixed integer programming
problem.
Gomorys All IPP Method
An optimum solution to an I. P. P. is first obtained by using simplex method
ignoring the restriction of integral values. In the optimum solution if all the
variables have integer values, the current solution will be the desired optimum
integer solution. Otherwise the given IPP is modified by inserting a new
constraint called Gomorys or secondary constraint which represents necessary
condition for integrability and eliminates some non integer solution without losing
any integral solution. After adding the secondary constraint, the problem is then
solved by dual simplex method to get an optimum integral solution. If all the
values of the variables in this solution are integers, an optimum inter-solution is
obtained, otherwise another new constrained is added to the modified L P P and
the procedure is repeated. An optimum integer solution will be reached
eventually after introducing enough new constraints to eliminate all the superior
non integer solutions. The construction of additional constraints, called
secondary or Gomorys constraints, is so very important that it needs special
attention.
Construction Of Gomorys Constraints
Consider a L P P for which an optimum non integer basic feasible solution has been
attained . With usual notations, let this solution be displayed in the following simplex
table.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Clearly the optimum basic feasible solution is given by x
B
= =
; max z = y
00
Since x
B
is a non- integer solution. We assume that y
10
is fractional.
The constraint equation is
y
10
= y
11
x
1
+ y
12
x
2
+ y
13
x
3
+ y
14
x
4
reduces to
y
10
= y
11
x
1
+ x
2
+ y
14
x
4
(1)
because x
2
and x
3
are basic variables ( which implies that y
12
= 1 and y
13
= 0 )
The above equation can be rewritten as
x
2
= y
10
- y
11
x
1
- y
14
x
4
Which is a linear combination of non-basic variables.
Now, since y
10
0 the fractional part of y
10
must also be non negative. We split over each of y
ij
in (1) into an integral part I
ij
, and a
non negative fractional part, f
1j
for j = 0,1,2,3,4. After this break up (1) may be written as
I
10 +
f
10
= (I
11
+ f
11
) x
2
+ (I
14
+ f
14
) x
4
Or
f
10
- f
11
x
2
f
14
x
4
= x
2
+ I
11
x
1
+ I
14
x
4
I
10
-
(2)
Comparing (1) and (2) we come to know that if we add an additional constraint in
such a way that the L.H.S. of (2) is an integer, then we shall be forcing the non-integer y
10
towards an integer. This is what is needed.
The desired Gomorys constraint is f
10
f
11
x
1
f
11
x
4
0.
Let it be possible to have f
10
f
11
x
1
f
11
x
4
= h where h > 0 is an integer. Then f
10
= h +
f
11
x
1
+ f
14
x
4
is greater than one. This contradicts that 0 < f
ij
< 1 for j = 0, 1, 2, 3, 4.
Thus Gomorys constraint is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
When G
sla
(1) is slack variable in the above first Gomory constraint.
This additional constraint is to be included in the given L.P.P. in order to move
further towards obtaining an optimum all integer solution. After the addition of
this constraint, the optimum simplex table looks like as given below
Since f
10
is negative. The optimal solution is infeasible and thus the dual simplex
method is to be applied for obtaining an optimum feasible solution. After obtaining this
solution, the above referred procedure is applied for constructing second Gomorys
constraint. The process is to be continued so long as an all integer solution has not been
obtained.
All I. P. P. Algorithm
The iterative procedure for the solution of an all integer programming problem is as
follows:
Step 1: Convert the minimization I.P.P. into that of maximization, if it is in the
minimization form. The integrality condition should be ignored.
Step 2: Introduce the slack or surplus variables, wherever necessary to convert the
inequations into equations and obtain the optimum solution of the given L.P.P. by using
simplex algorithm.
Step 3: Test the integrality of the optimum solution
If the optimum solution contains all integer values, an optimum basic feasible 1.
integer solution has been obtained.
If the optimum solution does not include all integer values then proceed onto next 2.
step.
Step 4: Examine the constraint equations corresponding to the current optimum solution.
Let these equations be represented by
Where denotes the number of variables and the number of equations.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Choose the largest fraction of ie to find
Let it be or write is as
Step 5: Express each of the negative fractions if any, in the k
th
row of the optimum
simplex table as the sum of a negative integer and a non-negative fraction.
Step 6: Find the Gomorian constraint
and add the equation
to the current set of equation constraints.
Step 7: Starting with this new set of equation constraints, find the new optimum solution
by dual simplex algorithm.
(So that G
sla
(1) is the initial leaving basic variable).
Step 8: If this new optimum solution for the modified L.P.P. is an integer solution. It is
also feasible and optimum for the given I.P.P. otherwise return to step 4 and repeat the
process until an optimum feasible integer solution is obtained.
Example:
Find the optimum integer solution to the following all I.P.P.
Maximize z = x
1
+ 2x
2
Solution:
Step 1: Introducing the slack variables, we get
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Step 2: Ignoring the integer condition, we get the initial simplex table as follows:
Introducing x
2
and leaving x
3
from the basis, we get
Introducing X
1
and leaving X
4
we get the following optimum table.
Optimum table
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The optimum solution these got is: .
Step 3: Since the optimum solution obtained above is not an integer solution, we must go
to next step.
Step 4: Now we select the constraint corresponding to the criterion
maxi (f
Bi
) = max (f
B1
, f
B2
, f
B3
)
= max
Since in this problem, the x
2
equation and x
1
-equation both have the same value of f
Bi
ie
, either one of the two equations can be used.
Now consider the first row of the optimum table . The Gomorys constraint to be added is
Adding this new constraint to the optimum table we get
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Step 5: To apply dual simplex method. Now, in order to remove the infeasibility of the
optimum solution:
, we use the dual simplex method.
i) leaving vector is G
1
(i.e. )
ii) Entering vector is given by
=
Therefore k = 3. So we must enter a
3
corresponding to which x
3
is given in the above
table. Thus dropping G
1
and introducing x
3
. We get the following dual simplex table.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Thus clearly the optimum feasible solution is obtained in integers.
Finally we get the integer optimum solution to the given IPP as x
1
= 4, x
2
= 3 and max z
= 10.
Learn application of Branch and Bound Techniques
The Branch and Bound Technique
Sometimes a few or all the variables of an IPP are constrained by their upper or lower
bounds or by both. The most general technique for the solution of such constrained
optimization problems is the branch and bound technique. The technique is applicable to
both all IPP as well as mixed I.P.P. the technique for a maximization problem is
discussed below:
Let the I.P.P. be
Further let us suppose that for each integer valued x
j
, we can assign lower and upper
bounds for the optimum values of the variable by
j = 1, 2, . r - (5)
The following idea is behind the branch and bound technique
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Consider any variable x
j
, and let I be some integer value satisfying L
j
I U
j
1. Then clearly an optimum solution to (1) through (5) shall also satisfy either
the linear constraint.
To explain how this partitioning helps, let us assume that there were no integer
restrictions (3), and suppose that this then yields an optimal solution to L.P.P. (1), (2),
(4) and (5). Indicating x
1
= 1.66 (for example). Then we formulate and solve two L.P.Ps each containing (1), (2)
and (4). But (5) for j = 1 is modified to be 2 x
1
U
1
in one problem and L
1
x
1
1 in
the other. Further each of these problems process an optimal solution satisfying integer
constraints (3)
Then the solution having the larger value for z is clearly optimum for the given
I.P.P. However, it usually happens that one (or both) of these problems has no
optimal solution satisfying (3), and thus some more computations are necessary.
We now discuss step wise the algorithm that specifies how to apply the
partitioning (6) and (7) in a systematic manner to finally arrive at an optimum
solution.
We start with an initial lower bound for z, say z
(0)
at the first iteration which is less than or
equal to the optimal value z*, this lower bound may be taken as the starting L
j
for some x
j
.
In addition to the lower bound z
(0)
, we also have a list of L.P.Ps (to be called master list)
differing only in the bounds (5). To start with (the 0
th
iteration) the master list contains a
single L.P.P. consisting of (1), (2), (4) and (5). We now discuss below, the step by step
procedure that specifies how the partitioning (6) and (7) can be applied systematically to
eventually get an optimum integer valued solution.
Branch And Bound Algorithm
At the t
th
iteration (t = 0, 1, 2,)
Step 0: If the master list is not empty, choose an L.P.P. out of it. Otherwise stop the
process, Go to step 1.
Step 1: Obtain the optimum solution to the chosen problem. If either
i) It has no feasible solution or
ii) The resulting optimum value of the objective function z is less than or equal to
z
(t)
, then let z
(t+1)
= z
(t)
and return to step 0 otherwise go to step 2.
Step 2: If the so obtained optimum solution satisfies the integer constraints (3) then
record it. Let z
(t+1)
be associated optimum value of z; return to step 0. Otherwise move to
step 3.
Step 3: Select any variable x
j
, j = 1, 2, ., p. that does not have an integer value in the
obtained optimum solution to the L.P.P. chosen in step 0. Let denote this optimal value
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
of x
j
. Add two L.P.Ps to the master list: these L.P.Ps are identical with the L.P.P. chosen
in step 0, except that in one, the lower bound on x
j
is replaced by + 1. Let z
(t+1)
= z
(t)
;
return to step 0.
Note: At the termination of the algorithm, if a feasible integer valued solution yielding z
(t)
has been recorded it is optimum, otherwise no integer valued feasible solution exists.
Example:
Use branch and bound technique to solve the following I.P.P.
Maximize z = 7x
1
+ 9x
2
(1)
x
1
+ 3x
2
< 6 - (2)
7x
1
+ x
2
< 35
0 < x
1
, x
2
< 7 (3)
x
1
, x
2
are integers - (4).
Solution:
At the starting iteration we can consider z
(0)
= 0 to be the lower bound, for x, since all x
j
=
0 is feasible. The master list contains only the L.P.P. (1) (2) and (3). which is designated
as problem 1. Choose it in step 0, and in step 1 determine the optimum solution.
z
0
= 63 (Solution to problem 1)
Since the solution is not integer valued, proceed from step 2 to step 3, and select x
1
. Then
since , place on the master list the following two additional problems.
Returning to step 0: with z
(1)
= z
(0)
= 0,
We choose problem 2 , Step 1 establishes that problem 2 has the feasible
solution
z
0
= 35, x
1
= 5 x
2
= 0 [solution to problem (2)] (5)
Since this satisfies the integer constraints, therefore at step 2 we record it by enclosing in
a rectangle and let z
(2)
= 35.
Returning to step 0 with z
(2)
= 35, we find that problem 3 is available. Step 1 determines
the following optimum feasible solution to it
Z
0
= 58, x
1
= 4, (Problem 3)
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Since the solution is not integer valued, proceed from step 2 to step 3 and select x
2
. Then
= 3. We add the following additional problems on the master list:
Problem 4: (1) (2) and
Problem 5: (1) (2) and 0 x
1
4; 0 x
2
3
Returning to step 0 with z
(3)
= z
(2)
= 35, choose problem 4 from step 1 we know that
problem 4 has no feasible solution and so we again return to step 0 with z
(4)
= z
(3)
= 35.
Only problem 5 is available in the master list. In step 1 we determine the following
optimum solution to this problem.
z
0
= 55 x
1
= 4 x
2
= 3 (solution to problems) - (6)
Since this satisfies the integer constraints, therefore at step 2. We record it by enclosing
inside a rectangle and let z
(5)
= 55.
Returning to step 0, we find that the master list is empty; and thus the algorithm
terminates.
Now, on terminating we find that only two feasible integer solution namely (5)
and (6) have been recorded. The best of these gives the optimum solution to the
given I.P.P. Hence the optimum integer solution to the given I.P.P. is
Z
0
= 55, x
1
= 4, x
2
= 3.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Summary
This chapter investigated the programming model in which the assumption of divisibility
was weakened. You learned two algorithms to determine the optimal solution for an
integer programming problem. One of these was the cutting plane algorithm devised by
Gomory and the other was the branch and bound algorithm developed by land and Doig.
Unit - 8-Infinite Queuing Models
Introduction
We are all too familiar with queues in our day to day existence and perhaps
there is no need to define what a queue is. On the way to place of work often we
have to wait for a bus, for the traffic lights to turn green and then at the office
premises for the lift. Cars waiting at petrol pumps for service, customers waiting
at the bank, telephone subscribers waiting for connections are all a common
sight. Aircrafts get delayed for want of a free runway. No so familiar to all, are the
cases of broken down machines waiting for repairs, workmen waiting for tools,
and goods in production shops waiting for cranes. Even inventory situations may
be regarded as queues not only in that customers wait in line for goods, but also
that goods form a waiting line awaiting consumption.
All the above examples have one common feature. Customers arrive at a service
centre and wait for service. The arrival of customers is not necessarily regular
and so is the time taken for service not uniform. Queues build up during hours of
demand and disappear during the lull period. Personally we do not like to wait.
In the commercial or industrial situations, it may not be economical to have
waiting lines. On the other hand, it may not be feasible or economical to totally
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
avoid queues. An executive dealing with the system then would like to find the
optimal facilities to be provided
Understand the Queuing Process
Queuing Theory
Queuing theory based on probability concepts gives an indication of the
capability of a given system and of the possible changes in its performance with
modification to the system. All the constraints of the process are not taken into
account in the formulation of a queuing model. There is no maximization or
minimization of an objective function. Therefore the application of queuing theory
cannot be viewed as an optimization process. With the help of queuing theory
the executive can at best make an informed guess of what the balance between
customer waiting time and service capability could be. He first considers several
alternatives, evaluates through queuing models, their effect on the system, and
makes his choice. The criteria for evaluation will be measures of efficiency of the
system like the average length of a queue, expected waiting time of a customer
and the average time spent by the customer in the system. In this approach, his
success primarily depends on the alternatives considered and not so much on
the queuing models developed.
It is essential for the executive to have a succinct understanding of the process
so that he will consider the right alternatives and formulate the correct models.
Analysis of a Queuing Process
Very often, the management is interested in finding out the quality of service
rendered to the customers. One of the indictors of efficiency of a system
involving flow of customers for service is how fast they are served. Augmenting
the physical facilities to provide service will in general improve service leading to
customer satisfaction. However, such facilities cannot be indiscriminately
increased as they could be quite expensive. If the queuing process involves
waiting of machines for service, it may be possible to find the economic level of
the maintenance facilities by balancing the cost of machine downtime against the
cost of maintenance facilities. Such evaluation in monetary terms may not
always be possible where the customers are people.
Then some measures of efficiency of the system as mentioned above may prove
helpful.
A typical investigation of a queuing system would comprise the following steps:
Step 1: Preliminary study
At this stage, an analysis of the process is made usually with the help of a flow
diagram and an attempt is made to identify the points which restrict service or
the characteristics which indicate scope for improvement.
Step 2: Exploration of the various alternatives
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
By introducing changes in the constituents of the queueing system, it should be
possible to effect improvement. The arrival pattern may be altered by
withdrawing service facilities to certain categories of customers or by introducing
an appointment system. The time taken for providing service may be improved
by increasing the capacity of the system; i.e. by increasing the number of service
channels or the working hours. It is also not uncommon to provide additional
service facilities to relieve congestion during peak periods. Modifying the queue
discipline may also change the characteristics of the system.
Priority may be given to arrivals involving high cost of waiting time. In multi
channel queues, separate channels of service may be provided for different
types of customers. The fixed change counter provided at the Church gate
Railway Station by the Western Railway is said to have cut down the waiting time
for a majority of customers.
Step 3: Collection of data and analysis
Direct observation of the system in terms of arrival time of customers, service
rate, length of queue and waiting time is done at this stage. The data is analyzed
in sufficient detail to determine the statistical pattern of the variables. The
measures of efficiency of the system are computed on the basis of queuing
theory. These are counterchecked with the results obtained through direct
observation to confirm the validity of the formulae applied. It may sometimes be
necessary to collect additional data regarding the process.
Step 4: Evaluation of alternatives
Effect of modifying the constituents of the system on the basis of the selected
alternatives is evaluated through application of queuing theory. Simulation
technique can also be used for the purpose. Based on the results obtained, the
best combination of the changes to be made in the existing system are selected.
Step 5: Implementation
The proposals formulated are implemented on a small scale and tested. If
necessary, further changes are made and rechecked before implementing them
on full scale. It is advisable to observe the functioning of the system periodically
to ensure that the results desired are maintained.
Constituents of A Queuing System
In specifying a queuing process one should know the following:
(a) Arrival Pattern: The average rate at which customers arrive as well as the
statistical pattern of arrivals
(b) Service facility: When service is available, number of customers that can be
served at a time and the statistical pattern of time taken for service
(c) Queue discipline: Method of choosing a customer for service from amongst
those waiting for service. (First come First served is a common basis but there are
also other methods.)
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
We shall now examine each of these constituents.
Arrival Pattern
The arrival of customers can be regular as in the case of an appointment system
of a dentist which is closely adhered to, or like the flow of components from a
conveyor. The regular pattern of arrivals is not very common nor is very easy to
deal with mathematically. We shall be primarily concerned with a pattern of
completely random arrivals, as it has useful applications.
Completely Random Arrivals
If the number of potential customers is infinitely large, the probability of an arrival in the
next interval of time will not depend upon the number of customers already in the system.
(The assumption is valid by and large, except for queues involving a small finite number
of customers.) Where the arrivals are completely random, they follow Poisson
distribution with mean equal to the average number of arrivals per unit time.
When dealing with arrivals it might sometimes be necessary to distinguish
between groups of customers like male and female callers, or large and small
aircraft.
There are several other types of arrival patterns which we shall not deal with due
to their restricted applications.
Service Facility
i) Availability of service
Apart from specifying the time span over which service is available it is also
necessary to examine if there are any constraints which reduce the number of
customers that can be served at a time. For instance, in a waiting line for a suburban
train, the probability distribution of the number of passengers that can be
accommodated in a train that arrives is relevant apart from the timings of the train
services.
ii) Number of Service Centers
If there is only one service centre referred to as a service channel, obviously only one
customer can be served at a time. Often there will be more than one service centre
and the behaviour of the queues will vary with the number of channels available for
service.
Multiple service channels may be arranged in series or in parallel. If the customer
has to go through several counters one after the other each providing a different part
of the service required, the arrangement is said to be in series.
Normally for withdrawals in a bank, the customer has to go to at least two counters
in series.
Provision of several counters offering the same type of service is a common
approach to reduce congestion. Ticket booths in a railway station have this type of
multiple channels with parallel arrangement.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
iii) Duration of service
This is the length of time taken to serve a customer. This can be constant or varying.
(a) Constant service time
This is not strictly realized in practice, but an assumption that service time
is constant holds true if the pattern of arrivals is very irregular
(b) Completely random service time
The service time can be considered completely random where the server
does not distinguish between the various arrivals, and he does not change
deliberately the duration of service on the basis of the time taken to serve the
previous arrival. The server forgets the time for which he has been serving a
customer. Under these conditions, the service time follows exponential
distribution with mean equal to reciprocal of the average rate of service.
(c) Service time following Erlang Distribution
In cases where the assumption of an exponential distribution for service
time is not valid, Erlang family of service time distributions is used.
S.No. Situation Arriving
Customers
Service Facility
1 Passage of customers through
supermarket checkout
Shoppers Checkout counters
2. Flow of automobile traffic through a road
network.
Automobiles Road Network
3. Transfer of electronic messages Electronic
messages
Transmission lines
4. Banking transactions Bank patrons Bank tellers
5. Flow of computer programmers through a
computer system.
Computer
Programmers
Central processing
unit
6. Sale of theatre tickets Theatre-goers Ticket booking
windows
7. Arrival of trucks to carry fruits and
vegetables from a central market
Trucks Loading crews and
facilities
8. Registration of unemployed at
employment exchange
Unemployed
personnel
Registration
assistants
9. Occurrences of fires Fires Firemen and
equipment
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
10.
Flow of ships to the seashore ships Harbor and docking
facilities
11 Calls at police control room Service calls policemen
Queue Discipline
The pattern of selection for service from the pool of customers is of two types.
The common pattern is to select in the order in which the customers arrive. First
come first served is a common example. In issuing materials from a stores
inventory sometimes the storekeeper follows the Last In First Out principle
because of the convenience it offers for removal from stocks and handling.
There can also be queues which accord priority to certain types of customers.
Here again there can be two approaches. In case of non pre-emptive priority the
customer getting service is allowed to continue with service till completion, even
if a priority customer arrives midway during his service. This is a common form
of priority. Pre-emptive priority involves stopping the service of the non priority
customers as soon as the priority customer arrives. Priority given to repairs of a
production holding machine over an auxiliary unit for allocation of maintenance
labour force is a typical example. Preference is given to larger ships over the
smaller ones irrespective of the order in which they arrive for allocation of berths.
Customer Behavior
a) Balking: Arriving customers are said to balk if they do not join a queue
because of their reluctance to wait.
b) Collusion: Customers may be in collusion in the sense that only one person
would join the queue but would demand service on behalf of several customers.
c) Reneging: Impatient customers who would not wait beyond a certain time and
leave the queue are said to renege.
d) Jockeying: Some customers keep on switching over from one queue to another
in a multiple service centre. This is called jockeying.
Server Behavior
Although the timings for service might have been specified, the server may
not be available through the entire span of time. For instance, after every
hour, he may leave the service centre for 5 minutes, to attend to his
personal needs.
Learn Mathematical Analysis of Queuing Process
Mathematical Analysis Of Queuing Process
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Statistical Equilibrium: In analyzing the queuing process, we are interested in developing
mathematical model which represent the system for a major part of the time. This implies
that changes occurring in the characteristics of the system are not to be considered if they
are of short durations. When specifying the statistical distributions of the arrivals or
service times, we are interested in an equilibrium state.
In a queuing process, at each point of time, there is a probability distribution for
the length of the queue. The number of customers will be very much different 15
minutes after opening the counter in a post office from that after one hour. After
the initial rush, one might reasonably expect to find the system with the same
type of probabilities of arrivals.
The probability distribution of the arrivals will then be different from that of the initial
state. The state during which the probability distribution remains the same, is called the
steady state and the system is said to have acquired a state of statistical equilibrium.
In the steady state, there will be variations in the queue from time to time but the
probability distributions representing the queueing process will remain the same and are
independent of the time at which the system is examined.
Properties of the System
In developing queuing models, we shall confine our discussion to queuing systems with
the following properties.
Arrivals
Customers are discrete entities
Population Finite/Infinite
No simultaneous arrivals
Pattern of arrivals in a time period t
0
follows Poisson distribution with average arrival
rate
x = 0, 1, 2 ..
Service
Single serve channel/Multiple channels
Single queue/Infinite capacity
Pattern discipline
First come First served
It may be noted that when the number of arrivals follows Poisson distribution
(discrete), the inter-arrival time i.e., the time between arrivals follows exponential
distribution (continuous).
Notations
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The Queuing Systems with which we are concerned are denoted by M/M/1 and
M/M/c where Ms stand for exponential inter arrival and exponential service time
distributions, and the third figure indicates the number of channels (or servers)
available (1 or c).
Learn Using Different Models to Solve Practical Problems
Single Channel Models
The formulae are listed in Tables 1 and 2. The examples which follow not only
illustrate the computation of various measures of efficiency of a queuing system
but give an idea of the areas of application as well.
Table 1
Formulae for Poisson Arrivals, Exponential Service, Single Channel Queuing Models
Infinite Population
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Table 2
Formulae for Poisson Arrivals, exponential Service, Single Channel
Queuing Models Number of customers limited to N
n = 0, 1, 2 . N
Example 1:
Patrons arrive at a small post office at the rate of 30 per hour. Service by the clerk on duty
takes an average of 1 minute per customer
a) Calculate the mean customer time
(i) spent waiting in line
(ii) spent receiving or waiting for service
b) Find the mean number of persons
(i) in line
(ii) receiving or waiting for service
Solution:
Mean arrival rate customers per hour
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
= customer per minute
Mean service rate / = 1 per minute
Traffic intensity
(a) (i) Mean customer time spent waiting in line minute
(ii) Mean customer time receiving or waiting for service
= 2 minutes
(b) (i) Mean number of persons in line
customer
(ii) Mean number of persons receiving or waiting for service
customer
Example 2:
The Tool Companys Quality Control Department is manned by a single clerk,
who takes an average of 5 minutes in checking parts of each of the machines
coming for inspection. The machines arrive once in every 8 minutes on the
average. One hour of machine is valued at Rs. 15 and a clerks time is valued at
Rs. 4 per hour. What are the average hourly queuing system costs associated
with the Quality Control Department?
Solution:
Mean arrival rate / = 1/8 per minute
= per hour
Mean service rate per minute
= 12 per hour
Average time spent by a machine in the system E(v) =
= hours
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Average queuing cost per machine is
An average arrival of machines per hour costs
= Rs. 25 per hour
Average hourly queuing cost = Rs. 25
Average hourly cost for the clerk = Rs. 4
Hence total hourly cost for the department = Rs. 29
Example 3:
A hospital emergency room can accommodate at most M = 5 patients. The patients arrive
at a rate of 4 per hour. The single staff physician can only treat 5 patients per hour. Any
patient overflow is directed to another hospital.
(a) Determine the probability distribution for the number of patients either waiting
for or receiving treatment at any given time.
(b) Determine the mean values for the number of patients in the emergency room ,
and the number of patients waiting to see the doctor.
Solution:
M = Maximum number of customers in the system = 5
/ = 4 per hour and = 5 per hour
The probability distribution for the number of patients in the system
The probability distribution for the number of patients either waiting for treatment
or receiving treatment is as follows.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(b) (i) The mean value for the number of patients in the emergency room =
0 (0.271) + 1 (0.217) + 2 (0.173) + 3 (0.139) + 4 (0.111) + 5 (0.089)
= 1.869
(ii) Average number of patients waiting to see the doctor E(m) = 0 0.271
+ 0 0.217 + 1 0.173 + 2 0.139 + 3 0.111 + 4 0.089 = 1.140.
Example 4:
Cars arrive at a toll gate on a frequency according to Poisson distribution with
mean 90 per hour. Average time for passing through the gate is 38 seconds.
Drivers complain of long waiting time. Authorities are willing to decrease the
passing time through the gate to 30 seconds by introducing new automatic
devices. This can be justified only if under the old system, the number of waiting
cars exceeds 5. In addition the percentage of gates idle time under the new
system should not exceed 10. Can the new device be justified?
Solution:
In the old system,
Mean arrival rate = 90 per hour
= 1.5 per minute
Mean service rate = per minute
Traffic intensity =
Expected number of waiting cars
Hence the new system is justified on the basis of the expected number of waiting
cars.
Under the new system, the probability of device being idle
= P
0
= 1
= 1 .75
= .25 which is greater than 10% required for the new device to be introduced.
As only one of the stipulated conditions is fulfilled the new automatic device is
not justified.
Example 5:
Customers arrive at a one-window drive-in bank according to a Poisson distribution with
mean 10 per hour. Service time per customer is exponential with mean 5 minutes. The
space in front of the window, including that for the serviced car can accommodate a
maximum of 3 cars. Other cars can wait outside this space.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(a) What is the probability that an arriving customer can drive directly to the space
in front of the window?
(b) What is the probability that an arriving customer will have to wait outside the
indicated space?
(c) How long is an arriving customer expected to wait before starting service?
(d) How many spaces should be provided in front of the window so that all the
arriving customers can wait in front of the window at least 20% of the time?
Solution:
(a) .
The probability that an arriving customer can drive directly to the space in
front of the window
= P
0
+ P
1
+ P
2
= P
o
+ pP
o
+ p
2
P
o
= P
o
(1+p + p
2
)
= (1 p) (1+p+p
2
)
Hence required probability = .42
(b) The probability that an arriving customer has to wait outside the indicated space
= Probability that there are at least 3 customers in the space in front of the window
= 1 (P
0
+P
1
+ P
2
+ P
3
)
= 1 (1 p) (1+p+p
2
+ p
8
)
= .48
(c) The average waiting time of a customer in queue.
=
(d) The existing space in front of the window can accommodate upto three cars.
Probability that a customer can wait in front of the window must be at least 0.20. This
is possible if the number of customers in the system is 0, 1 or 2 for at least 20% of the
time.
It seen from (a) that P
0
+ P
1
+ P
2
= 0.42
As even with one space in front of the window 42% of the times an arrival an wait in
the space, the number of spaces required is one or more.
Multiple Service Channels
The analysis of systems involving several service channels is more complex. However
these models are of major practical utility as one of the ways of improving service is to
provide additional service facilities. The list of formulae to be used is given in Table
below this is following by a few examples.
Table
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Formulae for Poisson Arrivals, Exponential Service
Multi channel Queuing Models Infinite Population
Example 6:
Ships arrive at a port at a rate of one in every 3 hours, with a negative exponential
distribution of interarrival times. The time a ship occupies a berth for unloading and
loading has a negative exponential distribution with an average of 12 hours. If the average
delay of ships waiting for berths is to be kept below 6 hours, how many berths should
there be at the port.
Solution:
For multichannel queues, < 1, to ensure that the queue does not explode.
. c > 4
Let us calculate the waiting time when c = 5
Substituting for n, c and p
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Average waiting time of a ship
= 6.65
Which is greater than 6 hours and inadequate.
When c = 6, P
o
=
And average waiting time of a ship E(w) = 1.71 hours.
Hence 6 berths should be provided at the port.
Example 7:
A tool crib is operated by M servers, and demands for service arrive randomly at mean
rate of 1.4 per minute. The mean service time per server is 1.25 minutes, and times are
exponentially distributed. Finite population effects may be ignored. If the average hourly
pay rate of a tool crib operator is Rs. 2 per hour and the average hourly pay rate of
production employee is Rs. 4 per hour, determine the optimum value of M. (You may
assume an 8 hour working day).
/ = 1.4 per minute
= .80 per minute
Substituting for and c,
P
0
will be 0.067, 0.156 and 0.170 for c = 2, 3 and 4 respectively. Average waiting time of
an arrival in the queue.
As the number of arrivals is per minute average waiting time of production
employee per day = E (w)
/
60 8
= E (v) 672
For a c channel system, one server will be idle if n = c 1, two will be idle for n = c 2
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
and so on. It can be shown that average idle time of service facilities .
For various values of c, the cost implications are presented in the Table below:
Table
The optimal number of servers is seen to be three
Example 8:
A bank has two counters for withdrawals. Counter one handles withdrawals of value less
than Rs. 300 and counter two Rs. 300 and above. Analysis of service time shows a
negative exponential distribution with mean service time of 6 minutes per customer for
both the counters. Arrival of customers follows Poisson distribution with mean eight per
hour for counter one and five per hour for counter two.
(a) What are the average waiting times per customer of each counter?
(b) If each counter could handle all withdrawals irrespective of their value how
would the average waiting time change?
Solution:
(a) For counter one
1
= 10 per hour, /
1
= 8 per hour.
For counter two,
2
= 10 per hour, per hour, /
2
= 5 per hour
(b) When each counter can handle all withdrawals irrespective of their value with
an average arrival rate of 13 per hour, mean service time of 10 per hour at each
counter and number of channels = 2,
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Average waiting time E(w) =
= 4.2 minutes
Example 9:
St. Peters Public Library wants to improve its service facilities in terms of reducing the
waiting time of the borrowers. At present there are two counters in operation and the
arrival of the borrowers follows Poisson distribution with one arriving every 4 minutes on
an average and the service time follows Poisson distribution with one arriving every 4
minutes on an average and the service time follows a negative exponential distribution
with a mean of 5 minutes. The library has relaxed its membership rules and a substantial
increase in the number of borrowers is expected . Find the number of additional counters
to be provided if the present arrival rate is expected to double and the average waiting
time of a borrower is to be half of the present value
Solution:
/ = 1/4 per minute, = 1/5 per minute,
= 5/4
The number of counters at present c = 2
= 4.33
P
0
= .23
Average waiting time of a customer
E (w) =
= 3.19 minutes.
As the arrival rate is expected to be doubled with the relaxation of membership
rules, / = 2/4 = 1/2 per minute and = 1/5 per minute.
Hence = 5/2
We shall examine the average waiting time of a customer for different number of
counters
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
When c = 3
P
0
1
= 22.25
P
0
= .04
E(w) = 7.02 minutes
This is more than the present average waiting time.
By adding one more counter, c = 4 and E(w) = 1.07 minute.
This is less than half of the present average waiting time. Thus it will be
necessary to provide 2 additional counters to fulfill the stipulated service
standards.
Erlang Family Of Distribution Of Service Times
Queuing processes discussed so far are mathematically simple. When assuming that the
service time follows negative exponential distribution, we were also taking that its
standard deviation is equal to its mean. As there would be situations where the mean and
the standard deviation substantially differ, the models have to be made more general by
using a distribution which conforms closely to the practical problems but yet retains the
simplicity of the properties of negative exponential distribution. A.K. Erlang had first
studied such a distribution.
Consider the distribution of a service time involving a fixed number of phases k,
each phase having a negative exponential distribution. If there are k phases, and
the average time taken by a customer through each phase is units, the service
time distribution f(t) is given by
The mean of this distribution is and standard deviation is
If k = 1, f (f) = e
1
which is the negative exponential distribution; same as the models
considered earlier.
The mode is located at t = .
If k = the variance is zero and this corresponds to a case where the service time is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
constant and has value
The Figure below shows the way the density functions varies as k increases.
Fig. 7.1
The measures of efficiency should take into account the number of customers getting
service, number having entered any one or more of the phases, and the number yet to join.
For arrivals following Poisson distribution, with mean / and service time following the
k
th
Erlang distribution with mean, the formulae applicable are given in Table below
Table
Formulae for Poisson Arrivals, Service with k phases, each phase having a negative
exponential distribution
Average queue length
Average number of units in the system
Average waiting time
Average time spent in the system
For constant service time, equating k to
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Example 10:
In a cafeteria at a bus depot the customers will have to pass through three counters. At
the first counter the customers buy Coupons at the second they select and collect the
snacks to be taken, and at the, third they collect coffee or tea as required. The server at
each counter takes on an average two minutes although the distribution of the time of
service is approximately exponential. If the arrivals of customers to the cafeteria are
approximately Poisson at an average rate of six per hour, what is the average time spent
by a customer waiting in the cafeteria ? What is the average time of getting the service ?
What is the most probable time spent in getting the service ?
Solution
This is a queuing process with service time following Erlang distribution.
No. of phases k = 3
. = 1/6 persons per minute
/ = 6 persons per hour
= 1/10 person per minute
(i) Average waiting time =
= 6 minutes
(ii) Average time spent in collecting coupons, snacks, etc
= = 6 minutes
(iii) Most probable time spent in collecting coupons, snacks etc
= Modal value
=
= 4 minutes
Summary
The flow of customers forms infinite population towards a service form secure on
account of lack of services facilities. The waiting line theory or Queuing Theory analysis
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
the number of facilities required and the cost of customers waiting time and suggest the
optimum service level. It contributes vital information required for balancing the cost of
service and cost associated with waiting time of the customer. There are different models
which are used under different conditions all these models were discussed in this unit.
Unit - 9-Finite Queuing Models
Introduction
The models discussed so far relate to situations involving infinite population of
customers i.e. the queue can increase indefinitely. There will be cases, where
the possible number of arrivals is limited and is relatively small. In a production
shop, if the machines are considered as customers requiring service from repair
crews or operators, the population is restricted to the total number of machines
in the shop. In a hospital ward, the probability of the doctors or nurses being
called for service is governed by the number of beds in the ward. Similarly, in an
aircraft the number of seats is finite and the number of stewardesses provided by
the airlines will be based on the consideration of the maximum number of
passengers who can demand service. As in the case of a queuing system with
infinite population, the efficiency of the system can be improved in terms of
reducing the average length of queues, average waiting time and time spent by
the customer in the system by increasing the number of service channels.
However, such increases mean additional cost and will have to be balanced with
the benefits likely to accrue. If the queuing system in a machine shop is under
study, the cost of providing additional maintenance crews or operators can be
compared with the value of additional production possible due to reduced
downtime of the machines. In cases where it is not possible to quantify the
benefits, the management will have to base its decisions on the desired
standards for customer service.
Understand the Difference between Finite and Infinite Queuing Model
The queue discipline in a finite queuing process can be:
(i) First come-first served
(ii) Priority e.g.: Machines of high cost may be given priority for
maintenance while others may be kept waiting even if they had broken
down before.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(iii) Random e.g.: in a machine shop if a single operator is attending to
several machines and several machines call for his attention at a time, he
may attend first to the one nearest to him
The analysis of Finite Queuing Models is more complex than those with infinite
population although the approach is similar. L.G. Peck and R. N. Hazelwood
have provided solutions to such problems in their book Finite Queuing Tables
(John Wiley & Sons Inc 1958)
Finite Queuing Tables
Notations used are different and are given below:
N = Population (machines, customers etc.)
M = Service channels (repairmen, telephone lines etc.)
T = Average service time (repair time, length of conversation on a telephone
etc.)
W = Average waiting time
U = Average running time (of machines) or mean time between calls for
service per unit
H = Average number of units being serviced
L = Average number of units waiting for service
J = Average number of units in operation
F = Efficiency Factor
X = Service Factor
D = Probability that if a unit calls for service, it will have to wait.
Let us consider a machine shop with N machines. The inter breakdown time of
these machines follows a negative exponential distribution with mean U. The
number of breakdowns follows Poisson distribution with mean
It is assumed that machines are kept running (or in operation) except when they are under
repairs or waiting for repair crew to attend.
If M repair crews are available, the time taken by any crew follows a negative
exponential distribution with mean T. Naturally, a machine which has broken
down will have to wait for repairs if all the repair crews are busy.
Learn about Different Mathematical Models Avaialble
Measures Of System Efficiency
Efficiency of the Repair System
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
For a given set of machines, the efficiency of the repair system may be judged
by the extent to which machines have to wait for repairs. If W is the average time
for which a machine has to wait, the efficiency factor F is defined as
At any point of time, a machine will either be running or under repair or waiting for
repairs. Therefore, the total number of machines N = J + H + L.
correspond to the probability that a machine is being
repaired, running or waiting for repairs respectively.
In the finite queuing tables, service factor X is defined as X is an indicator of the
utilisation of repair crew.
The formulae for other properties of the. system are given below:
Use Of Finite Queuing Tables:
The tables give the values of F and D for different values of N, M and X. They are
arranged in the ascending order of the values of the population. For each N, the
value of X increases from .001 to .950. For a given service factor X, several values
of M can be found.
For each value of X and M, values of D and F are tabulated.
The steps in the use of Finite Queuing Tables may be summarised as follows:
(i) Find mean service time T and mean running time U.
(ii) Compute the service factor
(iii) Select the table corresponding to the population N.
(iv) For the given population, locate the service factor value.
(v) Read off from tables, values of D and F for the number of service crews M. If
necessary, these values may be interpolated between relevant values of X.
(vi) Calculate the other measures L, W, H, and J from the formulae given.
The overall efficiency F of the system will increase with the number of service
channels (M) provided. As mentioned earlier, addition of service crews involves
cost, which should be justified by the increase in the efficiency of the system i.e.
additional running time of machines possible. However, it will be seen from the
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
tables that as M increases, the rate of increase in efficiency decreases. The
practical significance is that beyond a certain value of M, it is not worthwhile
increasing M as there would be no appreciable increase in the efficiency of the
system.
Example 1:
In a chemical factory there are five hoppers of identical size which feed material to
grinding mills. Due to changes in the requirement of material, there are variations in the
time taken for emptying the hoppers. On the basis of past experience this time was found
to follow negative exponential distribution with an average of 10 hours between getting
emptied. Whenever a hopper gets empty it has to be filled by a pay loader. Although the
capacity of the hoppers is the same, the time taken to fill the hoppers varies due to
different locations from which the material is to be loaded.
The time for filling the hopper also was found to follow negative exponential distribution
with an average of 2.5 hours. The company hires the pay loaders at a cost of Rs. 100 per
hour irrespective of whether it is operated or not. If the mill has to be stopped due to its
hopper getting empty it costs Rs. 1000 per hour in terms of loss of profits. Determine the
number of pay loaders which the company should engage to minimize overall cost.
Solution:
Since T = 2.5 and U = 10
For N = 5, X = 0.200 we have the following values from the tables:
M D F
3 0.028 0.998
2 0.194 0.976
1 0.689 0.801
We now prepare a table as below:
Table
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
As increasing the number of pay loaders beyond two reduces profits, the
company should engage only two pay loaders.
Summary
Queuing theory deals with situations where customers arrive, wait for the service,
get the service and leave the system. Customers, human or nor human, may
come individually or in groups from large/small population, at known/variable
times, from one or more queues and move in a certain order then to the service
station/s providing service whose speed may be fixed or variable. Queuing
systems are analyzed for determining the optimal service level, where the total
cost of providing service and waiting, is minimized. An increase in the service
level increases the cost of providing service but reduces the cost of waiting, while
a decrease in the service level induces opposite changes.
Unit - 10 -Simulation
Introduction
Generally in developing mathematical models of various systems or situations, it is
assumed that the statistical distribution of the variables conforms to a standard pattern.
This, however, is not always true. In a typical pricing problem, the management cannot
risk changing the price of the product without evaluating the various alternatives. Also,
representation of the reality in terms of a mathematical model becomes virtually
impossible because of the complexity of the interaction of several variables having a
bearing on the final outcome. One approach to the problem is to assign probabilities of
achieving various sales targets under different conditions of completion with changes in
price, demand, etc. and choose the alternative which gives the maximum profit. Where
formulating a mathematical model is difficult, simulation is of great help for decision
making.
Learn about the Basic Concepts of Simulation and how It is Applied in Solving
Business Problems
Basic Concepts
Simulation may be called experimentation in the management laboratory. In the
context of business problems, simulation is often referred to as Monte Carlo
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Analysis. The expression may be traced to two American mathematicians, Von
Neumann and Ulan, who in the late 1940s found a problem in the field of
nuclear physics too complex for analytical solution and too dangerous for actual
experimentation. Eventually they arrived at an approximate solution by sampling.
The method used by them somewhat resembles the manner in which gamblers
develop betting systems on the roulette table and the name Monte Carlos stuck.
Imagine a betting game in which the stakes are based on correct prediction of
the number of heads which occur when five coins are tossed. If it were only a
question of one coin; most people know that there is an equal likelihood of a
head or a tail occurring. i.e., the probability of a head is . However, without the
application of probability theory, it would be difficult to predict the chances of
getting various numbers of heads when five coins are tossed. We may take five
coins and toss them repeatedly. The outcomes may be noted for each toss and,
say, after every ten tosses the probabilities of various outcomes may be
estimated. As we know, the values of these probabilities will initially fluctuate but
they would tend to stabilise as the number of tosses is increased. This approach
in effect is a method of sampling but is not very convenient. Instead of actually
tossing the coins, we may carry out the experiment by using random numbers.
Random numbers have the property that any number is equally likely to occur
irrespective of the digit that has already occurred.
Let us estimate the probability of tossing of different numbers of heads with five
coins. We start with set random numbers given below:
78466 71923
78722 78870
06401 61208
04754 05003
97118 95983
By following a convention that even digits signify a head (H) and the odd digits
represent a tail (T), the tossing of a coin can be simulated. The probability of
occurrence of the first set of digits is and that of the other set is also a
condition corresponding to the probability of the occurrence of a head and the
probability of occurrence of a tail respectively.
It is immaterial which set of five digits should signify a head. The rule could be
that the digits 0, 1, 2, 3 and 4 represent a head and the digits 5, 6, 7, 8 and 9 a
tail. It is only necessary to take care that the set of random numbers allotted to
any event matches with its probability of occurrence. For instance, if we are
interested in allotting random numbers to three events A, B and C with
respective probabilities 0.24, 0.36 and 0.40 we choose two digit random
numbers 00 to 99.
The numbers 00 to 23 signify event A,
24 to 59 signify B and
60 to 99 signify C.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The first set of five random digits in the list of random numbers implies that the
outcome of the first toss of 5 coins is as follows:
Coin 1 2 3 4 5
Random number 7 8 4 6 6
Outcome T H H H H
Hence it is 4 heads and 1 tail.
Proceeding in the same way, we can tabulate the results of the first ten tosses.
Based on the ten tosses of the coins, the estimates of probabilities of occurrence of
different numbers of heads are:
0 Head 0
1 Head
2 Heads 0
3 Heads
4 Heads
5 Heads 0
As these estimates will come closer to the theoretical value with increasing
sample size, the experiment is to be continued further.
The results for obtaining 2 heads and 3 tails for 100 throws are shown below:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
In the first 10 throws 0
20 throws 6
30 throws 11
40 throws 14
50 throws 18
60 throws 19
70 throws 21
80 throws 22
90 throws 24
100 throws 27
Table 2 compares the final results at the end of 100 throws with the theoretical
probabilities.
It is observed that the results obtained with the large sample of 100 compare more
favourably with the theoretical values, than with a sample of ten sets of numbers.
Simulation Procedure
The approach adopted for solving a problem in gambling can be extended to
decision making in business where risk is a common feature. The probabilities
associated with the variables can be estimated on the basis of past data if
available, or by inputting subjective values.
In any simulation problems the variables to be studied will be given with
associated probabilities. The initial conditions will also be specified. We can
choose random number from table. However to get uniform results the random
numbers to be used will be specified. The first step is we code the data, i,e, we
assign random numbers to the variable. We identify the relationship between the
variables and run the simulation to get the results
Let us illustrate this by a simple example of a queuing process.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Example 1:
A sample of 100 arrivals of customers at a retail sales depot is according to the
following distribution:
A study of the time required to service customers by adding up the bills, receiving
payment, making change and placing packages in hand trucks, yields the
following distribution:
Estimate the average percentage customer waiting time and average percentage
idle time of the server by simulation for the next 10 arrivals.
Solution:
Step 1 : Convert the frequency distributions of time between arrivals and
service time to cumulative probability distributions.
Step 2 : Allocate random numbers 00 to 99 for each of the values of time
between arrivals and service time, the range allocated to each value
corresponding to the value of cumulative probability. (Tables 12 -3 and 12 4)
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Step 3 : Using random numbers from table, sample at random the tome of
arrival and service time for ten sets of random numbers.
Step 4 : Tabulate waiting time of arrivals and idle time of servers.
(Table 12-5)
Step 5 : Estimate the percent waiting time of arrivals and percent idle time
of servers corresponding to the ten samples.
Allocation of Random Numbers
TABLE 3
Allocation of Random Number Time between arrivals
TABLE 4
Allocation of Random Numbers Service Time
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(Note that the upper bound of random numbers allocated for each value of the parameter
is one less than the corresponding cumulative frequency since we have chosen a range of
random numbers form 00 to 99)
TABLE 5
The service facility is made available at clock time zero and the server has to be idle for
3.5 minutes when the service for first arrival starts. The service is completed at 5.0
minutes and again the server is idle for 2 minutes till the second arrival joins the system.
The first three arrivals get immediate service and they dont have to wait, as the server is
idle when they arrive. The fourth arrival who joins at 9.0 minutes has to wait for .5
minute when the service for the third is completed. Similarly the waiting time and idle
time can be computed for further arrivals.
Total elapsed time = 29 minutes
Waiting time of arrival = 1 minute
Percentage of waiting time =
Idle time for server = 14.5 minutes
Percentage of idle time =
Use of Random Number Tables
The random numbers could be selected by any random process, such as drawing
numbered chips from a hat. However, it is convenient to use a table of random numbers
which in fact is prepared on the basis of some such physical phenomenon. The grouping
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
of random numbers in the tables has no significance and one should be concerned with
individual digits only. The first random number could be picked at random from any
point in the tables and the subsequent ones are to be selected proceeding sequentially
either in a vertical or horizontal direction. Depending upon the number of digits required,
the random numbers will be chosen in sets of single digit, two digit numbers, etc.
Pseudo Random Numbers
Truly random numbers cannot be produced by an algorithm and hence random
numbers generated by using a recursive equation are referred to as pseudo
Random Numbers.
There are several methods of generating pseudo random numbers but we shall briefly
describe only the Mid Square Method . Operation starts with an arbitrary four digit integer
called the seed. To obtain the first random number, the seed is squared and all digits
except the middle four are ignored. The process is then repeated each time using the
previous random number as the new seed.
Seed
= 8695
= 75603025
Taking the middle 4 digits,
= 6030
= 36360900
= 3609
Repeating the above procedure:
One of the basic disadvantages of the mid square method is that the generated
numbers may start cycling after a short set of random numbers is obtained.
There are methods by which the seed can be chosen, so as to obtain a fairly long sequence
of numbers before cycling starts. Also statistical tests are available to check whether the
generated sequence is truly random.
Sample Size
As we have seen with the coin tossing experiment, the larger the number of
trials, the more confident we can be in our estimates. The question that arises is
how many trials for simulation? If the experiment is as simple as tossing a coin
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
involving only one variable, the sample size required for a given confidence level
at a specified degree of accuracy can be worked out.
Example 2
If it is needed to be 95% certain of being correct in an experiment with marginal
error of 1% of the true value, what should be the sample size?
Solution: Let be the proportion as percent of success.
Standard error will be
Where is the sample size.
The standard normal deviate value corresponding to 95% confidence level from
normal tables is 1.96.
Margin of error = 1.96
is maximum for = 50
Hence, the value of = 9600.
Usually, a simulation model involves several variables and it may not be possible to
determine the number of trials required to obtain the desired accuracy at a specified
confidence level.
One can only say that the accuracy associated with simulation improves as the square root
of the number of trials and hence there is a need for a large number of trials.
This calls for a great deal of computational effort and for most real life problems, the use
of computer becomes inevitable. In fact, special simulation languages such as GPSS and
SIMSCRIPT have been developed to save time and effort required to structure and debug
simulation models.
A practical indicator of when to stop simulation trials is given by the fact that the results
which violently fluctuate initially tend to stabilize as the simulation is continued. If the
successive cumulative results tally reasonable well, the simulation may be stopped. The
degree of accuracy required, of course, varies with the problem on hand and calls for the
judgment of the analyst.
Learn about Application and Limitations of Simulation
Application of Simulation
The range of application of simulation in business is extremely wide. Unlike the
other mathematical models, through abstract, simulation can be easily
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
understood by the users and thereby facilitates their active involvement. This, in
turn, makes the results more reliable and also ensures easy acceptance for
implementation. The degree to which a simulation model can be made close to
reality is dependent upon the ingenuity of the O.R team who should identify the
relevant variables as well as their behaviour.
We have already seen by means of an example how simulation could be used in a queuing
system. It can also be employed for a wide variety of problems encountered in production
systems the policy for optimal maintenance in terms of frequency of replacement of
spares or preventive maintenance, number of maintenance crews, number of equipment
for handling materials, job shop scheduling, routing problems, stock control and so forth.
The other areas of application include dock facilities, facilities at airports to minimize
congestion, hospital appointment systems and even management games.
As in the case of other O.R. models, with the help of simulation, the manager tries to
strike a balance between opposing costs of providing facilities (which usually mean long
term commitment of funds) and the opportunity and other costs of not providing them.
Limitations:
Simulation approach is recognized as a powerful tool for management decision making.
This does not mean that one should ignore the cost associated with a simulation study for
data collection, formation of the model and the computer time. Often this is quite
significant.
A simulation application is based on the premise that the behaviour pattern of relevant
variables is known, and this very premise sometimes becomes questionable. Not always
can the probabilities be estimated with ease or desired reliability. The results of
simulation should always be compared with solutions obtained by other methods
wherever possible, and tempered with managerial judgment.
Some Examples
Example 3
A bread vendor buys every morning loaves of bread at 0.45 each by placing his
order one day in advance (at the time of receiving his previous order) and sells
them at Rs. 0.70 each. Unsold bread can be sold the next day at Rs. 0.20 per
loaf and thereafter should be treated as of no value. The pattern of demand for
bread is given below:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The vendor adopts the following order rule. If there is no stock with him at the
end of the previous day, he orders 60 units. Otherwise he orders 50 or 55
whichever is nearest the actual fresh bread sale on the previous day. Starting
with zero stock and a pending order for 55 loaves, simulate for 10 days and
calculate the vendors profits.
Solution
TABLE 6
Allocation of random numbers
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
We can now construct a table to see, through simulation how the stocks and
sales fluctuate.
TABLE 7
Results of simulation
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
+ Represents lost sales as stock is limited
@ previous days closing stock is zero
Estimated profit = (549 x 0.70 + 2 x 0.20) 570 x 0.45 = Rs. 128.20
Example 4
The maintenance manager of a chemical company is interested in determining a
rational policy for maintenance of a pneumatic conveying equipment. The
equipment is a part of the process line and hence production holding. It has one
bearing each on the inlet side (A) and the outlet side (B). Whenever there is a
failure of any bearing, it has to be replaced immediately. The company has a
good system of maintaining records on performance of the equipment and the
following data is available:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The cost of bearing is Rs. 300 each for A and Rs. 500 each for B. The cost of
downtime of equipment is Rs. 700 per hour, and it takes 2 hours to replace one bearing
either at inlet or outlet side and 3 hours to replace both the bearings.
The three maintenance policies to be evaluated are
replace a bearing only when it fails, 1.
replace both the bearings if one fails, 2.
replace the bearing which fails plus the other one if it has been in use for more 3.
than its
estimated average service life i.e. 600 hours for bearing A and 860 hours for 4.
bearing B.
Find the best alternative through simulation.
Solution
It is assumed that the failure of a bearing is independent of the maintenance policy
followed.
Random numbers are allocated for different failure times:
TABLE 8
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
We can now select random numbers form the tables and generate a set of 12
bearing lives, for each of the bearings.
TABLE 9
Let us compare the costs of the three policies for the first 7200 hours.
Policy I. Replace a bearing only when it fails.
A requires replacement 10 times, and B 7 times during this period as seen from
the lives of successive bearings.
Total cost = Cost of bearing + cost of downtime
= (300 x 10 + 500 x 7) + (17 x 2 x 700)
= Rs. 30, 300
Policy II : Replace both the bearing if one fails.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
TABLE 10
There are 11 replacements during the period.
Total cost = 11 x (300 + 500) + 11 x 3 x 700
= Rs. 31, 900
Policy III: Replace the bearing which fails plus the other one in use for 600
or more hours for A and 860 or more hours in case of B.
Table 12 11 gives the analysis.
There are 4 replacements of A only on failure and 7 replacements both bearings.
Total cost = 4(300 + 2 x 700) + 7(800 + 3 x 700)
= Rs. 27,100
Policy III is the cheapest.
The simulation is limited to 7200 hours of operation since the purpose is only to
illustrate the method. With such small number of trials the results may turn out to
be erratic. Consider Policy I again. With a different set of random numbers and
extended simulation there may be occasions of both bearing failing at the same
time, thus affecting the cost of downtime for replacement. Simulation with
sufficiently large number of trials only can lead to dependable decisions.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
TABLE 11
This will be the total life(from Table 12 9) if newly fitted. For survivors from
previous replacement, this will be the balance life (col.8)
Example 5:
A factory produces 150 scooters. But the production rate varies with the following
distribution.
Production Rate 147 148 149 150 151 152 153
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Probability
0.05 0.10 0.15 0.20 0.30 0.15 0.05
At present it calls for a track which will hold 150 scooters. Using the following
random numbers determine the average number of scooters waiting for
shipment in the factory and average number of empty space in the truck.
Random Numbers 82, 54, 50, 96, 85, 34, 30, 02, 64, 47
Solution:
Production Rate Probability Cumulative
Probability
Random Number
assigned
147 0.05 0.05 00 04
148 0.10 0.15 05 14
149 0.15 0.30 15 29
150 0.20 0.50 30 49
151 0.30 0.80 50 79
152 0.15 0.95 80 94
153 0.05 1.00 95 99
Simulation work sheet
Trial
No
Random
No.
Simulated
Production Rate
Scooter Waiting
in the factory
Number of example
spaces in the truck
1 82 152 2 -
2 54 150 - -
3 50 150 - -
4 96 153 3 -
5 85 152 2 -
6 34 150 - -
7 30 150 - -
8 02 147 - 3
9 64 151 1 -
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
10
47 150 2 -
Total 8 3
. Average number of scooters waiting = 8 / 10 = 0.8 / day
Average number of empty space = 3 / 10 = 0.3 / day
Example 6:
Dr.Strung is a dentist. He gives appointment to patients every half-an hour.
However he does not know the nature if illness of patients arriving at this clinic.
From past record he has the following probability distribution and also know the
exact treatment timings. He starts his clinic at 8.00 am using the following
information determine the average waiting time of the customers and idle time of
the doctor.
Nature of
illness
Probability Time taken for
treatment (mts)
Filling 0.10 50
Check up 0.30 15
Crowning 0.15 40
Cleaning 0.30 15
Extraction 0.15 30
Random Numbers 56, 40, 26, 66, 87, 48, 17, 22, 04, 15
Solution:
Illness Probability Cum.Prob Random
No.assigned
Filling 0.10 0.10 00 09
Check up 0.30 0.40 10 39
Crowning 0.15 0.55 40 54
Cleaning 0.30 0.85 55 84
Extraction 0.15 1.00 85 99
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Simulation work sheet
Trial
No
Rando
m No.
Nature of
illness
Time
Taken
Patient
s
Arrival
Time
Treatment Doctor
identification
Patients
waiting
time
Starts Finisher
1 56 Cleaning 15 8.00 8.00 8.15 - -
2 40 Crowning 40 8.30 8.30 9.10 15 -
3 26 Check up 15 9.00 9.10 9.25 - 10
4 66 Cleaning 15 9.30 9.30 9.45 5 -
5 87 Extraction 30 10.00 10.00 10.30 15 -
6 48 Crowing 40 10.30 10.30 11.10 - -
7 17 Check up 15 11.00 11.10 11.25 - 10
8 22 Check up 15 11.30 11.30 11.45 5 -
9 04 Filling 50 12.00 12.00 12.50 15 -
10 15 Check up 15 12.30 12.50 13.05 - 20
Total 45 40
Doctors idle time = 45 mts
Patients Average waiting time = 40 / 10 = 4 mts
Summary
In this unit we studied the basics concepts concerned with simulations and then
we continued further to find simulations procedure. The allocation of random
numbers was discussed next with the use of random number tables. We studied
the sample size and the application of simulation at the end.
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Unit - 11-Project Scheduling And PERT-CPM
Introduction
A project such as construction of a bridge, highway, power plant, repair and maintenance
of an oil refinery or an air plane design, development and marketing a new product,
research and development etc., may be defined as a collection of inter-related activities
(or tasks) which must be completed in a specified time according to a specified sequence
and require resources such as personnel, money, materials, facilities etc.
The growing complexities of todays projects had demanded more systematic and more
effective planning techniques with the objective of optimizing the efficiency of executing
the project. Efficiency here implies effecting the utmost reduction in the time required to
complete the project while accounting for the economic feasibility of using available
resources.
Project management has evolved as a new field with the development of two analytic
techniques for planning, scheduling and controlling projects. These are the Critical Path
Method (CPM) and the Project Evaluation and Review Technique (PERT). PERT and
CPM are basically time-oriented methods in the sense that they both lead to the
determination of a time schedule.
Learn about PERT and CPM
Basic difference between PERT and CPM
Though there are no essential differences between PERT and CPM as both of them share
in common the determination of a critical path and are based on the network
representation of activities and their scheduling that determines the most critical activities
to be controlled so as to meet the completion date of the project.
PERT
1. Since PERT was developed in connection with an R and D work, therefore it had
to cope with the uncertainties which are associated with R and D activities. In PERT,
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
total project duration is regarded as a random variable and therefore associated
probabilities are calculated so as to characterise it.
2. It is an event-oriented network because in the analysis of network emphasis is
given an important stages of completion of task rather than the activities required to
be performed to reach to a particular event or task.
3. PERT is normally used for projects involving activities of non-repetitive nature in
which time estimates are uncertain.
4. It helps in pinpointing critical areas in a project so that necessary adjustment can
be made to meet the scheduled completion date of the project.
CPM
Since CPM was developed in connection with a construction project which 1.
consisted of routine tasks whose resources requirement and duration was known
with certainty, therefore it is basically deterministic.
CPM is suitable for establishing a trade-off for optimum balancing between 2.
schedule time and cost of the project.
CPM is used for projects involving activities of repetitive nature. 3.
Project scheduling by PERT-CPM
It consists of three basic phases: planning, scheduling and controlling.
1. Project Planning: The various steps involved during this phase are given below:
i) Identify various activities (task or work elements) to be performed in the
project.
ii) Determining requirement of resources such as men, materials, machines etc.,
for carrying out activities listed above.
iii) Estimating costs and times for various activities.
iv) Specifying the inter-relationship among various activities.
v) Developing a network diagram showing the sequential inter-relationships
between the various activities.
Scheduling: Once the planning phase is over, scheduling of the project, is when 1.
each of the activities required to be performed, is taken up. The various steps
involved during this phase are listed below:
Estimating the durations of activities, taking into considerations the 1.
resources required for these execution in most economic manner.
Based on these time estimates, preparing a time chart showing the start 2.
and finish times for each activity, and hence calculation of total project
duration by applying network analysis techniques such as forward
(backward) pass and floats calculation; identifying the critical path;
carrying out resource smoothing (or levelling) exercise for critical or
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
scarce resources including re-costing of the schedule taking into account
resource constraints.
Project Control: Project control refers to revaluating actual progress 3.
against the plan. If significant differences are observed then re-scheduling
must be done to update and revise the uncompleted part of the project.
PERT/CPM Network Components and Precedence Relationship
PERT/CPM networks consists of two major components as discussed below:
a) Events: An event represents a point in time that signifies the completion of some
activities and the beginning of new ones. The beginning and end points of an activity
are thus described by 2 events usually known as the Tail and head events. Events are
commonly represented by circles (nodes) in the network diagram. They do not
consume time and Resource
b) Activities: Activities of the network represent project operations or task to be
conducted. An arrow is commonly used to represent an activity, with its head
indicating the direction of progress in the project. Activities originating from a certain
event cannot start until the activities terminating at the same event have been
completed. They consume time and Resource.
Events in the network diagram are identified by numbers. Numbers are given to events
such that arrow head number must be greater than arrow tail number.
Activities are identified by the numbers of their starting (tail) event and ending (head)
event. An arrow (i .J) extended between two events, the tail event i represents the start of
the activity and the head event J represents the completion of the activity as shown in Fig.
9.1:
Fig. 9.1
Figure 9.2 shows another example, where activities (1, 3) and (2, 3) must be completed
before activity (3, 4) can start.
Fig. 9.2
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
The rules for constructing the arrow diagram are as follows:
Each activity is represented by one and only one arrow in the network. 1.
No two activities can be identified by the same head and tail events. 2.
To ensure the correct precedence relationship in the arrow diagram, the following 3.
questions must be answered as every activity is added to the network:
a) What activities must be completed immediately before these activity can
start ?
b) What activities must follow this activity ?
c) What activity must occur concurrently with this activity ?
This rule is self-explanatory. It actually allows for checking (and rechecking) the
precedence relationships as one progresses in the development of the network.
Example 1: Construct the arrow diagram comprising activities A, B, C .. and L such
that the following relationships are satisfied:
A, B and C the first activities of the project, can start simultaneously. 1.
A and B precede D. 2.
B precedes E, F and H. 3.
F and C precede G. 4.
E and H precede I and J. 5.
C, D, F and J precede K. 6.
K precedes L. 7.
I, G and L are the terminal activities of the project. 8.
Fig. 9.3
The dummy activities D1 and D2 are used (dotted lines) to establish correct precedence
relationships. D3 is used to identify activities E and H with unique end events. The events
of the project are numbered such that their ascending order indicates the direction of the
progress in the project.
Note: A dummy activity in a project network analysis has zero duration.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Critical Path Calculations
The application of PERT/CPM should ultimately yield a schedule specifying the start and
completion time of each activity. The arrow diagram is the first step towards achieving
that goal. The start and completion times are calculated directly on the arrow diagrams
using simple arithmetic. The end result is to classify the activities as critical or non
critical. An activity is said to be critical if a delay in the start of the course makes a delay
in the completion time of the entire project. A non-critical activity is such that the time
between its earliest start and its latest completion time is longer than its actual duration. A
non-critical activity is said to have a slack or float time.
Determination of the Critical Path
A critical path defines a chain of critical activities that connects the start and end events
of the arrow diagram. In other words, the critical path identifies all the critical activities
of the project.
The critical path calculations include two phases. The first phase is called the Forward
Pass where all calculations begin from the start node and move to the end node. At each
node a number is computed representing the earliest occurrence time of the corresponding
event. These numbers are shown in squares . In forward pass we note the number of heads
joining the event. We take the maximum earliest timing through these heads.The second
phase called the Backwards Pass, begins calculations from the end node and moves to
the start node. The number computed at each node is shown in a triangle ^ near end
point which represent the latest occurrence time of the corresponding event. Consider the
forward pass In backward pass we see the number of tails and take minimum value
through these tails.
Let ESi be the earliest start time of all the activities emanating from event i, i.e. ESi
represents the earliest occurrence time of event i, if i = 1 is the start event then
conventionally, for the critical path calculations, ESi = 0 , Let Dij be the duration of the
activity (i, j). Then the forward pass calculations are given by the formula:
ESi = maxi {ESi+Dij}, for all defined (i, j) activities with ESi=0. Thus in order to
compute EsJ for event j, Esi for the tail events of all the incoming activities (i, j) must be
computed first.
With the computation of all ESj, the forward pass calculations are completed. The
backward pass starts from the end event. The objective of this phase to calculate LCi,
the latest completion time for all the activities coming into the event i. Thus if i = n is
the end event LCn = ESn initiates the backward pass.
In general for any node i, LCi = min {LCj-Dij} for all defined activities are calculated,
which ends the calculation of backward pass.
The critical path activities can now be identified by using the results of the forward and
backward passes. An activity (i, j) lies on the critical path if it satisfies the following
conditions.
A) ESI = LCi
B) ESJ = LCJ
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
C) ESJ-ESI = LCJ-LCI = DiJ
These conditions actually indicate that there is no float or slack time between the earliest
stand and the latest start of the activity. Thus the activity must critical. In the arrow
diagram these are characterised by numbers in are the same at each of the head
and tail events and the difference between the number in at the head event and the
number in at the tail event in equal to the duration of the Activity.
Thus we will get a critical path, which is chain of connected activities, which spans the
network form start to end.
Example 2: Consider a network which stands from node 1 and terminate at node 6, the
time required to perform each activity is indicated on the arrows.
Fig. 9.4
Let us start with forward pass with ESi = 0.
Since there is only one incoming activity (1, 2) to event 2 with D12 = 3.
ES2 = ES1+ DS2 = 0+3=3.
Let us consider the end 3, since there only one incoming activity (2, 3) to event 3, with
D23 = 3.
ES3 = ES2+ D23 = 3+3 = 6.
To obtain ES4, since there are two activities A (3, 4) and (2,4 ) to the event 4 with D24 =
2 and D34 = 0.
ES4= maxi=2, 3 { ESi + De4}
= max { ES2 +D24, ES3 + D34}
= max {3+2, 6+0} = 6
Similary ES5 = 13 and ES6 = 19
Which completed first phase.
In the second phase we have
LC6= 19 = ES6
LC5 = 19-6 = 13
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
LC4= minJ = 5, 6 {LCJ D4J} = 6
LC3 = 6, LC2 = 3 and LC1 = 0
. activities (1, 2), (2, 3) (3, 4) (4, 5) (5, 6) are critical and (2, 4) (4, 6), (3, 6), are non-
critical.
Thus the activities (1, 2) (2, 3 ) (3, 4) (4, 5) and (5, 6) define the critical path which is the
shortest possible time to complete the project.
Determination of Floats
Following the determination of the critical path, the floats for the non-critical activities
must be computed. Note that for the critical activities this float is zero. Before showing
how floats are determined, it is necessary to define two new times that are associated with
each activity. There are Latest Start (LS) and the Earliest Completion (EC) times,
which are defined activity (i, J) by
LSeJ= LCJ DiJ
and ECeJ = ESi + DiJ
There are two important types of floats namely, Total Float (TF) and Free Float (FF). The
total float TFiJ for activity (i, J) is the difference between the maximum time available to
perform the activity (= LCJ ESi) and its duration (= DiJ )
That is,
TFiJ = LCJ ESI DiJ = LCJ ECiJ = LSiJ ESi
The free float is defined by assuming that all the activities start as early as possible. In
this case FFiJ for activity (i, J) is the excess of available time (= ESi ESi) over its
deviation (= DiJ );
that is, FFiJ = ESi ESi = DiJ .
Note that only for a critical activity must have zero total float. The free float must also be
zero when the total float is zero. The converse is not true, that is in the sense that a non-
critical activity may have zero free floats.
Let us consider the example taken before the critical path calculations together with the
floats for the non-critical activities can be summarized in the convenient form shown in
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Note: Total Cost = ESij = LFij ESij
Free float = Total float - Head slack
* Critical activity *
Example 3: A project consists of a series of tasks A, B, C, D, E, F, G, H, I with the
following relationships. (W < X,Y means X and Y cannot starts until W is completed,
X,Y < W means W cannot start until both X and Y are completed). With this notation
construct the network diagram having the following constraints A < D, E; B, D < F; C <
G, B < H; F,G < I.
Find also the minimum time of completion of the project, the critical path, and the total
floats of each task, when the time (in days) of completion of each task is as follows:
Task: A B C D E F G H I
Time: 23 8 20 16 24 18 19 4 10
Fig. 9.5
ES1 = 0, ES2 = 20, ES3 = 23, ES4 = 59, ES5 = 39, ES6 = 57, ES7 = 67
Activity (iJ) Duration
DiJ
Earliest Latest Table Float
TFiJ
Free
Float
FFiJ
Start
ESe
Finish
Eeij
Start
LJ-DiJ
Finish
LJ
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
(1, 2)
(1, 3)
(1, 4)
(2, 5)
(3, 4)
(3, 7)
(4, 5)
(4, 6)
(5, 6)
(5, 7)
(6, 7)
20
23
8
19
16
24
0
18
0
4
10
0
0
0
20
23
23
39
39
39
39
37
20
23
8
39
39
47
39
57
39
43
67
18
0
31
38
23
43
57
39
57
63
57
38
23
39
57
39
67
57
57
57
67
67
18
0*
31
18
0*
20
10
0*
18
24
0*
0
0
31
0
0
20
0
0
18
24
0
Critical path is 1 3 4 6 7.
Learn Using PERT for Project Management
Project Management PERT
Probability and Cost Consideration in Project Scheduling
The analysis in CPM does not take into the case where time estimates for the different
activities are probabilistic. Also it does not consider explicitly the cost of schedules. Here
we will consider both probability and cost aspects in project scheduling.
Probability considerations are incorporated in project scheduling by assuming that the
time estimate for each activity is based on 3 different values. They are
a = The optimistsic time, which will be required if the execution of the project
goes extremely well.
b = The pessimistic time, which will be required if everything goes bad.
m = The most likely time, which will be required if execution is normal.
The most likely estimate m need not coincide with mid-point of a and b. Then the
expected duration of each activity can be obtained as the mean of and 2 m. i.e.
.
This estimate can be used to study the single estimate D in the critical path calculation.
The variance of each activity denoted by V is defined by
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
variance V = .
The earliest expected times for the node i denoted by E(i) for each node i is obtained by
taking the sum of expected times of all activities leading to the node i, when more than
one activity leads to a node i, then greatest of all E(i) is chosen . Let i be the earliest
occurrence time of the event i, we can consider i as a random variable. Assuming that all
activities of the network are statistical independent, we can calculate the mean and the
variance of the i as follows E{i } = ESI and Var{i } = . Where K defines the
activities along the largest path leading to i.
For the latest expected time, we consider the last node. Now for each path move
backwords, substituting the for each activity (ij). Thus we have E(LJ) = E(a) and E(
i) = L(LJ) if only one path events from J to i or it is the minimum of {E[LJ) ]
for all J for which the activities (i, j) is defined.
Note: The probability distribution of times for completing an event can be approximated
by the normal distribution due to central limit theorem.
Since i represents the earliest occurrence time, event will meet a certain schedule time
STi (specified by an analyst) with probability
Pr (i STi) = Pr
= Pr (Z Ki)
where Z ~N(01) and Ki = . It is common practice to compute the probability that
event i will occur no later than its LCe such probability will then represent the chance that
the succeeding events will occur within the (ESe, LCe) duration.
Example 4: A project is represented by the network shown below and has the following
data.
Task A B C D E F G H I
Optimistic Time 5 18 26 16 15 6 7 7 3
Pessimistic Time 10 22 40 20 25 12 12 9 5
Most Likely Time 8 20 33 18 20 9 10 8 4
Determine the following:
a) Expected task time and their variance.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
b) The earliest and latest expected times to reach each event.
c) The critical path.
Fig. 9.6
a) The probability of an event occurring at the proposed completion data if the
original contract time of completing the project is 41.5 weeks.
b) The duration of the project that will have 96% channel of being completed.
Solution:
a) Using the formula we can calculate expected activity times and variance in the
following table
b)
Activity a B m v
1-2
1-3
1-4
2-5
2-6
3-6
4-7
5-7
6-7
5
18
26
16
15
6
7
7
3
10
22
40
20
25
12
12
9
5
8
20
33
18
20
9
10
8
4
7-8
20-00
33-0
18-0
20-0
9-0
9-8
8-0
4-0
0.696
0.444
5.429
0.443
2.780
1.000
0.694
0.111
0.111
Forward Pass:
E1 = 0 E2 = 7.8 E3 = 20 E4 = 33 E5 = 25-8 E6 = 29 E7 = 42.8
Backward Pass:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
L7 = 42.8 L6 = 38.8 L5 = 34.8 L4 = 33.0 L3 = 29.8 L2 = 16.8 L1 = 0.
The E-values and L-values are shown in Fig.
Fig. 9.7
c) The critical path is shown by thick line in fig. The critical path is 1-4-7 and the
earliest completion time for the project is 42.8 weeks.
d) The last event 7 will occur only after 42.8 weeks. For this we require only the
duration of critical activities. This will help us in calculating the standard duration of
the last event.
Expected length of critical path = 33+9.8 = 42.8
Variance of article path length = 5.429+0.694 = 6.123
Probability of meeting the schedule time is given by Pi (Z Ki) =
Pi (Z 0.52) = 0.30 (From normal distribution table)
Thus the probability that the project can be completed in less than or equal to 41.5
weeks is 0.30. In other words probably that the project will get delayed beyond 41.5
weeks is 0.70.
e) Given that P (Z Ki) = 0.95. But Z0.9S = 1.6 u, from normal distribution table.
Then 1.6 u =
Sji = 1.642.47+42.8 = 46.85 weeks.
Summary
Critical Path computations are quite simple, yet they provide valuable information
that simplifies the scheduling of complex projects. The result is that PERT-CPM
techniques enjoy tremendous popularity among practitioners in the field. The
usefulness of the techniques is further enhanced by the availability of specialized
computer systems for executing, analyzing and controlling network projects.
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Unit - 12-Game Theory
Introduction
Game theory was developed by John Von Newman. He worked on game theory
right from 1928. But, it gained prominence only after 1944 when he published
(along with Mrogenstren) the work Theory of games and economic behaviour.
This field of study is fast developing and it is highly resourceful.
Learn about Various Aspects of Game Theory
Competitive Situations
Competitive situations occur when two or more parties with conflicting interests
operate. The situations may occur as follows.
Marketing different brands of a commodity.
Two (or more) brands of detergents (soaps) try to capture the market by adopting various
methods (courses) such as advertising through electronic media, providing cash
discounts to consumers or offering larger sales commission to dealers.
Campaigning for elections.
Two (or more) candidates who contest an elections try to capture more votes by adopting
various methods (courses) such as campaigning through T.V., door to door
campaigning or campaigning through public meetings.
Fighting military battles.
Two forces fighting a war try to gain supremacy over one another by adopting various
courses of action such as direct ground attack on enemy camp, ground attack supported
by aerial attack or playing defensive by not attacking.
We consider each of the above situations to be a competitive game where the
parties (players) adopt a course of action (play the game).
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Characteristics of a Competitive Game
A competitive game has the following characteristics.
1. The number of players (competitors) is finite.
2. Each player has finite number of courses of action (moves).
3. The game is said to be played when each player adopts one of his course of
action.
4. Every time the game is played, the corresponding combination of courses of
action leads to a transaction (payment) to each player. The payment is called pay-off
(gain). The pay-off may be monetary (money) or some such benefit as increased sales,
etc.
5. The players do not communicate to each other.
6. The players know the rules of the game before starting.
n-Person Game
A game in which n players participates is called n-person game.
A game in which two players participate is called 2-person game (two-person game).
Zero-Sum Game
If a game is such that whenever it is played the sum of the gains (pay-off) of the players is
zero, it is called zero-sum game.
A zero-sum game which has two players is called two-person zero-sum game. It
is called rectangular game.
In a two-person zero-sum game, the gain of the one player is equal to the loss of
the other.
Two-Person Zero-Sum Game (Rectangular Game).
A two-person zero-sum game is a game in which
i) two players participate
ii) the gain of one player it the loss of the other.
In a two-person zero-sum game, let the players be A and B. Let be the m
courses of action for player A. Let be the n courses of action for player B. Let
be the pay-off (gain) of player A when he plays the course of
action, and player B plays the course of action . Then, thefollowing matrix is the pay-
off (gain) matrix of player A.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
This is a (read as m by n) game.
Here, is the gain of A. Also, is the loss of B. Therefore, (- ) is the gain of B. And so,
the pay-off matrix of B is obtained by writing (- ) in the place of in the above matrix
and then writing the transpose of the matrix.
Strategy
In a game, the strategy of a player is the predetermined rule by which he chooses his
course of action while playing the game.
The strategy of a player may be pure strategy or mixed strategy.
Pure Strategy
While playing a game, pure strategy of a player is his predecision to adopt a
specified course of action (say ) irrespective of the strategy of the opponent.
Mixed Strategy
While playing a game, mixed strategy of a player is his predecision to choose his course
of action according to certain pre-assigned probabilities.
Thus, if player A decides to adopt courses of action with perspective probabilities
0.4 and 0.6, it is mixed strategy.
Example 1: (2-finger morra game).
Two persons A and B play a game they should simultaneously raise their hand and
exhibit either one finger or two fingers. If both of them show one finger or if both show
two fingers, A should pay Rs. 10 to B. On the other hand, if one player shows one finger
and the other player shows two fingers, B should pay Rs. 5 to A.
Here, the pay-off matrix of A is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Here, suppose player A decides to show one finger , his strategy is pure strategy. On
the other hand, suppose A decides to play with probability 0.5 and with probability
0.5, his strategy is mixed strategy. (This means, if he is to play repeatedly, sometimes he
should play and at other times he should play . He should mix and randomly
almost equal number of times.)
Maximin Minimax Principle (of solving a two-person zero-sum game)
Suppose player A and player B are to play a game without knowing what the
other player would do. However, player A would like to maximize his profit and player B
would like to minimize his loss. And thus, each player would expect his opponent to be
calculative.
Suppose player A plays . Then, his gain would be according
as Bs choice is . Let . Then, is the minimum gain of A
when he plays . (Here, is the minimum pay-off in the first row.) Similarly, if A plays
, his minimum gain is which is the least pay-off in the second row. Thus proceeding,
we find that corresponding to As play , the minimum gains are the row
minimums . Suppose A chooses that course of which is maximum. This
maximum of the row minimum in the pay-off matrix is called maximin. The maximin
is
Similarly, when B plays, he would minimize his maximum loss. The maximum loss to B
when is . This is the maximum pay-off in the column. The minimum of
the column maximums in the pay-off matrix is called minimax. The minimax is
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
For some games, the maximin and the minimax are equal. That is,
. Such games are said to have saddle point.
On the other hand, if , the game does not have saddle point.
(Note that cannot be greater than ).
Saddle Point
In a two-person zero-sum game, if the maximin and the minimax are equal, we say that
the game has saddle point.
Saddle point is the position where the maximin (maximum of the row
minimums) and minimax (minimum of the column maximums) coincide.
If the maxmin occurs in the row and if the minimax occurs in the column, the
position (r, s) is the saddle point. Here, is the common value of the maximin and the
minimax. It is called the value of the game.
The value of a game is the expected gain of player A when both the players adopt
optimal strategy.
Note 1: If a game has saddle point, and if (r, s) is the saddle point, suggested
solution to both the players is pure strategy. For player A, the suggested solution
is . For player B, the suggested solution is .
Note 2: If a game does not have saddle point, the suggested solution is mixed
strategy.
Note 3:
A game is said to be fair if its value is zero.
Learn Application of Game Theory in Various Conflicting Situations
Solution to a Game with Saddle Point
Consider a two-person zero-sum game with players A and B. Let be the courses
of action for player A. Let be the courses of action for player B.
The saddle point of the game is found as follows.
1. The minimum pay-off in each row of the pay-off matrix is circled (marked with )
2. The maximum pay-off in each column is boxed (marked with )
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
3. In the above process, if any pay-off is circled as well as boxed, that pay-off is the
value of the game. The corresponding position is the saddle point.
Let (r, s) be the saddle point. Then, the suggested pure strategy for player A is .
The suggested pure strategy for player B is . The value of the game is .
Note: However, in the above procedure, if none of the pay-off is circled as well
as boxed, the game does not have saddle point. And so, the suggested solution for
the players is mixed strategy.
Example 2:
Verify whether the 2-finger morra game explained earlier has saddle point. If so, write
down the solution for the game.
Solution:
The pay-off matrix of player A is
1. The minimum pay-off in each row is circled.
2. The maximum pay-off in each column is boxed.
3. Since none of the pay-off is circled as well as boxed, the game does not have
saddle point. And so, the solution for the game is mixed strategy for both the players.
Example 3:
Two persons A and B, without showing each other, place a coin each on the
table. If the coins happen to be of the same denomination, player A will take both of
them. If they happen to be of different denominations, player B will take both of
them.Suppose player A has a few one-rupee coins and two-rupee coins. And suppose
player B has a few one-rupee, two-rupee and five-rupee coins.
i) Write down the pay-off matrix of A. Does the game have saddle point? If so, write
down the solution.
ii) What happens to the game if both the players play only with one-rupee and two-
rupee coins?
Solution:
i) The pay-off matrix of A is -
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
1. The minimum pay-off in each row is circled. (Here, the minimums repeat.)
3. The pay-off -1 is circled as well as boxed. Therefore, the game has a saddle
point. It is the position (1, 3).
The solution to the game is
a) Strategy for A is .
b) Strategy for B is .
c) Value of the game is v = -1 rupees.
ii) If both the players play only with one-rupee and two-rupee coins, the pay-off
matrix of A is
Here, the game does not have saddle point. Therefore, the suggested solution is mixed
strategy for the players.
Example 4:
A labour union of a firm is negotiating a new 5-year settlement regarding payments with
the management. The options the union has are : Aggressive bargaining, :
Bargaining with reasoning and : Conciliatory approach. The likely mode of response
from the management are : Aggressive bargaining, : Bargaining with reasoning,
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
: Legalistic approach and : Conciliatory approach. The gains to the union in each
case are as follows.
What strategy would you suggest for the two sides? What is the value of the
game?
Solution:
3. The value 12 is circled as well as boxed. And so, the game has a saddle point. It
is the position (1, 3).
Therefore, the solution to the game is
a) Strategy for the union is : Aggressive bargaining.
b) Strategy for the management is : Legalistic approach.
c) Value of the game is v = 12.
Example 5:
Solve the game
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Is the game fair?
Solution:
3. The value 7 is circled as well as boxed. And so, the game has a saddle point. It is
the position (2, 1).
d) The game is not fair because v is not equal to zero.
Example 6:
A two-person zero-sum game, has the following pay-off matrix. Solve the game.
Is the game fair?
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
3. The value 0 is circled as well as boxed. And so, the game has a saddle point. It is
the position (2, 2).
a) Strategy for the player is (second course of action).
b) Strategy for the opponent is (second course of action).
d) Since the value of the game is 0, the game is fair.
Dominance
In a rectangular game, suppose in the pay-off matrix of player A, each pay-off in one
specific row exceeds the corresponding pay-off in another specific row .
This means, whatever be the course of action adopted by player B, for A, the course of
action yields greater gains than the course of action . Therefore, is a better strategy
than irrespective of the strategy of B. And so, we say that dominates .
On the other hand, suppose each pay-off in a specific column is less than the
corresponding pay-off in another specific column . This means, for player B,
strategy has lesser loss than strategy irrespective of strategy of A. And so, we say
that dominates . Thus,
a) In the pay-off matrix, if each pay-off in the is greater than (or equal to
) the corresponding pay-off in the , dominates .
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
b) In the pay-off matrix, if each pay-off in the is less than
(or equal to) the corresponding pay-off in the , dominates .
Sometimes, a convex combination of two or more courses of action may dominate
another course of action.
Whenever a course of action (say or ) is dominated by others, that course of
action ( or ) can be deleted from the pay-off matrix. Such deletion will not affect
the choice of the solution.
Such deletion of courses of action reduces the order of the pay-off matrix.
Successive reduction of the order using dominance property helps us in
solving games.
Learn Solving Games Using Dominance
Solving Games Using Dominance
(Solving two-person zero-sum game with saddle point)
Consider a two-person zero-sum game with players A and B. Let be the
courses of action for player A. Let be the courses of action for player B.
Suppose the game has saddle point. Then, using dominance property, it is
possible to successively delete the courses of action of A as well as B
such that ultimately the pair comprising the saddle point alone remains.
The procedure in this regard is as follows.
a) In the pay-off matrix, if each pay-off in the is greater than (or equal to
) the corresponding pay-off in the , dominates . And so, is deleted.
b) In the pay-off matrix, if each pay-off in the is less than (or equal to)
the corresponding pay-off in the , dominates . And so,
is deleted.
c) The above steps are repeated in succession until the saddle point is reached.
And hence, the solutions is written down.
Note: Sometimes, a convex combination of two or more courses of action may
dominate another course of action.
Example 8:
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Solve the following game using dominance property.
Solution:
In the pay-off matrix, each pay-off in the first row exceeds the corresponding pay-off in
the third row. Therefore, dominates . And so, is deleted. The reduced matrix is
Here, each pay-off in the third column is less than corresponding pay-off in the first
column. Therefore, dominates . Similarly, dominates . Also, dominates .
Thus, the matrix reduces to
Here, since 12>8, dominates . And so, finally the matrix reduces to
Thus, (1, 3) is the saddle point. And so, the solution to the game is
c) Value of the game is v = 12
Example 9:
Solve the following zero-sum game and find its value.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
Solution:
In the pay-off matrix, each pay-off in the second column is less than (or equal to) the
corresponding pay-off in the third column. And so, the course of action Q dominates R.
Similarly, Q dominates S.
After deleting R and S, the reduced matrix is
Here, pay-off in the second row is greater than (or equal to) the corresponding pay-offs in
the first, third as well as fourth rows. Therefore, B dominates A, C and D.
After deleting A, C and D, the reduced matrix is
Here, 1<6. Therefore, Q dominates P.
After deleting P, the reduced matrix is
Thus, the solution to the game is
a) Strategy for Company X is B.
b) Strategy for Company Y is Q.
Example 10.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d
In a two-person zero-sum game, the pay-off matrix of A is
Write down the pay-off matrix of player B.
Solution:
The pay-off matrix of B is
Summary
In this unit of game theory we studied the concept of competitive situations
where the characteristics of competitive game and its strategy we considered.
The maximum minimum principle is discussed briefly saddle point and
Dominance is explained with clear cut examples.
.
.
.
.
P
D
F

C
o
n
v
e
r
t
e
r

P
r
o

1
0
.
0

U
n
r
e
g
i
s
t
e
r
e
d

MB0032 or Complete

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

MB0032 or Complete

Hochgeladen von

Copyright:

Verfügbare Formate

Unit 1- Introduction To OR

axis at the point B(0, 70) (put x

axis at the point D(0, 90).

Das könnte Ihnen auch gefallen