Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Management-Science-Module

Download as pdf or txt
Download as pdf or txt
You are on page 1of 136

Chapter I: INTRODUCTION TO MANAGEMENT SCIENCE

Introduction
Management science is applying a scientific approach to problem-solving management to
help managers make better choices. As suggested by this concept, management science includes
a variety of mathematically based strategies that have either been developed within the area of
management science or adopted from other disciplines such as mathematics, natural sciences,
statistics, and engineering. This module offers an introduction to the methods that make up the
philosophy of management and explains their contributions to crisis management.
Management science is a proven and recognized discipline in industry. Management
science techniques implementations are common, and they have also been credited with growing
the company firms' performance and profitability. Several suggest that they use management
science methods in different organization surveys, and the majority consider the findings to be
quite good. Management science (also referred to as operations research, systematic processes,
predictive study and decision-making sciences) is part of the core curriculum in most business
programs.

Learning Objectives
After studying this module, you should be able to:
1. Define management science and describe its importance in business.
2. Explain the major characteristics of management science and how it is applied in problem
solving.
3. Explain the importance of models in decision making process.
4. Explain and identify the appropriate management science technique in managerial
problems.

1.1 Definition and Characteristics of Management Science

Definition
Management science can be defined briefly as applying the scientific method to the
analysis and solution of problems with managerial decision. According to Turban and Meridith,
management science is applying analytical principles, procedures, and instruments to problems
concerning device processes in order to offer optimal answers to the problems for those in charge
of the processes.

Major Characteristics of Management Science


1. A main focus on decision making by the managers.
2. Usage of the analytical model to the decision-making phase.
3. Examining the condition of the judgment from a specific viewpoint
4. Using methods and expertise from multiple disciplines.
5. A dependency on the mathematical models of shape.
6. Computer widespread use.

Decision-making is a method of deciding between two or more possible courses of action


necessary to accomplish a particular goal. Decision-making is a systematic process and can be
described simply by defining the problem, searching for alternative courses of action, evaluating
alternatives and choosing one alternative.

System configuration comprises of three associated components: inputs, procedures, and


outputs. The inputs (people, raw materials, money) enter the system. Processes transform inputs
into outputs (processes may use resources, operating procedures, workers, machines). Outputs exit
the machine (Products, Customers served). Systems approach involves finding and defining all the
elements mentioned and the real world relationships between them. This method is the
presumption required to use management science methods, since they allow the precise definition
of a problem. System approach allows mathematical formulation of the problem.

1.2 Models and Modeling

Model is a generalized reflection of the natural world, and only significant and appropriate
objects or properties can be included in the model, whereas modeling is the method of treating and
explaining real-life problems in mathematical terms.

Management science follows a rational, structured approach to problem solving that


strongly resembles what is known as the analytical process of problem-solving. This approach, as
illustrated in a generally recognized and ordered series of steps follows: (1) observation, (2)
problem definition, (3) model construction, (4) model solution, and (5) solution implementation.

Observation
The first phase in the cycle of handling management science process is to recognize a
problem that occurs within the organization. The system must be monitored continuously and
closely so as to identify problems as soon as they arise.

Definition of the Problem


When a problem has been established, the problem must be described explicitly and in a
clear manner. Defining a problem incorrectly can easily lead to either no solution or inappropriate
solution. Hence the limitations of the problem and the extent to which it pervades other units of
the organization should be included in the definition of the problem. It's evident that we need to
know the production process details and all the appropriate information.

Model Construction
A concept in management science is a theoretical representation of an actual issue scenario.
It can be in the form of a graph or diagram, but most commonly a model in management science
consists of a series of mathematical links. The numbers and symbols form these mathematical
relationships. Two types of models exist: determinist and probabilistic. All aspects are known with
certainty in deterministic models. There is no question that deterministic models are ideal,
however they can give a fairly decent estimation of truth. There is a particular degree of ambiguity
in the probabilistic models. This is tempting to neglect the small degree of ambiguity and to use
deterministic models instead of probabilistic models. In case of high uncertainty we should
consider random variables rather than constants.

Model Solution
When models have been developed in management science, the methods described in this
text are used to solve them. A technique for a management science solution usually applies to a
particular type of model. Therefore, the form of concept and the process of solution are also part
of the scientific management research. We also mean problem solution when we refer to model
solution. The typical goal of the most problems is to find an optimal solution, that is, the best of
all feasible solutions (solutions that meet all the constraints).

Implementation of Solution Result


Implementation is the actual use of the model once it has been developed or the solution
to the problem the model was developed to solve. This is actually the main goal of management
and the original purpose of the whole process - not the model itself, but adjustment of reality
according to the recommendations ensuing from the results of the modeling process. In case we
do not use the results in the real production process, all our effort was absolutely vain. On the
contrary, if we constructed the model in a wrong way and we did not validate it, the applied results
could seriously harm the real system. In order to achieve the best management results, each step
must be carefully considered and cannot be skipped.
1.3 Management Science Techniques

Linear Programming
This is one of management science's best-known tools. This strategy describes the problem as
maximizing a linear function (minimizing) with reference to the set of linear constraints.

Integer Linear Programming


Additional variable values criteria are applied to the initial linear process ( i.e., process of
linear function and linear constraints). All or any of the values must be an integer. The particular
type of these variables is the binary variable (0-1 variable) with a value of 0 or 1. We think about
binary integer linear programming in such a situation. When only those variables are described as
integer (binary) in the model, then we are thinking about mixed integer linear programming.

Goal Programming
When several conflicting priorities need to be addressed concurrently, it needs more
effective resource. Goal programming is a specific methodology for coping with such situations,
usually in linear programming.

Distribution Models
A distribution problem is a specific category of problem in linear programming. There are
two major categories of distribution problems: the problem of transportation, and the problem of
assignment. The problem of transportation deals with shipping from a variety of suppliers to a
number of destinations while the problem of assignment deals with determining the right one-to-
one option for any of a number of prospective "candidates" to a number of possible "positions."

Nonlinear Programming
Models used in this field of management science are close to linear programming models;
but there is a major distinction between them: nonlinear models include nonlinear objective
function and/or other nonlinear constraints. Techniques employed in this field of management
science to solve problems are somewhat distinct from linear programming approaches.

Network Models
Many problems can be defined as a network (the collection of nodes and arcs) in graphic
terms. Typical circumstance is a network of transports: cities (nodes) are linked by roads (arcs) to
each other. When we evaluate the network (in this case we are involved, for example, in distances
among all the cities), the goal is always to locate the shortest distance from one city to all other
cities. Some forms of networks, instead of lengths, can be measured by capacities and then the
dilemma of maximal flow can be resolved. The most critical factor of several problems, addressed
by the help of network models, is unit cost and the aim is to find the lowest overall cost.

Project Management
Managers are responsible for organizing, managing and overseeing projects in certain
cases that consist of several different jobs or activities undertaken by a number of teams or persons.
It needs different time for a work to be completed. These problems are overcome by two simple
methods: CPM (Critical Path Method) and PERT (Program Analysis Review Technique). All
approaches allow the issue to be expressed on a network.

Inventory Models
Inventory control is one of the most common strategies, helping managers decide what to
buy and how much to buy. The primary purpose is generally to strike an accurate balance between
the expense of keeping inventory and the cost of implementing an order. Owing to the very various
actual distribution structures, there are several specific inventory models. We are considering two
distinct groups of models: deterministic and probabilistic. The demand rate in deterministic
models is constant over time, while the demand fluctuates over time in probabilistic inventory
models and can only be represented in probabilistic terms.
Waiting Line Models
This field of management science approaches deals with the circumstances in which a
variety of networks (e.g. vendors) ought to represent units ( e.g. customers). Since the number of
vendors is constrained, some units must wait in queue for the service. Thus the alternative name
for the models of the waiting line: Queuing Models. In actual life, the process of arrival of units
as well as the service times are irregular and the probabilistic approach is essential. Simple models
of waiting line can be obtained analytically (exact solution using derivative formulas), while the
simulation technique is needed for complex queuing systems.

Simulation
As management challenges are more complicated, it is always difficult to address them
using traditional methods (or not successful because of the time and expense spent). Simulation
solution is useful for this reason and is in many situations the best way to handle the issue.
Simulation is a programming experience with a simulation model intended to explain and measure
the actions of the real system-the machine software simulates the actual system. Complex waiting-
line models and inventory models are common conditions for effective simulation use.

Decision Analysis
Such methods may be used to choose appropriate approaches from a variety of alternatives
to decisions. According to the knowledge received by the manager, management challenges and
relevant methods are classified into three types: decisions under certainty (deterministic),
decisions under risk (probabilistic) and decisions under uncertainty. Of this function, we find
unique instruments: decision tables and trees.

Theory of Games
This field is an expansion of the decision making of two or more decision-makers to the
circumstances. All managers take simultaneous decisions (selected strategies) to execute an acts
that influences all decision-makers (players), i.e. their profits, costs etc. For certain disputes, two
or three decision-makers will collaborate, while battling with the others. We will consider a
standard case of a strategic game in economic theory-the oligopoly model.

Forecasting
Methods of forecasting support the manager predicting future elements of the business
operation. Statistics and econometrics deliver several time series and regression analysis methods
based. The key management task is to predict potential developments relative to the system's
previous behaviour. The well-known are the Moving Averages, Least Squares, Exponential
Smoothing processes, etc. Because the statistical significance for such models is very significant,
the manager's familiarity with validity hypotheses and statistical testing is relevant.

Multicriteria Decision Making


The decision maker needs to consider multiple criteria for many managerial problems. If
we find a solution which improves one criterion, it mostly aggravates some other criteria. It's
practically difficult to modify all the criteria at the same time. The Management's fair incentive is
to consider an acceptable solution. If the list of alternatives is small, we use alternative assessment
approaches. Few problems with set of constraints and set of objective functions are defined. In
this scenario multi-objective programming approaches have the remedy. Goal programming is the
unique type of this methodology in management science.

Markov Analysis
This approach can be used for explaining a system 's actions in a complex scenario (system
progression over time). If-at a given time point-the device is in one of the potential states, the
device may remain in the current state or switch into some other state at the following time point.
The transition probabilities are set for staying in the current state or moving to another state. The
manager will be confident in the possibility of the system being in the correct state at the present
time. Markov Analysis is a really strong management science method with a number of practical
applications.

Dynamic Programming
Management also needs to recognize a series of decisions where each judgment impacts
potential decisions in a major way. Dynamic programming lets administrators overcome certain
forms of challenges with these complex decision making. There is no common paradigm to solve
dynamic programming problems and, thus, the problems are divided into several classes. One
probable classification takes the deterministic and probabilistic models into consideration. Models
also use the sequential problems to describe the network. The Markov analysis may be viewed as
a complex programming probabilistic process.

Exercises and Activities

Discuss the following.


1. Differentiate management science from other mathematical discipline.

2. How do the goal of management science pay to the attainment of business objectives?

3. How can management science be useful in practice?

4. Explain which management science technique is appropriate in the following managerial


problems:
a. Inventory control
b. Facility design
c. Allocation of scarce resources
d. Investment decisions

5. Assess the limitations of management science in dropping risk of industry failures.

True or False
1. Decision making is a process of choosing between two or more available alternative
courses of action for the purpose of attaining a specific goal.
2. A solution that respects all the constraints is goal programming.
3. The application of the scientific method to the study and analysis of problems involving
large and complex systems, organizations or activities is called operations research.
4. Programming is an abstraction of reality.
5. Management science seeks for optimization of objective function, which looks for the
highest objective value.
6. Linear programming is an approach when several objective functions are considered
simultaneously.
7. The objective of goal programming is to minimize the undesirable deviations from the
goals.
8. Optimization of objective function, which looks for the lowest objective value is
minimization.
9. Optimal solution is a mathematical function expressed in terms of decision variables,
which is to be optimized.
10. A model that incorporates uncertainty in its functional relations and uncontrollable
variables is probabilistic model.
11. Implementation is the main goal of management and the original purpose of the whole
management science process according to the recommendations ensuing from the results
of the modelling process.
12. If we want to examine the impact of changes in inputs on changes in outputs, we have to
use sensitivity analysis.
13. Validation of the model is its comparison with the real system.
14. In dynamic programming, problems are classified into many groups.
15. In case of high level of uncertainty in model construction we should consider random
variables instead of constants.
16. The simpler the model, the easier its manipulation and solution will be.
17. Management science is an art.
18. A management science technique usually applies to a specific model type.
19. A model is a functional relationship that includes variables, parameters, and equations. T
20. Management science encompasses logical approach to problem solving. T
Chapter II: LINEAR PROGRAMMING

Introduction

Many big choices made by a company rely on the best path to accomplish the firm's
objectives, according to the operational climate constraints imposed on the manager. Such
constraints that take the form of finite capital, such as time, labour, energy, content, or money; or
they may be in the form of stringent instructions, such as a cereal or engineering design manual.
One of the company firms' most important objectives is to get as much income as possible or, in
other terms, to increase profit. The aim of specific business structures within a company (such as
a production or packaging department) is frequently to minimize costs.

Learning Objectives
After studying this module, you should be able to:
1. Understand and define linear programming.
2. Identify all necessary items that must be included in a model.
3. Write a verbal statement of the objective function and each constraint.
4. Define the decision variables.
5. Write the objective function in terms of the decision variables.
6. Write the constraints in terms of the decision variables.
7. Makes a decision about the solution of the problem.

Linear Programming

Linear Programming is a computer science technique to address issues of optimization


process. The word "linear" implies the linearity of all mathematical relationships within a model.
The standard model is the set of linear equations and/or inequalities (called constraints) and the
linear objective function (to be maximized or minimised). The constraints of non-negativity
(variables are negative or positive) are also included in the model. The manager's aim is to find
the right solution for the highest objective function value.2.1 Formulation of the Mathematical
Model

Example

ABC, Inc. creates 2 types of toys: car and boat. The car is priced at P550, and the boat at P700.
The cost of the car is P50, while P70 for the boat. The car needs 1 hour of woodwork labor and 1
hour of painting and assembling labor. The boat requires 2 hours of woodwork labor and1 hour of
painting and assembling labor. Cost of woodwork labor is P30 per hour, worth of painting and
assembling labor is P20 per hour. Monthly, ABC has 5000 existing hours of woodwork labor and
3000 hours of painting and assembling labor. There is an unlimited demand for boat, while an
average demand for car is at most 2000. ABC wants to get the best out of monthly profit (total
revenue - total cost).

Formulation

1. Decision variables

The variables would explain thoroughly the management's decisions to be made. To


optimize the income, the manager needs to determine how many cars and how many boats will be
generated per month. In this scenario, the variables to the decision are:

x1= number of cars produce each month;


x2= number of boats produce each month
2. Objective function

This function reflects the criteria of the management which must be maximized or minimized. The
management aims to optimize gross monthly income in the circumstance of the ABC as the
disparity between total monthly revenues and total monthly costs. Revenue and cost can be
represented as the function of the variables x1 and x2 for decision.

1. Total revenue (TR) = revenue from sold car + revenues from sold boat. The price of one car is
P550 and ABC produces x1 of car, the revenue resulting from the cars is 550x1. Likewise, the
revenue from sold boats is 700x2.

The total monthly revenue from production is expressed as:

TR= 500x1 + 700x2

2. Monthly wood cost (WC) = WC of produced cars + WC of produced boats. The wood cost of
production of one car (50 ) and the total number of produced car is x1, the monthly wood cost of
all produced cars is 50x1. Likewise, the monthly wood cost of all boats is 70x2.

Total monthly wood cost is :

WC= 50x1 + 70x2

3. Woodworrk labor cost (WLC) = WLC of produced cars + WLC of produced boats. One car
needs 1 hour of woodwork labor and cost of 1 hour of this labor is 30 , the unit cost is 30 . The
monthly cost of carpentry labor used for all produced cars is 30x1. One boat requires 2 hours, the
monthly cost of woodwork labor used for boats is 60x2.

Total monthly cost of woodwork labor:

WLC= 30x1 + 60x2

4. Painting and assembling labor cost (PALC) = PALC of produced cars + PALC of produced
boats. Both a piece of car and boat need 1 hour of painting and assembling labor. Cost of this labor
is 20 per hour. Therefore the total monthly cost of painting and assembling labor:

PALC= 20x1 +20x2

5. The total monthly cost can be expressed as:


TC= WC + WLC + PALC= (50x1 + 70x2) + (30x1 + 60x2) + (20x1 +20x2)
TC= 100x1 + 150x2

6. The total profit:


TP = TR − TC = (550x1 + 700x2) − (100x1 + 150x2 ) = 450x1 + 550x2

the objective function in the linear programming model is:


Maximize z = 450x1 + 550x2

450 and 550 in the function are termed objective function coefficients.

3. Constraints
If no restrictions are imposed, objective function (profit) can expand to infinity. There are three
limits (called constraints) to the development of toys, though:
1. ABC, Inc. has just 5000 hours of woodwork labor accessible per month.
2. No more than 3000 hours of finished labor can be used each month.
3. Because of minimal demand, there will be a maximum of 2000 cars delivered each month.

Expressing the constraints in mathematical way:

1. One car needs 1 hour of woodwork labor. If ABC manufactures monthly x1 of cars, x1 hours
of labor are consumed. Since one boat requires 2 hours and the manufacturing quantity equals x2,
the monthly consumption of woodwork labor is 2x2 hours.
Total consumption of woodwork labor for both products can be expressed as x1 + 2x2.
This is actual use of labor (in hours) that cannot be greater than available number of hours (5000).
With this, the constraint can be:
x1 + 2x2 ≤ 5000

2. The creation of the second constraint regarding painting and assembling labor is like to the
foregoing:

x1 + x2 ≤ 3000

3. The last constraint is easy to be construct. The number of produced cars x1 must be less than or
equal to 2000:
x1 ≤ 2000

Technological coefficients are coefficients of the decision variables in the constraints while
numbers 5000, 3000 and 2000 are termed right-hand side values.

4. Non-negativity Constraints
Both decision variables have rational sign restrictions: because the values of variables
reflect numbers of toys made, we would expect them not to be negative:

x1, x2 ≥ 0

Summary of mathematical model in standard form:

Maximize z = 450x1 + 550x2


subject to
x1 + 2x2 ≤ 5000,
x1 + x2 ≤ 3000 ,
x1 ≤ 2000 ,
x1, x2 ≥ 0 .

2.2 Graphical Solution of Linear Programming Problems

When a linear programming problem only involves two decision variables, it can be graphically
overcome. The next phase proceeds to solution:

Step 1 - Graphing a feasible area

That restriction (including non-negativity constraints) can be drawn very easily; because
all constraints have to be met concurrently, the mixture of drawn constraints defines the area that
is feasible. Each point from the feasible area suits the feasible solution.
There are 3 constraints and 2 nonnegativity constraints in the ABC’s problem.
Nonnegativity constraints are simple to graph, as all x1 and x2 variations must be in the first
quadrant of x1-x2 plane (Figure 2.1).

Linear inequality graphing (woodwork labor constraint)


x1 + 2x2 ≤ 5000

the borderline needs to be draw, corresponding to the linear equation


x1 + 2x2 = 5000 .

Coordinates of two points must be find to get the borderline. The simplest way is to look for the
borderline’s intersection points with axes x1 and x2. If x1 is set to zero (i.e. no cars are
manufactured), then x2 = 5000/2 = 2500 (2500 boats can be manufactured). This condition refers
to the point A (0, 2500) in Figure 2.2.

Also, if x2 is set to zero (no boats are manufactured), then x1 = 5000 (5000 cars can be
manufactured). It links to the point B (5000, 0) in Figure 2.2.

Linking points A and B by a straight get a graphical illustration of the equation x1 + 2x2 = 5000
We must now decide which portion of the first quadrant (divided by line AB) conforms to the x1
+ 2x2 5000 inequality.
For this reason the general approach is to pick an arbitrary point (of course not from the line) and
to test if it is feasible. When the point is feasible so the whole area is always feasible.

The best way is to take origin, (0 , 0) in our case. After these coordinates have been substituted by
the constraints, we get 5000 0 (the restriction is met). Therefore, the area of origin included is
feasible (shaded in Figure 2.2).
Figure 2.2 Woodwork Labor

Note: We omitted negative values of x1 and x2 due to nonnegativity constraints (Figure 2.1) in
graphing the feasible field.

Drawn in Figure 2.3 is the second inequality 3000 x1 + x2 ≤ (painting and assembling labor
constraint), the last constraint 2000x1 ≤ is comprehended in Figure 2.4. Feasible area can be seen
on the left side of the line while 2000x1 = is emphasized by a line parallel to the axis x2.

Figure 2.3 Painting and Assembling Labor


Step 2 - Combining the constraints

If all of the restrictions are drawn independently, they must be placed together in a single
graph. Practically, this final graph may be built by systematic addition of the individual feasible
areas from the beginning. The ultimate graph of the feasible area in the example is seen in figure
2.5. Any solution in this area is feasible as regards all the constraints.

Step 3 – Graphing an objective function

In the example the goal of the management is represented as the function of profit z= 450x1
+ 550x2. Because of the endless number of equations this feature can not necessarily be viewed
as a single line. The basic equation depends on z value.
Figure 2.6 introduces three of those equations of values 495,000, 1,485,000 and 2,475,000.
All the lines are parallel to each other and can be constructed in the same way as the constraints
borderline in Step 1. Both points on a specific line produce the same profit, and hence the lines
are typically called isoprofit lines. The arrow expressed income growth.

Step 4 – Finding the optimal solution

Integrating the feasible region with the isoprofit lines, we will consider the isoprofit line that is as
far from the origin as feasible, which also crosses the feasible field.

The dotted isoprofit line in Figure 2.7 correlates to the maximum profit z = 1,550 000. This line
intersects a viable area corner point, which is considered an optimal solution. It is point X (1000,
2000)

There are two potential ways of seeking the best approach using the model's graphical
representation:

1. The first approach has only been illustrated. We can determine the "best" point of the feasible
area, by finding a slope and the growth direction of the isoprofit line. This level is the optimal
solution, and the isoprofit line suits the ideal objective function optimal value. Within this section
of the text it is important to notice that a convex set is the feasible field of all linear programming
models.

A set of points S is a convex set if the section of the line connecting each pair of points in
S is entirely enclosed in S.

The disparity between convex and non-convex sets is shown in Figure 2.8. Although the first three
examples are convex sets (a), (b ) and ( c), the group (d) is non-convex.

The set (a) varies significantly from the sets (b ) and ( c), due to nonlinear border. If the
convex set boundary consists only of linear segments, as seen in cases (b ) and ( c), the set is called
a convex polyhedron, which reflects the standard feasible field of linear programming models. It
is the feature which is useful in linear programming methods.

The key problem is how to measure optimum solution of coordinates X(1000, 2000). As it
is apparent from Figure 2.6, this point is the borderline intersection of two constraints.
Determining the co-ordinates is therefore very simple by solving the series of two linear equations.
The borderlines compare in our situation are:
x1 + 2x2 = 5000 ,
x1 + x2 = 3000 .

Pretty simple solution: x1 = 1000 and x2 = 2000. Those values are point X coordinates.
Incorporating them into objective function z = 450x1 + 550x2 = 1 550 000.

2. It is important to identify a corner point before beginning to explain the second approach used
to find the best solution in the graph. With respect to the convex polyhedron definition it is possible
to describe the corner point clearly as follows:

A point P in convex polyhedron S is a corner point if it does not lie on any line
joining any pair of other (than P) points in S.

Keystone of the second approach is the basic linear programming theorem:

The optimal feasible solution, if it exists, will occur at one or more of the corner points.

In this theorem, what we need is to define the feasible area's corner points and determine the
objective values for all of those points. The optimal solution is the corner point and has the highest
objective value.
The feasible area in the example contains 5 corner points (Figure 2.9), of which the
coordinates can be observed in Table 2.1. All the coordinates were calculated in the manner
described above: after classification of the intersected lines, the set of equations is solved and the
x1 and x2 co-ordinates provide the solution. The objective value z is then calculated for each
solution by incorporating the co-ordinates into the objective function. From the table it is evident
that the optimal solution corresponds to the maximum objective value z = 1,550,000.

Corner point x1 x2 z
A 0 0 0
B 2000 0 900 000
C 2000 1000 1 450 000
D 1000 2000 1 550 000
E 0 2500 1 375 000

Simplex method is a general method used for solving linear programming problems. It is
an iterative algorithm for efficient searching for the optimal solution. The method uses the Gauss-
Jordan method of solving simultaneous equations. In addition, the method is based on the basic
linear programming theorem (the search can concentrate only on the corner points of the feasible
area). The run of algorithm starts in one of the corner points (usually in the origin) and moves to
adjacent corner that improves the value of the objective function. The process of movement
continues until no further improvement is possible. The simplex method and its variations have
been programmed and therefore even large linear programming problems can be easy solved.

2.3 Understanding the Optimal Solution

The appropriate and reliable analysis of the findings is almost as critical as the model itself.
After the solution has been obtained, it is necessary to go back to the start of the modeling process
and compare the variables with their real representations. In the case, x1 represents the number of
cars created each month, and x2 is the number of boats generated every month. Therefore it is
optimal to produce 1000 cars and 2000 boats every month as the optimal solution we have got x1
= 1000 and x2 = 2000. When the management maintains this output plan, ABC, Inc. generates
income of 1,550,000 per month (see the target role description in the model).
When the production varied from the measured approach but nevertheless followed all
product constraints, the profit would've been lower. By violating at least one limit, the
management can achieve greater income than 1 550 000. That will be an infeasible approach.

Analyzing the findings ABC would be involved in practical usage of available resources
(working hours) and meeting limited demand. If we integrate the optimal values of x1 and x2 into
all the restrictions, we can see whether they are fulfilled exactly as equations or as inequalities. As
for woodwork labor, the limit's left-hand side value is 1000 + 2(2000) = 5000, which is almost the
same as the right-hand side value 5000 (the constraint is obligatory). Since the right-hand side
relates to the hours of work available, equality means that no hour stays or all available hours are
being used for optimal production.

The painting and assembling labor condition is similar. The constraint's left-hand side
value is 1000 + 2000 = 3000, the right-hand side value is always equal to 3000; therefore, all
required painting and assembling labor hours are used. The last constraint relates to demand
limitation (the amount of cars created should be less than or equivalent to 2000). Because output
of 1000 cars is optimal (the constraint is non-binding), this is not the restriction that affects ABC,
Inc's production. The organization could increase cars production as it succeeded in having more
hours of work or developing the technologies ( i.e. reducing unit labor consumption).

Note: Slack variable is defined for the inequality of type “≤”. This variable assesses the
difference between the left-hand side and the right-hand side of the constraint. In both labor
constraints, zero is the slack variable in the example, as both sides of the constraints equal to each
other. In demand constraint, 1000 is the slack variable, as the right-hand side value is 2000 and
left-hand side value is 1000. In inequality of type “≥”, we express a surplus variable assessing the
difference between the left-hand side and the right-hand side of the constraint. For example, if
ABC produce at least 600 cars because of demand requirements (x1 ≥ 600) then ABC will actually
produce 1000 cars, therefore, 400 is the surplus variable.

The estimation of the effect on the optimal solution of the modifications in a particular
environment is the concept of a post-optimal (sensitivity) analysis. We have forecast future
increases in resource stocks (available labor hours) in technology, so it is important to examine
how shifts in commodity prices impact profit.

Figure 2.10 we compares the original objective function z=450x1+550x2 with the function
z=750x1 + 550x2 (the price of the cars changes from 550 to 850 ), getting a new optimal solution
(2000, 1000) with the profit z = 2,050,000.

Sensitivity analysis defines, typically for coefficients of objective function and right-hand
side values, the array of their potential changes that do not have the cardinal effect on the optimal
solution. Changing the car’s price from 550 to 649 (i.e. the objective function coefficient was
improved from 450 to 549), the optimal solution would be the original corner point (1000, 2000).
Only modification would be higher profit, while the production assembly would be alike. If the
price increased to 651, the optimal solution (2000, 1000) would be attained, and resulting in this
recommendation the business should change the production assembly from the base. The price
650 provides the business the option to manufacture either 1000 cars (and 2000 boats) or 2000
cars (and 1000 boats).

Modification in coefficients of objective function impacts only its slope, whereas the
feasible area remains static. The best corner point was determined by the slope of the function. If
right-hand side value is modified the feasible area is also modified.

Figure 2.11 Change in Painting and Assembling Labor

The constraint’s borderline’s movement was due to change (the new borderline is parallel
with the original). Smaller change in the right-hand side value means smaller change in the feasible
area while big changes in the feasible area makes the cardinal change as shown by Figure 2.11
(decrease in available painting and assembling labor from 3000 to 2000 hours). Corner points
original number of 5 decreases to 3. In this situation, new optimal solution is presented by the
corner point (0, 2000).

As the optimal corner point is out of the constraint of woodwork labor, there are several
idle hours of this labor, i.e. in this inequality the slack variable now is nonzero. The precise sum
remaining can be calculated simply by inserting the coordinates (0, 2000) into the constraint's left-
hand side. When 0 + 2(2000) = 4000, the amount of hours remaining unused is 5000-4000 = 1000.

Note: These modifications can be analyzed independently, i.e. only one parameter is modified (the
other is fixed), so we are evaluating the effect on the optimal solution and the profit. The basic
aspects of the sensitivity analysis are the lowered costs and the shadow prices along with the levels
of objective function coefficients and the ranges on right-hand hands.

2.4 Special Cases of Linear Programming Models

Only one optimal solution has been attained in the production problem of ABC, Inc. It is
a distinctive in most cases linear programming problems. Figure 2.12 express the feasible area
together with the objective function z (the arrow specifies the improvement direction’s of the
objective value). Unique optimal solution is the corner point A.
In the previous discussion, the price of the car is 650 (instead of 550), the management of
ABC, Inc. has two basic possibilities how to optimize the profit. Producing 1000 cars (and 2000
boats) or 2000 cars (and 1000 boats) would bring the identical optimal profit 1,650,000.

A linear programming problem with two or more optimal solutions is said to have
alternative (or multiple) optimal solutions.

This condition occurs in graphical depiction of the model when the objective function line
is parallel to the borderline of one constraint, as can be shown in Figure 2.13. Because the objective
function reflects the isoprofit line (in case of maximization) or the isocost line (in case of
minimization), all the solutions at the edge of the feasible area have the same objective value
(optimal). There are two optimal corner points (B and C) and the infinite number of optimal points
on line segment BC.

Figure 2.14 shows an interesting alternative of just described case. The difference is
evident – the feasible area is unbounded and there is only one optimal corner point D; the other
optimal points lie on the borderline running to infinity.
There is no optimal solution if the feasible solution area is unbounded and the objective
value is being enhanced in the direction of unboundedness, therefore, the optimal solution is
endless (see Figure 2.15). Since these instances are uncommon in reality, this finding typically
indicates mistake in formulation.

Infeasibility exists when no solution meets any of the constraints. There is no common
feasible area (no feasible solution) in case of two constraints graphed in Figure 2.16. In reality this
scenario sometimes occurs, particularly if the manager wants to be too accurate in model
formulation. In order to remove the infeasibility, the model must be simplify by for example
eliminating any vain restrictions.
While the two-variable model is not popular in practice and is used more for analytical purposes,
for multidimensional problems both theories may be simplified and transformed.

Exercises and Activities

I.Define the decision variables.

1. PR Developments is skilled in analyzing the response of customers to the latest goods and
services. A client, launching new form of body lotion, requested the PR Developments to plan a
campaign of specific door-to - door interviews about the opinion of households. Households with
children and without children should be interviewed; interviews will be held both daytime and
nighttime. A schedule for performing 1000 interviews with the following constraints is set:

A) There will be interviews with at least 300 households with children.


B) Interviews will be conducted for at least 400 households without children.
C) The total number of evening interviews should be greater than or equal to the number of
daytime interviews.
D) At least 35 percent of interviews should be conducted in households with children at evening.
E) At least 65 per cent of interviews will be performed in households without children evening
time

The cost varies according to interview type. Household interview with children is more
complicated and longer; evening interviewers are paid better than daytime interviewers.

The estimation of the cost is as follows:

Household Daytime interview Evening interview


with children 50 60
without children 40 50

PR Creations project manager can recommend interview schedule that would satisfy the
constraints for the lowest minimum total cost.

2InMe Co., is a business that is engaged in spending capital in the stocks of drink making firms.
The company's board is speculating about spending 2,000,000 in four available shares. There is a
need to spend any amount of money in government bond to stop a high-risk index. Projected from
a long-term financial market study, the following annual return expectations and risk indices:
Rate of return Risk index
Tropical Beer share 12 % 0.07
American Wine share 9% 0.09
American Brandy share 15 % 0.05
Tropical Milk share 7% 0.03
Government bond 6% 0.01

Such constraints were named at the managers’ meeting.


a) No more than 200,000 should be invested in Tropical Milk shares.
b) At least 20% of all investments must be covered by government bonds.
c) Neither alcohol-drink business might get more than 800,000 because of portfolio diversification.
d) Final portfolio risk index must be maximally 0.05.

Satisfying all the restrictions the management intends to maximize the annual return of the
portfolio.

II. Compute for the optimal solution of Problems # 1 and 2.

III. Multiple choice.

1. Surplus variable is
a. a nonnegative variable subtracted from the left-hand side of “≥” constraint that converts it
to equation
b. a negative variable subtracted from the left-hand side of “≥” constraint that converts it to
equation
c. a nonnegative variable subtracted from the right-hand side of “≥” constraint that converts
it to equation
d. a negative variable subtracted from the right-hand side of “≥” constraint that converts it to
equation

2. A feasible solution that maximizes or minimizes the objective function.


a. Optimal value
b. Optimal solution
c. Maximization
d. Optimization

3. The only one solution of the managerial problem that maximizes/minimizes the objective
function.
a. Bounded feasible solution
b. Unbounded feasible solution
c. Unique optimal solution
d. Optimal solution

4. Slack variable is
a. a negative variable added to the right-hand side of “≤” constraint that converts it to
equation
b. a nonnegative variable added to the right-hand side of “≤” constraint that converts it to
equation
c. a negative variable added to the left-hand side of “≤” constraint that converts it to
equation
d. a nonnegative variable added to the left-hand side of “≤” constraint that converts it to
equation
5. Nonnegativity constraint is the restriction specifying that decision variables in the model
must
a. be positive or zero
b. be negative or zero
c. be zero
d. either positive or negative

6. Deterministic parameters in the objective function that evaluate the usefulness of the
individual variables regarding the goal of the optimization.
a. Objective function
b. Objective function coefficient
c. Negativity constraints
d. Nonnegativity constraints

7. An inequality constraint for which, at a particular feasible point, left-hand side is


less/greater than the right-hand side
a. Negativity constraints
b. Nonnegativity constraints
c. Nonbinding constraints
d. Binding constraints

8. Linear programming model is a mathematical model with


a. a linear objective function, a set of linear constraints, and negative variables
b. a linear objective function, and negative variables
c. a linear objective function and a set of linear constraints
d. a linear objective function, a set of linear constraints, and nonnegative variables

9. A mathematical procedure for optimizing the linear objective function, respecting the sets
of linear constraints
a. Linear programming
b. Linear function
c. Linear objective
d. Linear programming model
10. An equation or inequality that rules out certain combinations of values for the variables
involved.
a. Problem
b. Constraints
c. Function
d. Surplus

11. A variable that represents managerial decision or component of decision that must be made
in a problem solving situation.
a. Slack variable
b. Surplus variable
c. Decision variable
d. Constraints variable

12. In defining the decision variable, the variables and their values represent the manager’s
decisions. These variables will be used in the mathematical model, therefore their
definition must be clear and unique.
a. Only the 1st statement is true
b. Only the 2nd statement is true
c. Both statements are false
d. Both statements are true
13. In transforming the verbal statement of each constraint into mathematical statement in the
form of equation or inequality,
a. the left-hand side of each constraint must be a linear function of the decision variables
and the right-hand side must be a nonnegative constant.
b. the right-hand side of each constraint must be a linear function of the decision variables
and the left-hand side must be a nonnegative constant.
c. the left-hand side of each constraint must be a nonnegative constant and the right-hand
side must be a linear function of the decision variables.
d. both hands must be a linear function of the decision variables

14. In case that the feasible solution area is unbounded and the objective value is being
improved in the direction of unboundedness, therefore
a. Optimal solution will be found
b. An error occurred in the formulation
c. The optimal solution is finite
d. All of the above will happen

15. The coefficients of the decision variables in the constraints are called
a. Right-hand side values
b. left-hand side values
c. technological coefficients
d. objective value
CHAPTER 3 – DISTRIBUTION MODELS

LEARNING OBJECTIVES
At the end of the lessons, the students are expected to:
1. formulate distribution models;
2. analyze and solve transportation problems; and
3. analyze and solve assignment problems.

Distribution Models
A distribution problem is a special type of linear programming problem. There are two
main types of distribution problems:
 Transportation Problem
 Assignment Problem

The transportation problem deals with shipments from number of sources to a number of
destinations while on the contrary, the assignment problem deals with finding the best one-to-one
match for each of a given number of viable candidates to variety of proposed “positions”. The
objective is to induce maximum reward or minimum cost.

Transportation Problem
Transportation problem deals with shipments (of materials, goods, people, etc.) from
a number of sources to a number of destinations. Every source has generally the restricted supply
and every destination has usually known demand. Unit shipping cost between each source and
destination is determined. The objective is to seek out a possible shipping schedule (shipped
quantities) with minimal total shipping cost. This model can also be used when a company is
deciding where to locate a new facility. Sensible monetary decisions regarding facility location
also attempt to minimize total transportation and production costs for the entire system.
If the total supply equals the total demand, the model is termed balanced transportation
model. In real-life situations the problem is usually unbalanced, i.e. the total demand exceeds the
total supply, or the total supply exceeds the total demand. Unbalanced models can be transformed
into balanced by introducing a dummy source or destination. Whereas shipments from a dummy
source correspond to unsatisfied requirements, shipments to a dummy destination represent
remains of suppliers.

Example #1
Star Coffee, Inc. operating in the Czech Republic is going to establish three subsidiaries
producing coffee. They should be located in following cities: Benešov, Jihlava and Tábor.
The main ingredient – coffee beans - will be supplied from two warehouses in Humpolec
and Pelhřimov (see Figure 1).

Fig.1 Star Coffee’s Distribution Problem


The management of the corporation has estimated the weekly requirements of the
companies. The warehouses’ capacities are limited. Coffee beans are transported once a week
from suppliers to destinations by train and it is possible to evaluate a unit shipping cost per ton.
All the values are given in the following table:

Destination Weekly
Benešov Jihvala Tábor Supply [t]
Warehouse
Humpolec 330 250 350 70
Pelhřimov 300 240 250 80
Weekly 150
Demand [t] 45 60 35 140
Tab. 1 Supply, Demand and Unit Shipping Cost
The objective is to ascertain such deliveries from warehouses to destinations that minimize
the total shipping value. This shipping schedule must satisfy the requirement of each
destination, and should not exceed supply of any warehouse.

Formulation

As it is apparent from Table 1 the problem is unbalanced - total demand (140 tons) is less
than the total supply (150 tons). The balanced model can be obtained after introducing a dummy
destination with demand 10 tons (computed as the difference between the values 150 and 140):

Destination Weekly
Benešov Jihvala Tábor
Dummy Supply [t]
Warehouse
Humpolec 330 250 350 0 70
Pelhřimov 300 240 250 0 80
Weekly 150
Demand [t] 45 60 35 10 150
Tab. 1.2 Balanced Transportation Model

Since the dummy destination does not exist, no physical shipping occurs and therefore
all the corresponding unit transportation costs equal zeroes.

1. Define the decision variables

In distribution models, it is reasonable to define the variables with two indices (generally
xij). In case of transportation model the variable xij corresponds to the amount transported from
the source i to the destination j.

x11 = amount of potatoes transported from Humpolec to Benešov [in tons],


x12 = amount of potatoes transported from Humpolec to Jihlava,

x13 = amount of potatoes transported from Humpolec to Tábor,


x14 = amount of potatoes remaining in Humpolec.
x21 = amount of potatoes transported from Pelhřimov to
Benešov, x22 = amount of potatoes transported from Pelhřimov
to Jihlava, x23 = amount of potatoes transported from Pelhřimov
to Tábor, x24 = amount of potatoes remaining in Pelhřimov.

2. The objective is to minimize the total shipping cost

If parameter cij is generally a unit shipping cost corresponding to the source i and the
destination j, then cij xij is the total shipping cost. If we evaluate the total shipping cost
corresponding to all the sources and destinations, their sum gives the total shipping cost for our
transportation problem. Thus, the objective function can be expressed as follows:

Minimize z = 330x11 + 250x12 + 350x13 + 300x21 + 240x32 + 250x33 .

3. The constraints

Real shipping must not exceed supply of warehouses and must satisfy demand
of destinations. In the example, there are 6 constraints:
x
1
1 + x12 + x13 + x14 = 70
x + x22 + x23 + x24
21 = 80
x
1
1 + x21 = 45
x
12 + x22 = 60
x
13 + x23 = 35
x
14 + x24 = 10

4. The nonnegativity constraints


x11 , x12 , ..., x24 ≥ 0 .
Optimal Solution
Table 1.3 shows the optimal solution. Every week, 60 tons is delivered from Humpolec
to Jihlava. From Pelhřimov, 45 tons are transported to Benešov and 35 tons to Tábor. This
shipping schedule minimizes the total weekly cost at the level 37 250 CZK. In Humpolec, 10
tons remain. This amount can be used for another purpose or the management should consider
increase in the production of the subsidiaries.

Destination Benešov Jihlava Tábor Dummy Weekly


Warehouse Supply [t]

Humpolec - 60 - 10 70

Pelhřimov 45 - 35 - 80

Weekly 150
Demand [t] 45 60 35 10 150

Tab. 1.3 Optimal Solution – Transported Amount


Initial solution - Northwest Corner Rule

The initial basic possible solution can be obtained by using Northwest Corner Rule. The
procedure for constructing an initial basic feasible solution selects the basic variables one at a time.
The method begins with an allocation at the top left-hand corner of the tableau and proceeds
systematically along either a row or a column and make allocations to subsequent cells until the
bottom right-hand corner is reached, by which time enough allocations will have been made to
constitute an initial solution. The procedure for constructing an initial solution using the North
West Corner rule is as follows:

North-West Corner Method (NWCM) Steps (Rule)


Step-1: Select the upper left corner cell of the transportation matrix and allocate min(s1,
d1).
Step-2: a. Subtract this value from supply and demand of respective row and column.

b. If the supply is 0, then cross (strike) that row and move down to the next cell.

c. If the demand is 0, then cross (strike) that column and move right to the next
cell.

d. If supply and demand both are 0, then cross (strike) both row & column and
move diagonally to the next cell.
Step-3: Repeat this steps until all supply and demand values are 0.
Example-2
1. Find Solution using North-West Corner method

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18
Demand 5 8 7 14

Solution:
TOTAL number of supply constraints : 3
TOTAL number of demand constraints : 4
Problem Table is

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18

Demand 5 8 7 14

The rim values for 𝑆1=7 and 𝐷1 =5 are compared.

The smaller of the two i.e. min(7,5) = 5 is assigned to 𝑆1 𝐷1


This meets the complete demand of 𝐷1 and leaves 7 - 5 = 2 units with 𝑆1

Table-1
D1 D2 D3 D4 Supply
S1 19(5) 30 50 10 2
S2 70 30 40 60 9
S3 40 8 70 20 18

Demand 0 8 7 14

The rim values for 𝑆1=2 and 𝐷2 =8 are compared.

The smaller of the two i.e. min(2,8) = 2 is assigned to 𝑆1 𝐷2

This exhausts the capacity of 𝑆1 and leaves 8 - 2 = 6 units with 𝐷2

Table-2

D1 D2 D3 D4 Supply
S1 19(5) 30(2) 50 10 0
S2 70 30 40 60 9
S3 40 8 70 20 18

Demand 0 6 7 14

The rim values for 𝑆2 =9 and 𝐷2 =6 are compared.

The smaller of the two i.e. min(9,6) = 6 is assigned to 𝑆2 𝐷2

This meets the complete demand of 𝐷2 and leaves 9 - 6 = 3 units with 𝑆2

Table-3
D1 D2 D3 D4 Supply
S1 19(5) 30(2) 50 10 0
S2 70 30(6) 40 60 3
S3 40 8 70 20 18

Demand 0 0 7 14

The rim values for 𝑆2 =3 and 𝐷3 =7 are compared.

The smaller of the two i.e. min(3,7) = 3 is assigned to 𝑆2 𝐷3

This exhausts the capacity of 𝑆2 and leaves 7 - 3 = 4 units with 𝐷3

Table-4

D1 D2 D3 D4 Supply
S1 19(5) 30(2) 50 10 0
S2 70 30(6) 40(3) 60 0
S3 40 8 70 20 18

Demand 0 0 4 14

The rim values for 𝑆3 =18 and 𝐷3 =4 are compared.

The smaller of the two i.e. min(18,4) = 4 is assigned to 𝑆3 𝐷3

This meets the complete demand of 𝐷3 and leaves 18 - 4 = 14 units with 𝑆3

Table-5

D1 D2 D3 D4 Supply
S1 19(5) 30(2) 50 10 0
S2 70 30(6) 40(3) 60 0
S3 40 8 70(4) 20 14

Demand 0 0 0 14

The rim values for 𝑆3 =14 and 𝐷4 =14 are compared.

The smaller of the two i.e. min(14,14) = 14 is assigned to 𝑆3 𝐷4

Table-6

D1 D2 D3 D4 Supply
S1 19(5) 30(2) 50 10 0
S2 70 30(6) 40(3) 60 0
S3 40 8 70(4) 20(14) 0

Demand 0 0 0 0

Initial feasible solution is

D1 D2 D3 D4 Supply
S1 19 (5) 30 (2) 50 10 7
S2 70 30 (6) 40 (3) 60 9
S3 40 8 70 (4) 20 (14) 18

Demand 5 8 7 14

The minimum total transportation cost =19×5+30×2+30×6+40×3+70×4+20×14=1015

Here, the number of allocated cells = 6 is equal to m + n - 1 = 3 + 4 - 1 = 6


∴ This solution is non-degenerate
Least Cost Method Rule

The Least Cost Method is another method used to acquire the initial feasible solution for
the transportation problem. Using this, the allocation begins with the cell which has the minimum
cost. The lower cost cells are selected over the higher-cost cell with the objective to have the least
cost of transportation.

Least Cost Method (LCM) Steps (Rule)

Step-1: Select the cell having minimum unit cost cij and allocate as much as possible,
i.e. min(si,dj).
Step-2: a. Subtract this min value from supply si and demand dj.

b. If the supply si is 0, then cross (strike) that row and If the demand dj is 0 then
cross (strike) that column.

c. If min unit cost cell is not unique, then select the cell where maximum allocation
can be possible

Step-3: Repeat this steps for all uncrossed (unstriked) rows and columns until all supply
and demand values are 0.

Stepping Stone method

The next step is determining whether the present allocation at any stage of the solution
process is ideal. The Stepping Stone method originates its name from the analogy of crossing a
pond using stepping stones. The occupied cells are analogous to the stepping stones, which are
used in making certain movements in this method. The Stepping Stone Method is used to check
the optimality of the initial feasible solution and a process for finding the potential of any non-
basic variables (empty cells) in terms of the objective function

The five steps of the Stepping-Stone Method are as follows:

Stepping Stone method

Step-1: Select an unused square to be evaluated.


Step-2: Beginning at this square, trace a closed path back to the original square via squares
that are currently being used (only horizontal or vertical moves allowed). You can
only change directions at occupied cells

Step-3: Beginning with a plus (+) sign at the unused square, place alternative minus (-)
signs and plus signs on each corner square of the closed path just traced.

Step-4: Calculate an improvement index, Iij by adding together the unit cost figures found
in each square containing a plus sign and then subtracting the unit costs in each
square containing a minus sign.
Step-5: Repeat steps 1 to 4 until an improvement index has been calculated for all unused
squares.

• If all indices computed are greater than or equal to zero, an optimal solution has
been reached.

• If not, it is possible to improve the current solution and decrease total shipping
costs.

Example-1
Find Solution using Voggel's Approximation method, also find optimal solution using stepping
stone method

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18
Demand 5 8 7 14

Solution:
TOTAL number of supply constraints : 3
TOTAL number of demand constraints : 4
Problem Table is

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18

Demand 5 8 7 14

Table-1

D1 D2 D3 D4 Supply Row Penalty


S1 19 30 50 10 7 9=19-10
S2 70 30 40 60 9 10=40-30
S3 40 8 70 20 18 12=20-8

Demand 5 8 7 14
Column
21=40-19 22=30-8 10=50-40 10=20-10
Penalty

The maximum penalty, 22, occurs in column 𝐷2 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐32 = 8.
The maximum allocation in this cell is min(18,8) = 8.
It satisfy demand of 𝐷2 and adjust the supply of 𝑆3 from 18 to 10 (18 - 8 = 10).

Table-2

D1 D2 D3 D4 Supply Row Penalty


S1 19 30 50 10 7 9=19-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20 10 20=40-20

Demand 5 0 7 14
Column
21=40-19 -- 10=50-40 10=20-10
Penalty

The maximum penalty, 21, occurs in column D1.


The minimum 𝑐𝑖𝑗 in this column is 𝑐11 = 19.
The maximum allocation in this cell is min(7,5) = 5.
It satisfy demand of 𝐷1 and adjust the supply of 𝑆1 from 7 to 2 (7 - 5 = 2).

Table-3

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10 2 40=50-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20 10 50=70-20

Demand 0 0 7 14
Column
-- -- 10=50-40 10=20-10
Penalty

The maximum penalty, 50, occurs in row S3.


The minimum 𝑐𝑖𝑗 in this row is 𝑐34 = 20.
The maximum allocation in this cell is min(10,14) = 10.
It satisfy supply of 𝑆3 and adjust the demand of 𝐷4 from 14 to 4 (14 - 10 = 4).
Table-4

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10 2 40=50-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 4
Column
-- -- 10=50-40 50=60-10
Penalty

The maximum penalty, 50, occurs in column 𝐷4 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐14 = 10.

The maximum allocation in this cell is min(2,4) = 2.


It satisfy supply of 𝑆1 and adjust the demand of 𝐷4 from 4 to 2 (4 - 2 = 2).

Table-5

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 0 --
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 2
Column
-- -- 40 60
Penalty

The maximum penalty, 60, occurs in column 𝐷4 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐24 = 60.
The maximum allocation in this cell is min(9,2) = 2.
It satisfy demand of 𝐷4 and adjust the supply of 𝑆2 from 9 to 7 (9 - 2 = 7).

Table-6

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 0 --
S2 70 30 40 60(2) 7 40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 0
Column
-- -- 40 --
Penalty

The maximum penalty, 40, occurs in row 𝑆2 .


The minimum 𝑐𝑖𝑗 in this row is 𝑐23 = 40.
The maximum allocation in this cell is min(7,7) = 7.
It satisfy supply of 𝑆2 and demand of 𝐷3 .

Initial feasible solution is

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 7 9 | 9 | 40 | 40 | -- | -- |
S2 70 30 40(7) 60(2) 9 10 | 20 | 20 | 20 | 20 | 40 |
S3 40 8(8) 70 20(10) 18 12 | 20 | 50 | -- | -- | -- |

Demand 5 8 7 14
21 22 10 10
21 -- 10 10
Column -- -- 10 10
Penalty -- -- 10 50
-- -- 40 60
-- -- 40 --

The minimum total transportation cost =19×5+10×2+40×7+60×2+8×8+20×10=779

Here, the number of allocated cells = 6 is equal to m + n - 1 = 3 + 4 - 1 = 6


∴ This solution is non-degenerate

Optimality test using stepping stone method...


Allocation Table is

D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 40 (7) 60 (2) 9
S3 40 8 (8) 70 20 (10) 18

Demand 5 8 7 14
Iteration-1 of optimality test

1. Create closed loop for unoccupied cells, we get

Unoccupied cell Closed path Net cost change


S1D2 S1D2→S1D4→S3D4→S3D2 30 - 10 + 20 - 8 = 32
S1D3 S1D3→S1D4→S2D4→S2D3 50 - 10 + 60 - 40 = 60
S2D1 S2D1→S2D4→S1D4→S1D1 70 - 60 + 10 - 19 = 1
S2D2 S2D2→S2D4→S3D4→S3D2 30 - 60 + 20 - 8 = -18
S3D1 S3D1→S3D4→S1D4→S1D1 40 - 20 + 10 - 19 = 11
S3D3 S3D3→S3D4→S2D4→S2D3 70 - 20 + 60 - 40 = 70

2. Select the unoccupied cell having the highest negative net cost change i.e. cell S2D2=-18.
and draw a closed path from S2D2.
Closed path is S2D2→ S2D4→ S3D4→ S3D2
Closed path and plus/minus allocation for current unoccupied cell S2D2

D1 D2 D3 D4 Supply
S1 19 (5) 30 [32] 50 [60] 10 (2) 7
S2 70 [1] 30 [-18] (+) 40 (7) 60 (2) (-) 9
S3 40 [11] 8 (8) (-) 70 [70] 20 (10) (+) 18

Demand 5 8 7 14

3. Minimum allocated value among all negative position (-) on closed path = 2
Subtract 2 from all (-) and Add it to all (+)

D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 (2) 40 (7) 60 9
S3 40 8 (6) 70 20 (12) 18

Demand 5 8 7 14

4. Repeat the step 1 to 3, until an optimal solution is obtained.


Iteration-2 of optimality test
1. Create closed loop for unoccupied cells, we get

Unoccupied
Closed path Net cost change
cell
S1D2 S1D2→S1D4→S3D4→S3D2 30 - 10 + 20 - 8 = 32
S1D3→S1D4→S3D4→S3D2→S2D2→S2 50 - 10 + 20 - 8 + 30 - 40 =
S1D3
D3 42
S2D1→S2D2→S3D2→S3D4→S1D4→S1 70 - 30 + 8 - 20 + 10 - 19 =
S2D1
D1 19
S2D4 S2D4→S2D2→S3D2→S3D4 60 - 30 + 8 - 20 = 18
S3D1 S3D1→S3D4→S1D4→S1D1 40 - 20 + 10 - 19 = 11
S3D3 S3D3→S3D2→S2D2→S2D3 70 - 8 + 30 - 40 = 52

Since all net cost change ≥0


So final optimal solution is arrived.

D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 (2) 40 (7) 60 9
S3 40 8 (6) 70 20 (12) 18

Demand 5 8 7 14

The minimum total transportation cost =19×5+10×2+30×2+40×7+8×6+20×12=743

Modified Distribution Method


The modified distribution method, is also known as MODI method or (u - v) method
provides a minimum cost solution to the transportation problems. MODI method is an
improvement over stepping stone method. 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 objectives are to develop and review
an integral transportation schedule that meets all demands from the inventory at a minimum total
transportation cost.
The concept of MODI can be further comprehended through an illustration given below:

MODI Method Steps (Rule)


Step-1: Find an initial basic feasible solution using any one of the three methods
NWCM, LCM or VAM.
Step-2: Find ui and vj for rows and columns. To start

a. assign 0 to ui or vj where maximum number of allocation in a row or column


respectively.

b. Calculate other ui's and vj's using cij=ui+vj, for all occupied cells.
Step-3: For all unoccupied cells, calculate dij=cij-(ui+vj), .
Step-4: Check the sign of dij

a. If dij>0, then current basic feasible solution is optimal and stop this
procedure.

b. If dij=0 then alternative soluion exists, with different set allocation and same
transportation cost. Now stop this procedure.

b. If dij<0, then the given solution is not an optimal solution and further
improvement in the solution is possible.
Step-5: Select the unoccupied cell with the largest negative value of dij, and included in
the next solution.
Step-6: Draw a closed path (or loop) from the unoccupied cell (selected in the previous
step). The right angle turn in this path is allowed only at occupied cells and at
the original unoccupied cell. Mark (+) and (-) sign alternatively at each corner,
starting from the original unoccupied cell.
Step-7: 1. Select the minimum value from cells marked with (-) sign of the closed path.

2. Assign this value to selected unoccupied cell (So unoccupied cell becomes
occupied cell).

3. Add this value to the other occupied cells marked with (+) sign.

4. Subtract this value to the other occupied cells marked with (-) sign.
Step-8: Repeat Step-2 to step-7 until optimal solution is obtained. This procedure stops
when all dij≥0 for unoccupied cells.

Example-1
Find Solution using Voggel's Approximation method, also find optimal solution using MODI
method

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18
Demand 5 8 7 14

Solution:
TOTAL number of supply constraints : 3
TOTAL number of demand constraints : 4
Problem Table is

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18

Demand 5 8 7 14

Table-1

D1 D2 D3 D4 Supply Row Penalty


S1 19 30 50 10 7 9=19-10
S2 70 30 40 60 9 10=40-30
S3 40 8 70 20 18 12=20-8

Demand 5 8 7 14
Column
21=40-19 22=30-8 10=50-40 10=20-10
Penalty

The maximum penalty, 22, occurs in column 𝐷2 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐32 = 8.
The maximum allocation in this cell is min(18,8) = 8.
It satisfy demand of 𝐷2 and adjust the supply of 𝑆3 from 18 to 10 (18 - 8 = 10).

Table-2

D1 D2 D3 D4 Supply Row Penalty


S1 19 30 50 10 7 9=19-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20 10 20=40-20

Demand 5 0 7 14
Column
21=40-19 -- 10=50-40 10=20-10
Penalty
The maximum penalty, 21, occurs in column 𝐷1 .
The minimum 𝑐𝑖𝑗 in this column is 𝑐11 = 19.
The maximum allocation in this cell is min(7,5) = 5.
It satisfy demand of 𝐷1 and adjust the supply of 𝑆1 from 7 to 2 (7 - 5 = 2).

Table-3

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10 2 40=50-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20 10 50=70-20

Demand 0 0 7 14
Column
-- -- 10=50-40 10=20-10
Penalty

The maximum penalty, 50, occurs in row 𝑆3 .


The minimum 𝑐𝑖𝑗 in this row is 𝑐34 = 20.
The maximum allocation in this cell is min(10,14) = 10.
It satisfy supply of 𝑆3 and adjust the demand of 𝐷4 from 14 to 4 (14 - 10 = 4).

Table-4

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10 2 40=50-10
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 4
Column
-- -- 10=50-40 50=60-10
Penalty

The maximum penalty, 50, occurs in column 𝐷4 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐14 = 10.
The maximum allocation in this cell is min(2,4) = 2.
It satisfy supply of 𝑆1 and adjust the demand of 𝐷4 from 4 to 2 (4 - 2 = 2).

Table-5

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 0 --
S2 70 30 40 60 9 20=60-40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 2
Column
-- -- 40 60
Penalty

The maximum penalty, 60, occurs in column 𝐷4 .


The minimum 𝑐𝑖𝑗 in this column is 𝑐24 = 60.
The maximum allocation in this cell is min(9,2) = 2.
It satisfy demand of 𝐷4 and adjust the supply of 𝑆2 from 9 to 7 (9 - 2 = 7).

Table-6

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 0 --
S2 70 30 40 60(2) 7 40
S3 40 8(8) 70 20(10) 0 --

Demand 0 0 7 0
Column
-- -- 40 --
Penalty

The maximum penalty, 40, occurs in row 𝑆2 .


The minimum 𝑐𝑖𝑗 in this row is 𝑐23 = 40.

The maximum allocation in this cell is min(7,7) = 7.


It satisfies supply of 𝑆2 and demand of 𝐷3 .

Initial feasible solution is

D1 D2 D3 D4 Supply Row Penalty


S1 19(5) 30 50 10(2) 7 9 | 9 | 40 | 40 | -- | -- |
S2 70 30 40(7) 60(2) 9 10 | 20 | 20 | 20 | 20 | 40 |
S3 40 8(8) 70 20(10) 18 12 | 20 | 50 | -- | -- | -- |

Demand 5 8 7 14
21 22 10 10
21 -- 10 10
Column -- -- 10 10
Penalty -- -- 10 50
-- -- 40 60
-- -- 40 --

The minimum total transportation cost =19×5+10×2+40×7+60×2+8×8+20×10=779

Here, the number of allocated cells = 6 is equal to m + n - 1 = 3 + 4 - 1 = 6


∴ This solution is non-degenerate

Optimality test using MODI method


Allocation Table is

D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 40 (7) 60 (2) 9
S3 40 8 (8) 70 20 (10) 18

Demand 5 8 7 14

Iteration-1 of optimality test


1. Find ui and vj for all occupied cells(i,j), where cij=ui+vj

1. Substituting, v4=0, we get


2.c14=u1+v4⇒u1=c14-v4⇒u1=10-0⇒u1=10
3.c11=u1+v1⇒v1=c11-u1⇒v1=19-10⇒v1=9
4.c24=u2+v4⇒u2=c24-v4⇒u2=60-0⇒u2=60
5.c23=u2+v3⇒v3=c23-u2⇒v3=40-60⇒v3=-20
6.c34=u3+v4⇒u3=c34-v4⇒u3=20-0⇒u3=20
7.c32=u3+v2⇒v2=c32-u3⇒v2=8-20⇒v2=-12

D1 D2 D3 D4 Supply ui
S1 19 (5) 30 50 10 (2) 7 u1=10
S2 70 30 40 (7) 60 (2) 9 u2=60
S3 40 8 (8) 70 20 (10) 18 u3=20

Demand 5 8 7 14
vj v1=9 v2=-12 v3=-20 v4=0
2. Find dij for all unoccupied cells(i,j), where dij=cij-(ui+vj)

1.d12=c12-(u1+v2)=30-(10-12)=32
2.d13=c13-(u1+v3)=50-(10-20)=60
3.d21=c21-(u2+v1)=70-(60+9)=1
4.d22=c22-(u2+v2)=30-(60-12)=-18
5.d31=c31-(u3+v1)=40-(20+9)=11
6.d33=c33-(u3+v3)=70-(20-20)=70

D1 D2 D3 D4 Supply ui
S1 19 (5) 30 [32] 50 [60] 10 (2) 7 u1=10
S2 70 [1] 30 [-18] 40 (7) 60 (2) 9 u2=60
S3 40 [11] 8 (8) 70 [70] 20 (10) 18 u3=20

Demand 5 8 7 14
vj v1=9 v2=-12 v3=-20 v4=0

3. Now choose the minimum negative value from all dij (opportunity cost) = d22 = [-18]
and draw a closed path from S2D2.
Closed path is S2D2→S2D4→S3D4→S3D2
Closed path and plus/minus sign allocation...

D1 D2 D3 D4 Supply ui
S1 19 (5) 30 [32] 50 [60] 10 (2) 7 u1=10
S2 70 [1] 30 [-18] (+) 40 (7) 60 (2) (-) 9 u2=60
S3 40 [11] 8 (8) (-) 70 [70] 20 (10) (+) 18 u3=20

Demand 5 8 7 14
vj v1=9 v2=-12 v3=-20 v4=0

4. Minimum allocated value among all negative position (-) on closed path = 2
Substract 2 from all (-) and Add it to all (+)

D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 (2) 40 (7) 60 9
S3 40 8 (6) 70 20 (12) 18

Demand 5 8 7 14
5. Repeat the step 1 to 4, until an optimal solution is obtained.

Iteration-2 of optimality test


1. Find ui and vj for all occupied cells(i,j), where cij=ui+vj

1. Substituting, u1=0, we get


2.c11=u1+v1⇒v1=c11-u1⇒v1=19-0⇒v1=19
3.c14=u1+v4⇒v4=c14-u1⇒v4=10-0⇒v4=10
4.c34=u3+v4⇒u3=c34-v4⇒u3=20-10⇒u3=10
5.c32=u3+v2⇒v2=c32-u3⇒v2=8-10⇒v2=-2
6.c22=u2+v2⇒u2=c22-v2⇒u2=30+2⇒u2=32
7.c23=u2+v3⇒v3=c23-u2⇒v3=40-32⇒v3=8

D1 D2 D3 D4 Supply ui
S1 19 (5) 30 50 10 (2) 7 u1=0
S2 70 30 (2) 40 (7) 60 9 u2=32
S3 40 8 (6) 70 20 (12) 18 u3=10

Demand 5 8 7 14
vj v1=19 v2=-2 v3=8 v4=10

2. Find dij for all unoccupied cells(i,j), where dij=cij-(ui+vj)

1.d12=c12-(u1+v2)=30-(0-2)=32
2.d13=c13-(u1+v3)=50-(0+8)=42
3.d21=c21-(u2+v1)=70-(32+19)=19
4.d24=c24-(u2+v4)=60-(32+10)=18
5.d31=c31-(u3+v1)=40-(10+19)=11
6.d33=c33-(u3+v3)=70-(10+8)=52

D1 D2 D3 D4 Supply ui
S1 19 (5) 30 [32] 50 [42] 10 (2) 7 u1=0
S2 70 [19] 30 (2) 40 (7) 60 [18] 9 u2=32
S3 40 [11] 8 (6) 70 [52] 20 (12) 18 u3=10

Demand 5 8 7 14
vj v1=19 v2=-2 v3=8 v4=10

Since all dij≥0.


So final optimal solution is arrived.
D1 D2 D3 D4 Supply
S1 19 (5) 30 50 10 (2) 7
S2 70 30 (2) 40 (7) 60 9
S3 40 8 (6) 70 20 (12) 18

Demand 5 8 7 14

The minimum total transportation cost =19×5+10×2+30×2+40×7+8×6+20×12=743

Assignment Problem
The assignment problem refers to the class of linear programming problems that include
determining the most efficient assignment of people to projects, salespeople to territories,
contracts to bidders, jobs to machines, and etc. The goal of the method is to minimize the total
costs or total time of performing the tasks at hand.
Consider the situation where some objects (service teams, jobs, employees, projects, etc.)
should be assigned on a one-to-one basis to other objects. Every assignment can bring profit or
incur cost. The objective is to maximize the total profit or minimize the total cost ensuing from
the assignment structure.

Example 6
VP Inc. gets four commissions for building family houses in various parts of Prague
(Michle, Prosek, Radlice, Trója). In the first step the company must solve the problem of
excavating the shafts for basements. Each excavation takes 5 days. Management of the company
decided to use four own excavators stored in four separated garages. The objective is to allocate
each excavator to exactly one excavation with minimal cost. Since the costs are derived from
distances [in km] between garages and destinations (Table 2.9), we can concentrate only on
these distances to define the objective.

Destination Michle Prosek Radlice Trója

Garage

Garage 1 5 22 12 18

Garage 2 15 17 6 10

Garage 3 8 25 5 20

Garage 4 10 12 19 12

Tab. 2.9 Distances Between Garages and Destinations


Formulation
As you can see, the structure of assignment problem is rather similar to the structure of
the transportation problem. However, here is one basic difference: all supplies and demands are
equal to 1. Thus, exactly one excavator leaves each garage and each destination needs exactly
one of the machines.

1. Define the decision variables


Whereas in transportation models the variable xij corresponds to the amount
transported from the source i to the destination j, in assignment model this variable acts
as an indicator of assignment of the object i to the object j. This binary variable equals 1
in case of assignment (matching) and 0 otherwise. In the example, xij = 1 if the excavator
from the garage i goes to the destination j, and xij = 0 if the excavator from the garage i
does not go to the destination j. It is quite evident that there are 16 bivalent variables in
the example:
xij = 1 or 0, i, j = 1, 2, 3, 4 .
2. The objective is to minimize the total distance necessary for all excavators’
movement

If parameter cij is a distance between the garage i and the destination j, the
objective function can be expressed as follows:
4 4
Minimize z = ∑∑cij xij ,

i =1 j=1
or using real unit cost:
Minimize z = 5x11 + 22x12 + ... + 12x44 .
The objective function corresponds to the sum of kilometers that all machines go
together.
3. The constraints
From each garage exactly one excavator goes away:
x11 + x12 + x13 + x14 = 1
x 21 + x22 + x23 + x24 = 1
x 31 + x32 + x33 + x34 = 1
x 41 + x42 + x43 + x44 = 1

Each destination needs exactly one excavator:


x11 + x21 + x31 + x41 = 1
x12 + x22 + x32 + x42 = 1
x13 + x23 + x33 + x43 = 1
x14 + x24 + x34 + x44 = 1
Notice that the model is balanced, because the number of garages is equal to the number
of destinations. Unbalanced model would need one or more dummy objects to become the
balanced model.

Optimal Solution
In Table 2.10 the optimal assignment is dark shaded. The excavators go from Garage 1 to
Michle, from Garage 2 to Trója, from Garage 3 to Radlice and from Garage 4 to Prosek. All
machines go together 32 km (5 + 10 + 5 + 12). If the excavators return every day to their
garages, the total daily distance is 64 km and in 5 days all machines would go 320 km.
This value can be a basis for evaluating the total transportation cost.

Destination Michle Prosek Radlice Troja

Garage

Garage 1 5 22 12 18

Garage 2 15 17 6 10

Garage 3 8 25 5 20

Garage 4 10 12 19 12

Tab. 2.10 Optimal Assignment

Hungarian Algorithm
The Hungarian algorithm uses the following theorem for polynomial runtime complexity
and guaranteed optimality:
If a number is added to or subtracted from all of the entries of any one row or column of
a cost matrix, then an optimal assignment for the resulting cost matrix is also an optimal
assignment for the original cost matrix.
By using the Hungarian theorem, we decrease our original weight matrix to contain zeros.
We try to assign tasks to agents such that each agent is doing only one task and the penalty
incurred in each case is zero.
Core of the algorithm (assuming square matrix):

Step 1- For each row of the matrix, find the smallest element and subtract it from every
element in its row.
Step 2- Do the same (as step 1) for all columns.
Step 3- Cover all zeros in the matrix using minimum number of horizontal and vertical lines.
Step 4- Test for Optimality: If the minimum number of covering lines is n, an optimal
assignment is possible and we are finished. Else if lines are lesser than n, we haven’t
found the optimal assignment, and must proceed to step 5.
Step 5- Determine the smallest entry not covered by any line. Subtract this entry from each
uncovered row, and then add it to each covered column. Return to step 3.

ACTIVITIES. Solve the following.

1. Find Solution using Voggel's Approximation method, also find optimal solution using MODI
method

D1 D2 D3 D4 Supply
S1 23 35 60 10 7
S2 80 20 45 50 11
S3 50 9 60 30 18
Demand 4 7 9 24

2. Find Solution using Least Cost method

D1 D2 D3 D4 Supply
S1 19 30 50 10 7
S2 70 30 40 60 9
S3 40 8 70 20 18
Demand 5 8 7 14

3. Mr. Mandy works as a manager for a chip manufacturer, and he has currently 3 people on
the road meeting clients. His salespeople are in Jaipur, Pune and Bangalore, and he want them
to fly to three other cities: Delhi, Mumbai and Kerala. The table below shows the cost of airline
tickets in INR between the cities:

Delhi Kerala Mumbai


Jaipur 2500 4000 3500
Pune 4000 6000 3500
Bangalore 2000 4000 2500

Possible assignment: Cost = 11000 INR

Delhi Kerala Mumbai


Jaipur 2500 4000 3500
Pune 4000 6000 3500
Bangalore 2000 4000 2500

Other Possible assignment: Cost = 9500 INR and this is the best of the 3! possible
assignments.
Delhi Kerala Mumbai
Jaipur 2500 4000 3500
Pune 4000 6000 3500
Bangalore 2000 4000 2500

Below is the cost matrix of example given in above diagram.


2500 4000 3500
4000 6000 3500
2000 4000 2500
References:
Chand, S. (2014, April 12). Assignment Problem in Linear Programming : Introduction and
Assignment Model. Your Article Library.
https://www.yourarticlelibrary.com/ergonomics/operation- research/assignment-
problem- in-linear-programming-introduction-and-assignment-model/34712#:%7E:text=
%3A,profit%20or%20sale%20is%20maximum.

Goyal, R. (2020). Hungarian Algorithm for Assignment Problem | Set 1 (Introduction). Retrieved
from https://www.geeksforgeeks.org/hungarian-algorithm-assignment-problem-set-1-
introduction/

Least Cost Method. Business Jargons. Retrieved from https://businessjargons.com/least-cost-


method.html#:~:text=Definition%3A%20The%20Least%20Cost%20Method,the%20leas
t %20cost%20of%20transportation.

Limbore, N. (2013). Review Of Modified Distribution Method In Current Business. Retrieved


fromhttps://www.researchgate.net/publication/259332000_REVIEW_OF_MODIFIED_DISTRI
B UTION_ METHOD_IN_CURRENT_BUSINESS

Shah, P. N. (2020). Transportation Problem using Least Cost Method Algorithm & Example-1.
AtoZMath. Retreived from https://cbom.atozmath.com/example/CBOM
/Transportation.aspx?he=e&q=lcm

Transportation and Assignment Problems. (2012). Retrieved from


http://www.maths.unp.ac.za/coursework/MATH331/2012/transportation_assignment.pdf

What is Modified Distribution Method. (2016, June 13). Business Jargons. Retrieved from
https://businessjargons.com/modified-distribution-method.html
CHAPTER 4 – NETWORK MODELS

LEARNING OBJECTIVES
At the end of the lessons, students are expected to:
1. formulate network models;
2. analyze and solve shortest path problems;
3. analyze and solve traveling salesperson problems;
4. analyze and solve network problems using minimal spanning tree; and
5. analyze and solve maximum flow problems.

Network Models

A few issues can be described graphically as a network (the set of nodes and arcs). Typical
circumstance is a transportation network: cities (nodes) are connected with one another by roads
(arcs). If we assess the network (in this case we are interested e.g. in distances between all the
cities), the undertaking is often to locate the minimal distance from one city to all other cities.
Some types of networks may be evaluated by capacities rather than distances and afterward the
problem of maximal flow can be solved. In numerous issues, solved with utilization of the network
models, the most significant value is unit cost and the objective is to find the minimal total cost

Some administrative problems can be described graphically as a network. It is a set of


nodes and arcs that diagrams the relationships between objects of the real system. Network models
can represent e.g. a transportation system where nodes are the cities and arcs are the connections
between them (roads). The network expression is also appropriate for description of computer
systems (Local Area Network, World Wide Web, etc.), the company organization and production
process, piping systems, or projects. In the following section the basic terminology and notation
for network problems and some applications will be discussed.
The network model describes conformations of flow in a connected system in which the
flow can include material, people, funds, and so on. These configurations are appropriately
described with flow diagrams that support in the development of effective spreadsheet models.
The opportunity of undertaking some of the model building with a diagram makes network models
a special category of linear programs.

3.1 Network Terminology


As we mentioned earlier, the network is a set of nodes and arcs. The arc is a connector
between a pair of nodes and can be directed (oriented) or undirected. A directed arc specifies
which node is considered as the point of origin. The orientation of the arc is marked with the
arrowhead.

i j i j
Fig. 3.1 Undirected and Directed Arc

If a network contains directed arcs it is called a directed network. Otherwise it is called


an undirected network.
The network is a specific sequence of arcs in which the initial node of each arc is identical
with the terminal node of the preceding arc in the sequence is a path. This sequence must cross
the different nodes. In Figure 3.2 the example of such a path is shown (the sequence across the
nodes 1 – 2 – 4 – 5 – 6).
Fig. 3.2 Path Between Nodes 1 and 6

The path in Figure 3.2 is an open path, since it starts and ends in different nodes. If the
path starts and ends in the same node (closed path) it is termed a circuit (cycle).

Fig. 3.3 Circuit

If there is a path connecting every pair of nodes in the network, it is called connected.
The network shown above is connected. If we removed the arcs (1, 2) and (1, 3), there would
no longer have the connected network.
The connected network without any circuit is called a tree. If we denote the total number
of nodes in the network as n it is evident that the tree involving all nodes must contain exactly
(n – 1) arcs. If we add any arc to the tree, the circuit will appear.
A spanning tree is a tree including all the nodes from the original network. The tree
shown in Figure 3.4 is an example of spanning tree. Removing of any arc from the spanning
tree changes the tree into unconnected network.

Fig. 3.4 Spanning Tree

In real situation the network is evaluated. The values can be added to nodes or/and to
arcs and can represent time, distance, cost, capacity, etc. In the following section, several
examples of network problems are described.
3.2 Basic Network Applications
Before describing the frequent applications of network theory - project management we
mention some simple circumstances where the network representation is effectively utilized.
However, it is not possible to describe all the methods used for solving problems and we will
concentrate just on some simpler methods.

3.2.1 Shortest Path Problem


The problems in this category concern situations where we have to locate the shortest
path from an origin to a destination. Typically, there is no direct connection between these two
points and the path crossing numerous other points must be found. A network represents the
possible connections (arcs) between all points (nodes). Distance between the pair of connected
nodes is attributed to each arc.
The shortest-path problem is a specific network model that has gotten a lot of consideration
for both practical and theoretical reasons. The essence of the problem can be specified as follows:
Given a network with distance ci j (or travel time, or cost, etc.) related with each arc, discover a
path through the network from a particular origin (source) to a particular destination (sink) that
has the shortest total distance. The simplicity of the statement of the problem is somewhat
confusing, since a number of significant applications can be formulated as shortest- (or longest-)
path problems where this formulation is not obvious at the outset. These include problems of
equipment replacement, capital investment, project scheduling, and inventory planning. The
theoretical interest in the problem is due to the fact that it has a special structure. Moreover to
being a network, that results in very efficient solution procedures. Further, the shortest-path
problem often occurs as a subproblem in more complex situations, such as the subproblems in
applying decomposition to traffic-assignment problems or the group-theory problems that arise in
integer programming

Example 3.1
In Figure 3.5 there is an example with 6 cities. Direct distances between the cities are in
kilometers. The goal is to run from the city 1 to the city 6 through shortest path. It is easy to locate
the path across the cities 2 and 3. The length of the shortest path is 40 km.

Fig. 3.5 Shortest Path Between Nodes 1 and 6

The issue can be conveyed as finding the shortest paths not only from one origin to one
destination, but from each node to all the other nodes (Table 3.1). Because the network is
undirected the matrix of shortest paths is symmetric (e.g. the length of the path from 1 to 6
equals the length of the path from 6 to 1).
To 1 2 3 4 5 6
From

1 0 14 24 26 32 40

2 14 0 10 12 18 26

3 24 10 0 15 28 16

4 26 12 15 0 23 15

5 32 18 28 23 0 30

6 40 26 16 15 30 0

Tab. 3.1 Shortest Paths Between All Pairs of Nodes

3.2.2 Traveling Salesperson Problem (TSP).


The travelling salesman problem asks the question: "With a list of cities and the distances
between each pair of cities, what is the shortest possible route that visits each city and returns to
the origin city?" It is a problem in combinatorial optimization, important in theoretical computer
science and operations research.

Example 3.2
A salesperson has to visit a specified group of cities and come back to the origin (home
city). This tour should be as short as possible in terms of the total distance. In Figure 3.6 such a
tour (circuit) is displayed (it crosses the nodes 1, 2, 5, 4, 6, 3, 1). The optimal total distance is
111km.

Fig. 3.6 Traveling Salesperson Problem


If we tabulate the direct distances between nodes, we can illustrate the similarity to the
assignment problem, in which the objective is to mark 6 cells respecting the following rule:
exactly one cell in each row and in each column is marked. Table 3.2 shows the optimal solution
of TSP (dashes are used for unidentified connections).

To 1 2 3 4 5 6
From
1 - 14 25 - - -
2 14 - 10 12 18 -
3 25 10 - 15 - 16
4 - 12 15 - 23 15
5 - 18 - 23 - 30
6 - - 16 15 30 -

Tab. 3.2 Traveling Salesperson Problem


If the undertaking had been answered as the classical assignment problem numerous
solutions would have been found, infeasible in case of traveling salesperson problem. One of
those solutions is strongly shaded in Table 3.3. When we graph this solution (Figure 3.7), we
can see two sub-tours 1-2-3-1 and 4-5-6-4. Whereas it is feasible solution of assignment
problem, it is infeasible in traveling salesperson problem. Hence in formulation of the model,
it is indispensable to add constraints that prevent building sub-tours.

To 1 2 3 4 5 6

From

1 - 14 25 - - -

2 14 - 10 12 18 -

3 25 10 - 15 - 16

4 - 12 15 - 23 15

5 - 18 - 23 - 30

6 - - 16 15 30 -

Tab. 3.3 Assignment Problem

Fig. 3.7 Infeasible Sub-Tours in TSP

3.2.3 Minimal Spanning Tree


In some circumstances, instead of finding the shortest path or the shortest tour, the
objective is to guarantee a connection between all nodes in the network. Assuming n nodes in the
network, a spanning tree is such sub-network that comprises exactly (n-1) arcs and no circuits. If
we have the evaluated network (e.g. by distances), the minimal spanning tree is a spanning tree
with the minimal sum of values.
A minimum spanning tree is a subset of the edges of a connected, edge-weighted
undirected graph that links all the vertices together, without any cycles and with the minimum
possible total edge weight. It is a spanning tree whose sum of edge weights is as small as possible.
Example 3.3
We illustrate one of the possible algorithms on this elementary example. Assume that the
managerial problem is to connect 9 locations of an exhibition area with the source of electricity
power. The goal is to minimize the cost of all the extensions. The direct distances (in meters)
between locations can be found in Figure 3.8. The node 1 is the source of power. The price per 1
meter of a cable is 10 CZK.

Fig. 3.8 Electricity Power Distribution in the Exhibition Area

Solution
1. In the first step we find two arcs with minimal distances. These are the arcs (8, 10) and
(3, 5) with direct distances 35 and 40 meters (Figure 3.9).

Fig. 3.9 First Step – Finding Two Arcs with Minimal Lengths
2. In the second step we need to find for another arc (from the set of all the
remaining arcs in the network) with the minimal distance. This arc is added into the set of
the arcs selected before. We must note the relevant rule that no circuit can happen after
adding a new arc. If the circuit appeared we would ignore this arc in the searching process
Fig. 3.10 Second Step – Adding a New Arc

3. We repeat the second step until a minimal spanning tree is found.


Figure 3.11 illustrates the optimal solution (minimal spanning tree). The total length of
the cable necessary for all the extensions is 490 meters. Thus, total cost of distribution of
electricity power in the exhibition area is 4900 CZK.

Fig. 3.11 Optimal Solution – Minimal Spanning Tree

3.2.4 Maximum Flow Problem


In some network design applications the concern is to send as much flow as possible
through the network, rather than to send flow at lowest cost. This alternative is readily handled
by dropping the balance constraints at the origins and destinations of flow, while substituting
an objective that stands for total flow in some sense.

To find a maximal quantity of gas, fluid, traffic, information, people and so on that can
be transported through a capacitated network is the objective. This network comprises two
relevant nodes – a starting node called the source and a terminal node called the sink. However
in directed networks the movement must note the orientation of the arcs (movement in the
reverse direction is restricted), in undirected networks movement in both directions is allowed.
The maximum flow problems include finding a feasible flow through a flow network that
acquires the maximum possible flow rate. The maximum flow problem can be seen as a special
case of more complex network flow problems, such as the circulation problem.
There are five relevant rules that must be considered during the construction of the
mathematical model:
1. The quantity flowing through an arc must be less than or equal to the capacity of that
arc.
2. For every node, except the source and the sink, the quantity flowing out of a node is
equal to the quantity flowing into it.
3. Total quantity flowing into the source is zero.
4. Total quantity flowing out of the sink is zero.
5. Total quantities flowing out of the source and into the sink are equal.

Example 3.4
The board of White Lake City, which is located on the edge of a small lake, solves the
problem of minimizing disruptive effects of possible flood. One of solutions is to rebuild the
system of drains that would pump the water out of the lake into great reservoir. Since there are
two alternatives of a drain system, the city board must decide which of them will be used.

Fig. 3.12 Drain System — “Northern Channel”

Fig. 3.13 Drain System— “Southern Channel”

The objective is to maximize the quantity of water being pumped in one hour. In
both networks (Figures 3.12 and 3.13) the capacities of all the drains (in cubic meters per
hour) are known.

Optimal Solution
If the city used the first alternative “Northern Channel”, it would allow to pump
maximally 1930 cubic meters per hour. Figure 3.14 displays for each drain the quantity
of water and the direction in which the water will be flowing. Capacity of some drains is
not saturated (e.g. (1, 2) or (6, 7)); some drains are not used at all ((2, 3), (3, 4) and (7,
8)).

Note: A maximum flow problem has usually multiple optimal solutions. Using
other algorithms could lead to another solution with the maximal flow 1930 cubic meters
per hour, however with a different structure of individual flows.

In Figure 3.15 you can find the optimal solution for the second alternative
“Southern Channel”. Since the maximum flow equals 2270 cubic meters per hour, the
city board should prefer this alternative to protect the White Lake City from the possible
flood.

ACTIVITIES

1. Ships are available at three ports of origin and need to be sent to four ports of destination.
The number of ships available at each origin, the number required at each destination, and
the sailing times are given in the tableau below.

Our objective is to minimize the total number of sailing days.


a) Find an initial basic feasible solution.
b) Show that your initial basis is a spanning tree.
c) Find an initial basic feasible solution using the other two methods presented in the text.
Solve completely, starting from the three initial solutions found in parts (a) and (c).
Compare how close these solutions were to the optimal one.
d) Which of the dual variables may be chosen arbitrarily, and why?
e) Give an economic interpretation of the optimal simplex multipliers associated with the
origins and destinations.

2. A distributing company has two major customers and three supply sources. The
corresponding unit from each supply center to each customer is given in the following
table, together with the total customer requirements and supply availabilities.

Customer Available Supply center 1 2 supplies 1 −1 3 300 2 1 6 400 3 1 5 900 Customer


requirements 800 500 Note that Customer 1 has strong preferences for Supplier 1 AND
will be willing not only to absorb all the transportation costs but also to pay a premium
price of $1 per unit of product coming from Supplier 1.
a) The top management of the distributing company feels it is obvious that Supply Center
1 should send all its available products to Customer 1. Is this necessarily so? (Hint.
Obtain the least-cost solution to the problem. Explore whether alternative optimal
solutions exist where not all the 300 units available in Supply Center 1 are assigned to
Customer 1.)
b) Assume Customer 2 is located in an area where all shipments will be subject to taxes
defined as a percentage of the unit cost of a product. Will this tax affect the optimal
solution of part (a)?
c) Ignore part (b). What will be the optimal solution to the original problem if Supply
Center 1 increases its product availability from 300 units to 400 units?

3. Consider the following transshipment problem where goods are shipped from two plants
to either a warehouse or two distribution centers and then on to the two end markets.

The production rates in units per month are 250 and 450 for plants 1 and 2, respectively.
The demands of the two markets occur at rates 200 and 500 units per month. The costs of
shipping one unit over a particular route is shown adjacent to the appropriate arc in the
network.
a) Redraw the above transshipment network as a network consisting of only origins
and destinations, by replacing all intermediate nodes by two nodes, an origin and
destination, connected by an arc, from the destination back to the origin, whose
cost is zero.
b) The network in (a) is a transportation problem except that a backwards arc, with
flow xii, from newly created destination i to its corresponding newly created
source, is required. Convert this to a transportation network by substituting x 0 ii =
B − xii. How do you choose a value for the constant B?
c) Certain arcs are inadmissible in the original transshipment formulation; how can
these be handled in the reformulated transportation problem?
d) Interpret the linear-programming tableau of the original transshipment network and
that of the reformulated transportation network.
e) Can any transshipment problem be transformed into an equivalent transportation
problem?
References
Baker, K. (2011). Linear Programming: Network Models. Retrieved from
https://onlinelibrary.wiley.com/doi/10.1002/9780470949108.ch3#:~:text=The%20networ
k%20model%20describes%20configurations,%2C%20funds%2C%20and%20so%20on.
&text=The%20possibility%20of%20doing%20some,special%20category%20of%20linea
r%20programs.
Fabry, J. (2003). Management Science. University of Economics Prague.
Fourer, R. (2003). Network Linear Programs. Retrieved from
https://ampl.com/BOOK/CHAPTERS/18-network.pdf
Network Models. Retrieved from http://web.mit.edu/15.053/www/AMP-Chapter-08.pdf
CHAPTER 5 – PROJECT MANAGEMENT

LEARNING OBJECTIVES

At the end of the lessons, students are expected to:

1. define and explain the basic concepts in project management;


2. analyze and solve project management problems using critical path method (CPM); and
3. analyze and solve project management problems using program evaluation review
technique (PERT).

Project Management

The managers are responsible for planning, scheduling and controlling undertakings that
comprise of many separate jobs or assignments performed by a variety of departments or
individuals in several circumstances. An execution of each job takes specific time. There are
two basic methods for solving those problems: CPM (Critical Path Method) and PERT
(Program Evaluation Review Technique). The two methods require the network representation
of the problem and both are developed independently but are now often used interchangeably.

Project Management comprises of a set of techniques that helps management accomplish


large-scale projects; projects that usually involve coordination of numerous jobs throughout the
organization.

Example 1
Figure 3.16 illustrates an example with two activities — designing a new model of car
and making its prototype. Event 1 is the start point of Activity A. Whereas event 2 can be
considered as the finish point of Activity A and simultaneously as the start point of
Activity B. Meanwhile event 3 is the finish point of Activity B. The sequence of both activities
is manifested: the prototype cannot be made until the design of the car is finished (Activity A
immediately precedes Activity B, or Activity B immediately succeeds Activity A).

In adding a new activity (Computer Presentation), we must specify its predecessors. The
public presentation of the car can start only after the finishing the designing works. Nevertheless,
this activity is independent of creating the prototype. The network representing the existing
activities is shown in Figure 3.17. After the car has been designed, both making the prototype
and computer presentation may start.
The last two activities that was added in the illustrative example are Prototype
Presentation and Prototype Test. Presentation of the prototype must succeed creating the
prototype and should follow the computer presentation. However, test of the prototype does not
essentially succeed computer presentation of the new model and it depends only on
finishing the prototype. Introducing a dummy activity into the network is the only way how to
express this requirement (Figure 3.18). This activity assures that the undertakings and events are
in proper sequence. From the figure, Activity D follows both Activity B and Activity C, however,
Activity E succeeds only Activity B.

Fig. 3.18 Dummy Activity

Although the explanation of project can be easily understood, the structure of


representative network could be rather difficult. The correct graphical expression
of all interrelated activities is hence the key step in project management. The sequence of
all activities can be concisely described in the following table:

Activity Activity Immediate


Description Predecessors
A Design a Car None
B Make a Prototype A
C Computer Presentation A
D Prototype Presentation B, C
E Prototype Test B

Tab. 3.4 New Car Project

The final step must be checking the constructed network according to the table.
Extra relationships are prohibited and no sequence can be lost.
In the following sections, time-oriented situations where the key issue of the project
is its duration and scheduling of all the activities will be discussed. For this purpose, two
basic techniques were developed: Critical Path Method (CPM) and Program Evaluation
Review Technique (PERT). The three important phases of project management are the
following:

1. Planning. In planning, the entire project is decomposed into individual activities. It is


needed to know the duration of each activity and its interrelation with other
activities as it was shown in Example 3.5. The outcome of the planning phase is the
constructed network as the graphical representation of the real project.
2. Scheduling. In scheduling, to propose, using CPM or PERT, start and finish times for
each activity as well as the duration of the entire project is the objective of this phase.
Some activities need a attention since their possible delay causes delay of the whole
project. These activities are called critical and their sequence makes a critical path. Non-
critical activities might be delayed; however an upper bound of possible delay must be
computed.

3. Controlling. This is the final phase in project management. Managers should combine
the network (as the result of the planning phase) and the time chart (as the result of the
scheduling phase) to make periodic progress reports. Real performance of the project
according to the proposed schedule needs everyday control and analysis. The network
can be updated throughout time and a new schedule for the remaining part of the project
should be then recalculated. Controlling phase is highly substantial especially in case of
large and long-term projects.

Critical Path Method

Projects are made up of tasks that have to follow to a schedule to meet a deadline. Without
mapping the work, it can quickly get out of hand and you’ll find your project off track. There are
numerous techniques that project managers use to schedule their project activities in order that
everything lands where it is supposed to and the project fits the approved timeline. The critical
path method is one of these.

The Critical Path Method is defined in the Project Management Body of Knowledge
(PMBOK) as the sequence of scheduled activities that determines the duration of the project. It
is the longest path (in time) from Start to Finish that indicates the minimum time necessary to
complete the entire project. The objective of the problem is determination of the following items:

Moreover, this method is fundamentally an algorithm for decision making. This algorithm
takes a task's start time, its duration, and finish time to figure out which activities deserve the
most attention (i.e. are "critical" for the project).

These scheduled activities must be performed if the project is to be considered a success.


Moreover, they must be completed in a specific order. If you’re building a house, you can’t
construct the walls and then dig the foundation; you have to do it in a sequence.
1. Finish time of the project.
2. Start and finish times of each activity.
3. A critical path consisting of critical activities.
4. Non-critical activities and their possible delay.
The method is explained on the following example.

Example 2
Project manager of direct marketing company Music DM, Inc. has received the direction
to prepare the Christmas compilation of Czech carols. The offer should be sent to the address of
company’s customers, who are interested in carols. Because of this, a thorough analysis should
be carried out to select the best customers for the promotion. The project manager (PM) must
consider the following activities to realize the project.

1. Songs Selection. In this first step the Project Manager examines the questionnaires in
which a sample of customers was asked about their interests. In assistance of an external
music expert and experts on royalties the PM arranges a list of songs that will be gathered
and describes medium sources of these songs.

2. Mastering. After the songs selection the CD can be mastered.

3. Promotion Material Elaborating. Promotion material is the suggestion being sent to


each customer. It comprises of a personal letter, the product presentation brochure (thus,
songs selection must be a predecessor) and the order form. After explaining the promotion
material, the correction process must be followed. In the example, these two steps are
combined into one activity.

4. Customers’ Analysis. This is a vital step of the marketing process, since the accurate
product must be offered to the right people. Detailed analysis of customers’ interests
according to their previous behavior or their reactions on accomplished tests should lead
to success of the project. As the result of the analysis, the Project Manager acquires the
criteria for customers’ selection and the estimated number of customers to be promoted.

5. Promotion Material Production. After promotion material is elaborated and the


number of letters is valued, the PM can order the production of all promotion
materials.

6. Make CD Copies. Number of CD Copies is valued according to the customers’


analysis. Since only a segment of all promoted customers will be interested in the
offered product, the PM makes a decision about the quantity of CDs to be created.

7. Customers Selection. After the selection criteria are defined, the final selection of
the best customers for the successful promotion can be advanced.

8. Data to Printing House. All selected customers (their name and address) are
directed to printing house.

9. Promotion Material to Printing House. Produced promotion material is directed


to printing house where the individual letters will be printed.

10. Laser Print. Selected names and addresses are printed on the personal letters in the
printing house. To simplify the example, all the materials, devoted to be sent to the
customers must be completed in the printing house. In practice these jobs carry out
in different places.

11. Mailing. After all the promotion materials are put into envelopes, they are
disseminated to local post offices and then sent directly to final addressees.

All the envelopes have to be delivered until Thursday, December 12. The described
activities together with their duration (in working days) and their immediate predecessors
can be found in the following table.

Activity Activity Duration Immediate


Description Predecessors
A Songs Selection 15 None
B Mastering 8 A
C Promotion Material Elaborating 6 A
D Customers Analysis 7 A
E Promotion Material Production 4 C, D
F Promotion Material to Printing House 5 E
G Customers Selection 3 D
H Make CD Copies 12 B, D
I Data to Printing House 3 G
I Laser Print 9 F, I
K Mailing 8 H, J

Tab. 3.5 Christmas Compilation Project

The Project Manager must now set the starting date of the project such that it will
be finished exactly on the specified mail date (December 12).

Solution
1. Construct the network
In the first step we have to create a network that graphically signifies the project
(Figure 3.19). Two dummy (fictitious) activities 𝐷1 , 𝐷2 have been added to guarantee
accurate interrelations between all real activities. Their durations are zeros.

Fig. 3.19 Direct Marketing Project

While constructing the network, we should respect the following rules:


• A network must have one start node and one finish node.
• Each activity must be represented just by one arc.
• Two nodes are connected maximally by one arc.
• The node representing the completion of an activity has higher number than the
node representing the start of this activity — the rule prevents circuits in the
network.
2. Event analysis
For this purpose, we must introduce some new terms:

The earliest event time for a node is the earliest time at which all the preceding
activities have been completed. The earliest event time for completing a project
is the earliest time of the last event.
This definition corresponds to the rule that all activities starting in a node can be
realized after all predecessors have been completed.

The latest event time for a node is the latest time at which the event can occur
without delaying the determined completion time of the project. The latest event
time for completing a project can be set in the planning process or can be
considered as the earliest event time for completing a project.

All the earliest event times and the latest event times are computed in two phases:
forward pass and backward pass.

Forward pass
We calculate the earliest event times for all the nodes. The computation runs
through the nodes according to their numbers (in ascending order), from the start of the
project to its finish. For better understanding this phase, one needs to compute all the times
directly in the network. Because of this, we divide each node into three parts (Figure 3.20).

Fig. 3.20 Node Definition for Using CPM

In the top part of the node there is its number. Figure 3.20 depicts a general node
with the number i. Bottom part of the node is divided into the left and the right part.
Whereas in the left part we will compute the earliest event time for the node (𝐸𝑇𝑖 ), in the
right part we get the latest event time (𝐿𝑇𝑖 ).

Hence, in the first phase we must calculate 𝐸𝑇𝑖 for each node in the network. Figure
3.21 shows the forward pass for the Example 3.6. We explain the process of computation
in the following steps:

1. Set 𝐸𝑇𝑖 = 0. This is the start time of the activity A (1,2) and of the project itself.

2. The earliest event time 𝐸𝑇2 determines the possible start of all the activities
coming out from the node 2. All of them — D (2,3), B (2,4), C (2,5) — can start
after all previous activities had been finished (in this case it is the only activity
A (1,2)). Earliest time is 0 + 15 = 15. Thus, 𝐸𝑇2 — 15.

3. 𝐸𝑇3 — 15 + 7 = 22. The earliest event time of the node 3 is computed similarly
to the node 2.
4. Since the node 4 represents the finish node of two activities—B (2,4) and D (3,4)-

the earliest event time 𝐸𝑇4 must be computed as the maximum of possible finish

times of these activities: 𝐸𝑇4 =max (15 + 8, 22 + 0) = 23.

This rule can be generally expressed as follows:

𝐸𝑇𝑗 = max(𝐸𝑇𝑖 + 𝑡𝑖𝑗 ) , (3.1)

where
𝐸𝑇𝑗 . . . earliest event time of the node y that is the finish node of the
activity (i,j),
𝐸𝑇𝑖 . . . earliest event time of the node i that is the start node of the activity (i,j),

𝑡𝑖𝑗 . . . duration of the activity (i,j).

5. 𝐸𝑇5 = max (22 + 0, 15 + 6) = 22.


6. 𝐸𝑇6 = 22 + 4 = 26.

7. 𝐸𝑇7 = 22 + 3 = 25.

8. 𝐸𝑇8 = max (26 + 5, 25 + 3) = 31.


9. 𝐸𝑇9 = max (23 + 12, 31 + 9) = 40.
10. 𝐸𝑇10 = 40 + 8 = 48. This is the earliest event time for completing the project. It is
evident that the entire project cannot be completed earlier than in 48 days. We
denote this time as T.

Fig. 3.21 CPM— Forward Pass

Note: The earliest event time for completing the project ETi is determined by the longest
path leading from the node 1 to the node 10.

Backward pass

This phase of CPM involves calculating the latest event times for all the nodes. However
the forward pass crosses the nodes in ascending order according to their numbers, backward pass
goes from the finish of the project to its start in the opposite (descending) order. Backward pass is
shown in Figure 3.22.

1. First, it is essential to set the latest event time for finishing the project. There are two
possibilities:
a. We set LT10 = ET10 . The latest event time of completing the project equals to its
earliest event time.

b. LT10 = TPL > ET10. A duration of the entire project can be planned by the
management of the company to TPL.

i. In the example, we select the first possibility LT10 = ET10 = T. The value
of T corresponds to the shortest duration of the project.

2. The latest event time LT9 determines the latest possible start of the activity K (9,10) without
delaying the project. Since the activity duration is 8 days and the project must be completed
in 48 days, LT9 = 48 – 8 = 40.

3. Likewise, for the node 8, the latest event time is computed as LT8 = 40 – 9 = 31. It follows
from the simple rule: if the project should not be delayed, all successive latest event times
(computed before) must be satisfied. If the project should be completed in 48 days, event
number 9 should happen in 40 days and hence, the activity J (8,9) must start at the latest in
31 days.
4. LT7=31–3=28.
5. LT6=31–5=26.
6. LT5=26–4=22.
7. LT4=40–12=28.
8. Computing the latest event time of the node 3 is not as easy as the previous
computations. LT3 depends not only on one successive latest event times, but even on
three of them (those events correspond to the nodes 4, 5 and 7). Therefore the latest event
time must be computed according to the following formula:

LTi = min(LT j − tij ) , (3.2)

where

LTi … latest event time of the node i that is the start node of the activity (i, j),

LTj … latest event time of the node j that is the finish node of the activity (i, j),

tij … duration of the activity (i, j).


Using the formula for the node 3, we get LT3 = min (28 – 0, 22 – 0,

28–3)=22.

If we did not respect this rule and we computed, e.g. LT3 = 28 – 3 = 25, it would not
be possible to meet LT5 = 22.

9. Similarly, LT2 = min (22 – 7, 28 – 8, 22 – 6) = 15.


10. LT1=15–15=0.

Fig. 3.22 CPM – Backward Pass

3. Activity analysis

After computing the earliest event times and the latest event times for all the nodes, we
must approach the last step of the project analysis – determination of critical and noncritical
activities. Critical activities are such activities, the delay of which causes delay in completion of
the entire project. On the other hand, each noncritical activity can be delayed. Nevertheless, its
delay is limited. For this purpose, the total float (also called the slack) for each activity must be
calculated:

TFij = (LT j − ETi − tij ) , (3.3)

where

TFij … total float of the activity (i, j),

LTj … latest event time of the node j that is the finish node of the activity (i,j),

ETi … earliest event time of the node i that is the start node of the activity (i, j),

tij … duration of the activity (i, j).


Total floats of critical activities are zeroes.

Total floats of noncritical activities are positive.

Note: If the planned duration of the project is considered in the backward pass as TPL > T, total
floats of all the critical activities are equal to (TPL – T).

A sequence of critical activities determines a critical path (this term is also used in the
name of CPM – critical path method). The critical path is the longest path leading from the start
of a project to its finish. In many projects multiple critical paths exist. As mentioned above, the
determination of the critical path needs the computation of the total float for each activity. In
Figure 3.23, these numbers are calculated in brackets. The critical activities (with zero total
floats) are highlighted.

Fig. 3.23 CPM – Total Floats and Critical Activities

There is the only one critical path in the project consisting of following 6 critical activities
(as the critical activity D1 is dummy, it is not in the list):

Activity Description of Activity Earliest Earliest Latest

Start Time Start Date Finish Date

A Songs Selection 0 Oct 4 Oct 24

D Customers Analysis 15 Oct 25 Nov 5

E Promotion Material Production 22 Nov 6 Nov 11

F Promotion Material to Printing House 26 Nov 12 Nov 18


J Laser Print 31 Nov 19 Nov 29

K Mailing 40 Dec 2 Dec 11

Tab. 3.6 Critical Activities

There is no critical activities that can be delayed if the project should be completed in T
= 48 days. As the management decided to deliver all the envelopes until December 12, the
project has to be started on October 4 (48 working days ago). It is obvious that starting the
project on October 3 causes increasing in all the total floats by 1 day (the planned project
duration TPL = 49 days).

The positive total floats determine noncritical activities (the dummy activity D2 is not in
the list):

Activity Description of Activity Earliest Earliest Latest Total

Start Time Start Date Finish Date Float

B Mastering 15 Oct 25 Nov 13 5

C Promotion Material Elaborating 15 Oct 25 Nov 5 1

G Customers Selection 22 Nov 6 Nov 13 3

H Make CD Copies 23 Nov 7 Nov 29 5

I Data to Printing House 25 Nov 11 Nov 18 3

Tab. 3.7 Noncritical Activities

Total floats give three possibilities of delay of each noncritical activity:

1. A start of the activity can be postponed.


2. Duration of the activity can be extended (the activity can be even interrupted).
3. The combination of 1 and 2.
All postponements and extensions, of course, must respect the total floats.
PERT - Program Evaluation Review Technique

While in CPM only one estimate of each activity’s duration is deliberated, PERT
(Program Evaluation Review Technique) uses three estimates to form a weighted average of
the expected completion time:

aij = optimistic estimate of duration – the shortest possible time, in which the activity (i,
j) can be accomplished (if execution goes extremely well),

bij = pessimistic estimate of duration – the longest time that the activity (i, j) could take
(when everything goes wrong), mij = most likely estimate of duration – the time that would occur
most often (if the activity (i, j) were repeated under exactly the same conditions many times), or
the time estimated by experts.

Fig. 3.24 Activity Time β-distribution

If we consider the β probabilistic distribution of activity time (Figure 3.24), the expected
completion time of the activity (i, j) can be then computed as a weighted average of the three
time estimates using the following formula:

a
µij = ij + 4mij + bij (3.4)

Furthermore it is possible to calculate the standard deviation for each activity:

σ ij = bij − aij (3.5)

6
The standard CPM is used (tij = µij) to determinate a critical path after calculating the
expected completion time µij for all the activities. While in a deterministic model the value of T
(the total duration of the project) can be computed as the sum of durations of the activities on
the critical path (CP), in a probabilistic model we use their expected completion times for
calculating the expected duration of the project:

M= ∑µij (3.6)

(i, j)∈CP

Likewise, the variance of the project duration can be computed as follows:

σ2= ∑σ ij2 (3.7)

(i, j)∈CP

Real duration of the project is a continuous random variable with the mean M and the
variance σ2 (or the standard deviation σ). It is possible to show (on the base of central limit
theorem) that this random variable tends to be normally distributed. Then, two simple
managerial questions can be answered:

1. What is the probability of project completion within a desired time TD?


The normal distribution N (M, σ) can be transformed into the standard normal
distribution N (0, 1) using the following formula:

TD −M

z= . (3.8)

σ
For the purpose of the required probability we just look up the value of the standard
normal distribution corresponding to the value z.

2. What is the completion time TD in which the project will be finished with
a desired probability p?
We look up the value zp corresponding to the probability p and then we compute the
required completion time TD as follows:
TD = M + z p σ . (3.9)

Note: The table of the standard normal distribution values can be found in Appendix.

Example 3

To show PERT, we change Example 3.6 by introducing three duration estimates for each
activity (Table 3.8).

Activity Description of Activity aij mij bij µij Immediate

Predecessors

A Songs Selection 11 15 19 15 None

B Mastering 7 8 9 8 A

C Promotion Material Elaborating 5 6 7 6 A

D Customers Analysis 5 7 9 7 A

E Promotion Material Production 2 3 10 4 C, D

F Promotion Material to Printing House 3 4 11 5 E

G Customers Selection 2 3 4 3 D

H Make CD Copies 8 11 20 12 B, D

I Data to Printing House 2 3 4 3 G

J Laser Print 6 8 16 9 F, I

K Mailing 6 8 10 8 H, J

Tab. 3.8 Christmas Compilation Project – Three Estimations of Activities Duration


Solution

1. Construct the network


The considered change, of course, does not affect the network representing the original
project in terms of the interrelations between all the activities (Figure 3.19).
2. Determine the expected completion time for each activity
All the expected completion times are computed using Equation 3.4. For example, the
weighted average for the activity A (1,2) is:

µ12 = 11+ 4(15) +19 = 15.

Likewise, we must determine these numbers for all the other activities. The estimations
for each activity in Table 3.8 were intentionally set so as the computed values µij equal to the
duration values specified in Table 3.5, which allows us to compare the results (CPM and PERT).

3. Event analysis
This phase is identical with the calculation in Example 3.6. We use µij instead of tij in
each step. While Figure 3.22 displays the results of the computation, it is essential to consider
probabilistic feature of the project. All the computed numbers (the earliest and the latest event
times) are actually expected values (averages) of the associated random variables. The expected
duration of the project is M = 48 days.

4. Activity analysis
Using Equation 3.3 it is likely to compute total float for each activity, or somewhat its
expected value. The activities with zero total floats are critical. The only one critical path
(identical with the critical path in Example 3.6) consists of the following activities:
Activity Description of Activity aij mij bij µij σij

A Songs Selection 11 15 19 15 8/6

D Customers Analysis 5 7 9 7 4/6

E Promotion Material Production 2 3 10 4 8/6

F Promotion Material to Printing House 3 4 11 5 8/6

J Laser Print 6 8 16 9 10/6

K Mailing 6 8 10 8 4/6

Tab. 3.9 Critical Activities in PERT


Standard deviations σij in Table 3.9 are computed using Equation 3.5 (the values are
expressed in fractions).

Using formulas (3.6) and (3.7) we get the expected duration and the variance (standard
deviation) of duration of the project:

M =15+7+4+5+9+8= 48,

σ 2 = (8/6)2 + (4/6)2 + (8/6)2 + (8/6)2 + (10/6)2 + (4/6)2 = 9,

σ = 3.

As we mentioned above, a real duration of project is the random variable with


(approximately) the normal distribution N (M, σ). Hence, in the example, it is the distribution
N (48, 3).

Fig. 3.25 Normal Distribution of the Project Completion Time

Let us suppose that the management of the company wants to find the probabilities of the
project completion within 45 days (i.e. until December 9) and within 54 days (i.e. until
December 20). First we calculate transformed values of standard normal variables:

z1 = 45 − 48 = −1 ,

z2 = 54−48 =2.
3
In the table of the standard normal distribution N (0, 1) we look up the distribution
values corresponding to the required probabilities:

p1 = 0.1587 ,

p 2 = 0.9772 .

Thus, the probability with which the project will have been completed by December 9 is
0.1587. The probability of finishing the project by December 20 is 0.9772.

As mentioned above, the management may solve the opposite question: to discover the
date by which the project will have been finished, e.g. with a reliability of 95%. For this purpose,
we must look up the value zp corresponding to the probability p = 0.95. The value found in the
table is zp = 1.645. Introducing this value into Equation 3.9 we get:

TD = 48 +1.645 (3) = 52.935.

If we round the result, we can take 53 days as the duration of the project corresponding
to the desired probability. Hence, the project will have been finished (with the reliability of
95%) by December 19.

ACTIVITIES

Solve the critical path analysis using critical path method technique.
The Project Manager must now set the starting date of the project such that it will
be finished exactly on the specified mail date (January 15). Construct the network
and event analysis.

Activity Activity Duration Immediate


Description Predecessors
A Songs Selection 17 None
B Mastering 9 A
C Promotion Material Elaborating 4 A
D Customers Analysis 5 A
E Promotion Material Production 7 C, D
F Promotion Material to Printing House 10 E
G Customers Selection 6 D
H Make CD Copies 20 B, D
I Data to Printing House 8 G
I Laser Print 9 F, I
K Mailing 12 H, J

All the envelopes have to be delivered until Thursday, January 15. The
described activities together with their duration (in working days) and their
immediate predecessors can be found in the following table.

Solve the PERT with the given table and answer the following question.

Activity Description of Activity aij mij bij µij Immediate

Predecessors

A Songs Selection 12 16 19 15 None

B Mastering 9 5 10 8 A

C Promotion Material Elaborating 7 9 8 6 A

D Customers Analysis 9 6 7 7 A

E Promotion Material Production 3 6 13 4 C, D

F Promotion Material to Printing House 6 9 12 5 E

G Customers Selection 8 2 7 3 D

H Make CD Copies 0 8 23 12 B, D
I Data to Printing House 5 2 6 3 G

J Laser Print 6 9 16 9 F, I

K Mailing 8 8 10 8 H, J

a. What is the probability of project completion within a desired time TD?


b. What is the completion time TD in which the project will be finished with
a desired probability p?

References

Cohen, E. (2018). How to Use the Critical Path Method for Complete Beginners. Retrieved from
https://www.workamajig.com/blog/critical-path-method

Fabry, J. (2003). Management Science. University of Economics Prague.

Larsen, J. Modelling with linear programming. Retrieved from


http://www2.imm.dtu.dk/courses/02737/lpModelling.pdf

Levy, F. K. The ABCs of the Critical Path Method. Retrieved from https://hbr.org/1963/09/the-
abcs-of-the-critical-path-method#:~:text=The%20critical%20path%20(or%20paths,
Jr.%2C%20and%20Morgan%20R.

Ray, S. (2019). Understanding Critical Path in Project Management (Example Included).


Retrieved from https://www.projectmanager.com/blog/understanding-critical-path-
project-management
Inventory Models

Inventory represents the goods or material that must be held by a company for use sometime
in the future. The usual examples of inventories are:
 raw material,
 finished goods,
 semi-finished products,
 spare parts.

The key reasons for maintaining inventory are protection against fluctuating demand,
protection against delayed supply, savings on ordering cost, benefits on large quantities,
protection against inflation, etc. All these reasons (and many others) lead to following two
questions for a company’s management:

1. How much to order?


2. When to order?

To be able to answer these questions, the most important types of cost associated with
inventory management must be considered in the appropriate analysis.

Inventory Terminology and Models Classification

Since the objective in the inventory model is typically the minimization of the total cost,
we have to identify the most important types of partial inventory cost:

1. Ordering and setup cost are all the necessary expenses of placing an order. This item
represents the fixed charge that includes e.g. the cost of paper work, billing cost and
supplier’s fixed cost associated with the order. More frequent ordering of smaller quantities
results in a higher cost during the period than less frequent ordering of larger quantities.

2. Unit purchasing cost is the variable cost associated with purchasing a single unit. This
item becomes an important factor in case the price of the single unit depends on the size of
the order (quantity discount – unit price decreases with increasing order quantity).

3. Holding or carrying cost includes:


 storage cost (maintaining the storage space),
 store keeping operations (taking physical inventory),
 insurance and taxes on inventory,
 interest (paid on the capital invested in inventories),
 the opportunity cost (e.g. the interest gained from money saved in the bank, the
return from money invested in shares or bonds, or the yield gained by possible
alternative use of funds),
 cost due to the possibility of spoilage or obsolescence.
Holding and carrying cost can be expressed either in CZK per unit per time
period (usually a year) or in percentage of the annual value of the inventory.
4. Shortage (or stockout) cost is significant if shortage of items affects the company’s
activities, revenue, profit, employment inside the company etc. In general, shortage leads
to unsatisfied demand for the required items. Following examples of such cost can be
considered:

 cost of idled production, cost of idled machines (in case of the spare-parts storage),
 cost of placing special expensive expediting order to restore the inventory,
 loss of customers due to late deliveries of the finished goods,
 other cost, often incalculable (the shortage of blood or ambulance in garage).

Consideration of these four cost items depends on the real specifications of the analyzed
inventory system and thereby also on the used inventory model. Before providing a classification
of inventory models and analyzing some of them, several additional terms are necessary to be
explained (Turban E., and J. R. Meredith, 1988)

Inventory Level
This is the available size of the inventory, i.e. the number of stocked items (the number of
units) or the amount of stocked material (in kilograms, liters, meters, etc.).

Demand and Depletion


The purpose of making inventory corresponds to specific demand. In production process,
for example, the production technology determines amount of items required within specific
period. The rate of demand then determines the inventory depletion rate (and hence, the inventory
level). The higher the rate of demand is, the quicker the inventory level is being reduced. The rate
of demand can be constant throughout time or can fluctuate.

Reordering, Reorder Point, and Lead Time


Since the inventory is being depleted, stocked items must be replenished periodically.
When the inventory level is reduced to a signal level called the reorder point, the replenishment
order must be placed to restore the inventory on time. The time period between placing the order
and receiving the shipment is called the lead time. This period can be constant or variable.

Shortages, Surpluses, and Safety Stock


In an ideal case, the shipment arrives at the moment when the last unit of inventory has
been depleted. Wrong timing of real replenishment order may cause the future unexpected
shortage of the inventory (if we make the order after the optimal reorder date) or surplus of units
in the warehouse (if we make the order before the optimal reorder date). To prevent the shortage
(its full elimination or partial reduction) company’s management can keep a buffer in the form of
safety stock that is used in case where shortage event would occur. This issue is the most significant
in the models with probabilistic demand or lead time.

Average Inventory
Concept of an average inventory is used in almost all the inventory models. As a short
illustration of this very important term of inventory management, we introduce the following
example:
For example:

Monitoring the inventory level during a five-day period, we get the following numbers:
Days Units
Monday 200
Tuesday 150
Wednesday 100
Thursday 50
Friday 0
The average inventory level is then is 100 units computed as follows:

200 + 150 + 100 + 50 + 0


Average Inventory = 5 = 100 units

Figure 4.1 shows the depletion of the inventory in time (decrease of the inventory level occurs at
the beginning of each day). Since demand is constant in time, the average inventory can be also
computed as (200+0)/2 = 100 units.

Classification of Inventory Models

As mentioned above, a type of demand determines the formulation of the problem and,
consequently, the applied model. From this point of view, the inventory models can be divided
into deterministic models and probabilistic models.

In deterministic models the rate of demand (and the inventory depletion rate) is known
with certainty. As an example we can consider the production process with the constant production
rate. This rate (e.g. number of units produced per day) determines the constant rate of demand for
the inventory (e.g. semi-products) and hence, the uniform decrease in the inventory level. These
are some of such models:

 Basic economic order quantity model (EOQ model).


 EOQ model with back orders allowed (with planned shortages).
 Economic production lot size model.
 EOQ model with quantity discounts.

In probabilistic models the rate of demand (and the inventory depletion rate) is probabilistic
(described as a random variable). To illustrate such models let us consider a large supermarket
with complex inventory system. Demand for any goods offered in the supermarket is obviously
random. Demand level (in day, week or month) actually determines real depletion of the inventory.
It is evident that monitoring and analyzing the demand fluctuation in time is a complex and never-
ending process for a highly experienced team of managers and database analysts.

Two basic models will be described later:

 Probabilistic model with continuous demand.


 Single-period decision model.

Figure 4.2 illustrates that it is possible to extend the classification of the inventory models
according to demand from another point of view (Taha, 1982)

1. In case of a static demand, its rate is known with certainty (deterministic model) and in
addition, it does not change from one time period to the next. This type of demand is
considered in the simplest inventory models.
2. Although in case of a dynamic demand its rate is known with certainty, it is not constant
throughout time. Thus, we know the demand value (or its rate) for each period.
3. A stationary probabilistic demand corresponds to a random variable with the probability
density function, remaining unchanged over the time.
4. In a non - stationary case of demand, the corresponding probability density function varies
within time period. Considering this type of demand leads to the most complex inventory
models in which the simulation process is often used as the last chance of successful
inventory managing.
Deterministic Inventory Models
As defined before, in the deterministic models we assume known rate of demand. Although
the rate in such models may change throughout time (the change is known with certainty), we
concentrate only on the static deterministic models. Thus, the constant demand (without any
change from one time period to the next) is considered in all the models described in this section.

The Economic Order Quantity Model (EOQ)


EOQ model is the simplest of all the inventory models. Two basic questions associated
with the inventory management (how much and when to order) can be extended to the following
list of questions:

1. How much (optimum quantity) should be ordered?


2. When (in optimum reorder point) should the order be placed?
3. What is the total cost?
4. What is the average inventory level?
5. What is the maximum inventory level?

Because of the complexity of each real inventory system, we must introduce some
assumptions for each constructed model.
Assumptions of EOQ model:
 single item is considered,
 demand for the item is known and constant over the time,
 lead time is known and constant over the time,
 uniform depletion of the inventory is supposed,
 order quantity is constant,
 purchasing cost is independent of the order quantity,
 unit holding cost is independent of the order quantity,
 replenishment is executed exactly on the point when the shipment arrives and the
inventory level reaches zero

Respecting all the assumptions we can graph the behavior of the EOQ model as it is shown
in Figure 4.3. The ordering and replenishment processes are organized periodically in constant
cycles (we consider three cycles) and all the order quantities are equal. Each cycle begins with the
inventory replenishment and finishes with the total depletion of the store.
Considering the assumptions, the inventory level is being decreased with the uniform rate
between the beginning and the end of the cycle. Thus, this process corresponds to the downward
lines in Figure 4.3. Since each order arrives exactly on the point of the total depletion, no shortages
occur in the inventory process.
Two important inventory levels are outlined in the graph: the maximum level and the
average level. In the EOQ model the maximum level equals to the order quantity (denoted as q).
Hence, the average level is q/2. Within each cycle, of course, a reorder point could be found (for
better readability of the figure these points and corresponding lead times are not marked in the
graph).
For

example, the private brewery produces monthly 4,000 hl of beer. 25% of the production is planned
to be filled into glass bottles. The empty bottles are stored in plastic cases (each case contains 20
bottles) and the average annual holding cost per case is 20 CZK. The carrier, transporting the cases
into the brewery’s store, charges a fixed cost associated with each order for 11,000 CZK. In
addition, the brewery’s own fixed cost of 1,000 CZK per each order is necessary to be involved
into the final calculation. The lead time between the placing each order and its delivery is 1/2 of
month. Since filling of the bottles is the uniform process the inventory depletion rate is uniform as
well.

Management of the brewery decided to analyze the inventory system in order to minimize the total
cost associated with inventory replenishment and holding of the bottles in the store.

These questions should be answered:


1. What is the optimum quantity order?
2. What are the maximum level and the average level of inventory?
3. What is the minimum total annual cost?
4. When to order?
5. How many orders are placed per year?

The solution would have to be:

1. Define input parameters & decision variables


In the example, following parameters of the model can be recognized:

Q – annual demand for cases,


c1 – average annual holding cost per case,
c2 – ordering cost per order,
d – lead time (in years) between placing an order and receiving delivery.
Referring to the managerial questions, we introduce the following variables into the model:
q – order quantity (number of cases in each order),
n – number of orders placed within a year,
OC – total annual ordering cost,
HC – total annual holding cost,
TC – total annual cost,
qmax – maximum inventory level,
qavg – average inventory level,
t – time interval (in years) between two consecutive orders (length of the inventory
cycle),
r – reorder point (level of inventory at reordering).

First, we must calculate (or recalculate) the parameters in order to make them comparable in terms
of period (annual values) and unit of measurement (a case). All the variables, of course, must
respect this notation.

(a) Annual demand Q. The monthly production is 4 000 hl = 400 000 liters of beer. Only 25%
of the production is to be filled into the bottles, i.e. 100 000 liters. Considering the half-
liter bottles, the monthly demand corresponds to 200 000 bottles, i.e. 10 000 cases. Hence
the annual demand Q = 120 000 cases.
(b) Annual holding cost c1. This parameter was specified (in the definition of the problem)
as c1 = 20 CZK per case.
(c) Ordering cost c2. Since placing each order involves 11 000 CZK and 1 000 CZK, the
calculated cost is c2 = 12 000 CZK per order.
(d) Lead time d. As mentioned above, all considered time periods are calculated in years.
Because the lead time is 1/2 of month, we get d = 1/24 of year.

2. Define the function of the total annual cost


The total annual cost TC, including the total annual holding cost and the total annual
ordering cost, can be expressed as follows:

TC = HC + OC (4.1)

Before analyzing this equation in detail, we define two other variables in the model. As it is shown
in Figure 4.3 (following all assumptions of the EOQ model), the average inventory level qavg can
be calculated as:
qavg = q/2 (4.2)

and the maximum inventory level qmax equals to the order quantity:

qmax = q (4.3)

The total annual holding cost HC in Equation 4.1 is then given as the annual holding cost
per case, multiplied by the average inventory level (average number of cases on hand):
(4.4)

Since the number of orders placed within a year is:


n= Q/q (4.5)

the total annual ordering cost OC in (4.1) can be expressed as ordering cost per order,
multiplied by the number of orders:

(4.6)

Thus, introducing (4.4) and (4.6) into (4.1) the total annual cost TC is the function of the only
variable q:

(4.7)

Figure 4.4 shows all the three total annual cost functions HC, OC and TC. Whereas HC is in direct
proportion with q (a linear function), OC is in indirect proportion with q (a hyperbola). The
function TC is then the graphical sum of these two curves. In the graph we marked three possible
points (total annual cost) corresponding to different order quantities 10,000, 60,000 and 120,000
cases. Detailed cost calculation for three prospective policies can be found in Table 4.1.

Table 4.1 Annual Inventory Cost Calculation


Policy I Policy II Policy III
Annual demand Q 120,000 120,000 120,000
Order quantity q 10,000 60,000 120,000

Annual holding cost per case c1 20 20 20


Average inventory level qavg = q/2 5,000 30,000 60,000
Total annual holding cost HC 100,000 600,000 1,200,000

Ordering cost per order c2 12,000 12,000 12,000


Number of orders n = Q/q 12 2 1
Total annual ordering cost OC 144,000 24,000 12,000
Total annual cost TC 244,000 624,000 1,212,000
For better

understanding of the delivery cycles and depletion of the inventory in proposed policies, see Figure
4.5 (time is measured in months). While Policy I seems to be quite reasonable, Policy III might be
even irrealizable because of the brewery’s store capacity. The lower ordering frequency is, the
more of store space is occupied on the point of replenishment.

Fig. 4.5 Different Ordering Policies

3. The objective is to minimize the total annual cost

As it has been already shown (Equation 4.7), the total annual cost can be expressed as:
Minimizing this function, we will find the optimum order quantity. Since TC(q) is a continuous
function with the only one minimum point (see Figure 4.4), the optimum value of q is obtained by
solving the following equation (of the only one unknown q):

(4.8)

Thus, the optimum order quantity is:

(4.9)

Note: to be mathematically precise, we should verify whether this value corresponds to the
minimum point of the analyzed function. Since the second derivative of TC(q) is positive, the point
is the minimum point indeed.

Introducing the values of all the parameters (Q, c1 and c2) into Equation 4.9 we obtain the optimum
order quantity in the example:

√[2(120,000)(12,000)]
q* = = 12,000 cases
20

If the brewery’s management (considering the annual demand of 120,000 cases) wants to minimize
the total annual cost associated with replenishment and holding bottles in the store, 10 orders of
12 000 cases should be placed per year (Equation 4.5).\

Substitution of (4.9) into (4.7) leads, after the simplification, to the formula of the optimum total
annual cost:

(4.10)

Using this equation in the example, the optimum value of the total annual cost is:

Figure 4.6 apparently shows that the optimum value TC* occurs at the intersection of HC and OC.
Hence, an alternative approach to determination of the optimum order quantity is solving the
following equation of unknown q:

By easy manipulation of this equation we get the formula (4.9).


Fig. 4.6 Optimum Order Quantity & Total Annual Cost

The optimum ordering frequency n* (number of orders placed within a year) determines the
optimum length of the inventory cycle t*- this is a reciprocal value of n*:

(4.11)

Since in the example n* = 10, the optimum length t* = 1/10 of year. Supposing a continuous
production process throughout a year, the management would place orders in the intervals of
365/10 = 36.5 days. In Figure 4.7 all 10 delivery cycles are graphed in real time (in days). See the
maximum and the average inventory level in the graph.

Fig 4.7 Optimum Delivery Cycles

The last management’s decision concerns a reorder point. As defined before, this is the alarming
inventory level that requires placing a new order. This value depends, of course, on the lead time
between placing order and delivery. Figure 4.8 shows the relations between all the values
necessary for determination of the optimum reorder point r*. Using the elementary mathematics
(the parallel triangles theorem) we can write:

r*/d = q*/ t*

Hence, the optimum reorder point r* is:

r*=d q*/ t* =dQ (4.12)

Thus, in the example, we get:


r*=(1/24) 120,000 = 5,000 cases

If the inventory level decreases to 5 000 cases, the management must place a new order to prevent
discontinuity of the production process due to cases shortage in the future.

Figure 4.8 Reorder Point

Equation 4.12 is valid only in case of d ≤ t*. If d = t*, a new order should be placed exactly in the
moment when the previous order is delivered. However, if d > t*, formula (4.12) is inconsistent
with a logical assumption r* ≤ q*, and should be modified as follows:

(4.13)
The operator “modulo” returns the remainder for division of the first argument by the second.
Suppose d = 1/6 instead of 1/24 (the lead time is 2 months instead of 1/2 of month). The calculation
of (4.13) gives r* = 20 000 mod 12 000 = 8 000 cases.

Summary
This is the summary for the brewery’s management to make the optimal decisions in the
company’s inventory process.

Each year 10 orders will be placed. The order quantity is 12,000 cases. The interval between
placing two consecutive orders is 36.5 days. This value represents the length of delivery cycle.
When the inventory in the store decreases to 5,000 cases a new order must be placed.

Following this optimal policy, the brewery’s total annual cost associated with the inventory of
cases will be 240,000 CZK.

The average inventory level is 6,000 cases, the maximum level is 12,000 cases.

Note 1: As mentioned above, the available space in the store might be a significant restriction of
the model in case that the maximum (and optimum) inventory level would be higher than (or very
close to) the store capacity. Mostly, in the real process it is not possible to implement exactly 36.5
days between two consecutive orders, but this is not a major problem. More important issue seems
to be the minute of replenishment. In a real company the replenishment is being started before the
total depletion of the inventory. Thus, in the real life the maximum inventory level might be rather
higher than the computed value to keep the continuity of the production process.
Note 2: The computed order quantity (or other results as e.g. the number of ordering cycles) in the
EOQ model may be non - integer. After rounding the result both down and up we calculate the
total cost and then select the better alternative.

The EOQ Model with Planned Shortages

In the elementary EOQ model we assume that the store is replenished exactly on the point when
the inventory level reaches zero. Thus, no shortage occurs in such situation. However, in many
situations it is possible to allow shortages, or even to plan them rather then to build a safety stock.
Since each shortage affects the P&L (profit and loss) of a company, specific cost associated with
the stockout must be considered in the inventory management. Such cost may be a loss of the
profit, or an extra charge that must be paid if, for example, other activities (e.g. continuous
production process) depend on the inventory.

Suppose, we will modify Example 4.2, in which the brewery’s inventory process was analyzed.
We allow the shortage of cases in the company’s store. Since the production (filling bottles) is
continuous process that cannot be interrupted, in case of stockout it is possible (in a short term) to
use a small flexible store with a safety stock. Cases from this house can be supplied into the
production process directly and in a uniform flow (similarly to the large store). Using this
possibility, the company must pay extra annual charge (150 CZK per case) corresponding to the
average annual operating cost on the inventory in the small store (no holding host is considered).
For simplicity we will call the large store as the primary store and the smaller store as the
secondary store.

The brewery’s management should make decisions on the order quantity and the period in which
reorders would be placed to minimize the total cost.

To Solve:
1. Define input parameters & decision variables
As Figure 4.9 shows, the considered inventory process is quite different from the previous
example.

Some of the parameters and variables, considered in the elementary EOQ model, must be
redefined. In addition, some new parameters and variables must be introduced.

Figure 4.9 EOQ Model with Planned Shortage

The following parameters are defined in the brewery’s model with planned shortages:
Q – annual demand for cases,
c1 – average annual holding cost per case (in the primary store),
c2 – ordering cost per order,
c3 – average annual operating cost per case (in the secondary store),
d – lead time (in years) between placing an order and receiving delivery.

The list of variables follows:


q – order quantity (number of cases in each order),
s – shortage level (inventory level in the secondary store),
n – number of orders placed within a year,
OC – total annual ordering cost,
HC – total annual holding cost (concerning with the primary store),
SC – total annual shortage cost (operating cost associated with the secondary store)
TC – total annual cost,
qmax – maximum inventory level,
qavg – average inventory level,
smax – maximum shortage level (maximum inventory level in the secondary store),
savg – average shortage level (average inventory level in the secondary store),
t – time interval (in years) between two consecutive orders (length of the inventory cycle),
t1 – time period (in years) during which no stockout occurs in the primary store,
t2 – time period (in years) during which the inventory in the secondary store is used,
r – reorder point (level of inventory at reordering).

2. Define the function of the total annual cost

Referring to Figure 4.9, we will carry out the analysis of one inventory cycle.

The average inventory level qavg in the cycle can be expressed as:

qavg = q – s / 2 (4.14)

and the average shortage level savg (the average inventory level in the secondary store):

savg = s /2 (4.15)

The maximum inventory level qmax equals to the order quantity minus the shortage level:

qmax = q – s (4.16)

The maximum shortage level smax corresponds, in the example, to the maximum inventory level
in the secondary store.

Since the holding cost in the primary store relates to the period t1 (only in this period the store is
not empty), the total holding cost HC in one cycle can be calculated as follows:

(4.17)

Similarly, the total shortage cost SC in one cycle corresponding to the period t2 (only in this
period the secondary store is used) is:

(4.18)
Note: In the beginning of the period t1 the secondary store is logically replenished (the inventory
level is smax) and then left waiting (consider that there is no holding cost) all this period without
any use (the door is being locked). After the primary store is totally depleted the door of the
secondary store is open and the inventory starts to be used with the only operating cost. Within the
period t2 the inventory in this store is depleted as well and replenished again (together with the
primary store) at the end of the inventory cycle.

Analyzing the total ordering cost OC in one cycle, we should realize that each cycle is actually
determined by one order. Thus, we can simply write:

OC = c2 (4.19)

Total cost associated with one cycle (and with one order) can be calculated as HC + SC + OC.

If we consider that the number of orders placed within a year is:

n=Q/q (4.20)

the total annual cost TC can be then expressed as:

TC = (HC + SC +OC) n

or as the function of two variables q and s:

(4.21)

The variables t1 and t2 in Equation 4.21 can be expressed (using the parallel triangles theorem
and considering t = 1/n = q/Q) as follows:

t1 = q – s / Q ,

t2 = s / Q

Introducing these formulas into (4.21) we get (after the simplification):

(4.22)

3. The objective is to minimize the total annual cost


The function (4.22) is actually the function of the only two variables q and s. In order to find the
minimum of TC, its partial derivatives with respect to q and s are set to zero.
Solving two generated equations simultaneously, we get as the results the optimum order
quantity:

(4.23)

and the optimum (maximum) shortage level:

(4.24)

In the example, the optimum order quantity is:

√[2(120,000)(12,000)] √[20+150]
q* = = = 12,774.98 cases
20 150

and the optimum shortage level:

s* = 12,774.98 [20 / (20+150)] = 1,502.94 cases

The optimum number of orders placed in the year is:

n* = Q / q* = 9.39

and the optimum length of the delivery cycle:

t* = q* / Q = 0.11 years = 38.86 days

These non - integer values come into conflict with practical recommendations how often and how
much to order. The successful manager cannot direct the inventory department to place orders of
12 774.98 cases in intervals of 38.86 days. To make reasonable decisions the manager should
adjust the results. Whatever corrections will be made we lose the optimum solution. However, we
can suggest such strategy that will be as close as possible to the optimal (but impracticable)
strategy.

4. Correction of non - integer results

4.1. First, we round the optimum length of the delivery cycle. Two possibilities
(rounding down or up) are considered:

t (A) = 38 days
t (B) = 39 days

4.2. Following the first integer alternative (strategy A), the company should make an
order of “exactly” this quantity:
An alternative method how to calculate this value is based on the parallel triangle theorem (Figure
4.10):

(4.25)

Figure 4.10 Optimum and Integer Lengths of Inventory Cycle

As the value A q is non - integer, the management should round it down or up. Rounding down
leads to infeasible solution: 12,493 (365/38) < 120,000. Rounding up to 12 494 cases satisfies (and
exceeds) the annual demand (surplus is almost 9 cases).

Thus, the order quantities for both the strategies are computed and rounded up:

q (A) = 12, 494 cases


q (B) = 12, 822 cases

3. Similarly to Equation 4.25, we can simply determine the maximum shortage level corresponding
to 38 days (39 days):

(4.26)
(4.27)

After substitution of the exact values into (4.26) and (4.27) we get non - integer values:

As in the brewery’s inventory process these numbers are considered to be integer as well, the
management has two possibilities for each strategy. Strategy A is branched into strategies A1 and
A2, strategy B into strategies B1 and B2 as follows:

Further in the text we are going to combine the integer order quantities with the integer shortage
levels to compute the total cost and to make the best real decision.

4. Since the optimum length of the inventory cycle has been rounded (down or up), the
used strategy causes that the last cycle in the current year is not the full cycle. Hence, the delivery
will not come exactly at the end of the year and some cases remain either in the primary or in the
secondary store.

Before cost calculations we compute the number of the inventory cycles:

n A = 365 / 38 = 9.61
n B = 365 / 39 = 9.36

Rounding down these values we get the number of full cycles:

n(A) = n(B) = 9

5. Now we can project total annual cost as follows:

TC = TCF 365/nt (4.28)


where:

TC … total annual cost,


TCF … total cost relating to all the full periods,
n … number of the full periods,
t … length of the full period.

For calculation of TCF we use Equation 4.22 (however, we must take Q = nq instead of Q =
120,000). Of course, we should consider all four strategies:

Strategy t q n Q = nq s qmax = q-s TCF TC


A1 38 12,494 9 112,446 1,469 11,025 218,241 232,918
A2 38 12,494 9 112,446 1,470 11,024 218,241 232,918
B1 39 12,822 9 115,398 1,508 11,314 221,135 229,956
B2 39 12,822 9 115,398 1,509 11,313 221,135 229,956

Table 4.2 Total Cost Estimation

In Table 4.2 the exact values of TCF and TC are rounded to integers. Whereas differences between
total cost for strategies A1 and A2 (or B1 and B2) are not quite significant, crosswise differences
could be considerable for the company.

Since the management’s objective is to minimize the total annual cost, strategy B (either B1 or B2)
should be used in the company’s inventory process. Total cost is estimated in the amount of
approximately 229,956 CZK.
For the calculation of optimum total annual cost we use Equation 4.22, in which the optimum (non
- integer) order quantity and optimum (non - integer) shortage level are substituted:

An application of the practicable “integer” strategy B increases total annual cost in about 2% (in
comparison with the impracticable optimum total annual cost).

The maximum inventory levels qmax in the primary store for each strategy can be found in Table
4.2.

6. The last decision concerns with the reorder point. For the EOQ model with planned
shortages we must change the elementary EOQ model’s formula (4.13) as follows:

(4.29)

If we decide to follow the strategy B1, the reorder point is:

r = [(1/24) 12,822) / (39/365) mod 12,822] – 1,508 = 5,000 – 1,508 = 3,492 cases

Summarizing, the brewery’s management should place orders periodically in the intervals of 39
days. This value also corresponds to the length of delivery cycle. The order quantity is 12,822
cases. The maximum inventory level in the primary store is 11,314 cases. When the inventory
level decreases to 3,492 cases, a new order must be placed. On the point of delivery, 11,314 cases
are stored in the primary store and 1 508 cases are stored in the secondary store.

Following this strategy, the brewery’s total annual cost associated with the inventory of cases is
approximately 225,441 CZK.
Note: Comparison of the total annual cost in case the shortages are allowable (225,441 CZK), with
the cost in situation where no shortage may occur (240 000 CZK in Example 4.2) shows a
possibility for the company’s management to improve its decisions. Finding a reasonable solution
(building and using the secondary store) leads to improvement of the objective (decreasing total
cost).

The EOQ Model with Quantity Discounts

In real markets suppliers usually offer to customers’ quantity discounts as an incentive for large
order quantities. The larger order quantity is, the lower purchase price per item a customer must
pay. Unit holding cost is often derived from the purchase price, mostly as its percentage. Hence,
larger lots lead to lower unit holding cost (the elementary EOQ model’s assumption of fixed unit
holding cost is not valid any more). In practice, quantity discounts are scheduled in several
purchase price intervals.

Say for instance, suppose now that the brewery in Example 4.2 takes the cases from a supplier,
offering quantity discounts scheduled in Table 4.3. Annual unit holding cost is calculated as 50%
of unit purchase price.

Order Size Purchase Price Unit Holding Cost


Discount Category (Cases) (CZK per Case) (CZK per Case)
1 0 to 499 46 23
2 500 to 14,999 40 20
3 Over 15,000 36 18
Table 4.3 Quantity Discounts Schedule

The brewery’s management should again decide for the order quantity that would lead to the
minimum total annual cost. No shortages are allowed.
Solution:

1. Define input parameters & decision variables

Definitions of all the parameters and variables can be taken from the elementary EOQ
model in Example 4.2. In addition the parameter cq and the variable PC are considered in the
model:
cq – unit purchase price (per case) – for the discount category corresponding to
the order quantity q
PC – total annual purchase cost.

2. Define the function of the total annual cost

The total annual cost TC, including the total annual holding cost, the total ordering annual cost
and total purchase cost, can be expressed as follows:

TC = HC + OC + PC (4.30)
Since the total purchase cost PC is calculated as the unit purchase price, multiplied by the annual
demand, the total annual cost TC can be expressed as the function of the variable q:

(4.31)

3. The objective is to minimize the total annual cost

Similarly to the elementary EOQ model, for each discount category we compute the optimum
order quantity:

(4.32)

Table 4.4 shows the computed EOQ using the various unit holding cost c1 associated with three
discount categories.

1. In the first category, the optimum order quantity is q*1 = 11,191 (rounded up), what is
not a relevant value, because the upper bound for this category is 4,999 cases. Thus, for the
calculation of the total cost we take q1 = 4,999 cases instead of 11,191 cases.

2. The optimum order quantity for the second category q*2 = 12,000 (an exact value) falls
within the interval <5 000; 14 999> and we can take q2 = 12,000 for the calculation of minimum
total cost.

3. In the last category, the optimum order quantity q*3 = 12,650 (rounded up) is lower than
the lowest possible quantity (15,000 cases) to qualify for a discount. Therefore we take q3 = 15,000.

Discount Category Order Size q* q TC


1 0 to 499 11,191 4,999 5,865,546
2 500 to 14,999 12,000 12,000 5,040,000
3 Over 15,000 12,650 15,000 4,551,000
Table 4.4 Total Cost for Quantity Discount

The total annual cost for each discount category is calculated using the formula (4.31):

In summary, the objective is to minimize the total annual cost, the management should use the
third discount category. The optimum order quantity is 15 000 cases. Qualifying for the offered
discount the purchase price is 36 CZK per case. The total annual cost is 4,551,000 CZK.

All the inventory characteristics (using formulas for the elementary EOQ model) could be, of
course, determined.
The Economic Production Lot Size Model

The last deterministic model, presented in this textbook, concerns the situations where the
inventory is not replenished through the single batch, but gradually over a time period. As there
are many variations of those models, we describe the best known of them. The inventory cycle in
this model consists of two phases: production phase and nonproduction phase (Figure 4.11).

In the production phase (within a period t1), the store is uniformly replenished from the production
batch. Simultaneously, the inventory is being depleted in this phase. In Figure 4.11 a gradual
inventory buildup is obvious – it is produced more than it is demanded (a production rate is greater
than a demand rate). When the peak of the inventory level is reached, the production run is stopped
and the inventory is depleted only (within a period t2). In the point of the total depletion of the
inventory a new production batch is started and the inventory cycle is repeated. No shortages are
allowed in this model.

Figure 4.11 Economic Production Lot Size Model

In the model we consider two cost items: holding cost and production setup cost. The
definition of the holding cost is the same as in the EOQ model. Production setup cost is a fixed
cost connected with a preparation of the production batch. This cost is independent of the
production lot size, similarly, as the ordering cost is independent of the order quantity in the EOQ
model.

The objective is to determine such production lot size q (economic lot size) that minimizes
the total cost. It is evident that the maximum inventory level is always lower than the production
lot size, and the average inventory level is one-half of the maximum inventory level.
For instance, modification of example 4.2 consists now in the process of preparing the
bottles for their filling. Instead of their simple ordering, the bottles are processed on the cleaning
line. From this line the bottles are being transported (in cases) into the store and then from the store
(when they are needed) directly to the filling process. We are not interested, in the example, in the
processes preceding to the cleaning bottles (how and when the bottles are ordered, or where they
are stored, etc.). However, we know the calculation of the cost connected with a preparation of the
cleaning process (e.g. cost of servicing, setting up and washing the cleaning machine and the whole
line). This cost was calculated to 12,000 CZK per one successive cleaning process. Suppose the
original unit holding cost (annually 20 CZK per case) and the annual demand (120,000 cases). The
cleaning line’s daily output is 8,000 bottles. Preparation of the cleaning line takes 1/2 of month.

The brewery’s management wants to determine the size of cleaning batch to minimize the total
annual cost. Following questions should be answered to schedule production runs:

1. What is the optimum lot size?


2. What is the maximum level of the inventory?
3. What is the minimum total annual cost?
4. How long time does a cleaning run take?
5. When to start the preparation process (setup) for the cleaning run?

To solve, although the brewery does not produce the bottles (or cases of bottles) - it just cleans
them - we will call the process the production, since the models like this are actually used for
production-consumption processes.

1. Define input parameters & decision variables

We consider following parameters:


Q – annual demand for cases,
c1 – average annual holding cost per case,
c2 – fixed setup cost per cleaning lot,
p – production rate (number of cleaned cases, e.g. per year),
h – demand rate (number of demanded cases, e.g. per year),
d – lead time (in years), necessary to prepare a new production run,

and these variables:


q – production lot size (number of cleaned cases in a cleaning lot),
n – number of lots within a year,
SC – total annual setup cost,
HC – total annual holding cost,
TC – total annual cost,
qmax – maximum inventory level,
qavg – average inventory level,
t1 – length of a production period (in years), when the production runs,
t2 – length of a depletion period (in years), when only depletion proceeds,
t – cycle - time period (in years) between starts of two consecutive cleaning lots,
r – starting setup point (level of the inventory when the setup of the cleaning line should
be started).

First, we must recalculate the parameters:

(a) Production rate p. This parameter was specified as 8 000 cleaned bottles per day.
Since our time unit is one year and the quantity unit is one case, we get 8 000*365/20
cleaned cases per year (if the cleaning line runs without any interruption). Thus p =
146,000 cases per year.
(b) Lead time d. The lead time is 1/2 of month, i.e. d = 1/24 of year.

Note: In the example, if considered time unit is one year, the demand rate h equals to the annual
demand Q. Generally, whatever time units of parameters p and h are used, they must be identical
(mostly one day).

2. Define the function of the total annual cost

The total annual cost TC consists of the total annual holding cost and the total annual setup cost:

TC = HC + SC (4.33)

In the time period t1 the total number of cleaned cases is pt1. Since the demanded number of cases
is ht1, after this period the maximum inventory level is pt1 - ht1 = (p - h) t1. Considering pt1 = q (i.e.
t1 = q/p), the maximum inventory level qmax can be expressed as follows:

qmax = (p – h / p) q (4.34)

and the average inventory level qavg can be calculated as follows:

(4.35)

The total annual holding cost HC is defined as the annual holding cost per case, multiplied by
the average inventory level:

(4.36)

Since the number of production lots within a year is:

n=Q/q (4.37)

the total annual setup cost SC in (4.33) can be expressed as setup cost per cleaning lot, multiplied
by the number of lots:

SC = c2n = c2 Q / q (4.38)
Thus, substituting (4.36) and (4.38) into (4.33), the total annual cost TC is the function of the
variable q:

(4.39)

3. The objective is to minimize the total annual cost

After setting the first derivative of (4.39) to zero we get the optimum (economic) lot size:

(4.40)

Introducing the values of all the parameters into Equation 4.40 we obtain the optimum lot size in
the brewery:

√[2(120,000)(12,000)] √146,000
q* = = 28,436.16 cases
20 146,000−120,000

Substituting the value of q* into (4.34), the maximum inventory level is q*max = 5,064 cases.

Substitution of (4.40) into (4.39) leads, after the simplification, to the formula of the optimum
total annual cost:

(4.41)

Using this equation in the example, the optimum value of the total annual cost is:

Since pt1 = q, the optimum length of the production period can be expressed as follows:

(4.42)

Thus, in the example, t *1 = 0.1948 years that corresponds to 71.1 days.

Similarly, because of ht2 = qmax, the optimum length of the depletion period is:

(4.43)

The value of this period, in the example, is t*2 = 0.0422 years (i.e. 15.4 days).
Since t = t1 + t2, the optimum cycle (time between starts of two consecutive cleaning lots) is t* =
0.1948 + 0.0422 = 0.237 years (86.5 days). For determination of this value, of course, the well-
known formula can be used:

t* = q* / Q (4.44)

The last question concerns with the optimum starting setup point. Similarly to the EOQ model
(where this point is called reorder point), it is an alarm inventory level to start preparations for a
new cleaning run (a new order in the EOQ model). The way of calculation of the starting setup
point depends on the value of the lead time d which is necessary to prepare a new run (Figure
4.12).

Figure 4.12 Starting Setup Point (“Reorder Point”)

1. If *d ≤ t2 , the setup is started within the depletion period and the alarm inventory level
equals exactly to the quantity demanded during the period d:

r* = hd (4.45)

2. If t * ≥ d > t, the setup is started within the production period and the starting setup point
corresponds to the quantity produced (and stored) during the period t * − d:

r* = (p – h) (t* - d) (4.46)

Thus, monitoring the inventory level, the management must, in addition, detect whether the
production runs or does not.

Since in the example, d = 1/24 = 0.0417 < t*2 = .0422. Equation (4.45) must be used for the
calculation of the optimum starting setup point:
r* = hd = 5,000 cases

Note: If d > t *, the operator “modulo” must be applied in (4.45) and (4.46), similarly to the EOQ
model.

Summarizing, the brewery’s management should schedule the cleaning lots of 28,436.16 cases
after 86.5 days. Each cleaning lot takes 71.1 days. When a lot is finished, 5,064 cases are in the
store and a depletion period begins. Then the management should monitor the inventory level and
when it decreases to 5,000 cases, a new setup must be started.

Following this optimal schedule, the brewery’s total annual cost associated with the cleaning
bottles and their holding in the store is approximately 101,280 CZK.

Note: Since almost all the values are non - integer, the results should be corrected using similar
procedure as we presented in Section 4.2.2.

Probabilistic Inventory Models

In all models described in Section 4.2, deterministic demand and uniform inventory
depletion have been assumed. In presented brewery’s situation, this assumption seems to be
reasonable because of the constant rate of filling bottles. However, in many real situations, both
demand and depletion rate cannot be predicted exactly and only a probabilistic distribution of them
is available.

In most probabilistic models three following basic questions should be discussed:

1. When to order?
2. How much to order?
3. How much to store in a safety stock?

The first and the second question can be found in deterministic models. Probabilistic
models must deal, in addition, with the third question. Safety stock is a buffer stock, built up to
avoid the inventory shortages that can occur in the situations with uncertain demand or fluctuating
depletion rate.

We present two probabilistic inventory models: the model with continuous demand and the single-
period decision model.

Probabilistic Model with Continuous Demand

Except for several differences in assumptions, a probabilistic model with continuous demand is
rather close to the EOQ model presented in Section 4.2.1. Therefore, sometimes it is called as the
EOQ Model with Safety Stock.
Assumptions:
� single item is considered,
� demand for the item is not known with certainty (the probabilistic distribution of demand is
given),
� lead time is known and constant over the time,
� continuous, but not uniform depletion of the inventory is supposed,
� unit purchasing cost is independent of the order quantity,
� unit holding cost is independent of the order quantity,
� no additional cost are considered in case of shortage,
� replenishment is executed exactly on the point when the shipment arrives (if shortage occurs,
the unsupplied demand is filled immediately on the delivery point).

As Figure 4.13 shows, probabilistic character of demand may cause an overstock (Cycle I)
or a stockout (Cycle II):

1. In the first situation a delivery arrives when some inventory remains in the store. In this
case the real demand during the lead time is lower than a reorder point (an inventory level
when a new reorder is placed).
2. In the second situation a shortage occurs, as the real demand during the lead time exceeds
a reorder point.

Figure 4.13 Inventory Level in Model with Probabilistic Demand

To be able to analyze this probabilistic model we must know:


 Type of continuous probability distribution of demand - we can suppose the normal
distribution (Figure 4.14).
 Mean of demand μQ.
 Standard deviation of demand σQ.
Figure 4.14 Normal Distribution of Demand

Suppose now that the brewery’s production of beer (in Example 4.2) follows consumers’
thirst that depends on probabilistic processes (e.g. weather changes, tourism, etc.). From the
company’s sale statistics the annual demand is estimated at 120,000 cases and standard deviation
at 12,000 cases. The brewery’s managers suppose the normal probability distribution of demand.
All the other parameters of the model are identical to the parameters in Example 4.2.
The objective is to schedule the inventory process with respect to possible shortages
occurrence.

To solve:
1. Calculation of optimum order quantity and optimum reorder point

For this purpose, we can use the equations from Section 4.2.1, however instead of
a deterministic value of demand Q we use the mean of probabilistic demand μQ. Since
μQ = 120,000, an estimation of optimum order quantity is q* = 12,000 cases.
Since the annual demand is a random variable, demand within the lead time (lead-
time demand) is a random variable as well. The mean and the standard deviation of
demand within the lead time d can be expressed as follows:

(4.47)

Hence, the optimum reorder point estimation is:

(4.48)

Considering in the example μQ = 120 00 cases, σQ = 12 000 cases and d = 1/24 of year,
the optimum reorder point r* = (1/24) 120 000 = 5 000 cases. The standard deviation
of the lead-time demand is σd = (1/24) 12 000 = 500 cases.

The lead-time demand is supposed to have the normal probability distribution (Figure
4.15), similarly to the total annual demand.
Figure 4.15 Normal Distribution of Lead-Time Demand

2. Building a safety stock

In Section 4.2.2 the deterministic model with planned shortages was introduced. Since in
the probabilistic model, demand is uncertain, the shortages occur randomly and thus, they cannot
be planned. To avoid or reduce a risk of stockout, in most of probabilistic models a safety stock is
being built up.

For this purpose, we provide three alternative interpretations of the key term in probabilistic
models – service level:

1. Service Level is the probability with which demand will be met within the inventory cycle.
2. Service Level is the probability with which shortage will not occur within the inventory cycle.
3. Service Level is the percentage of time that all demand is met.

Since the safety stock is built up to protect the company against the shortages, its level
depends on the required service level. Suppose 20 planned orders (cycles) and management’s
decision that only one stockout can be tolerated (5%). Hence, the service level is 95%. The
objective is to determine the safety stock that avoids other stockouts. Of course, the higher the
safety stock is, the more money a company must pay as the holding cost. Therefore, a compromise
decision must be made in such situations.

Building the safety stock actually means a decision to place an order before reaching the
optimum reorder point. Thus, the implemented reorder point can be expressed as follows:

r p = r* + w (4.49)

where:
rp … reorder point for the given service level p,
r* … optimum reorder point (4.48),
w … safety stock level.
As mentioned above, use of the safety stock increases the holding cost, what leads to increasing
the mean of the total cost:

(4.50)

The objective is now to find such a safety stock level w that satisfies the given service level and
minimizes the formula (4.50).

Denoting Qd as the real lead-time demand, the probability with which this value will be less than
or equal to the reorder point rp, must equal at least the service level p:

(4.51)

Note that the probability on the left-hand side of the inequality (4.51) is a probability that the
inventory is not totally depleted before a new delivery. This probability, of course, must satisfy
the required service level.

According to the initial assumption, the real lead-time demand Qd has the normal probabilistic
distribution with the mean (4.48) and the standard deviation (4.47): N (r*, σd). This distribution is
transformed into the standard normal distribution N (0, 1) using the following formula:

(4.52)

Formula (4.52) can be written in the form:

(4.53)

After substitution of (4.53) and (4.49) into (4.51) we get:

and then, after the simplification:

(4.54)

Finally, Equation 4.54 determines the minimum level of the safety stock w that must be built up to
satisfy the service level p:

(4.55)
1. Suppose that the brewery’s management decided to keep the service level at 95%. In the
table of the standard normal distribution N (0, 1) in Appendix we look up the value zp
corresponding to the given probability p = 0.95. The corresponding value found in the table
is z0.95 = 1.645. Thus, for the value of the safety stock we get:

w ≥ 1.645 (500) cases

Minimizing the mean of the total cost (4.50), the optimum safety stock level is:
w* = 823 cases

2. If the service level is 99%, the corresponding distribution value is approximately z0.99 =
2.327 and the optimum safety stock level is:
w* = 2.327 (500) = 1,164 cases

In summary, because of probabilistic demand, the brewery’s management must consider the
possibility of shortages that may occur within the year.

If managers require 95% service level, the safety stock of 823 cases must be built in each cycle,
i.e. the reorder point is 5,000 + 823 = 5,823 cases. Total inventory cost is 240,000 + 20(823) equals
to 56,460 CZK.

If the service level is being set to 99%, the safety stock of 1,164 cases must be built in each cycle
and hence the reorder point 5,000 plus 1,164 equals 6,164 cases. Total inventory cost is 240,000
plus 20 multiplied by 1,164 equals 263,280 CZK.

Single-Period Decision Model

Up to this point we have dealt with the systems operating continuously and having many
inventory cycles. The single-period inventory model refers to the situations in which only one
order is placed for the considered time period and the inventory is out of stock or there is a surplus
of units at the end of that period. There is a penalty associated with overstock or stockout. Rarely,
demand is equal exactly to the supply and no penalty is counted.

The single-period models concern seasonal or perishable items that cannot be kept for
future periods. As a typical example of such goods we can mention newspapers that do not keep
their recency to the next day. Therefore, a problem dealing with the single-period decision model
is often known as the Newsboy Problem. Other seasonal or perishable goods are, for example,
seasonal clothing, Christmas trees, Halloween pumpkins, flowers, bread, some fruits, salads,
reserved rooms or seats, etc.

Since in all of the mentioned situations, demand is mostly uncertain, we will develop the
probabilistic single-period model. To be able to solve such problems, we must know the
probabilistic distribution of demand. Knowledge of (either continuous or discrete) the distribution
can be stated e.g. from the history (as the statistical values), from the marketing research or
analysis, etc. Placing the order results in building the initial inventory at the beginning of the
period. Within this period the real demand is less than, greater than or equal to the initial inventory
level.

For example, A bakery department’s manager of a new supermarket Happyland should optimize
everyday order of rolls. The supermarket buys a roll for 1 CZK and sells it to the final consumers
for 2 CZK. If, in the evening some rolls remain in the store, the bakery department changes them
into crumbs that will be sold later for 12 CZK per sack. For filling one sack of crumbs 20 rolls are
needed.

From the experience of opening comparable supermarkets in comparable areas, the


supermarket’s analyst recommends to consider the normal probability distribution of the daily
demand with a mean of 10,000 rolls and a standard deviation of 500 rolls.

Solving the given example:

1. Unit cost calculation

Before specifying the cost we define:

q – daily quantity of ordered rolls (initial inventory level),


Q – real daily demand for rolls,
μ – mean value of daily demand,
σ – standard deviation of daily demand.

Because of the uncertainty of the daily demand, three situations can occur in the
supermarket:

1. The real demand is less than the order quantity (Q < q).
In this case, in the evening (q – Q) rolls remain in the store and they are used for
crumbs.

Unit cost in this situation corresponds to the marginal loss that can be expressed as
follows:

ML = purchase cost – salvage value (4.56)

In the example, the purchase cost is 1 CZK per roll. The salvage value is 0.6 CZK
per roll as each roll corresponds to 1/20 of a sack of crumbs (and each sack’s value
is 12 CZK). Hence the marginal loss per a roll:

ML = 0.4 CZK

2. The real demand is greater than the order quantity (Q > q).
In this situation, (Q – q) rolls are demanded (by customers) but not sold because of
their shortage. Unit cost in this case equals the marginal profit loss:

MPL = selling price – purchase cost (4.57)

Since, in the supermarket, the selling price is 2 CZK per roll and the purchase cost
is 1 CZK per roll, the marginal profit loss per roll due to the unsatisfied demand is:

MPL = 1 CZK

3. The real demand equals the order quantity (Q = q).


In this ideal, but rather hypothetic situation no loss is generated.

2. Determination of the optimum initial inventory level

If we denote p as the probability with which no stockout occurs, an expected marginal loss per
roll is then p (ML). Since (1 - p) is the probability of stockout, an expected marginal profit loss per
roll is (1 - p) MPL. The following equation corresponds to the optimum expected cost:

p(ML) = (1 - p) MPL

Hence, the probability with which no stockout occurs can be expressed as follows:

p = MPL / (ML + MPL) (4.58)

Similarly to the previous model (see the inequality (4.51)), the initial inventory level q (daily order
quantity) must satisfy the following inequality:

P{Q<q}>p (4.59)

The probability p in (4.59) represents the optimum service level (the probability with which
demand will be satisfied within the given period).

Since the demand Q in Happyland has the normal probabilistic distribution with the mean μ = 10
000 rolls and the standard deviation σ = 500 rolls, we must transform the general normal
distribution N (μ, σ) into the standard normal distribution N (0,1) using known formula:

zp = (Q – μ) / σ (4.60)

Extracting Q from (4.60) and introducing it into (4.59) we get the inequality that must be satisfied
to keep the service level p:

Q > μ + Zp σ (4.61)

First, in the example, we count the value of the optimum service level:
p = 1 / (0.4 + 1) = 0.7143

Then, for this probability we look up the value of the standard normal distribution N (0, 1):

Z0.7143 = 0.566

Thus, for the initial inventory level we can write:

q > 10,000 + 0.566 (500) = 10, 283 rolls

The optimum initial inventory level (the optimum daily order quantity) is:

q* = 10, 283 rolls

Suppose now that the discrete probabilistic distribution of the daily demand Q is available instead
of the description in the form of the continuous distribution. In Table 4.5, the probability of each
demand level is estimated. In addition, cumulative probabilities (with which demand is less than
or equal to the given level) are calculated.
Demand Level Probability Cumulative Probability
8,000 0.01 0.01
8,200 0.02 0.03
8,400 0.04 0.07
8,600 0.06 0.13
8,800 0.07 0.20
9,000 0.09 0.29
9,200 0.10 0.39
9,400 0.11 0.50
9,600 0.12 0.62
9,800 0.10 0.72
10,000 0.09 0.81
10,200 0.07 0.88
10,400 0.05 0.93
10,600 0.04 0.97
10,800 0.02 0.99
11,000 0.01 1.00
Tables 4.5 Discrete Probabilistic Distribution for Demand
Since the calculated value of the service level p = 0.7143 falls into the interval <0.62; 0.72> in the
column of the cumulative probabilities, the optimum daily order quantity can be found in the
interval <9 600; 9 800> rolls. If the manager may order rolls in boxes of 200 rolls (i.e. 150 rolls,
for example, cannot be ordered), the optimum order quantity will be set at the level:

q* = 9,800 rolls

In summary, When the bakery department’s manager has the above-mentioned information about
the continuous probability distribution of demand, and wants to keep the service level 0.7143, the
optimum initial inventory level is 10 283 rolls.
Considering the table of the discrete probabilistic distribution of demand, the manager should order
9 800 rolls each day.

Glossary:

Cycle Time – The length of time between the placing of two consecutive orders.

Demand Rate – Inventory demanded within a specific time period.

Depletion – The process of reduction of the inventory, or reduction of the inventory to a zero
point.

Deterministic Inventory Model – Inventory model in which demand within a time period is
known with certainty.

Economic Order Quantity (EOQ) – An order quantity that minimizes the annual holding cost
plus annual ordering cost.

Holding Cost – Variable cost associated with storing inventory.

Inventory Level – The current (available) amount of inventory.

Lead Time – The time between placing an order and receiving the shipment (delivery).

Marginal Loss – Unit cost (purchase cost minus salvage value) that occurs when demand is less
than the order quantity.

Marginal Profit Loss – Unit loss (selling price minus purchase cost) that occurs when demand is
greater than the order quantity.

Ordering Cost – Fixed cost of placing one order.

Probabilistic Inventory Model – Inventory model in which demand fluctuates throughout time
period and it is described as a random variable.

Production Lot Size – Amount of items being produced in one cycle.

Production Rate – Number of items being possibly produced within specific time period.

Quantity Discount – A discount on the unit purchasing cost offered by supplier to a buyer willing
to buy in large lots.

Reorder Point – The inventory level at which a new order is placed.

Safety Stock – Inventory maintained specifically to reduce shortages.


Service Level – The probability with which demand is met within the inventory cycle. The
percentage of time that all demand is met on request.

Setup Cost – Fixed cost associated with preparation of a production run (lot).
Shortage (Stockout) – Inability to provide the units from stock. Available inventory is insufficient
to meet demand.

Shortage (Stockout) Cost – Variable cost associated with shortage of inventory.

Surplus – Available inventory exceeds demand.

Unit Purchasing Cost – Variable cost associated with purchasing a single unit of the inventory
(unit price).
WAITING LINE MODELS

In this topic we deal with the real-life situations in which customers come to service
system with their requirements in order to be fully satisfied. In general, both the customers and
the servers in service systems may be people, machines, jobs etc. In the following table you can
find several examples of such situations.

Service System Customer Server


Doctor’s Consultancy Room Patient Doctor
Service Shop Machine Mechanic
Bank Client Clerk
Crossing Car Traffic Light
Telephone Exchange Call Switch Board
Airport Airplane Runway
Airport Passenger Passport Control
First – Aid Station Car Accident Ambulance
Fire Station Fire Emergency Unit
Service Station Car Petrol Pump
Restaurant Consumer Seat
World Wide Web PC Station Web Server
PC Station Job Control Unit
Table 5.1 Examples of Waiting Line Systems

Because of server’s limited capacity, customers may wait until the service of the customers who
came before is finished. Since, in presented systems, queues are typically formed, the waiting line
models are sometimes called queuing models and the used theory is called queuing theory.

Introduction to Queuing Theory

As mentioned above, in each waiting line system two basic elements are distinguished: a
customer and a server (service facility). Figure 5.1 shows the general structure of a queuing system.

Figure 5.1 Waiting Line System

The source represents the group of potential customers (e.g. the whole population). When
some of them require a service, they become real customers who enter the system through the
arrival process. If the server providing the required service is available, the customer is served
immediately. Otherwise, the customer has to wait in a queue until the server finishes the service
of the previous customers. After finishing the service, the customer exits the system.

The Arrival Process

The source of customers might be either finite of infinite (unlimited). Actually, no source is really
infinite, however we may consider the source as infinite (for the system) if it is so large that the
probability of a second arrival is not significantly changed by the first arrival. As the example of
such customers we can mention e.g. the visitors of Prague Castle. The finite source may be, for
example, the enginery in a factory (limited number of machines). Customers may enter the system
either in batches (goods’ delivery, passengers at the airport gate, castle visitors, etc.) or
individually (patients, cars in a car-wash line, train arrivals, etc.).

Arrivals of customers may be scheduled (trams coming to a tram stop) or unscheduled


(patients requiring the emergency aid). A time period between arrivals of two sequential customers
is called the inter - arrival time. If customers’ arrivals are monitored through any time unit (hour,
day, etc.), the average inter - arrival time can be counted. The reciprocal value of the average inter
- arrival time is the average arrival rate (the average number of arrivals per time unit). In case of
unscheduled arrivals, the average arrival rate is a random variable, mostly described by a
probability distribution.

For appropriate description of the arrival process the Poisson discrete probability
distribution may be used. It describes the average number of customers coming per specific time
unit (e.g. per hour). The mean of this probability distribution is denoted λ. The inter - arrival times
then follow the continuous exponential distribution with the mean 1/λ. Both distributions are
possible to be used in case of independent arrivals of the customers, i.e. if the arrival of a customer
is independent of the arrival of previous customers and simultaneously does not significantly affect
the future arrivals.

The Service Process

Similarly to the inter - arrival time between coming customers, we are interested in the
time the customer spends at the service facility, called service time, which may be either
deterministic or probabilistic. In probabilistic (fluctuating) case, the length of the service action
mostly follows the exponential distribution. The average service time is usually designated by
mean time 1/μ, where μ determines the average service rate (it measures the average server’s
capacity – number of customers being possibly served per time unit). From the above
consideration, it ensures that the service rates follow the Poisson distribution.

The service part of the waiting lines system (Figure 5.1) is determined by the type, number and
arrangement of the service facilities. We offer basic possibilities of the service configurations:

1. Single facility. Only one server is available to serve the customers (e.g. dentist’s chair, shop
assistant behind the shop board).
Figure 5.2 Single Service Facility

2. Multiple, parallel, identical facilities. In the service system a customer can find several
servers providing identical service within identical average service time. Either single
queue or multiple queues may be considered in such systems. Whereas in case of single
queue (Figure 5.3) the customer mostly has to respect the rule: go to the first available
server (e.g. bank counters, shop assistants behind the shop board, ambulance), in case of
multiple queues (Figure 5.4) customers are allowed to choose the server (e.g. with the
shortest queue) by themselves (petrol pumps in a gasoline station, checkout counters in a
supermarket, carriageway at traffic lights).

Exit
Arrival

Queue Servers

Figure 5.3 Parallel, Identical Service Facilities (Single Queue)

Figure 5.4 Parallel, Identical Service Facilities (Multiple Queues)


3. Multiple, parallel, but not identical facilities. Although in many situations servers provide
the same service (in terms of satisfying identical needs), they may differ, for example, in
the value of the average service time (e.g. express and regular mechanics) or in the method
of payment (cash, credit card, cheque), in the way of service (declare or nothing to declare
in the airport), etc. In such situations the customers must (or may) choose the server (queue)
and therefore multiple queues are typically formed.
Figure 5.5 Parallel, Non - Identical Service Facilities (Multiple Queues)

4. Serial facilities. Customer must go through all the facilities arranged in a series. The
servers, of course, are seldom identical. Good example is the tour of the patient with the
broken leg to be treated in the policlinic. First, he must wait for a doctor who sends him to
the X-ray photography. Then the patient waits both for taking photography and for making
it. When he turns back to the doctor he may wait until the previous patient is served, and
after the doctor decides that the leg must be fixed in the plaster, patient is taken to the
special department for making the plaster bandage. In case the patient cannot walk, he must
wait for the ambulance car to take him home. On the tour the patient goes through many
service facilities and must queue quite often.

Figure 5.6 Serial Service Facilities

5. Combination of facilities. The basic configurations described above may be combined into
a complex structure of service facilities. If, in the previous example of the patient’s tour,
the whole hospital were considered instead of the single orthopedics, the complexity of
such waiting line system would be even more evident.

The Waiting Line

Waiting area (Figure 5.1) is the part of the system where customers must spend some time
when the service facility is busy. After finishing the service of the customer coming before, next
customer leaves the waiting area and proceeds to the server. In order to make clear who is the next,
we must understand the discipline of the queue being formed in the waiting area:

1. FCFS (First-come, first-served). Customers are served in the same order they have arrived,
i.e. the first-coming customer is served the first from all who are waiting in the queue. An
alternative name of this queue discipline is FIFO (First-in, first-out). This rule is typical
for most of the waiting line systems (shops, banks, hospitals, tram stops, printers, traffic
lights, service stations, some assembly lines, etc.) and it is assumed in the models described
in the following sections.
2. LCFS (Last-come, first-served), or LIFO (Last-in, first-out). Last arrivals are served first.
This queue discipline can be found, for example, in warehouses or in production systems,
where it enables to reduce handling and transportation.

3. PRI (a priority system). In many situations customers may use the advantage of the priority.
It is a predefined rule that determines the order in which customers proceed to the server.
For example, handicapped or old people, children and women may be selected to precede
the young men at fire escape. People living in dangerous areas may have higher priority to
the others in case of vaccination against icterus. In many situations, V.I.P. (Very Important
People) have the highest priority.

In emergency (preemptive priority) systems, the importance of an arriving customer may


be so high that the service process of a less important customer can be even interrupted.
For example, if a serious problem occurs, the doctor-expert may be withdrawn from the
less important job. Inspector will probably interrupt questioning the booster after the bank
robbery has been suddenly announced.

If people with exactly the same level of priority are proceeding to the server (e.g. the same-
aged children), a further rule must be used for the selection of the first of them (e.g. FCFS).

4. SIRO (Selection in random order). There is no predefined order of service, i.e. customers
are selected randomly. For example, at the press conference the chairperson (mostly)
selects the journalists’ questions without any specific rule, passengers on sinking boat must
draw lots to decide the order to embark the lifeboats, etc.

Analysis of Waiting Line Models

The structure and attributes of the waiting line systems have been discussed. However,
managers are interested in optimizing the system’s behavior. In order to be able to expertise the
quality of the system (e.g. to find weaknesses in the waiting line system), the manager requires the
essential cost and other characteristics describing performance of the model.

Cost consideration

Similarly to other managerial situations, the objective of decision making in waiting line
models is to minimize the total cost. Although the manager does not operate on the customers’
side, he should be interested in their satisfaction with the service process. The increasing service
level may lead to an increase in profit, or to a decrease in waiting cost. From this point of view
the successful manager determines the number of servers to assure full satisfaction of the
customers. However, this decision sharply increases the service cost. For example, if there are too
long queues in a supermarket in front of the checkers, customers would prefer an alternative
supermarket where they would be served without waiting. In this case the managers should
evaluate the profit loss and consider a decision of hiring additional checkers to reduce waiting
time. Of course, additional wages (fixed cost) are connected with this decision.
The service cost (facility cost) generally includes:

 Cost of construction (capital investment). When, e.g. a new production line is being
constructed, all the fixed cost connected with this project must be considered.

 Cost of operation. This variable cost includes material, labor and energy cost required for
operations (service providing). In case of the mentioned production line the company must
consider operator’s wages, and power or fuel consumption.

 Cost of maintenance and repair. Maintenance of the production line and all its repairs must
be calculated.

 Other costs. For example, insurance and rental of space for the production line must be
paid. Similarly to the inventory models, the compromise must be found in waiting line
models as well.

Time characteristics

When a customer is making a decision on entrance the waiting line system he may be interested
in the total time he would spend in the system. Because of probabilistic character of the models
only estimations or average values can be calculated. Two basic time measures of the system’s
performance are:

 average waiting time in the queue,


 average waiting time in the system (time spent waiting for the service and being served).
Of course, not only customers are interested in these two characteristics, but also managers who
are responsible for the optimal performance of the system.

Number of customers

There are two other important estimations, evaluating the behavior of the analyzed system:

 average number of customers in the queue,


 average number of customers in the system (i.e. customers in the queue and in the service
process).

Probability characteristics

 probability of an empty service facility is the probability that there are no customers in the
system (the facility is idle),
 probability of the service facility being busy equals the probability that an arriving
customer has to wait for service (e.g. in case of system with the single server it is the
probability of not finding an empty system); this measure refers to the utilization of the
service facility,
 probability of finding exactly N customers in the system (waiting in the queue or being
served),
 probability that the number of customers in the system (N) will be larger than a specified
number (n),
 probability of being in the system longer than time t; this characteristic might be, from the
customer’s point of view, rather important in the situation where several alternative
services are being offered.

Basic Waiting Line Models

As the computation of all the measures mentioned in the previous section depends on the
structure of the analyzed waiting line system, we will first outline the models classification in this
section. Then we will present two basic waiting line models: standard single-server model and
standard multi-server model.

Classification of Waiting Line Models

The Kendall’s notation uses the standard sequence of symbols for the waiting line models
classification:
A/B/C/D/E/F

Each of the six positions in the pattern corresponds to one of six model’s attributes. All together
enable perfect description of basic waiting line systems. The meaning of the attributes is following:

A describes the arrival process, i.e. the type of the probability distribution corresponding to the
inter - arrival time. For example, symbol M in this position is used for the exponential distribution,
D for constant time period, U for the uniform probability distribution, N for the normal distribution
etc.

B refers to the probability distribution of the service time. Exactly the same symbols are used as
for the distributions of the inter - arrival time.

C determines the number of parallel servers.

D is used for the queue discipline (FCFS, LCFS, PRI, SIRO).

E is the maximum length of queue. If there is no limit for the queue, symbol ∞ is used, otherwise
given number n is written in this position.

F specifies the size of the customers’ source. For the infinite source we use ∞, whereas in case of
the finite source we write the given number n.

Using the Kendall’s notation we offer several examples of the most common waiting line systems:

M/M/1/FCFS/∞/∞ standard single-server model,


M/M/K/FCFS/∞/∞ standard multi-server model,
M/M/1/PRI/∞/∞ priority service, single-server model,
M/M/1/FCFS/n/∞ limited queue, single-server model,
M/M/K/FCFS/∞/n finite source, multi-server model.

Note: If only A/B/C notation is used for the model, it means that FCFS queue discipline, no limit
for the queue and infinite source are assumed.

Standard Single-Server Exponential Model M/M/1

The simplest model (shown in Figure 5.2) among waiting line models is predicated on the
following assumptions:

Assumptions of M/M/1 model:


 waiting line has a single server,
 inter - arrival times are described by exponential probability distribution with the mean =
1/λ,
 service times follow exponential probability distribution with the mean = 1/μ,
 infinite source,
 unlimited length of queue,
 the queue discipline is FCFS.

As described in Section 5.1, all waiting line models can be characterized by several measures
of performance. We do not present the derivation of all the formulas because of its complicacy.
We use the following example to illustrate the calculation of the measures.

For example, In the small grocery store there is only one shop board with only one shop assistant.
In the period from 8 a.m. to 6 p.m. 18 customers per hour (on the average) come to the grocery.
The assistant is able to serve (on the average) 25 customers per hour. The assistant’s salary is
15,000 CZK per month. The grocery’s owner tested another assistant who was able to serve 29
customers per hour. Because of faster service in this case 20 customers per hour (on the average)
came to the grocery. However, this assistant requires the salary of 22,000 CZK per month. The
average profit is 50 CZK per sale. The grocery’s owner considers hiring the faster assistant instead
of the slower one.

Solving the above example, since it will be quite interesting to compare both alternatives to each
other step by step, we will calculate all the characteristics simultaneously. All non - integer values
are rounded to appropriate number of decimal figures.

1. Definition of input parameters

The average arrival rate λ


The number of coming customers follows the Poisson probability distribution (the inter -
arrival times are described by the exponential distribution). Whereas in the first alternative
the average arrival rate is 18 customers per hour, in the second alternative it is 20
customers per hour.

Slow Assistant Fast Assistant


λ 18 20

The average service rate c


The service times follow the exponential probability distribution. The average service rate
is the reciprocal value of the average service time. In the example, the average service rates
are 25 and 29 customers per hour.

Slow Assistant Fast Assistant


μ 25 29

The salary
The monthly salaries of the assistants are 15 000 and 22 000 CZK.

Slow Assistant Fast Assistant


S 15,000 22,000

The sale profit


The average sale profit is the same for both the assistants: SP = 50 CZK per sale.

2. Calculating measures of performance

The utilization of the system


This characteristic is calculated as the ratio of the average arrival rate and the average
service rate:

ρ=λ/μ (5.1)

If ρ ≥ 1, the length of the queue would increase without any bound. Therefore only the
assumption of ρ < 1 is acceptable to management.

Slow Assistant Fast Assistant


ρ 0.72 0.69

The utilization of the system is the probability that the server is busy, or the probability
that there is at least one customer in the system.

The probability of an empty facility (server is idle)


Considering this value as the probability that there are no customers in the system, we can
use Equation 5.1:

P (0) = 1 – ρ (5.2)

In the example, the values are computed as follows:

Slow Assistant Fast Assistant


P (0) 0.28 0.31

The average waiting time in the system


This is an estimation of time a customer spends in the system (waiting for the service and
being served):
W = 1 / (μ – λ) (5.3)

Slow Assistant Fast Assistant


W (in hours) 0.143 0.111
W (in minutes) 8.6 6.7

This characteristic may affect the customers in their decision to enter the grocery. In case
of the faster assistant the customers save (on the average) almost 1/4 of their time (spent in
the grocery) in comparison with the slower assistant.

The average waiting time in the queue


This is the average time a customer waits in the queue before the service starts:

Wq = W – (1/ μ) = [λ / μ(μ – λ)] (5.4)

It is quite easy to calculate this measurement, if we consider that the average waiting time
in the system W is the average waiting time in the queue Wq, plus the average service time
1/μ.
Slow Assistant Fast Assistant
Wq (in hours) 0.103 0.077
Wq (in minutes) 6.2 4.6
Whereas in the first situation the customers wait in the queue more than 6 minutes (on the
average), in the second situation they wait less than 5 minutes.

The average number of customers in the system


Generally, all the customers in the system are those being served together with those who
are waiting in the queue. The number of customers in the system can be calculated as the
average arrival rate (the number of arrived customers per time unit), multiplied by the
average waiting time (average time that the arrived customers spend in the system):

L = λW = λ / (μ - λ) (5.5)

Slow Assistant Fast Assistant


L 2.57 2.22

The average number of customers in the queue


Similarly to the previous calculation, the number of customers in the queue is the average
arrival rate (the number of arrived customers per time unit), multiplied by the average
waiting time in the queue:

Lq = λWq = λ2 / [μ (μ – λ)]
Slow Assistant Fast Assistant
Lq 1.85 1.53

The probability of finding exactly N customers in the system


When N customers are in the system, one customer is being served and (N-1) customers
are waiting in the queue.

P(N) = P(0) ρN = (1 – ρ) ρN (5.7)

In the following table we present the probabilities of finding 0, 1, 2, 3, 4, 5 customers in


the grocery:

Slow Assistant Fast Assistant


P(0) 0.280 0.310
P(1) 0.202 0.214
P(2) 0.145 0.148
P(3) 0.105 0.102
P(4) 0.075 0.070
P(5) 0.054 0.048

The probabilities P (0), P (1) and P (2) are lower for the slow assistant, since in this case
fewer customers come to the grocery. However, further probabilities P (3), P (4) and P (5)
are lower for the fast assistant because of his perfect service ability (Figure 5.7).

Figure 5.7 The probability of finding N customers in the system


The probability that the number of customers in the system (N) will be larger than a
specified number of customers (n)

P{N>n} = ρn+1 (5.8)


Similarly to the previous characteristic we draw the following table:

Slow Assistant Fast Assistant


P(N > 0) 0.720 0.690
P(N >1) 0.518 0.476
P(N >2) 0.373 0.328
P(N >3) 0.269 0.226
P(N >4) 0.193 0.156
P(N >5) 0.139 0.108

The probability P{N > 0} is the probability that the system is not empty, i.e. that the server
is busy. With the probability P{N > 0} the coming customer will be waiting in the queue.

The probability of being in the system longer than specified time (t)
Time spent in the system (T) consists of the waiting time in the queue and the service time.
The probability that this time is greater than specified time (t) can be expressed as follows:

P{T > t} = e( λ - μ)t

In the following table the probabilities (5.9) are computed for several specified times.

Slow Assistant Fast Assistant


P{T > 1 min} 0.890 0.861
P{T > 2 min} 0.792 0.741
P{T > 3 min} 0.705 0.638
P{T > 4 min} 0.627 0.549
P{T > 5 min} 0.558 0.472
P{T > 1 min} 0.890 0.861

Note: Although all times in the table are expressed in minutes, the calculation of (5.9) must
respect the chosen time unit (hour, in our case). Thus, parameter t in the formula must be
1/60, 2/60, etc.

3. Cost and profit analysis

If the grocery’s owner considers the replacement of the slower assistant, he should be
interested in the number of sales, which determines the daily profit. Considering 22
working days within a month (each day has 10 working hours), simple formula can be used
for the calculation of total monthly profit:

Where:
TP … the total monthly profit,
λ … the average number of customers coming to the grocery per hour,
SP … the average profit per sale.
In the following table monthly assistants’ salary S can be found together with the calculated
total monthly profit TP

Slow Assistant Fast Assistant


S 15,000 CZK 22,000 CZK
TP 198,000 CZK 220,000 CZK

In summary, The grocery’s owner should hire the fast assistant, as he increases the monthly profit
of 22,000 CZK and required salary will be higher only by 7,000 CZK. Although the cost and
profit analysis is quite simple, several interesting measures of performance have been calculated.
We present some of them:

Characteristics Slow Assistant Fast Assistant


Utilization of the grocery 0.720 0.690
Probability that the assistant is idle 0.280 0.310
Average waiting time 8.6 min 6.7 min
Average waiting time in the queue 6.2 min 4.6 min
Average number of customers in the grocery 2.57 2.22
Average number of customers in the queue 1.85 1.53
Probability of being in the grocery more than 1 min 0.890 0.861

As it is apparent from the table, the fast assistant improves in total the performance of the grocery
and its attractiveness for the customers.

Standard Multi-Server Exponential Model M/M/K

In the previous model the single server has been assumed. However, many waiting line
systems (e.g. petrol station, bank, supermarket) have more than one server (e.g. petrol pumps,
counters). For this purpose, we add in the waiting line model further parallel servers providing the
identical service. In such a model, the following assumptions are:

Assumptions of M/M/K model:


� system has K parallel, identical servers,
� one waiting line (queue) exists,
� inter - arrival times are described by exponential probability distribution with the
mean = 1/λ,
� service times of each server follow exponential probability distribution with the
mean = 1/μ,
� infinite source,
� unlimited length of queue,
� the queue discipline is FCFS.
The assumption of the only one queue for all servers corresponds to the model in Figure 5.3.
Customers who wait in the queue move to the first available service facility. This situation is
typical for bank counters, shop assistants, seats in a restaurant, etc.

As μ is the average service rate for each server, Kμ is the average service rate for the system.
Similarly to the single-server model, the constraint λ/(Kμ) < 1 prevents the system from creating
the infinite queue.

Suppose, we modify the previous example. Suppose now that the grocery’s owner
considers employing two assistants instead of one. Although the shop board is divided into two
separate parts, the only one waiting area is in the shop. The customers proceed from the queue to
the first available assistant. The second assistant is supposed to work at the same ability level with
the first shop assistant, i.e. he is able to serve 25 customers per hour. Therefore his salary is 15 000
CZK per month as well. Because of the faster (doubled) service more people come to the grocery.
The estimation of the average arrival rate is 30 customers per hour.

To solve the given problem, the derivation of all equations is, of course, more complex for
the multiple-server model than for the single-server model, and the equations themselves are more
complex at all. We will calculate all the values for the two-assistant grocery and in the summary
we will compare them with the results from the single-assistant model.

1. Definition of input parameters

The average arrival rate λ


The estimation of this value is λ = 30 customers per hour.

The average service rate μ


This value is the average number of customers that each assistant is able to serve per one hour.
Ensuing from the definition of the problem, the average service rate is μ = 25 customers per
hour.

The number of the assistants K


As in the grocery two assistants should be working, K = 2.

The salary
Since the monthly salary of one shop assistant is 15,000 CZK, the total salary of two assistants
is 30,000 CZK.

One (1) Assistant Two (2) Assistants


S 15,000 30,000

The sale profit


The average sale profit is the same both for the single-assistant and for the two-assistant
grocery: SP = 50 CZK per sale.
2. Calculating measures of performance

The average service rate of the system Kμ


This characteristic is the average number of customers that the system (all the servers) is able
to serve per time unit.

In the example, two assistants (together) can serve Kμ = 2(25) = 50 customers per hour.

The utilization of the single-server system


This measure is the ratio of the average arrival rate and the average service rate of the server:

Ρ=λ/μ (5.10)

The utilization is ρ = 1.2.

The utilization of the system


This characteristic can be expressed as the ratio of the average arrival rate and the average
service rate of the system:

(5.11)

The utilization of the grocery is ρ = 0.6 . Since this value is less than 1, formation of the
infinite queue is avoided.

The utilization of the system is the probability that all the servers are busy, or the
probability that there are at least K customers in the system.

The probability of finding no customers in the system (all servers are idle)

(5.12)

The probability of finding empty grocery is P(0) = 0.250.

The probability of finding exactly N customers in the system


Three possible situations may occur in the system:

1. N < K. All customers are being served and (K – N) servers are idle.
2. N = K. All customers are being served and all servers are busy.
3. N > K. All servers are busy and (N – K) customers are waiting in the queue.
Hence, two different formulas are derived for this characteristic:
(5.13)
.

The following table presents the probabilities of finding 0, 1, 2, 3, 4, 5 customers in the


grocery:

Slow Assistant
P(0) 0.250
P(1) 0.300
P(2) 0.180
P(3) 0.108
P(4) 0.065
P(5) 0.039

Note: The probability of waiting in the queue can be expressed, in the example, as follows:

P{N > 1}= 1− (P(0) + P(1))

Thus P{N > 1} = 1 – (0.250 + 0.300) = 0.450.\

The average number of customers in the queue

(5.14)

In the example, the average number of customers in the queue is Lq = 0.675.

The average number of customers in the system

L = Lq + ρ (5.15)

The average number of customers in the grocery is L = 1.875.

The average waiting time in the queue


This is the average time a customer waits in the queue before the service starts:

Wq = (Lq / λ) (5.16)

The customers wait in the queue (on the average) Wq = 0.0225 hours i.e. 1.35 minutes.

The average waiting time


This is the estimation of time a customer spends in the system (waiting for the service and
being served):

(5.17)

The customers spend in the grocery (on the average) W = 0.0625 hours i.e. 3.75 minutes.

3. Cost and profit analysis

Similarly to Example 5.1 we present the calculation of cost and profit both for the single-
assistant model and for the two-assistant model. The total monthly profit can be expressed as
follows:

Where:
TP … the total monthly profit,
λ … the average number of customers coming to the grocery per hour,
SP … the average profit per sale.

In the following table the monthly salary S can be found together with the calculated total
monthly profit TP:

One (1) Assistant Two (2) Assistants


S 15,000 CZK 30,000 CZK
TP 198 000 CZK 330 000 CZK

Summarizing, regarding the cost and profit analysis, there is no doubt about employing two
assistants instead of one. The comparison of both alternatives in terms of their basic characteristics
seems to be more interesting:

Characteristics One (1) Assistant Two (2) Assistants


Utilization of the grocery 0.720 0.600
Probability that the grocery is empty 0.280 0.250
Average waiting time 8.6 min 3.8 min
Average waiting time in the queue 6.2 min 1.4 min
Average number of customers in the grocery 2.57 1.88
Average number of customers in the queue 1.85 0.68
Probability of waiting in the queue 0.720 0.450

Computer Simulation in Waiting Line Models

In many real situations it is entirely impossible to solve the problem analytically. Even in
simple systems similar to the grocery with two assistants, all the derived equations seem to be
quite complex. Both the structure of a waiting line system and the parameters (e.g. type of the
probability distribution used in the model) have significant impact on the complexity of solution.
When the analytical approaches abort, simulation experiments are the only way how to treat the
managerial problem.

Computer simulation can be defined as a special method using computer experiments with
the model of a real system. Simulation approach can be successfully applied, of course, especially
to complex waiting line models. First, we present a slight introduction into the terminology used
in simulation models.

Entity is an object that goes through the model.


Resource is an agent required by the entity.

In the waiting line models entities correspond to the customers who enter the system in
order to be served and resources are the servers providing the required service.

Event is a significant change of the system.


Activity is a process between two events.

Three basic types of events occur in the waiting line models: a customer’s entry to the
system, a start of the service and a finish of the service. Hence, two different activities run in such
a model: waiting in the queue and the service.

If the behavior of a system is to be simulated, we must be able to generate the events in the
system. For this purpose, it is necessary to know the probability distribution of all the random
variables in the model. Thus, generated random values may correspond, for example, to the inter
- arrival times or service times. Simulation time is the real time period throughout which the
simulated system’s performance is being monitored.

The most widespread computer simulation languages are GPSS/H, SLX, SIMSCRIPT, SLAM,
SIMPLE++, AWESIM, etc. Simulation software has, as the basic component, a built-in statistics
collection and printout of the results. In addition, cost analysis enables to optimize the system.
Many software packages, especially the tutorial products, provide the graphical animation of the
running simulation.

Since the computer simulation is just an experiment with the model, the obtained results
differ from the values being calculated analytically (of course, in case the analytical solution exists
at all). The difference between the analytical values and the values estimated from the simulation
run become generally lower considering the longer simulation time.

For example, the simulation model has been developed for the example with two assistants
in the grocery. Comparison of the results calculated analytically in the previous example and the
estimations obtained in the simulation run is presented in the following table (we have simulated
one working month). It is evident that the computer simulation offers perfect approximation of the
grocery’s performance.
Characteristics Analytical Results Simulation
Utilization of the grocery 0.600 0.612
Average waiting time 3.8 min 3.8 min
Average waiting time in the queue 1.4 min 1.3 min
Average number of customers in the grocery 1.88 1.90
Average number of customers in the queue 0.68 0.67
Total monthly profit 330,000 CZK 333,250 CZK

Glossary
Arrival Rate – The average number of customers arriving in the waiting line system per unit of
time.
Computer Simulation – Numerical technique using a computer experiment with the model of a
real system.
Customer – An object entering the system with its service requirements.
Interarrival Time – The time between two consecutive arrivals.
Queue (Waiting Line) – One ore more customers waiting for service.
Server (Service Facility) – An object in the system providing service for entering customers.
Service Rate – The average number of customers being served in the waiting line system per unit
of time.
Service Time – The time required for completion of a service.
Source of Customers – A group of potential customers who can enter the system.
Utilization of the System – The fraction of time the server is busy (computed as the ratio of the
arrival rate to the service rate).
Waiting Line Models – Models describing systems with service facilities being required by
arriving customers.
Because of the server’s limited capacity, queues occur in the system.
References:

Concepts, Theories and Examples retrieved from: Fabry, J. (2003). Management Science.
University of Economics Prague pdf

[1] Anderson, D. R., D. J. Sweeney, and T. A. Williams. An Introduction to Management


Science: Quantitative Approaches to Decision Making, 7th edition, West Publishing,
Minneapolis 1994.

[2] Anderson, D. R., D. J. Sweeney, and T. A. Williams. Quantitative Methods for


Business, 6th edition, West Publishing, Minneapolis 1995.

[3] Dlouhý, M. Simulace pro ekonomy, VŠE v Praze 2001.

[4] Eppen, G. D., F. J. Gould, and C. P. Schmidt. Introductory Management Science, 4th
edition, Prentice Hall, Englewood Cliffs, New Jersey 1993.

[5] Jablonský, J. Operační výzkum: Kvantitativní modely pro ekonomické


rozhodování, Professional Publishing, Praha 2002.

[6] Markland, R. E., and J. R. Sweigart. Quantitative Methods: Applications to Managerial


Decision Making, John Wiley, New York 1987.

[7] Taha, H. A. Operations Research: An Introduction, 3rd edition, MacMillan Publishing,


New York 1982.

[8] Turban E., and J. R. Meredith. Fundamentals of Management Science, 4th edition,
BPI/Irwin, Homewood 1988.

[9] Winston, W. L. Operations Research: Applications and Algorithms, 3rd edition,


Wadsworth Publishing, Belmont 1994.

You might also like