Unit 1 Introduction to Operations Research Structure: 1.1 Introduction Objectives 1.2 Historical Background Definitions of operations research 1.3 Scope of Operations Research 1.4 Features of Operations Research 1.5 Phases of Operations Research 1.6 Types of Operations Research Models A broad classification of OR models 1.7 Operations Research Methodology Definition Construction Solution Validation Implementation 1.8 Operations Research Techniques and Tools 1.9 Structure of the Mathematical Model 1.10 Limitations of Operations Research 1.11 Summary 1.12 Glossary 1.13 Terminal Questions 1.14 Answers 1.15 Case Study 1.1 Introduction Welcome to the unit on operations research management. Operations research management focuses on the mathematical scoring of consequences of a decision aiming to optimise the use of time, effort, and resources to avoid blunders. The act of obtaining best results under any given circumstances is known as optimising. The key purpose of Operations Research (OR) is to do preparative calculations that aid the decision-making process. Operations Research Unit 1 Sikkim Manipal University Page No. 2 Hence, decision-making is a key part of our daily life. The ultimate goal of all decisions is to maximise benefits and to minimise effort and time. OR gives decision makers the power to make effective decisions and improve day-today operations. Decision makers consider all the available options, study the outcomes, and estimate the risks. In simple situations, common sense and judgement can be used to take decisions. For example, if you are buying a microwave or washing machine, the decision-making process is not very complicated. You can simply compare the price, quality, and durability of the well-known brands and models in the market and take a decision based on it. However, in complex situations, although it is possible to take decisions based on ones common sense, a decision backed by mathematical calculations reduces the risk factor and increases the probability of success. Some such situations, where decision-makers have to depend on mathematical scoring and reasoning, are finding an appropriate product mix amidst competitors products or planning a public transportation network in a city. Objectives: After studying this unit, you should be able to: - describe the historical background of OR - list the significant features of OR - describe the methodology of OR - define the structure of a mathematical model in OR - describe the significance of the function of OR 1.2 Historical Background During the World War II, scientists from United Kingdom studied the strategic and tactical problems associated with air and land defence of the country. The aim of this study was to determine the effective utilisation of limited military resources to win the battle. The technique was named operations research. After World War II, operations research techniques were developed and deployed in the decision-making process in complicated situations in various fields, such as industrial, academic, and government organisations. Operations Research Unit 1 Sikkim Manipal University Page No. 3 1.2.1 Definitions of operations research Churchman, Aackoff, and Aruoff defined operations research as the application of scientific methods, techniques and tools to the operation of a system with optimum solutions to the problems where 'optimum' refers to the best possible alternative. The objective of OR is to provide a scientific basis to the decision-makers for solving problems involving interaction with various components of the organisation. This can be achieved by employing a team of scientists from different disciplines to work together for finding the best possible solution in the interest of the organisation as a whole. The solution thus obtained is known as an optimal decision. You can also define operations research as The use of scientific methods to provide criteria for decisions regarding man, machine, and systems involving repetitive operations. Self Assessment Questions 1. The main objective of OR is to provide a _______ ________ to the decision-makers. 2. OR employs a team of _________ from _________ __________. 1.3 Scope of Operations Research Any problem, either simple or complicated, can use OR techniques to find the best possible solution. This section will explain the scope of OR by analysing its application in various fields of everyday life. - In defence operations In modern warfare, the three major military components namely, Air Force, Army, and Navy carry out the defence operations. The activities in each of these components can be further divided in four sub-components - administration, intelligence, operations, training and supply. The applications of modern warfare techniques in each of the components of military organisations require expert knowledge in respective fields. Furthermore, each component works to drive maximum gains from its operations and there is always a possibility that the strategy beneficial to one component may be unfeasible for another component. Thus in defence operations, there is a requirement to co-ordinate the activities of various components. This Operations Research Unit 1 Sikkim Manipal University Page No. 4 gives maximum benefit to the organisation as a whole, having maximum use of the individual components. A team of scientists from various disciplines gets together to study the strategies of different components. After appropriate analysis of the various courses of actions, the team selects the best course of action, known as the optimum strategy. - In industry The system of modern industries is so complex that an individual cannot intuitively judge the optimum point of operation in its various components. The business environment is always changing and any decision useful at one time may not be suitable some time later. There is always a need to check the validity of decisions continuously against the situations. The industrial revolution with increased division of labour and introduction of management responsibilities has made each component an independent unit having its own goals. For example, production department minimises the cost of production but maximises output. Marketing department maximises the output, but minimises cost of unit sales. Finance department tries to optimise the capital investment and personnel department appoints good people at minimum cost. Thus, each department plans its own objectives and all these objectives of various departments or components come to conflict with one another and may not agree to the overall objectives of the organisation. The application of OR techniques helps in overcoming this difficulty by integrating the diversified activities of various components to efficiently serve the interest of the organisation as a whole. OR methods in industry can be applied in the fields of production, inventory controls and marketing, purchasing, transportation, and competitive strategies. - Planning In modern times, it has become necessary for every government to carefully plan, for the economic development of the country. OR techniques can be fruitfully applied to maximise the per capita income, with minimum sacrifice and time. A government can thus use OR for framing future economic and social policies. - Agriculture With increase in population, there is a need to increase agriculture output. However, this cannot be done arbitrarily. There are several restrictions. Hence, the need to determine a course of action that serves the best under the given restrictions. You can solve this problem by applying OR techniques. Operations Research Unit 1 Sikkim Manipal University Page No. 5 - In hospitals OR methods can solve waiting problems in outpatient department of big hospitals and administrative problems of the hospital organisations. - In transport Different OR methods can be applied to regulate the arrival of trains and processing times, to minimise the passengers waiting time and reduce congestion, and to formulate suitable transportation policy, thereby reducing the costs and time of transshipment. - Research and development OR methodologies can be applied in the field of R&D for several purposes, such as to control and plan product introductions. Self Assessment Questions 3. A government can thus use OR for framing future ______ and _______ 4. In hospital OR methods can solve waiting problems in ______ department of big hospitals and ______ problems of the hospital organisations. 1.4 Features of Operation Research Some key features of OR are as follows: - OR is system-oriented. OR scrutinises the problem from an organisations perspective. The results can be optimal for one part of the system, while the same can be unfavourable for another part of the system. - OR imbibes an interdisciplinary team approach. Since no single individual can have a thorough knowledge of all the fast developing scientific know-how, personalities from different scientific and managerial cadre form a team to solve the problem. - OR uses scientific methods to solve problems. - OR increases effectiveness of the managements decision-making ability. - OR uses computers to solve large and complex problems. - OR offers a quantitative solution. - OR also takes into account the human factors. Operations Research Unit 1 Sikkim Manipal University Page No. 6 Self Assessment Questions 5. OR ________ inter-disciplinary approach. 6. OR increases the effectiveness of ________ ability. 1.5 Phases of Operations Research The scientific method in OR study generally involves three phases. Figure 1.1 depicts the three phases of OR. Fig. 1.1: Phases of Operations Research Let us now study the phases in detail. Judgment phase This phase includes the following activities: - Determination of the operations - Establishment of objectives and values related to the operations - Determination of suitable measures of effectiveness - Formulation of problems relative to the objectives Research phase This phase utilises the following methodologies: - Operation and data collection for a better understanding of the problems - Formulation of hypothesis and model - Observation and experimentation to test the hypothesis on the basis of additional data - Analysis of the available information and verification of the hypothesis using pre-established measure of effectiveness - Prediction of various results and consideration of alternative methods Operations Research Unit 1 Sikkim Manipal University Page No. 7 Action phase This phase involves making recommendations for the decision process. The recommendations can be made by those who identify and present the problem or by anyone who influences the operation in which the problem has occurred. Self Assessment Questions 7. Action phase involves making recommendations for the decision process. (True/False) 8. One of the OR phases is judgement phase. (True/False) 1.6 Types of Operations Research Models A model is an idealised representation or abstraction of a real-life system. The objective of a model is to identify significant factors that affect the reallife system and their interrelationships. A model aids the decision-making process as it provides a simplified description of complexities and uncertainties of a problem in a logical structure. The most significant advantage of a model is that it does not interfere with the real-life system. 1.6.1 A broad classification of OR models You can broadly classify OR models into the types depicted in figure 1.2. Fig. 1.2: Classification of Models Operations Research Unit 1 Sikkim Manipal University Page No. 8 Let us now study the models in detail. Physical models These models include all forms of diagrams, graphs, and charts. They are designed to tackle specific problems. They bring out significant factors and interrelationships in pictorial form to facilitate analysis. There are two types of physical models. They are: - Iconic models - Analogue models Let us now study the two types of physical models in detail. Iconic models are primarily images of objects or systems, represented on a smaller scale. These models can simulate the actual performance of a product. Analogue models are small physical systems having characteristics similar to the objects they represent, such as toys. Mathematical or symbolic models These models employ a set of mathematical symbols to represent the decision variable of the system. The variables are related by mathematical systems. Some examples of mathematical models are allocation, sequencing, and replacement models. By nature of environment These models can be further classified as follows: - Deterministic models - These are the models in which everything is defined and the results are certain, such as an EOQ model. - Probabilistic models - These are the models in which the input and output variables follow a defined probability distribution, such as the games theory. By the extent of generality These models can be further classified as follows: - General models These are the models which you can apply in general to any problem. For example, linear programming. - Specific models - These are the models that you can apply only under specific conditions. For example, you can use the sales response curve or equation as a function in the marketing function. Operations Research Unit 1 Sikkim Manipal University Page No. 9 Self Assessment Questions 9. Diagram belongs to physical models. (True/False) 10. Allocation problems are represented by iconic models. (True/False) 1.7 Operations Research Methodology The basic dominant characteristic feature of operations research is that it employs mathematical representations or models to analyse problems. This distinct approach represents an adaptation of the scientific methodology used by the physical sciences. The scientific method translates a given problem into a mathematical representation which is solved and retransformed into the original context. Figure 1.3 depicts the OR approach to problem solving. Fig. 1.3: Steps in the OR methodology As shown in figure 1.3, OR methodology consists of five steps. They are - defining the problem, constructing the model, solving the model, validating the model, and implementing the result. Let us now study the steps in detail. 1.7.1 Definition The first and the most important step in the OR approach of problem solving is to define the problem. One needs to ensure that the problem is identified properly because this problem statement will indicate the following three major aspects: - Description of the goal or the objective of the study - Identification of the decision alternative to the system Operations Research Unit 1 Sikkim Manipal University Page No. 10 - Recognition of the limitations, restrictions, and requirements of the system 1.7.2 Construction Based on the problem definition, you need to identify and select the most appropriate model to represent the system. While selecting a model, you need to ensure that the model specifies quantitative expressions for the objective and the constraints of the problem in terms of its decision variables. A model gives a perspective picture of the whole problem and helps in tackling it in a well-organised manner. Therefore, if the resulting model fits into one of the common mathematical models, you can obtain a convenient solution by using mathematical techniques. If the mathematical relationships of the model are too complex to allow analytic solutions, a simulation model may be more appropriate. Hence, appropriate models can be constructed. 1.7.3 Solution After deciding on an appropriate model, you need to develop a solution for the model and interpret the solution in the context of the given problem. A solution to a model implies determination of a specific set of decision variables that would yield an optimum solution. An optimum solution is one which maximises or minimises the performance of any measure in a model subject to the conditions and constraints imposed on the model. 1.7.4 Validation A model is a good representation of a system. However, the optimal solution must work towards improving the systems performance. You can test the validity of a model by comparing its performance with some past data available from the actual system. If under similar conditions of inputs, your model can reproduce the past performance of the system, then you can be sure that your model is valid. However, you will still have no assurance that future performance will continue to duplicate the past behaviour. Secondly, since the model is based on careful examination of past data, the comparison should always reveal favourable results. In some instances, this problem may be overcome by using data from trial runs of the system. One must note that such validation methods are not appropriate for non-existent systems because data will not be available for comparison. Operations Research Unit 1 Sikkim Manipal University Page No. 11 1.7.5 Implementation You need to apply the optimal solution obtained from the model to the system and note the improvement in the performance of the system. You need to validate this performance check under changing conditions. To do so, you need to translate these results into detailed operating instructions issued in an understandable form to the individuals who will administer and operate the recommended system. The interaction between the operations research team and the operating personnel reaches its peak in this phase. 1.8 Operations Research Techniques and Tools The different techniques and tools used in OR are as follows: - Linear programming You can use linear programming to find a solution for optimising a given objective. The objective may be to maximise profit or to minimise cost. You need to ensure that both the objective function and the constraints can be expressed as linear expressions of decision variables. You will learn about the various uses of linear programming in Unit 2. - Inventory control methods The production, purchasing, and material managers are always confronted with questions, such as when to buy, how much to buy, and how much to keep in stock. The inventory model aims at optimising these inventory levels. - Goal programming In linear programming, you take a single objective function and consider all other factors as constraints. However, in real life there may be a number of important objective functions. Goal programming has several objective functions, each having a target value. Programming models are developed to minimise deviations from these targets. - Queuing model The queuing theory is based on the concept of probability. It indicates the capability of a given system and the changes possible in the system when you modify the system. In formulating a queuing model, you need not take into account all the constraints. There is no maximisation or minimisation of an objective function. Therefore, the application of queuing theory cannot be viewed as an optimisation process. You can use the queuing theory to estimate the required balance between customer waiting time and the service capability of the Operations Research Unit 1 Sikkim Manipal University Page No. 12 system. You need to first consider several alternatives, evaluate them through queuing models, study their effect on the system, and then make a choice. The criteria for evaluation will be measures of efficiency of the system, such as the average length of a queue, expected waiting time of a customer, and the average time spent by the customer in the system. In this approach, your success primarily depends on the alternatives considered and not much on the queuing models developed. - Transportation model The transportation model is an important class of linear programs. The model studies the minimisation 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 objective of the model is to develop an integral transportation schedule that meets all the demands from the inventory at a minimum total transportation cost. The transportation problem involves m sources, each of which has ai (i = 1, 2, ..,m) units of homogeneous product and n destinations available, and each of which requires bj (j = 1, 2., n) units of products. Here ai and bj are positive integers. The cost cij of transporting one unit of the product from the ith source to the jth destination is given for each i and j. It is assumed that the total supply and the total demand are equal.
= = = n j 1 m i 1 ai bj (1) Condition (1) is guaranteed by creating either a fictitious destination with a demand equal to the surplus if total demand is less than the total supply or a (dummy) source with a supply equal to the shortage if total demand exceeds total supply. The cost of transportation from the fictitious destination to all sources and from all destinations to the fictitious sources are assumed to be zero so that the total cost of transportation will remain the same. - In addition to the above, there are tools such as the sequence model, the assignment model, and network analysis, which you will learn in detail in the later units. Operations Research Unit 1 Sikkim Manipal University Page No. 13 Self Assessment Questions 11. OR methodology consists of definition, solution, and validation only. (True/False) 12. The interaction between the OR team and management reaches peak level in the implementation phase. (True/False) 1.9 Structure of the Mathematical Model Many industrial and business situations are concerned with planning activities. In each case of planning, there are limited sources, such as men, machines, material, and capital at the disposal of the planner. One has to take decisions regarding these resources to maximise production, minimise the cost of production, or maximise the profit. These problems are referred to as the problems of constrained optimisation. Linear programming is a technique for determining an optimal schedule of interdependent activities, for the given resources. Therefore, you can say that programming refers to planning and the process of decision-making about a particular plan of action from a given set of alternatives. Any business activity or production activity to be formulated as a mathematical model can best be discussed through its parts, which are as follows: - Decision variables - Objective function - Constraints Let us now study the parts in detail. Decision variables Decision variables are the unknowns, which you need to determine from the solution of the model. The parameters represent the controlled variables of the system. Objective function The objective function defines the measure of effectiveness of the system as a mathematical function of its decision variables. The optimal solution to the model is obtained when the corresponding values of the decision variable yield the best value of the objective function whilst satisfying all constraints. Therefore, you can say that the objective function acts as an indicator for the achievement of the optimal solution. Operations Research Unit 1 Sikkim Manipal University Page No. 14 While formulating a problem, the desire of the decision-maker is expressed as a function of n decision variables. (This function is a linear programming problem, that is, each of its items will have only one variable raised to the power one). Some of the objective functions in practice are: - Maximisation of contribution or profit - Minimisation of cost - Maximisation of production rate or minimisation of production time - Minimisation of labour turnover - Minimisation of overtime - Maximisation of resource utilisation - Minimisation of risk to environment or factory Constraints To account for the physical limitations of the system, you need to ensure that the model includes constraints, which limit the decision variables to their feasible range or permissible values. These are expressed as constraining mathematical functions. For example, in chemical industries, there are restrictions from the government regarding the releasing of gases into the environment. Restrictions from sales department about the marketability of some products are also treated as constraints. Thus, a linear programming problem has a set of constraints in practice. The mathematical models in OR may be viewed generally as determining the values of the decision variables x J, where J = 1, 2, 3, ------ n, which will optimise Z = f (x 1, x 2, ---- x n). Subject to the constraints: g i (x 1, x 2 ----- x n) ~ b i, i = 1, 2, ---- m And xJ > 0 j = 1, 2, 3 ---- n where ~ is s, >, or =. The function f is called the objective function, where xj ~ bi, represent the ith constraint for i = 1, 2, 3 ---- m where bi is a known constant. The constraint xj > 0 is called the non-negativity condition, which restricts the variables to zero or positive values only. Operations Research Unit 1 Sikkim Manipal University Page No. 15 Example, Diet problem Formulate the mathematical model for the following: VitaminA and VitaminB are found in food1 and food2. One unit of food1 contains 5 units of vitaminA and 2 units of vitaminB. One unit of food2 contains 6 units of vitaminA and 3 units of vitaminB. The minimum daily requirement of a person is 60 units of vitaminA and 80 units of VitaminB. The cost per one unit of food1 is Rs.5 and one unit of food2 is Rs.6. Assume that any excess units of vitamins are not harmful. Find the minimum cost of the mixture (food1 and food2) which meets the daily minimum requirements of vitamins. Mathematical model of the diet problem: Suppose, x1 = the number of units of food1 in the mixture x2 = the number of units of food2 in the mixture. Let us formulate the constraint related to vitamin-A. Since each unit of food 1 contains 5 units of vitamin A, we have that x1 units of food1 contains 5x1 units of vitamin A. Since each unit of food 2 contains 6 units of vitaminA, we have that x2 units of food2 contains 6x2 units of vitaminA. Therefore, the mixture contains 5x1 + 6x2 units of vitamin-A. Since the minimum requirement of vitamin A is 60 units, you can say that 5x1 + 6x2 > 60. Now lets formulate the constraint related to vitaminB. Since each unit of food1 contains 2 units of vitaminB we have that x1 units of food1 contains 2x1 units of vitamin-B. Since each unit of food2 contains 3 units of vitaminB, we have that x2 units of food2 contains 3x2 units of vitaminB. Therefore the mixture contains 2x1 + 3x2 units of vitaminB. Since the minimum requirement of vitaminB is 80 units, you can say that 2x2 + 3x2 > 80 Next lets formulate the cost function. Given that the cost of one unit of food1 is Rs.5 and one unit of food2 is Rs.6. Therefore, x1 units of food1 costs Rs.5x1, and x2 units of food2 costs Rs.6x2. Therefore, the cost of the mixture is given by cost = 5x1 + 6x2. If we write z for the cost function, then you can write z = 5x1 + 6x2. Since cost has to be minimised, you can write min z = 5x1 + 6x2. Operations Research Unit 1 Sikkim Manipal University Page No. 16 Since the number of units (x1 or x2) are always non-negative, you have x1 > 0, x2 > 0. Therefore, the mathematical model is: 5x1 + 6x2 > 60 2x1 + 3x2 > 80 x1 > 0, x2 > 0, min z = 5x1 + 6x2. 1.10 Limitations of Operations Research The limitations are more related to the problems of model building, time, and money factors. The limitations are: - Magnitude of computation Modern problems involve a large number of variables. The magnitude of computation makes it difficult to find the interrelationship. - Intangible factors Nonquantitative factors and human emotional factors cannot be taken into account. - Communication gap There is a wide gap between the expectations of managers and the aim of research professionals. - Time and money factors When you subject the basic data to frequent changes then incorporating them into OR models becomes a costly affair. - Human factor Implementation of decisions involves human relations and behaviour. Self Assessment Questions 13. OR imbibes _________ team approach. 14. Linear programming is tool of _______. 15. The three phases of OR are ________. 16. To solve any problem through OR approach, the first step is _______. 17. _________ represents a real life system. 18. _________ represents the controlled variables of the system. Operations Research Unit 1 Sikkim Manipal University Page No. 17 1.11 Summary Let us recapitulate the important concepts discussed in this unit: - The objective of OR is to provide a scientific basis to the decisionmakers for solving problems involving interaction with various components of the organisation - The scope of OR is in various fields such as defence, industry, government, agriculture, hospitals, transport and research and development. - Some key features of OR are OR is system oriented, it imbibes inter disciplinary team approach and uses scientific methods to solve problems. - The three phases of OR are Judgement phase, research phase, and action phase. - You can broadly classify OR models according to physical models, mathematical models , models by nature of environment and models by extent of generality. - The steps in OR methodology are problem definition, model construction, model solution, model validation and result implementation. - Linear programming, inventory control methods, goal programming, queuing model and transportation model are the different tools and techniques used in OR. - Any business activity or production activity has to be formulated as a mathematical model. - Some of the limitations of OR are magnitude of computation, intangible factors, communication gap, time and money factors and human factors 1.12 Glossary Probability: possible outcomes of an event Hypothesis: unproved theory Network analysis: a mathematical representation of the problem by lines and nodes joined to form a network Operations Research Unit 1 Sikkim Manipal University Page No. 18 1.13 Terminal Questions 1. Define OR. 2. What are the features of OR? 3. What is a model in OR? Discuss the different models available in OR. 4. Write short notes on the different phases of OR. 5. What are the limitations of OR? 1.14 Answers Self Assessment Questions 1. Scientific basis 2. Scientists, different disciplines 3. economic , social policies 4. Outpatient, administrative 5. Imbibes 6. Decision making 7. True 8. True 9. True 10. False 11. False 12. False 13. Inter-disciplinary 14. OR 15. Judgement phase, research phase, and action phase 16. Define the problem 17. Model 18. Parameters Terminal Questions 1. OR is defined as the application of scientific methods, techniques and tools to the operation of a system with optimum solutions to the problems - refer 1.2.1 2. Some key features of OR are OR is system oriented, it imbibes inter disciplinary team approach and uses scientific methods to solve problems - refer 1.4 3. Types of operations research models - refer 1.6 Operations Research Unit 1 Sikkim Manipal University Page No. 19 4. Phases of operations research are judgement, research and action phase - refer 1.5 5. Limitations of OR - refer 1.10 1.15 Case Study IMB Optimiser System IBM was considering integrating its national network of spare parts inventories to improve service support for their customers. They developed a model for their inventory system that improved customer service while reducing the value of IBMs inventories by $ 250 million and saving an additional $20 million per year through improved operational efficiency. A particularly interesting aspect of the model validation phase of this study was the way the future users of the inventory system had been incorporated into the testing process. Because these future users were sceptical about the system being developed, representatives were appointed to a user team to serve as advisors to the OR team. After a preliminary version of the new system had been developed, a pre-implementation test of the system was conducted. Extensive feedback from the user team led to major improvements in the proposed system. Large computer system was used to apply this model. The system developed was called optimiser. It provided optimal control of service levels and spare-parts inventories throughout IBMs U.S. parts distribution network, which included two central automated warehouses, dozens of field distribution centres, and many thousands of outstation locations. The parts inventory maintained in this network is valued in billions of dollars. Optimiser consists of four major modules. - A forecasting system module contains a few programs for estimating the failure rates of individual types of parts. - A data delivery system module consists of approximately 100 programs that process over 15 gigabytes of data to provide the input for the model. - A decision system module then solves the model on a weekly basis to optimise control of the inventories. - The fourth module includes six programs that integrate the optimiser into IBMs Parts Inventory Management System (PIMS). PIMS is a Operations Research Unit 1 Sikkim Manipal University Page No. 20 sophisticated information and control system that contains millions of lines of code. Careful planning was required to implement the complex optimiser system for controlling IBMs national network of spare-parts inventories. Three factors proved to be especially important in achieving a successful implementation. By the time implementation phase was reached, operational managers had a strong sense of ownership and had become ardent supporters for installing optimiser in their functional areas. A second success factor was a very extensive user acceptance test whereby users could identify problem areas that needed to be rectified prior to implementation. The third key was that a new system was phased in gradually, with careful testing at each phase, so the major bugs could be eliminated before the system went live nationally. Discussion Questions: 1. Analyse the need for OR team at IBM. 2. Explain optimiser with its modules. 3. How did this new system help and what were the reasons for its success. Reference: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. - Cohen M., & Kamesan P. V., & Kleindorfer P., & Lee H., & Tekerian A. (Jan-Feb. 1990). Optimizer: IBMs Multi-Echelon Invenentory System for Managing Service Logistics. E- References: - newagepublishers.com. http://www.newagepublishers.com/samplechapter/001012.pdf. Operations Research Unit 2 Sikkim Manipal University Page No. 21 Unit 2 Linear Programming Structure: 2.1 Introduction Objectives 2.2 Linear Programming Problem 2.3 Requirements of LPP Basic assumptions of LPP 2.4 Mathematical Formulation of LPP General form of LPP Canonical forms 2.5 Case Studies of LPP Solved problem 2.6 Applications of LPP 2.7 Advantages of LPP 2.8 Limitations of LPP 2.9 Summary 2.10 Glossary 2.11 Terminal Questions 2.12 Answers 2.13 Case Study 2.1 Introduction In the previous unit, we dealt with operations research. We discussed the background, scope, features, phases, types, limitations, techniques, and tools of operations research. We also analysed the operations research methodology and the structure of the mathematical model. In this unit, we will deal with linear programming. Linear programming focuses on obtaining the best possible output (or a set of outputs) from a given set of limited resources. Minimal time and effort and maximum benefit coupled with the best possible output or a set of outputs is the mantra of any decision-maker. Today, decision-makers or managements have to tackle the issue of allocating limited and scarce resources at various levels in an organisation, in the best possible manner. Man, money, machine, time, and technology are some of Operations Research Unit 2 Sikkim Manipal University Page No. 22 the common resources. The managements task is to obtain the best possible output (or a set of outputs) from these given resources. The output can be measured in terms of profit, cost, social welfare, and the overall effectiveness. In several situations, you can express the output (or a set of outputs) as a linear relationship among several variables. The amount of available resources can be expressed as a linear relationship among various system variables. The managements dilemma is to optimise (maximise or minimise) the output or the objective function subject to the set of constraints. Optimisation of resources in which both the objective function and the constraints are represented in a linear form is known as Linear Programming Problem (LPP). Objectives: After studying this unit, you should be able to: - explain linear programming problem - analyse a feasible region - analyse the applications of LPP 2.2 Linear Programming Problem Linear Programming (LP) is a mathematical technique designed to help managers in their planning and decision-making. It is usually used in an organisation that is trying to make the most effective use of its resources. Resources typically include machinery, manpower, money, time, warehouse space, and raw materials. A few examples of problems in which LP has been successfully applied are: - Developments of a production schedule that will satisfy future demands for a firms product and at the same time minimise total production and inventory costs. - Establishment of an investment portfolio from a variety of stocks or bonds that will maximise a companys return on investment. - Allocation of a limited advertising budget among radio, TV, and newspaper spots in order to maximise advertising effectiveness. - Determination of a distribution system that will minimise total shipping cost from several warehouses to various market locations. Operations Research Unit 2 Sikkim Manipal University Page No. 23 - Selection of the product mix in a factory to make best use of machine and man hours available while maximising the firms profit. 2.3 Requirements of LPP The common requirements of LPP are as follows. - Decision variables and their relationship - Well-defined objective function - Existence of alternative courses of action - Non-negative conditions on decision variables Let us now study the basic assumptions of LPP. 2.3.1 Basic assumptions of LPP The basic assumptions of LPP are: - Linearity - Both objective functions and constraints can be expressed as linear inequalities. - Deterministic - All co-efficients of decision variables in the objective and constraints expressions are known and finite. - Additivity - The value of the objective function and the total amount of each resource used must be equal to the sum of respective individual contributions by decision variables. - Divisibility - The solution values of decision variables can be nonnegative values including fractions. Self Assessment Questions 1. Both the objective function and constraints are expressed in _____ forms. 2. LPP requires existence of _______, _______, of _______. 3. Solution of decision variables can also be ____________. 2.4 Mathematical Formulation of LPP The procedure for mathematical formulation of a linear programming problem consists of the following major steps: Operations Research Unit 2 Sikkim Manipal University Page No. 24 Fig. 2.1: Procedure of mathematical formulation of a linear programming problem Solved problem 1: A company has three operational departments (weaving, processing, and packing) with a capacity to produce three different types of clothes - suits, shirts, and woollens, yielding a profit of Rs.2, Rs.4, and Rs.3 per meter respectively. One meter of suiting requires 3 minutes of weaving, 2 minutes of processing, and 1 minute of packing. Similarly, one meter of shirting requires 4 minutes of weaving, 1 minute of processing, and 3 minutes of packing. One meter of woollen requires 3 minutes in each department. In a week, total run time of each department is 60, 40, and 80 hours for weaving, processing, and packing respectively. Formulate the linear programming problem to find the product mix to maximise the profit. Operations Research Unit 2 Sikkim Manipal University Page No. 25 Mathematical formulation: The data of the problem is summarised below in table 2.1: Table 2.1: summary of the data of the problem Departments Profit Weaving (in min) Processing (in min) Packing (in min) (Rs. per meter) Suits 3 2 1 2 Shirts 4 1 3 4 Woollens 3 3 3 3 Availability (min) 60*60 40*60 80*60 Let us solve this problem according to procedure outlined in fig. 2.1 Step 1: The key decision is to determine the weekly rate of production for the three types of clothes. Step 2: Let us designate the weekly production of suiting, shirting, and woollens as x1 meters, x2 meters, and x3 meters respectively. Step 3: As it is not possible to produce negative quantities, feasible alternatives are sets of values of x1, x2, and x3 satisfying x1, x2, and x3 0. Step 4: The constraints are the limited availability of three operational departments. One meter of suiting requires 3 minutes of weaving. The quantity being x1 meters, the requirement for suiting alone will be 3x1 units. Similarly, x2 meters of shirting and x3 meters of woollen will require 4x2 and 3x3 minutes respectively. Thus, the total requirement of weaving will be 3x1+4x2+3x3, which should not exceed the available 3600 minutes. Therefore, the labour constraint becomes 3x1+4x2+3x3 3600. Similarly, the constraints for the processing department and packing departments are 2x1+x2+3x3 2400 and x1+3x2+3x3 4800 respectively. Step 5: The objective is to maximise the total profit from sales. Assuming that whatever is produced is sold in the market, the total profit is given by the linear relation z= 2x1+4x2+3x3. Operations Research Unit 2 Sikkim Manipal University Page No. 26 The linear programming problem can thus be put in the following algebraic format: Find x1, x2, and x3 so as to maximise Z = 2x1+4x2+3x3 Subject to constraints: 3x1+4x2+3x3 3600 2x1+x2+3x3 2400 3x1+3x2+3x3 4800 x1, x2, and x3 0 2.4.1 General form of LPP The LPP is a class of mathematical programming where the functions representing the objectives and the constraints are linear. Optimisation refers to the maximisation or minimisation of the objective functions. The general linear programming model can be defined as follows: Maximise or minimise Z = c1 x1 + c2 x2 + - - - - + cn xn Subject to the constraints, a11 x1 + a12 x2 + + a1n xn ~ b1 a21 x1 + a22 x2 + + a2n xn ~ b2 - am1 x1 + am2 x2 + - + amn xn ~ bm and x1 0, x2 0, xn 0 Where cj, bi, and aij (i = 1, 2, 3 m, j = 1, 2, 3 n) are constants determined from the technology of the problem and xj (j = 1, 2, 3 - n) is the decision variable. Here ~ is either (less than), (greater than), or = (equal). In terms of the above formulation, the coefficients cj, bi, and aij are interpreted physically as follows. If bi is the available amount of resources i, where aij is the amount of resource that must be allocated to each unit of activity j, the worth per unit of activity is equal to cj. Operations Research Unit 2 Sikkim Manipal University Page No. 27 2.4.2 Canonical forms You can represent the general LPP mentioned above in the canonical form as follows: Maximise Z = c1 x1+c2 x2 + + cn Subject to a11 x1 + a12 x2 + + a1n xn b1 a21 x1 + a22 x2 + + a2n xn b2
am1 x1+am2 x2 + + amn xn bm x1, x2, x3, xn 0 The following are the characteristics of the canonical form. - All decision variables are non-negative. - All constraints are of type. - The objective function is of the maximisation type. Any LPP can be represented in the canonical form by using five elementary transformations, which are as follows: 1. The minimisation of a function is mathematically equivalent to the maximisation of the negative expression of this function. That is, Minimise Z = c1 x1 + c2x2 + . + cn xn is equivalent to Maximise Z = c1x1 c2x2 cn xn. 2. Any inequality in one direction ( or ) may be changed to an inequality in the opposite direction ( or ) by multiplying both sides of the inequality by 1. For example, 2x1+3x2 5 is equivalent to 2x13x2 5. 3. An equation can be replaced by two inequalities in opposite direction. For example, 2x1+3x2 = 5 can be written as 2x1+3x2 5 and 2x1+3x2 5 or -2x1-3x2 5 and 2x1 3x2 5. Operations Research Unit 2 Sikkim Manipal University Page No. 28 4. An inequality constraint with its left hand side in the absolute form can be changed into two regular inequalities. For example, |2x1+3x2| 5 is equivalent to 2x1+3x2 5 and 2x1+3x2 5 or 2x1 3x2 5. 5. The variable, which is unconstrained in sign, is equivalent to the difference between two non-negative variables. For example, if x is unconstrained in sign then x = (x+ x) where x+ 0, x 0. Case-let An automobile company has two units X and Y, which manufactures three different models of cars - A, B, and C. The company has to supply 1500, 2500, and 3000 cars of type A, B, and C respectively per week (6 days). It costs the company Rs.1,00,000 and Rs.1,20,000 per day to run the units X and Y respectively. In a day, unit X manufactures 200, 250, and 400 cars and unit Y manufactures 180, 200, and 300 cars of A, B, and C respectively. The operations manager has to decide on how many days per week should each unit be operated to meet the current demand at minimum cost. The operations manager along with his/her team uses a LPP model to arrive at the minimum cost solution. 2.5 Case Studies of LPP Solved problem 1 A firm engaged in producing 2 models - model A and model B, performs only 3 operations painting, assembly, and testing. Table 2.1 depicts the relevant data. Table 2.1: Unit Sale Price and Hours Required for Each Unit Unit sale price Hours required for each unit Assembly Painting Testing Model A Rs. 50.00 Model B Rs. 80.00 1.0 1.5 0.2 0.2 0.0 0.1 Total numbers of hours available each week are as under assembly 600, painting 100, and testing 30. The firm wishes to determine the weekly product-mix to maximise revenue. Operations Research Unit 2 Sikkim Manipal University Page No. 29 Solution: Let us first write the notations as under: Z: Total revenue x1: Number of Units of Model A x2: Number of Units of Model B x1, x2: Decision variables Since the objective (goal) of the firm is to maximise its revenue, the model can be stated as follows. The objective function Z = 50x1 + 80x2 is to be maximised subject to the following constraints: 1.0x1 + 1.5x2 600, (assembly constraints) 0.2x1+0.2x2 100, (painting constraints) 0.0x1+0.1x2 30, (testing constraints) x10, x20, (the non-negativity constraints) Solved problem 2 A milk distributor supplies milk in bottles to houses in three areas A, B, and C in a city. His delivery charge per bottle is 30 paise in area A, 40 paise in area B, and 50 paise in area C. He has to spend on an average, 1 minute to supply one bottle in area A, 2 minutes per bottle in area B, and 3 minutes per bottle in area C. He can spare only 2 hours 30 minutes for milk distribution but not more than 1 hour 30 minutes for area A and B together. He can deliver at the most 120 bottles. Find the number of bottles that he has to supply in each area so as to maximise the supply. Construct a mathematical model. Operations Research Unit 2 Sikkim Manipal University Page No. 30 Solved problem 3 An oil company has two units A and B, which produce three different grades of oil super fine, medium, and low grade. The company has to supply 12, 8, and 24 barrels of super fine, medium, and low-grade oil respectively per week. It costs the company Rs.1000 and Rs.800 per day to run the units A and B respectively. In a day, unit A produces 6, 2, and 4 barrels and unit B produces 2, 2, and 12 barrels of super fine, medium, and low grade respectively. The manager has to decide on how many days per week should each unit be operated in order to meet the requirement at minimum cost. Formulate the LPP model. Solution: Table 2.2 depicts the summary of the given data. Table 2.2: Capacity and Requirements Details of an Oil Company Product Capacity Requirements Super fine Medium Low grade Cost Unit A Unit B 12 8 24 - 6 2 4 Rs. 1000 2 2 12 Rs. 800 Let x1 and x2 be the number of days that unit A and B must be operated per week respectively. Then the objective of the manager is to: Minimise the cost function Z = 1000 x1 + 800 x2 Subject to the constraints 6x1 + 2x2 12 (super fine) 2x1 + 2 x2 8 (medium) 4x1 + 12 x2 24 (low grade) and x1 >0, x2 0 2.5.1 Solved problem A toy company manufactures two types of doll, a basic version-doll A and a deluxe version-doll B. Each doll of type B takes twice as long to produce one doll of type A. Thus, the company would have time to make a maximum Operations Research Unit 2 Sikkim Manipal University Page No. 31 of 2000 types of A dolls per day. The supply of plastic is sufficient to produce 1500 dolls per day (both A and B combined). The deluxe version requires a fancy dress of which only 600 is available per day. If the company makes a profit of Rs.3.00 and Rs.5.00 per doll, respectively on doll A and B, then how many of each doll should be produced per day in order to maximise the total profit. Formulate the problem. Formulation Let x1 and x2 be the number of dolls produced per day of type A and B respectively. Let doll A require t hrs. So, doll B require 2t hrs. The total time to manufacture x1 and x2 dolls should not exceed 2,000t hrs. Therefore, tx1+2tx2 2000t Other constraints are simple. The linear programming problem becomes: Maximise p= 3x1+5x2 Subject to restrictions x1+2x2 2000(time constraint) x1+x2 1500(plastic constraint) x2 600 (dress constraint) and non-negativity restrictions x1 0, x20 Self Assessment Questions 4. One of the characteristics of canonical form in the objective function must be maximisation. (True/False) 5. 2x 3y 10 can be written as -2x + 3y -10. (True/False) Operations Research Unit 2 Sikkim Manipal University Page No. 32 2.6 Applications of LPP Linear programming is a powerful tool for selecting alternatives in a decision problem. Consequently, it has been applied in a wide variety of problem settings. We will indicate a few applications covering the major functional areas of a business organisation. - Finance - The problem of the investor could be a portfolio-mix selection problem. In general, the number of different portfolios can be much larger than what the example indicates. More and different kinds of constraints can be added. Another decision problem involves determining the mix of funding for a number of products when more than one method of financing is available. The objective may be to maximise total profits, where the profit for a given product depends on the method of financing. For example, funding may be done with internal funds, short-term debt, or intermediate financing (amortised loans). There may be a limitation on the availability of each of the funding options as well as financial constraints requiring certain relationships between the funding options so as to satisfy the terms of bank loans or intermediate financing. There may also be limitations on the production capacity for the products. The decision variables would be the number of units of each product to be financed by each funding option. - Production and operations management - Quite often, in the process industries, a given raw material can be converted into a wide variety of products. For example, in the oil industry, crude oil is refined into gasoline, kerosene, home-heating oil, and various grades of engine oil. Given the present profit margin on each product, the problem is to determine the quantities of each product that should be produced. The decision is subject to numerous restrictions such as limits on the capacities of various refining operations, raw-material availability, demands for each product, and any government-imposed policies on the output of certain products. Similar problems also exist in the chemical and food-processing industries. - Human resources - Personnel planning problems can also be analysed with linear programming. For example, in the telephone industry, demands for the services of installer-repair personnel are seasonal. The problem is to determine the number of installer-repair personnel and line-repair personnel required each month as part of the work force, Operations Research Unit 2 Sikkim Manipal University Page No. 33 considering the total costs of hiring, layoff, overtime, and regular-time wages are minimised. The constraints set includes restrictions on the service demands that must be satisfied, overtime usage, union agreements, and the availability of skilled people for hire. This example runs contrary to the assumption of divisibility. However, the work-force levels for each month would normally be large enough that rounding off to the closest integer in each case would not be detrimental, provided the constraints are not violated. - Marketing - Linear programming can be used to determine the proper mix of media to use in an advertising campaign. Suppose that the available media are radio, television, and newspapers. The problem is to determine how many advertisements to place in each medium. Of course, the cost of placing an advertisement depends on the medium chosen. We wish to minimise the total cost of the advertising campaign, subject to a series of constraints. Since each medium may provide a different degree of exposure of the target population, there may be a lower bound on the total exposure from the campaign. Also, each medium may have a different efficiency rating in producing desirable results. Thus there may be a lower bound on efficiency. In addition, there may be limits on the availability of each medium for advertising. - Distribution - Another application of linear programming is in the area of distribution. Consider a case in which there are m factories that must ship goods to n warehouses. A given factory could make shipments to any number of warehouses. Given the cost to ship one unit of product from each factory to each warehouse, the problem is to determine the shipping pattern (number of units that each factory ships to each warehouse) that minimises total costs. This decision is subject to the restrictions that demand at each factory cannot be more than it has the capacity to produce. Self Assessment Questions 6. Personnel planning problems can also be analysed with ___________. 7. Linear programming is a powerful tool for selecting alternatives in a _________________. 8. Linear programming can be used to determine the _______________ to use in an advertising campaign. Operations Research Unit 2 Sikkim Manipal University Page No. 34 9. Quite often in the process industries, a given _________ can be made into a wide variety of products. 2.7 Advantages of LPP The advantages of linear programming techniques may be outlined as follows: - Linear programming technique helps in making the optimum utilisation of productive resources. It also indicates how decision makers can employ productive factors most effectively by choosing and allocating these resources. - The quality of decisions may also be improved by linear programming techniques. The user of this technique becomes more objective and less subjective. - Linear programming technique provides practically applicable solutions because there might be other constraints operating outside the problem. These constraints must also be taken into consideration. Just because so many units must be produced does not mean that all those can be sold. So the necessary modification of its mathematical solution is required for the sake of convenience to the decision maker. - In production processes, high lighting of bottlenecks is the most significant advantage of this technique. For example, when bottlenecks occur, some machines cannot meet the demand while others remain idle for some time. 2.8 Limitations of LPP In spite of wide area of applications, some limitations are associated with linear programming techniques. These are stated below: - In some problems, objective functions and constraints are not linear. Generally, in real life situations concerning business and industrial problems, constraints are not linearly treated as variables - There is no guarantee of integer valued solutions. For example, in finding out how many men and machines would be required to perform a particular job, rounding off the solution to the nearest integer will not give an optimal solution. Integer programming deals with such problems. - A linear programming model does not take into consideration the effect of time and uncertainty. Thus the model should be defined in such a way Operations Research Unit 2 Sikkim Manipal University Page No. 35 that any change due to internal as well as external factors can be incorporated. - Sometimes, large-scale problems cannot be solved with linear programming techniques even when the computer facility is available. Such difficulty may be removed by breaking the main problem into several small problems and then solving them separately. - Parameters appearing in the model are assumed to be constant. But, in real life situations they are neither constant nor deterministic. - Linear programming deals with only single objective, whereas in real life situations, problems come across with multi-objectives. Goal programming and multi-objective programming deal with such problems. 2.9 Summary Let us recapitulate the important concepts discussed in this unit: - In LPP, first identify the decision variables with economic or physical quantities whose values are of interest to the management. - The problems must have a well-defined objective function expressed in terms of the decision variable. - The objective is to maximise the resources when it expresses profit or contribution. If the objective function indicates cost, then it has to be minimised. - The decision variables interact with each other through some constraints. These constraints arise due to limited resources, stipulation on quality, technical, legal, or a variety of other reasons. 2.10 Glossary Decision variables - set of unknown quantities in LPP Bottleneck - A point in the production process that causes delay. 2.11 Terminal Questions 1. Explain LPP by giving two examples. 2. Write any three applications of LPP. 3. What are the advantages of LPP? 4. Explain the basic assumptions of LPP. 5. State the general form of LPP. Operations Research Unit 2 Sikkim Manipal University Page No. 36 6. Three grades of coal A, B, and C contain ash and phosphorous as impurities. In a particular industrial process, a fuel obtained by blending the above grades containing not more than 25% ash and 0.03% phosphorous is required. The maximum demand of the fuel is 100 tons. Percentage impurities and costs of the various grades of coal are shown below. Assuming that there is an unlimited supply of each grade of coal and there is no loss in blending, formulate the blending problem to minimise the cost. Coal grade Ash% Phosphorus% Cost per ton in Rs. A 30 0.02 240 B 20 0.04 300 C 35 0.03 280 7. A firm manufactures two products. The net profit on product 1 is Rs.3 per unit and the net profit on product 2 is Rs.5 per unit. The manufacturing process is such that each product has to be processed in two departments D1 and D2. Each unit of product 1 requires processing for 1 minute at D1 and 3 minutes at D2; each unit of product 2 requires processing for 2 minutes at D1 and 2 minutes at D2. Machine time available per day is 860 minutes at D1 and 1200 minutes at D2. How much of product 1 and product 2 should be produced every day so that the total profit is maximum. Formulate this as a problem in LPP. 2.12 Answers Self Assessment Questions 1. Linear 2. Alternate course of action 3. Fractions 4. True 5. True 6. Linear programming 7. Decision problem 8. Proper mix of media 9. Raw material Operations Research Unit 2 Sikkim Manipal University Page No. 37 Terminal Questions 1. LPP is a mathematical technique for planning and decision making. Examples: development of production schedule and establishment of an investment portfolio. For more information refer 2.2 2. Applications of LPP in the area of finance, productions and operations management, human resources, marketing and distributions. For more information refer 2.6 3. Advantages of LPP: optimum utilisation of productive resources, improves quality of decisions, provides practically applicable solutions. For more information refer 2.7 4. Basic assumptions of LPP: Linearity, deterministic, additivity, divisibility etc. For more information refer 2.3 and 2.3.1 5. The general linear programming model can be defined as follows: Maximise or minimise Z = c1 x1 + c2 x2 + - - - - + cn xn For more information refer 2.4.1 6. [Hint: x1= tons of grade A coal, x2= tons of grade B coal & x3= tons of grade C coal]. For more information refer 2.4 and 2.5. 7. Maximise 3x1 + 5x2, subject to x1 + 2x2 800 (minutes) 3x1 + 2x2 1200 (minutes) x1, x2 0 For more information refer 2.4 and 2.5. 2.12 Case Study Airport Taxi Service A transport company situated in New Delhi finds great demand for transport services from the airport. Though there is taxi service from the airport, there is yet heavy demand for more transport. Hence, the company is considering the purchase of new vehicles for transportation between Delhi airport and hotels in the city. After studying the different types of vehicles available, there were three vehicles under consideration station wagons, mini buses, and large buses. The purchase price would be Rs. 2,45,000 for each station wagon, Rs. 3,50,000 for a mini bus, and Rs. 5,00,000 for a large bus. The Board of Directors have authorised a maximum amount of Rs. 50,00,000 for these purchases. Because of the heavy air travel involved, the new vehicles would be utilised at maximum capacity, regardless of the type of vehicles Operations Research Unit 2 Sikkim Manipal University Page No. 38 purchased. The expected net annual profit would be Rs. 15,000 for the station wagon, Rs. 35,000 for the mini bus, and Rs. 45,000 for the large bus. The company has hired 30 new drivers for the new vehicles. They are qualified drivers for all the three types of vehicles. The maintenance department has the capacity to handle an additional 80 station wagons. A mini bus is equivalent to 5/3 station wagons and each large bus is equivalent to two station wagons in terms of their use of the maintenance department. Discussion Questions: 1. Formulate a linear programming model, which would help to determine the optimum number of each type of vehicle to be purchased in order to maximise profit. [Hint: x1= no. of station wagons to purchase, x2= no. of minibus to purchase & x3= no. of large buses to purchase]. References : - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E-References: - Linear Optimization. http://www.mirrorservice.org/sites/home.ubalt.edu/ntsbarsh/Businessstat/ opre/partVIII.htm. Operations Research Unit 3 Sikkim Manipal University Page No. 39 Unit 3 Graphical Analysis of Linear Programming Problems Structure: 3.1 Introduction Objectives 3.2 Graphical Analysis 3.3 Some Basic Definitions 3.4 Graphical Methods to Solve LPP Working rule Classification of feasible points Example on maximisation of LPP Solved problems on mixed constraints LP problem 3.5 Some Exceptional Cases Solved problem for alternative optimal solution Solved problem for unbounded solution Solved problem for inconsistent solution 3.6 Important Geometric Properties of LPP 3.7 Summary 3.8 Glossary 3.9 Terminal Questions 3.10 Answers 3.11 Case Study 3.1 Introduction In the previous unit, we dealt with Linear Programming Problem (LPP). We analysed the definition, requirements, mathematical formulation, case studies, applications, advantages, and disadvantages of LPP. In this unit, we will deal with the graphical analysis of LPP. Today, many resources needed as inputs to operations are in limited supply. Operations managers must understand the impact of this situation on meeting their objectives. Linear Programming (LP) is one way through which the operations managers can determine how best to allocate their scarce resources. Once the linear programming model has been formulated on the basis of the given objective and associated constraint functions, the next step is to solve Operations Research Unit 3 Sikkim Manipal University Page No. 40 the problem and obtain the best possible or the optimal solution using various mathematical and analytical techniques that can be employed for solving the linear programming model. The graphical solution procedure is one of the methods of solving two variable linear programming problems. Objectives: After studying this unit, you should be able to: - solve linear programming problems graphically - identify some exceptional cases involved in LPP - analyse the important geometric properties of LPP 3.2 Graphical Analysis You can analyse linear programming with two decision variables graphically. Example Let us consider the following illustration: Maximise z = 700 x1+500 x2 Subject to 4x1+3x2 210 2x1+x2 90 x1 0, x2 0 Let the horizontal axis represent x1 and the vertical axis x2 as shown in figure 3.1. First, draw the line 4x1 + 3x2 = 210, (by replacing the inequality symbols by the equality) which meets the x1-axis at the point A (52.50, 0) (put x2 = 0 and solve for x1 in 4x1 + 3x2 = 210) and the x2axis at the point B (0, 70) (put x1 = 0 in 4x1 + 3x2 = 210 and solve for x2). Operations Research Unit 3 Sikkim Manipal University Page No. 41 Fig. 3.1: Linear Programming with 2 Decision Variables Any point on the line 4x1+3x2 = 210 or inside the shaded portion will satisfy the restriction of the inequality, 4x1+3x2 <= 210. Similarly, the line 2x1+x2 = 90 meets the x1-axis at the point C(45, 0) and the x2 axis at the point D(0, 90) as shown in figure 3.2. Fig. 3.2: Linear Programming with 2 Decision Variables Operations Research Unit 3 Sikkim Manipal University Page No. 42 Figure 3.3 depicts how the area can be sketched by combining the two graphs. Fig. 3.3: Feasible Region The three constraints including non-negativity are satisfied simultaneously in the shaded region OCEB. This region is called feasible region. 3.3 Some Basic Definitions The table 3.1 gives the some basic definitions of the terms in graphical analysis. Table 3.1: Basic Definitions in Graphical Analysis No. Terms Definitions 1. Feasible Region Any non-negative value of (x1, x2) (i.e., x1 0, x2 0) is a feasible solution of the LPP if it satisfies all the constraints. The collection of all feasible solutions is known as the feasible region. 2. Convex A set X is convex if for any point x1, x2 in X, the line segment joining these points is also in X. By convention, a set containing only a single point is also a convex set. Operations Research Unit 3 Sikkim Manipal University Page No. 43 3. Half Plane A linear inequality in two variables is known as a half plane. The corresponding equality or the line is known as the boundary of the half plane. 4. Convex Polygon A convex polygon is a convex set formed by the intersection of finite number of closed half-planes. Refer figure 3.4 and figure 3.5 5. Redundant Constraint A redundant constraint is a constraint which does not affect the feasible region. 6. Basic Solution A basic solution of a systems of m equations and n variables (m < n) is a solution where at least n-m variables are zero. 7. Basic Feasible Solution A basic feasible solution of a system of m equations and n variables is where all basic variables are non-negative. 8. Optimal Feasible Solution Any feasible solution that optimises the objective function is called an optimal feasible solution. Note: The objective function is maximised or minimised at one of the extreme points referred to as optimum solution. Extreme points are referred to as vertices or corner points of the convex regions. Self Assessment Questions 1. The collection of all feasible solutions is known as the ________ region. 2. A linear inequality in two variables is known as a _________. Operations Research Unit 3 Sikkim Manipal University Page No. 44 3.4 Graphical Methods to Solve LPP While obtaining the optimal solution to an LPP by the graphical method, the statement of the following theorems of linear programming is used: - The collection of all feasible solutions to an LPP constitutes a convex set whose extreme points correspond to the basic feasible solutions. - There are a finite number of basic feasible regions within the feasible solution space. - If the convex set of the feasible solutions of the system of simultaneous equation is a convex polyhedron, then at least one of the extreme points gives an optimal solution. - If the optimal solution occurs at more than one extreme point, the value of the objective function will be the same for all convex combination of these extreme points. 3.4.1 Working rule The method of solving an LPP on the basis of the above analysis is known as the graphical method. The working rule for the method is as follows. Step 1: Formulate the problem in terms of a series of mathematical equations representing objective function and constraints of LPP. Step 2: Plot each of the constraints equation graphically. Replace the inequality constraint equation to form a linear equation. Plot the equations on the planar graph with each axis representing respective variables. Step 3: Identify the convex polygon region relevant to the problem. The area which satisfies all the constraints simultaneously will be the feasible region. This is determined by the inequality constraints. Step 4: Determine the vertices of the polygon and find the values of the given objective function Z at each of these vertices. Identify the greatest and the least of these values. These are respectively the maximum and minimum value of Z. Step 5: Identify the values of (x1, x2) which correspond to the desired extreme value of Z. This is an optimal solution of the problem. Operations Research Unit 3 Sikkim Manipal University Page No. 45 3.4.2 Classification of the feasible points Figure 3.6 depicts the feasible points of any non-empty LP feasible region, which can be classified as, interiors, boundaries, or vertices. Source: http://www.mirrorservice.org/sites/home.ubalt.edu/ntsbarsh/ Business-stat/opre/partVIII.htm Fig. 3.6: Classification of Feasible Region Points The point B in the above two-dimensional figure, for example, is a boundary point of the feasible set because every small circle is centred at the point B. However small it may be, it contains both points, some inside the set and some points outside the set. The point I is an interior point because the I circle, all smaller circles, and some larger ones contain exclusively points in the set. The collection of boundary points belongs to boundary line (segment), e.g. the line segment cd. The intersections of boundary lines (segments) are called the vertices; if feasible it is called the corner point. In three-dimensional space and higher, the circles become spheres and hyper-spheres. It must be noted that the LP constraints provide the vertices and the cornerpoints. A vertex is the intersection of 2-lines, or in general n-hyper planes in LP problems with n-decision variables. Operations Research Unit 3 Sikkim Manipal University Page No. 46 3.4.3 Example on maximisation of LPP Solved problem 1 Solve the given LPP using the graphical method. Maximise z =50x1 + 80x2, subject to the constraints 1.0x1 + 1.5x2 600 1 0.2 x1 +0.2x2 100 2 0.0x1 + 0.1x2 30 3 and x1 0, x2 0 Solution The horizontal axis represents x1 and the vertical axis x2. Plot the constraint lines and mark the feasibility region as depicted in figure 3.7. The shaded region is the feasibility region carried out by the constraints operating on the objective function. This depicts the limits within which the values of the decision variables are permissible. The intersection point C and D can be solved by the linear equations x2 = 30; x1 + 1.5 x2 = 600, and 0.2x1 + 0.2x2 = 100 and x1 + 1.5x2 = 600 that is, C (150, 300) and D(300, 200). The next step is to maximise revenues subject to the feasible region shown in figure 3.7. Fig. 3.7: Showing Feasible Region Operations Research Unit 3 Sikkim Manipal University Page No. 47 Final solution: A(0,300) Z= 24000 B(500,0) Z= 25000 C(150, 300) Z= 31500 D(300,200) Z= 31000 We choose point C (150, 300) as the final solution which maximises the given problem. 3.4.4 Solved problems on mixed constraints LP problem Solved problem 2 Use the graphical method to solve the following LPP. Minimise z = 20 x + 10 y, subject to the constraints x + 2y 40 1 3 x + y 30 2 4x + 3y 60 3 and x, y 0 Solution: Plot each constraint on a graph by first writing it as a linear equation. Then use the inequality condition of each constraint to mark the feasible region as depicted in figure 3.8. The coordinates of the extreme points of the feasible region are: A= (15, 0), B= (40, 0), C= (4, 18), and D= (6, 12). The values of the objective function at each of these extreme points are as depicted in table 3.2: Table 3.2: Extreme Point Values of the Objective Function Extreme Point Co-ordinates (x, y) Objective Function Value, z = 20x + 10 y A (15, 0) 20 (1) + 10 (0) = 300 B (40,0) 20 (40) + 10 (0) = 800 C (4,18) 20 (4) + 10 (18) = 260 D (6,12) 20 (6) + 10 (12) = 240 The minimum value of the objective function Z = 240 occurs at the extreme point D (6, 12). Hence the optimal solution is x = 6, y = 12, and min Z = 240. Operations Research Unit 3 Sikkim Manipal University Page No. 48 Fig. 3.8: Showing Feasible Region 3.5 Some Exceptional Cases In the section 3.4, we discussed some linear programming problems which may be called as well-behaved problems. In each case, a solution was obtained, in some cases it took less effort while in some others it took a little more. But a solution was finally obtained. It should not be taken as a rule. There may be an LPP for which no solution exists or for which the only solution obtained is an unbounded one. Though such problems seldom occur in real situations, it will be an omission, if at this stage the reader is not exposed to such exceptional cases. Here we consider four different cases that arise in the application of graphical method. - A unique optimal solution - In this case, only one optimum solution will be obtained in a graphical solution method. - Multiple optimal solutions/Alternative solution -: When the objective function is parallel to a binding constraint (i.e., a constraint that is satisfied as an equation by the optimal solution), the objective function will assume the same optimum value at more than one solution point. For this reason, they are called alternative optima. - An unbounded solution - When the values of the decision variables may be increased indefinitely without violating any of the constraints, the Operations Research Unit 3 Sikkim Manipal University Page No. 49 feasible region is unbounded. The value of objective function, in such cases, may increase (for maximisation) or decrease (for minimisation) indefinitely. Thus, both the solution space and the objective function value are unbounded. - Infeasible problem - When the constraints cannot satisfy simultaneously, the linear programming problem has no feasible solution. 3.5.1 Solved problem for alternative optimal solution Solved problem 3 Use graphical method to solve the following LPP. Maximise z = 2 x + 4 y, subject to the constraints: x + 2 y 5 1 x + y 4 2 x, y 0 Solution The problem is depicted graphically in figure 3.9. The extreme points of the feasible region are O, A, B, and C. We observe that our objective function is parallel to the line BC, which forms the boundary of the feasible region. Thus, as we move the objective function away from the origin, it coincides with the portion BC of the constraint line which forms the boundary of the feasible region. This shows that any point including the extreme points B and C on the same line between B and C is an optimal solution. Therefore, in fact an infinite number of values of x and y give the same values of objective function. Now, the value of objective function at each of the extreme points is evaluated as follows: Table 3.3: Extreme Point Values of the Objective Function Extreme Point Co ordinates (x, y) Objective Function Value, z = 2x + 4 y O (0, 0) 0 A (4,0) 8 B (3,1) 10 (Max) C (0,2.5) 10 (Max) Since any point on the line segment BC gives the maximum value (z=10) of the objective function, there exists an alternative optima. Operations Research Unit 3 Sikkim Manipal University Page No. 50 Fig. 3.9: Graphical Representation of Solved Problem 3 3.5.2 Solved problem for unbounded solution Solved problem 4 Solve the given LPP using graphical method. Maximum z= 3x1+2x2 Subject to x1-x21, x1+ x23, and x1, x20 Graphical solution: Figure 3.10 depicts the region of feasible solution, which is the shaded area. Fig. 3.10: Feasible Region of Two-dimensional LP Operations Research Unit 3 Sikkim Manipal University Page No. 51 It is clear from this figure that the line representing the objective function can be moved far even parallel to itself in the direction of increasing z, and still has some points in the region of feasible solutions. Hence z can be made arbitrarily large, and the problem has no finite maximum value of z. Such problems are said to have unbounded solutions. Infinite profit in practical problems of linear programming cannot be expected. If LPP has been formulated by committing some mistake, it may lead to an unbounded solution. 3.5.3 Solved problem for inconsistent solution Solved problem Solve the given LPP using graphical method Maximise z= 3x1-2x2 Subject to constraints x1+x2 1, 2x1+2x2 4, and x1,x20 Solution: Figure 3.11 depicts the problem graphically. Fig. 3.11: Feasible Region of Two Dimensional LPP The figure shows that there is no point in (x1, x2) which satisfies both the constraints simultaneously. Hence the problem has no solution because the constraints are inconsistent. Operations Research Unit 3 Sikkim Manipal University Page No. 52 Self Assessment Questions 3. The feasible region is a convex set. (True/False) 4. The optimum value occurs anywhere in feasible region. (True/False) 5. An unbounded solution means that there exist a finite number of solutions to the given problem. (True/False) 3.6 Important Geometric Properties of LPP Geometric properties of LP problems, which are observed while solving them graphically, are summarised below: - The feasible region has an important property which is called the convexity property of geometry, provided the feasible solution for the problem exists. - Convexity means that the region of feasible solutions has no holes in them, that is, they are solids, and they have no cuts as on the boundary. - This fact can be expressed more precisely by saying that the line joining any two points in the region also lies in the region. - The boundaries of the feasible regions are lines or planes. - There are corners or extreme points on the boundary and there are edges joining various corners. - The objective function can be represented by a line or a plane for any fixed value of Z. - At least one corner of the region of feasible solutions will be an optimal solution whenever the maximum or minimum value of z is finite. - If the optimal solution is not unique, there are points other than corners that are optimal but in any case at least one corner is optimal. - A different situation is found when the objective function can be made arbitrarily large. Of course, no corner is optimal in that case. - Infeasible region. Self Assessment Questions 6. The feasible region has an important property which is called the convexity property of geometry (True/False) Operations Research Unit 3 Sikkim Manipal University Page No. 53 7. The boundaries of the feasible regions are lines or planes. (True/False) 8. If the optimal solution is not unique, there are points other than corners that are optimal but in any case at least one corner is optimal. 3.7 Summary Let us recapitulate the important concepts discussed in this unit: - Any LPP with two decision variables can be solved graphically. - Any non-negative solution satisfying all the constraints is known as a feasible solution of the problem. - The collection of all feasible solutions is known as a feasible region. - The feasible region of an LPP is a convex set. - The value of the decision variables, which maximises or minimises the objectives function, is located on the extreme point of the convex set formed by the feasible solutions. - Sometimes the problem may be infeasible indicating that no solution exists for the problem. 3.8 Glossary Inequality constraints: condition of LPP that does not hold exactly Constraint functions: function that describes the set of conditions of LPP Optimal solution: best solution from all feasible solutions 3.9 Terminal Questions 1. Use the graphical method to solve the following LPP: Maximise z= 5x1 + 3x2 Subject to: 3x1 + 5x2 15 5x1 + 2x2 10 x1, x2 0 2. Use graphical method to solve the following LPP: Maximise z= 2x1+4x2 Subject to constraints: x1+2x2 5, x1+x2 4, x1, x2 0 Operations Research Unit 3 Sikkim Manipal University Page No. 54 3. Use graphical method to solve the following LPP: Maximise z= 6x1+x2 Subject to constraints: 2x1+x2 3, x1-x20, and x1, x2 0 3.10 Answers Self Assessment Questions 1. Feasible 2. Half-plane 3. True 4. False 5. False 6. True 7. True 8. True Terminal Questions 1. x1 = 20/19, x2 = 45/19, and max z = 235 / 19. Refer 3.4 2. Any point on the line segment BC gives the maximum value (z=10) of the objective function. There exists an alternative solution. Refer 3.4 3. Unbounded solution. Refer 3.5.2 3.11 Case Study Travel Agents Problem A local travel agent is planning a charter trip to a famous sea resort. The eight-day seven-night package includes the round-trip fare, surface transportation, boarding and lodging, and selected tour options. The charter trip is restricted to 200 persons and the past experience indicates that there will not be any problem in arranging 200 passengers. The problem faced by the travel agent is to determine the number of Deluxe, Standard, and Economy tour packages to offer for this charter. All three plans differ in terms of their seating and flight services, quality of accommodation, meal plans, and tour options. The following table summarises the estimated prices of the three packages and the corresponding expenses of the travel agent. The travel agent has hired an aircraft for a flat fee of Rs.2, 00,000 for Operations Research Unit 3 Sikkim Manipal University Page No. 55 the entire trip. The price per person and the cost for the three packages are as follows: Table 3.4: Summary of the Package Prices and Expenses of the Travel Agent Tour plan Price(Rs.) Hotel cost(Rs.) Meals and expenses(Rs.) Deluxe 10,000 4,750 3,000 Standard 7,000 2,500 2,200 Economy 6,500 2,200 1,900 In planning the trip, the following considerations must be taken into account: - At least 10% of the packages must be of the deluxe type. - At least 30% but not more than 70% must be of the standard type. - At least 30% must be of the economy type. - The maximum number of deluxe packages available in any aircraft is restricted to 60. - The hotel desires that at least 120 tourists should be for the deluxe and standard packages together. The travel agent wishes to determine the number of packages to offer in each type of trip so as to maximise the total profit. Discussion Questions: 1. Formulate the problem as LPP. [Hint: x1= no. of deluxe packages, x2= no. of standard packages & x3= no. of economy packages]. 2. Find the optimum solution using graphical methods for the LPP and interpret your results. [Hint: Plot constraint lines and mark feasibility region] References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. Operations Research Unit 3 Sikkim Manipal University Page No. 56 E- References: - Linear Optimization http://www.mirrorservice.org/sites/ home.ubalt.edu/ntsbarsh/Business-stat/opre/partVIII.htm. Operations Research Unit 4 Sikkim Manipal University Page No. 57 Unit 4 Simplex Method Structure: 4.1 Introduction Objectives 4.2 Standard Form of LPP Fundamental theorem of LPP 4.3 Solution of LPP Simplex Method Initial basic feasible solution of an LPP To solve an LPP in canonical form by simplex method 4.4 The Simplex Algorithm Steps 4.5 Penalty Cost Method or Big M-method 4.6 Two Phase Method 4.7 Solved Problems on Minimisation 4.8 Summary 4.9 Glossary 4.10 Terminal Questions 4.11 Answers 4.12 Case Study 4.1 Introduction In the previous unit we dealt with graphical analysis of linear programming problems. We discussed some basic definitions, graphical methods to solve LPP, some exceptional cases, advantages, limitations, and important geometric properties of LPP. In this unit, we will deal with the simplex method, which focuses on solving an LPP of any enormity involving two or more decision variables. The simplex algorithm is an iterative procedure for finding the optimal solution to a linear programming problem. The objective function controls the development and evaluation of each feasible solution to the problem. If a feasible solution exists, it is located at a corner point of the feasible region determined by the constraints of the system. The simplex method simply selects the optimal solution amongst the set of feasible solutions of the problem. The efficiency of this algorithm is because it considers only those feasible solutions which are provided by the corner points, and that too not all of them. Operations Research Unit 4 Sikkim Manipal University Page No. 58 Objectives: After studying this unit, you should be able to: - create a standard form of LPP from the given problem - apply the simplex algorithm to the system of equations - find solution using big M-technique - explain the importance of the two phase method 4.2 Standard Form of LPP The characteristics of the standard form of LPP are: - All constraints are equations. The right-hand side element of each constraint equation is non-negative. - All the variables are non-negative. - The objective function is of maximisation type. The inequality constraints of equations are obtained by adding or subtracting the left-hand side of each such constraint by a non-negative variable. 1. Introduce slack variables (Si) for s type of constraint. 2. Introduce surplus variables (Si) and artificial variables (Ai) for > type of constraint. 3. Introduce only artificial variable for = type of constraint. 4. Cost (Cj) of slack and surplus variables will be zero and that of artificial variable will be M. To make the right-hand side of a constraint equation positive, multiply both sides of the resulting equation by -1. Use the elementary transformations introduced with the canonical form to achieve the remaining characteristics. Any standard form of the LPP is given by Maximise or Minimise z Cixi n i =1 = Subject to: aij x j Si bi (bi 0)i 1,2...........m. n j 1 = > = = and xj> 0, j = 1, 2, --- n S i> 0, i = 1, 2, --- m Operations Research Unit 4 Sikkim Manipal University Page No. 59 4.2.1 Fundamental theorem of LPP A set of m simultaneous linear equations in n unknowns/variables, where n > m, AX = b, with r (A) = m. If there is a feasible solution X > 0, then there exists a basic feasible solution. Self Assessment Questions 1. We add surplus variable for constraint. (True/False) 2. The right-hand side element of each constraint is non-negative in the standard form. (True/False) 4.3 Solution of the LPP Simplex Method Consider an LPP given in the standard form. To optimise z = c1 x1 + c2 x2 + ---+ cnxn Subject to a11 x1 + a12 x2 + -- + an x n S1 = b1 a21 x1 + a22 x2 + ----+ a2nxn S2 = b2 . am1 x1 + am2 x2 + -- + amnxnSm = bm x1, x2, --- xn, S1, S2 ---, Sm> 0 To each of the constraint equations, add a new variable called an artificial variable on the left-hand side of every equation, which does not contain a slack variable. Subsequently, every constraint equation will contain either a slack variable or an artificial variable. The introduction of slack and surplus variables does not alter the equations of constraints or the objective function. Therefore, such variables can be incorporated in the objective function with zero coefficients. However, the artificial variables do change the constraints as these are added only to the left-hand side of the equations. The newly derived constraint equation is equivalent to the original equation, only if all the artificial variables have a value zero. Artificial variables are incorporated into the objective function with very large positive coefficient M in the minimisation program and very large negative coefficientM in the maximisation program guaranteeing that the artificial variable takes the zero Operations Research Unit 4 Sikkim Manipal University Page No. 60 value in optimal solutions. The large positive and negative coefficients represent the penalty incurred for making a unit assignment to the artificial variable. Thus the standard form of LPP can be given as follows: Optimise Z = CT X Subject to AX = B, and X > 0 Where X is a column vector with decision, slack, surplus, and artificial variables, C is the vector corresponding to the costs; A is the coefficient matrix of the constraint equations and B is the column vector of the righthand side of the constraint equations. Solved problem 1: Minimise Z = 4x1 + x2 Subject to 3x1 + x2 = 3 4x1 + 3x2> 6 x1 + 2x2s 3 x1, x2> 0 Solution: Rewriting in the standard form, Minimise Z = 4x1 + x2 + 0.S1 + 0.S2 + M (A1 + A2) Subject to 3x1+ x2 + A1 = 3 4x1 + 3x2 S1 + A2 = 6 x1 + 2x2 + S2 = 3 x1, x2, S1, S2, A1, A2> 0 Operations Research Unit 4 Sikkim Manipal University Page No. 61 When S2 is a slack variable, S1 is a surplus variable, and A1 and A2 are artificial variables. Representing this program in matrixes, you have Z = (4 1 0 0 M M) | | | | | | | | | | | | . |
\ | 2 1 2 1 2 1 A A S S x x > 0 4.3.1 Initial basic feasible solution of an LPP Consider a system of m equations in n unknowns x1, x2 - - xn, a11 x1 + a12 x2 + - - + a1nxn = b1 a21 x1 + a22 x2 + - - + a2nxn = b2
am1 x1 + am2 x2 + - - + amnxn = bn Where m s n To solve this system of equations, you must first assign any of n m variables with value zero. The variables assigned the value zero are called Operations Research Unit 4 Sikkim Manipal University Page No. 62 non-basic variables, while the remaining variables are called basic variables. Solve the equation to obtain the values of the basic variables. If one or more values of the basic variables are valued at zero, then the solution is said to degenerate, whereas if all basic variables have non-zero values, then the solution is called a non-degenerate solution. A basic solution satisfying all constraints is said to be feasible. Consider an LPP given in the standard form Optimise Z = CTX Subject to AX = B, X > 0 To arrive at the initial solution of such a problem denoted by X0, treat all decision and surplus variables as non-basic variables. Assign value zero; all slack and artificial variables as basic variables are assigned values, which are on the right-hand side of the corresponding constraint equations. Solved problem 2 To obtain basic solution of the system of equations 2x1 + x2 x3 = 2 3x1 + 2x2 + x3 = 3 where x1, x2, x3 > 0 Solution: At the onset, assign 3 2 = 1 variable the value zero since there are three variables and two equations. Case (i) Let x3 = 0 i.e., x3 be a non-basic variable. Then, the equation becomes 2x1 + x2 = 2 3x1 + 2x2 = 3 Solving, we get x1 = 1, x2 = 0. Therefore, the solution degenerates, but is feasible. Case (ii) Let x2 be a non-basic variable i.e., x2 = 0, then solution is x1 = 1 and x3 = 0. Here also the solution degenerates but is feasible Operations Research Unit 4 Sikkim Manipal University Page No. 63 Case (iii) Let x1 be non-basic i.e., x1= 0 Solution is x2 = 5/3 , x3 = 3 1 The solution non-degenerates, but is not feasible. 4.3.2 To solve an LPP in canonical form by simplex method To solve an LPP problem in canonical form by the simplex method, follow the steps below. 1. Zj = sum of [cost of variable x its coefficients in the constraints profit or cost coefficient of the variable]. 2. Convert the problem into standard form. 3. Find Zj - Cj for each variable. 4. Slack and artificial variables will form basic variable for the first simplex table. Surplus variable will never become basic variable for the first simplex table. 5. Select the most negative value of Zj - Cj. That column is called key column. The variable corresponding to the column will become basic variable for the next table. 6. Divide the RHS quantities by the corresponding values of the key column to get ratios; select the minimum ratio. This becomes the key row. The basic variable corresponding to this row will be replaced by the variable found in step 5. 7. The element that lies both on key column and key row is called pivotal element. 8. Once an artificial variable is removed as basic variable, its column will be deleted from the next iteration. 9. For maximisation problems, decision variables coefficient will be same as in the objective function. For minimisation problems, decision variables coefficients will have opposite signs as compared to objective function. 10. Values of artificial variables will always be M for both maximisation and minimisation problems. 11. The process is continued till all Zj - Cj > 0. Operations Research Unit 4 Sikkim Manipal University Page No. 64 Self Assessment Questions 3. A basic solution is said to be a feasible solution if it satisfies all constraints. (True/False) 4. If one or more values of basic variable are zero then solution is said to be degenerated. (True/False) 5. The element that lies both on key column and key row is called pivotal element. (True/False) 4.4 The Simplex Algorithm To test for optimality of the current basic feasible solution of an LPP uses the following algorithm called simplex algorithm. Lets also assume that there are no artificial variables existing in the program. 4.4.1 Steps Figure 4.1 depicts the steps to solve the simplex algorithm. Fig. 4.1: Steps to Solve Simple Algorithm Operations Research Unit 4 Sikkim Manipal University Page No. 65 Let us now study the steps in detail. 1) Locate the negative number in the last row of the simplex table. Do not include the last column. The column that has negative number is called the pivot column. 2) Next, form ratios by dividing each positive number in the pivot column, excluding the last row into the element in the same row and last column. Assign that element to the pivot column to yield the smallest ratio as the pivot element. If more than one element yields the same smallest ratio, choose the elements randomly. The program has no solution, if none of the element in the pivot column is non-negative. 3) To convert the pivot element to unity (1) and then reduce all other elements in the pivot column to zero, use elementary row operations. That is to divide each element in key row by key element, to find new values for that row. 4) To make other elements in key column zero, use the formula: Number in new row = Number in old row [ Number above or below key element * Corresponding number in the new row] 5) Replace the x-variable in the pivot row and first column by x-variable in the first row pivot column. The variable to be replaced is called the outgoing variable and the variable that replaces it is called the incoming variable. This new first column is the current set of basic variables. 6) Repeat steps 1 through 4 until all the negative numbers in the last row excluding the last column are exhausted. 7) You can obtain the optimal solution by assigning the value to each variable in the first column corresponding to the row and last column. All other variables are considered as non-basic and are assigned value zero. The associated optimal value of the objective function is the number in the last row and last column for a maximisation program, but the negative of this number for a minimisation problem. Operations Research Unit 4 Sikkim Manipal University Page No. 66 Case-let A manufacturing company discontinued production of an unprofitable product line. This created excess production capacity. The companys management is considering devoting this excess capacity to one or more of the other ongoing products. Knowing the capacity of the machines, the number of machine hours required to produce one unit of each product and the unit profit per product, the management needs to find out how much of each product the company should produce to maximise profit. The management can use the simplex method to arrive at the required solution. Solved problem 3 Maximise z = x1+ 9x2 + x3 Subject to x1 + 2x2 + 3x3s 9 3x1 + 2x2 + 2x3s 15 x1, x2, x3 > 0 Rewriting in the standard form Maximise z = x1 + 9x2 + x3 + 0.S1 + 0.S2 Subject to the conditions x1 + 2x2 + 3x3 + S1 = 9 3x1 + 2x2 + 2x3 + S2 = 15 x1, x2, x3, S1, S2> 0 Where S1 and S2 are the slack variables. Solution: The initial basic solution is x1, x2, x3 =0, S1 = 9, S2 = 15 Therefore, X0 = | | | . |
\ | 2 1 S S , C0 = | | | . |
\ | 0 0 Table 4.1 depicts the initial simplex table. Operations Research Unit 4 Sikkim Manipal University Page No. 67 Table 4.1: Initial Simplex Table X1 1 X2 9 X3 1 S1 0 S2 0 Ratio S1 0 S2 0 1 3 2* 2 3 2 1 0 0 1 9 15 2 9 = 4.5 2 15 = 7.5 Zj Cj 1 9 | 1 0 0 Work column* pivot element S1 outgoing variable, x2 incoming variable Since the three Zj Cj are negative, the solution is not optimal. Choose the maximum negative value that is 9. The corresponding column vector x2 enters the basis replacing S1, since ratio is at minimum. You can use the elementary row operations to reduce the pivot element to 1 and other elements of work column to zero. Divide row 1 with the key element 2 and get the new row. Now make all other elements in the key column zero. First iteration The variable x2 becomes a basic variable replacing S1 and you obtain the table 4.2. Table 4.2: First Iteration Table, Variable x2 Becomes A Basic Variable X1 X2 X3 S1 S2 Ratio 1 9 1 0 0 X2 9 2 1 1 2 3 2 1 0 2 9 S2 0 2 0 1 1 1 6 Zj Cj 7/2 0 2 25 2 9 0 2 81 Since all the elements of the last row are non-negative, the optimal solution is obtained. The maximum value of the objective function Z is 2 81 , achieved for x2= 2 9 S2 = 6, are the basic variables (refer the shaded column of table 4.2.). All other variables are non-basic. Operations Research Unit 4 Sikkim Manipal University Page No. 68 Solved problem 4 Use simplex method to solve the LPP Maximise Z = 2x1 + 4x2 + x3 + x4 Subject to x1 + 3x2 + x4 s 4 2x1 + x2 s 3 x2 + 4x3 + x4 s 3 x1, x2, x3, x4 > 0 Rewriting in the standard form Maximise Z = 2x1 + 4x2 + x3 + x4 + 0.S1 + 0.S2 + 0.S3 Subject to x1 + 3x2 + x4 + S1 = 4 2x1 + x2 + S2 = 3 x2 + 4x3 + x4 + S3 = 3 x1, x2, x3, x4, S1, S2, S3 > 0 Solution: The initial basic solution is x1, x2, x3, x4 = 0, S1 = 4, S2 = 3, S3 = 3 Therefore, X0 = | | | . |
S2 0 2 1 0 0 0 1 0 3 1 3 S3 0 0 1 4 1 0 0 1 3 1 3 Zj Cj 2 4 1 1 0 0 0 0 | pivot column * pivot element Operations Research Unit 4 Sikkim Manipal University Page No. 69 S1 is the outgoing variable and x2 is the incoming variable to the basic set. The first iteration gives the table 4.4: Table 4.4: The First Iteration Table - X2 is the Incoming Variable X1 X2 X3 X4 S1 S2 S3 Ratio 2 4 1 1 0 0 0 X2 4 3 1 1 0 3 1 3 1 0 0 3 4 S2 0 3 5 0 0 3 1
3 1 1 0 3 5 S3 0 3 1 0 4* 3 2
3 1 0 1 3 5 12 5
Zj Cj 3 2 0 1 3 1 3 4 0 0 3 16 | pivot column X3 enters the new basic set replacing S3, the second iteration gives the table 4.5: Table 4.5: Second Iteration Table X3 Replaces S3 X1 X2 X3 X4 S1 S2 S3 Ratio 2 4 1 1 0 0 0 X2 4 3 1 1 0 3 1 3 1 0 0 3 4 4 S2 0 3 5 * 0 0 3 1
3 1 1 0 3 5 1 x3 1 12 1 0 1 6 1
12 1 0 4 1 12 5 Zj Cj 4 3 0 0 2 1 4 5 0 4 1 4 23 | pivot column X1 enters the new basic set replacing S2, the third iteration gives the table 4.6: Operations Research Unit 4 Sikkim Manipal University Page No. 70 Table 4.6: Third Iteration Table X1 Replaces S2 X1 X2 X3 X4 S1 S2 S3 Ratio 2 4 1 1 0 0 0 X2 4 0 1 0 5 2 5 2
5 1 0 1 X1 2 1 0 0 5 1
5 1 5 3 0 1 X3 1 0 0 1 20 3
10 1 20 1 4 1 2 1 Zj Cj 0 0 0 27/20 10 11 20 9 4 1 2 13 Since all the elements of the last row are non-negative, the optimal solution is Z = 2 13 , which is achieved for x2 = 1, x1 = 1, x3 = 2 1 , and x4 = 0. Case Study A manufacturing firm has discontinued production of a certain unprofitable product line. This created considerable excess of production capacity. The management is considering devoting this excess capacity to one or more of the three products, called product 1, 2, and 3. Table 4.7 depicts the available capacity on the machines, which might limit output. Table 4.7: Capacity of the Machine for Production Machine Type Available Time (in machine hours per week) Milling Machine 250 Lathe 150 Grinder 50 Table 4.8 depicts the number of machine-hours required for each unit of the respective product. Operations Research Unit 4 Sikkim Manipal University Page No. 71 Table 4.8: Machine Hours for Each Unit Productivity (in Machine hours/Unit) Machine Type Product 1 Product 2 Product 3 Milling Machine 8 2 3 Lathe 4 3 0 Grinder 2 0 1 The unit profit would be Rs.20, Rs.6, and Rs.8 for products 1, 2, and 3. Find how much of each product the firm should produce in order to maximise profit? Let x1, x2, and x3 units of products 1, 2, and 3 be produced in a week. Then the total profit from these units is Z = 20 x1 + 6 x2 + 8 x3 To produce these units, the management requires 8x1 + 2x2 + 3x3 machine hours of milling machine 4x1 + 3x2 + 0 x3 machine hours of lathe and 2x1 + x3 machine hours of grinder Since the time available for these three machines are 250, 150, and 50 hours respectively, we have 8x1 +2x2 + 3x3 s 250 4x1 + 3x2 s 150 2x1 + x3 s 50 Obviously x1, x2, x3 > 0 Thus the problem is to Maximise Z = 20x1 + 6x2 + 8x3 Subject to 8x1 + 2x2 + 3x3 s 250 4x1 + 3x2 s 150 2x1 + x3 s 50 x1, x2, x3 > 0 Operations Research Unit 4 Sikkim Manipal University Page No. 72 Rewriting in the standard form, Maximise Z = 20x1 + 6x2 + 8x3 + 0S1 + 0S 2 + 0S 3 Subject to 8x1 + 2x2 + 3x3 + S1 = 250 4x1 + 3x2 + S2 = 150 2x1 + x3 + S3 = 50 x1, x2, x3, S1, S2, S 3 > 0 Solution: The initial basic solution is x1, x2, x3 =0, X0 = | | | . |
20 6 8 0 0 0 0 | pivot column* pivot element Operations Research Unit 4 Sikkim Manipal University Page No. 73 X1 enters the basic set of variables replacing the variable S3. The first iteration gives the table 4.10: Table 4.10: The First Iteration Table- X1 Replaces S3 X1 X2 X3 S1 S2 S3 Ratio 20 6 8 0 0 0 S1 0 0 2 1 1 0 4 50 2 50 = 25 S2 0 0 3* 2 0 1 2 50 3 50 X1 20 1 0 2 1 0 0 2 1 25 Zj Cj 0 6 2 0 0 10 500 | pivot column * pivot element X2 enters the basic set of variables replacing the variable S2. The second iteration gives the following table: Table 4.11: The Second Iteration Table- X2 Replaces S2 X1 X2 X3 S1 S2 S3 Ratio 20 6 8 0 0 0 S1 0 0 0 3 1 1 3 2
3 8 3 50 50 X2 6 0 1 3 2 0 3 1
3 2 3 50
X1 20 1 0 * 2 1 0 0 2 1 25 50 Zj Cj 0 0 2 0 2 6 600 | Work column * pivot element. X3 enters the basic set of variables replacing the variable X1. The third iteration yields the following table: Operations Research Unit 4 Sikkim Manipal University Page No. 74 Table 4.12: The Third Iteration Table- X3 Replaces X1 X1 X2 X3 S1 S2 S3 Ratio 20 6 8 0 0 0 S1 0 3 2 0 0 1 3 2 3 0 X2 6 3 4 1 0 0 3 1 0 50 X3 8 2 0 1 0 0 1 50 Zj cj 4 0 0 0 2 8 700 Since all zj cj> 0 in the last row, the optimum solution is 700. i.e., the maximum profit is Rs.700, which is achieved by producing 50 units of product 2 and 50 units of product 3. Self Assessment Questions 6. The element that lies both on key column and key row is called pivotal element (True/False) 7. Pivotal element lies on the crossing of key column and key row. (True/False) 8. In the simplex algorithm, the variable to be replaced is called the outgoing variable and the variable that replaces it is called the incoming variable (True/False) 4.5 Penalty Cost Method or Big-M Method Consider an LPP in which at least one of the constraints is of type > or =. While expressing in the standard form, add a non-negative variable to each of such constraints. These variables are called artificial variables. Addition of artificial variables causes violation of the corresponding constraints as they are added to only one side of an equation. The new system is equivalent to the old system of constraints only if the artificial variables are valued at zero. To guarantee such assignments in the optimal solution, you can incorporate artificial variables into the objective function with large positive or negative coefficients in minimisation and maximisation programs respectively. You denote these coefficients by M. Whenever artificial variables are a part of the initial solution X0, the last row of simplex Operations Research Unit 4 Sikkim Manipal University Page No. 75 table will contain the penalty cost M. You can make the following modifications in the simplex method to minimise the error of incorporating the penalty cost in the objective function. This method is called big Mmethod or penalty cost method. Fig. 4.2: Steps in Big M-method Let us now study the steps in detail. 1) The last row of the simplex table is decomposed into two rows, the first of which involves those terms not containing M, while the second involves those containing M. 2) The step 1 of the simplex method is applied to the last row created in the above modification and followed by steps 2, 3, and 4 until this row contains no negative elements. Then step 1 of simplex algorithm is applied to those elements next to the last row that are positioned over zero in the last row. 3) Whenever an artificial variable ceases to be basic, it is removed from the first column of the table as a result of step 4. It is also deleted from the top row of the table as is the entire column under it. 4) The last row is removed from the table whenever it contains all zeroes. Operations Research Unit 4 Sikkim Manipal University Page No. 76 5) If non-zero artificial variables are present in the final basic set, then the program has no solution. In contrast, zero valued artificial variables in the final solution may exist when one or more of the original constraint equations are redundant. Solved problem 5 Use penalty cost method to Maximise z = 2x1 + 3x2 Subject to x1 + 2x2s 2 6x1 + 4x2> 24 x1, x2> 0 Solution: Rewriting in the standard form, we have Maximise Z = 2x1 + 3x2 + 0S1 + 0S2 M A1 Subject to x1 + 2x2 + S1 = 2 6x1 + 4x2 S2 + A1 = 24 x1, x2, S1, S2, A1 > 0 Table 4.13 depicts the initial simplex table. Table 4.13: The Initial Simplex Table X1 X2 S1 S2 A1 2 3 0 0 M Ratio S1 0 1* 2 1 0 0 2 2 1 2 = A1 M 6 4 0 1 1 24 4 6 24 = 2 3 0 0 0 0 6M 4M 0 M 24M | Work column The first iteration gives the following table: Operations Research Unit 4 Sikkim Manipal University Page No. 77 Table 4.14: The First Iteration Table X1 X2 S1 S2 A1 2 3 0 0 M X1 2 1 2 1 0 0 2 Ai M 0 8 -6 1 1 12 0 1 2 0 0 4 0 8M 6M 1M 0 12M The process is complete as all the elements of the last two rows are nonnegative. But existence of non-zero artificial variables in the basic solution shows that the problem has no solution (infeasible). Self Assessment Questions 9. The value of artificial variable is M. (True/False) 10. Artificial variables are entered as basic variables. (True/False) 4.6 Two Phase Method The drawback of the penalty cost method is the possible computational error resulting from assigning a very large value to the constant M. To overcome this difficulty, a new method is considered, where the use of M is eliminated by solving the problem in two phases. Phase I: Formulate the new problem. Start by eliminating the original objective function by the sum of the artificial variables for a minimisation problem and the negative of the sum of the artificial variables for a maximisation problem. The simplex method optimises the objective of maximisation or minimisation with the constraints of the original problem. If a feasible solution is arrived, the optimal value of the new objective function is zero (suggestive of all artificial variables being zero). Subsequently proceed to phase II. If the optimal value of the new objective function is non-zero, it means there is no solution to the problem and the method terminates. Phase II: Start phase II using the optimum solution of phase I as the base. Then take the objective function without the artificial variables and solve the problem using the simplex method. Operations Research Unit 4 Sikkim Manipal University Page No. 78 Solved problem 6 Use the two phase method to Maximise z = 3x1 x2 Subject to 2x1 + x2 > 2 x1 + 3x2 s2 x2 s4, x1, x2 > 0 Rewriting in the standard form, Maximise z = 3x1 x2 + 0S1 MA1 + 0.S2 + 0.S3 Subject to 2x1 + x2 S1 + A1 = 2 x1 + 3x2 + S2 = 2 x2 + S3 = 4, x1, x2, S1, S2, S3, A1 > 0 Phase 1: Consider the new objective, Maximise Z* = A1 Subject to 2x1 + x2 S1 + A1 = 2 x1 + 3x2 + S2 = 2 x2 + S3 = 4, x1, x2, S1, S2, S3, A1 > 0 Table 4.15 depicts the initial simplex table arrived through the simplex method. Table 4.15:The Initial Simplex Table X1 X2 S1 A1 S2 S3 0 0 0 1 0 0 Ratio A1 1 2* 1 1 1 0 0 2 2 2 = 1 S2 0 1 3 0 0 1 0 2 1 2 = 2 S3 0 0 1 0 0 0 1 4 2 1 1 0 0 0 2 | Work column * pivot element X1 enters the basic set replacing A1. Operations Research Unit 4 Sikkim Manipal University Page No. 79 The first iteration gives the following table: Table 4.16: The First Iteration Table X1 X2 S1 A1 S2 S3 0 0 0 1 0 0 X1 0 1 2 1
2 1 2 1 0 0 1 S2 0 0 2 5 2 1 (1/2) 1 0 1 S3 0 0 1 0 0 0 1 4 0 0 0 1 0 0 0 Phase I is complete since there are no negative elements in the last row. The optimal solution of the new objective is Z* = 0. Phase II: Consider the original objective function, Maximise z = 3x1 x2 + 0S1 + 0S2 + 0S3 Subject to x1 + 2 x2 2 S1 =1 2 5 x2 + 2 S1 + S2 =1 x2 + S3 = 4 x1, x2, S1, S2, S3> 0 Table 4.17 depicts the corresponding simplex table with the initial solution x1 = 1, S2 = 1, S3 = 4. Table 4.17: The Corresponding Simplex Table X1 X2 S1 S2 S3 3 1 0 0 0 Ratio X1 3 1 2 1
2 1 0 0 1 S2 0 0 2 5 2 * 1 1 0 1 2 1 1 = 2 S3 0 0 1 0 0 1 4 0 2 5 -3/2 0 0 3 Operations Research Unit 4 Sikkim Manipal University Page No. 80 | Work column * pivot element Proceeding to the next iteration, we get the following table: Table 4.18: The Phase II Iteration Table X1 X2 S1 S2 S3 3 1 0 0 0 X1 3 1 3 0 1 0 2 S1 0 0 5 1 2 0 2 S3 0 0 1 0 0 1 4 0 10 0 3 0 6 Since all the elements of the last row are non-negative, the current solution is optimal. The maximum value of the objective function is Z = 6, which is attained for X1 = 2, X2 = 0. Solved problem 7 Maximise z = 3x1 + 2x2, Subject to 2x1 + x2 s2, 3x1 + 4x2 S2 + A1> 12, x1, x2 > 0 Rewriting in the standard form, Maximise z = 3x1 + 2x2 + 0S1 + 0.S2 MA1 Subject to 2x1 + x2 + S1 = 2 3x1 + 4x2 S2 + A1 = 2 x1, x2, S1, S2, A1 > 0 Solve the above using the two phase method. Phase I: Consider the new objective function Maximise z* = A1 Subject to 2x1 + x2 + S1 =2 3x1 + 4x2 S2 + A1 = 12, Operations Research Unit 4 Sikkim Manipal University Page No. 81 x1, x2, S1, S2, A1 > 0 Table 4.19 depicts the initial simplex table. Table 4.19: The Initial Simplex Table X1 X2 S1 S2 A1 Ratio 0 0 0 0 1 S1 0 2 1* 1 0 0 2 1 2 = 2 A1 1 3 4 0 1 1 12 4 12 = 3 3 4 0 1 0 12 | Work column The first iteration gives the following table: Table 4.20: The First Iteration Table X1 X2 S1 S2 A1 0 0 0 0 1 X2 0 2 1 1 0 0 2 A1 1 5 0 4 1 1 4 5 0 4 1 0 4 Since all the elements of the last row are non-negative, the procedure is complete. But the existence of non-zero artificial variables in the basic set indicates that the problem is infeasible 4.7 Solved Problems on Minimisation Solved problem 8 Minimise = Z = 3x1 + 8x2 Subject to x1 + x2 = 200 x1> 80 x2s 60 x1, x2 > 0 Solution: In the standard form Operations Research Unit 4 Sikkim Manipal University Page No. 82 Minimise Z = 3x1 + 8x2 + MA1 + OS1 + MA2 + OS2 Subject to x1 + x2 +A1 = 200 x1 S1 + A2 = 80 x2 + S2 = 60 S1, S2, A1, A2> 0 Table 4.21: Simplex Table 1 Cj - 3 - 8 0 0 - M - M C.B B.V X1 X2 S1 S2 A1 A3 Qty Ratio - M A1 1 1 0 0 1 0 200 200 - M A2 1 P.E 0 - 1 0 0 0 80 80 K R 0 S2 0 1 0 1 0 1 60 o Zj - Cj -2M +3 | K C -M + 8 M 0 0 0 Table 4.22: Simplex Table 2 Cj - 3 - 8 0 0 - M C.B B.V X1 X2 S1 S2 A1 Qty Ratio Transformation - M A1 0 1 1 0 1 120 120 R1 1 = R1-R2 1 - 3 x1 1 0 -1 0 0 80 o R2 1 = R2 0 S2 0 1 P.E 0 1 0 60 60 K R R3 1 = R3 Zj - Cj 0 -M + 8| K C -M + 3 0 0 Table 4.23: Simplex Table 3 Cj - 3 - 8 0 0 C.B B.V X1 X2 S1 S2 A1 Qty Ratio Transformation - M A1 0 0 1 P.E - 1 1 60 60 R1 1 = R1 - 3 x1 1 0 - 1 0 0 80 -ve R2 1 = R2 - 8 x2 0 1 0 1 0 60 o R3 1 = R3 R1 1 Zj - Cj 0 0 - M + 3 | KC M - 8 0 0 Operations Research Unit 4 Sikkim Manipal University Page No. 83 Table 4.24: Simplex Table 4 Cj - 3 - 8 0 0 C.B B.V X1 X2 S1 S2 Qty Ratio Transfor-mation 0 S1 0 0 1 - 1 60 - ve R1 1 = R1 - 3 x1 1 0 0 - 1 140 -ve R2 1 = R2 + R1 1 - 8 x2 0 1 0 1 P.E 60 80 K R R3 1 = R1 1 Zj - Cj 0 0 0 - 5 Table 4.25: Simplex Table 5 Cj - 3 - 8 0 0 C.B B.V X1 X2 S1 S2 Qty Ratio Transformation 0 S1 0 1 1 0 120 R1 1 = R1 + R3 1 - 3 x1 1 1 0 0 200 R2 1 = R2 + R3 1 0 S2 0 1 0 1 60 R3 1 = R3 Zj - Cj 0 5 0 0 Since all Zj - Cj > 0, the optimum solution is x1 = 200, x2 = 0, and min Z = 600. 4.8 Summary Let us recapitulate the important concepts discussed in this unit: - The simplex algorithm is an iterative procedure for finding the optimal solution to a linear programming problem - The simplex method simply selects the optimal solution amongst the set of feasible solutions of the problem. - The inequality constraints of equations are obtained by adding or subtracting the left-hand side of each such constraint by a non-negative variable - The introduction of slack and surplus variables does not alter the equations of constraints or the objective function. 4.9 Glossary Slack variable: variable added to the left of s type of constraint to change it to equality. Operations Research Unit 4 Sikkim Manipal University Page No. 84 Surplus variable: variable subtracted from the left of the right > type of constraint to change it to equality. Simplex Table: table used in the simplex algorithm to keep trace of computations made at every iteration. 4.10 Terminal Questions 1. Maximise z = 3x1 x2 Subject to the constraints 2x1 + x2 > 2 x1 + 3x2 s 3 x2 s 4 x1, x2 > 0 2. Minimise z = 6x1 + 7x2 Subject to the constraints x1 + 3x2 > 12 3x1 + x2 > 12 x1 + x2 > 8 x1, x2 > 0 4.11 Answers Self Assessment Questions 1. False 2. True 3. True 4. True 5. True 6. True 7. False 8. True 9. True 10. True Operations Research Unit 4 Sikkim Manipal University Page No. 85 Terminal Questions 1. Rewrite in the standard form. The maximum value of the objective function is Z = 9 which is attained for X1 = 3, X2 = 0. Refer 4.6 2. In the standard form minimize Z, Optimum solution is Z = 5.8, X1= 8/3, X2 = 6. Refer 4.7 4.12 Case Study New Plant Construction The state electricity board is planning to construct a new plant for the next 10 years. It is possible to construct four types of electric power facilities steam plants using coal for energy, hydro electric plants with no reservoir, hydroelectric plants with small reservoirs, and hydroelectric plants with large reservoirs. Consumption of electricity is based on three characteristics. The first is the total annual usage the requirement in the area is estimated to be 4,000 billion kilowatt-hours by the 10th year. The second characteristic is the peak usage of power usually on a hot summer day at about 2 P.M. Any plan should provide enough peaking capacity to meet a projected peak need of 3,000 million kilowatts in the 10th year. The third characteristic is guaranteed power output measured as the average daylight output in midwinter when the consumption is high and water levels for hydroelectric power are low. The 10-year requirement is for 2,000 million kilowatts of guaranteed power. The various possible power plants vary in terms of how they satisfy characteristics. For example, hydro electric plants with reservoirs provide substantial peaking capacity, whereas steam plants and hydroelectric plants with no reservoirs are poor in this respect. The characteristics of the various types of plants are shown in the table below. Each is measured in terms of a unit of capacity. The unit of capacity is defined to be the capacity to produce 1 billion kilowatt-hours per year. The annual operating costs of the various types of plants also vary considerably. The final column in the table shows the discounted total costs, including both investments costs and the discounted annual operating costs. Operations Research Unit 4 Sikkim Manipal University Page No. 86 Characteristics of electric plants per unit (1 bn KWh) of annual output. Type Guaranteed Output (millions of kilowatt) Peak Output (million of kilowatt) Investment Cost (Rs. 000) Discounted Total cost (Rs. 000) Steam 0.15 0.20 1200 2600 Hydroelectric: no reservoir 0.10 0.10 1600 1680 Hydroelectric: small reservoir 0.10 0.40 2400 2560 Hydroelectric : large reservoir 0.80 0.90 4000 4400 Discussion Questions: 1. Help the company in developing a 10-year plan that would detail the capacity of each type of plant that it should build. 2. Develop an LP model and solve it to minimise the total discounted cost. However, there is a restriction that no more than Rs.14,000 million can be used for investment in plants over the 10 years. References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E- Reference : - The Design of Manufacturing Systems. http://www.springerlink.com Operations Research Unit 5 Sikkim Manipal University Page No. 87 Unit 5 Duality in Linear Programming Problem Structure: 5.1 Introduction Objectives 5.2 Importance of Duality Concepts 5.3 Formulation of Dual Problem 5.4 Economic Interpretation of Duality Economic interpretation of dual variables 5.5 Sensitivity Analysis Changes in Cj of a non-basic variable Change in Cj of a basic variable Change in available resources Calculating the range 5.6 Summary 5.7 Glossary 5.8 Terminal Questions 5.9 Answers 5.10 Case Study 5.1 Introduction In the previous unit, simplex method, we studied about the standard form and the solution for linear programming problem. The simplex method focuses on solving linear programming problem of any enormity involving two or more decision variables. The simplex algorithm is an iterative procedure for finding the optimal solution to a linear programming problem. The objective function controls the development and evaluation of each feasible solution to the problem. If a feasible solution exists, it is located at a corner point of the feasible region determined by the constraints of the system. In this unit, duality in linear programming problem, we will study about the importance of duality concepts and the formulation of dual concepts. We will also learn about the sensitivity analysis. Every Linear Programming Problem (LPP) is associated with another linear programming problem involving the same data and optimal solutions. The two problems are said to be duals of each other. One problem is called the primal, while the other problem is called the dual. Operations Research Unit 5 Sikkim Manipal University Page No. 88 The dual formulation is derived from the same data and solved in a manner similar to the original 'primal' formulation. In other words, you can say that dual is the 'inverse' of the primal formulation due to the following reasons: - If the primal objective function is 'maximisation' function, then the dual objective function is 'minimisation' function and vice-versa. - The column coefficient in the primal constraint is the row coefficient in the dual constraint. - The coefficients in the primal objective function are the Right Hand Side (RHS) constraint in the dual constraint. - The RHS column of constants of the primal constraints becomes the coefficient of the dual objective function. The concept of duality is useful to obtain additional information about the variation in the optimal solution. These changes could be affected in the constraint coefficient, in resource availabilities and/or objective function coefficient. This effect is termed as post optimality or sensitivity analysis. Objectives: After studying this unit, you should be able to: - describe the duality concept - solve a dual problem - describe the economical interpretation - apply sensitivity analysis 5.2 Importance of Duality Concepts The importance of duality is due to two main reasons: - If the primal contains a large number of constraints and a smaller number of variables; the labour of computation can be considerably reduced, by converting it into the dual problem and then solving it. - The interpretation of the dual variable from the loss or economic point of view proves extremely useful in planning future decisions. Characteristics of dual solutions If the primal problem possesses a unique non-degenerate, optimal solution, then the optimal solution to the dual is unique. However, dual solutions arise under a number of other conditions which are as follows: Operations Research Unit 5 Sikkim Manipal University Page No. 89 - When the primal problem has a degenerate optimal solution, the dual has multiple optimal solutions. - When the primal problem has multiple optimal solutions, the optimal dual solution is degenerate. - When the primal problem is unbounded, the dual is infeasible. - When the primal problem is infeasible, the dual is unbounded or infeasible. Self Assessment Questions 1. Dual linear programming problem always reduces the amount of computation. (True/False) 2. It is possible to reverse the dual linear programming problem to primal linear programming problem. (True/False) 5.3 Formulation of Dual Problem Consider the following linear programming problem: Maximise Z = c1x1 +c2x2 + . . . +cnxn Subject to the constraints a11 x1 + a12 x2 + . . . + a1n xn b1 a21 x1 + a22 x2 + . . . + a2n xn b2 am1 x1 + am2 x2 + . . . + amn xn bm x1, x2, . . ., xn 0 To construct a dual problem, we must adopt the following guidelines: - The maximisation problem in the primal becomes a minimisation problem in the dual and vice versa. - () type of constraints in the primal becomes () type of constraints in the dual and vice versa. - The coefficients c1, c2,...,cn in the objective function of the primal become b1, b2,,bm in the objective function of the dual. - The constants b1, b2,,bm in the constraints of the primal become c1, c2,...,cn in the constraints of the dual. - If the primal has n variables and m constraints the dual will have m variables and n constraints. - The variables in both the primal and dual are non-negative. Operations Research Unit 5 Sikkim Manipal University Page No. 90 Thus the dual problem will be as follows: Minimise W = b1 y1 + b2 y2 + . . . +bmym Subject to the constraints a11 y1 + a21 y2 + . . . + am1ym c1 a12 y1 + a22 y2 + . . . + am2ym c2 a1n y1 + a2n y2 + . . . + amnym cn y1, y2, . . .,ym 0 Formation of dual linear programming problem is easier; when the canonical form of linear programming problem for maximisation problem must contain type of constraints, while for minimisation problem, it must contain type of constraints. Solved problem 1 Write the dual of Max Z = 4x1 + 5x2 Subject to 3x1 + x2 15 x1 +2 x2 10 5x1 + 2x2 20 x1, x2, 0 Solution The given problem is in its standard form. Therefore, its dual is: Min W = 15y1 + 10 y2 + 20 y3 Subject to 3y1 + y2 + 5 y3 4 y1 + 2y2 + 2y3 5 y1, y2, y3, 0 Solved problem 2 Write the dual of Min Z = 10x1 + 12x2 Subject to 2x1 +3 x2 10 5x1 +6 x2 20 x1 + 2x2 15 2x1 + 3x2 12 x1, x2, 0 Solution The given problem is in its standard form. Therefore, its dual problem is: Max W = 10y1 + 20 y2 + 15 y3 + 12y4 Operations Research Unit 5 Sikkim Manipal University Page No. 91 Subject to 2y1 + 5y2 + y3 + 2y4 12 6y1 + 2y2 + 3 y3 10 y1, y2, y3,y4 0 Solved problem 3 Write the dual of Max Z = 100x1 + 200x2 Subject to 3x1 - 10 x2 15 +4x1 +15 x2 20 x1, x2, 0 Solution First we convert 3x1 - 10x2 15 to type as shown here. -3x1 + 10x2 -15 Therefore, its dual is: Min Z = -15y1 + 20 y2 Subject to -3y1 + 4y2 100 10y1 + 15y2 200 y1, y2, 0 Solved problem 4 When the constraints contain = sign, write the dual of Max Z = 40x1 + 30x2 Subject to 10x1 +6 x2 15 5x1 +7x2 -10 x1 + x2 = 9 x1 + x2 10 x1, x2, 0 Solution Firstly, 5x1 7x2 -10 Operations Research Unit 5 Sikkim Manipal University Page No. 92 is rewritten as -5x1 +7x2 10 Secondly, x1 + x2 = 9 is written as x1 + x2 9 Also x1 + x2 9 this is same as -x1 - x2 -9 Therefore, the given problem is: Max Z = 40x1 + 30x2 Subject to 10x1 + 6 x2 15 - 5x1 +7 x2 10 x1 +x2 9 - x1 x2<=- 9 -x1 x2 -10 Therefore, its dual is: Min W = 15y1 + 10 y2 + 9y3 1 9y3 11 -10y4 Subject to 10y1 - 5y2 + y3 1 y3 11 y4 40 6y1 + 7y2 + y3 1 y3 11-y4 30 y1, y2, y3 1 y3 11,y4 0 Let y3 1 y3 11= y3 Then the dual becomes Min W =- 15y1 +10 y2 +9 y3 -10y4 Subject to 10y1 - 5y2 + y3-y4 40 6y1 + 7y2 + y3 +y4 30 y1, y2,y4 0 y3 is unrestricted in sign Operations Research Unit 5 Sikkim Manipal University Page No. 93 Solved problem 5 Write the dual of Min Z = 12x1 + 15x2 Subject to 5x1 +3 x2 10 x1 + x2 5 x1 0 x2 is unrestricted in sign Solution Let x2 =x2 1 x2 11 0. Therefore, its standard form is: Min Z = 12x1 + 15x2 1 -15x2 11 Subject to 5x1 +3(x2 1- x2 11) 10 - [x1 +(x2 1- x2 11)] -5 Or - x1 - x2 1+ x2 11 -5 x1, x2 1, x2 11 0 Therefore, dual is: 5y1 y2 12 .(1) 3y1 y2 15 .(2) -3y1 + y2 - 15 (3) No.3 constraint is 3y1 y2 15 (4) Constraints 2 and 4 give 3y1 y2 = 15 The dual problem is: Max W = 10y1 5 y2 Subject to 5y1 y2 12 3y1 y2 =15 y1 0 y2 unrestricted in sign Solved problem 6 Write the dual of the following linear programming problem: Minimise Z = 3x1 2x2 + 4x3 Subject to 3x1 + 5x2 + 4x3 7 6x1 + x2 + 3x3 4 7x1 - 2x2 - x3 10 x1 - 2x2 + 5x3 3 4x1 + 7x2 - 2x3 2, x1, x2, x3 0 Operations Research Unit 5 Sikkim Manipal University Page No. 94 Solution Since the problem is of minimisation, all constraints should be of type. Therefore, you have to multiply the third constraint throughout by -1 so that -7x1 + 2x2 + x3 -10. Let y1, y2, y3, y4 and y5 be the dual variables associated with the above five constraints. Then the dual problem is given by: Maximise W = 7y1 + 4 y2 10 y3 + 3 y4 + 2y5 Subject to 3y1 + 6y2 7 y3 + y4 + 4y5 3 5y1 + y2 + 2y3 2y4 + 7y5 -2 4y1 + 3y2 + y3 + 5y4 2y5 4 y1, y2, y3, y4, y5 0 Solved problem 7 Find the dual of: Maximise 12x1 + 10x2 Subject to 2x1 + 3x2 18 2x1 + x2 14 x1, x2 0 Solution The 'dual' formulation for this problem will be as follows: Minimise 18y1 + 14y2 Subject to 2y1 + 2y2 12 3y1 + y2 10 y1>= 0, y2 0 Points to be noted - The column coefficients in the primal constraint namely (2, 2) and (3, 1) have become the row coefficients in the dual constraints. - The coefficient of the primal objective function namely 12 and 10 have become the constants in the right hand side of the dual constraints. - The constants of the primal constraints, namely 18 and 14, have become the coefficient in the dual objective function. - The direction of the inequalities has been reserved. - While the primal is a maximisation problem, the dual is a minimisation problem. Operations Research Unit 5 Sikkim Manipal University Page No. 95 Solved problem 8 Obtain the dual problem of the following primal formulation: Maximise Z = 2x1 + 5x2 + 6x3 Subject to 5x1 + 6x2 -x3 3 -2x1 + x2 + 4x3 4 x1 - 5x2 + 3x3 1 -3x1 - 3x2 + 7x3 6 x1, x2, x3 0 Solution Step 1 - Write the objective function of the dual. As there are four constraints in the primal, the objective function of the dual will have 4 variables. Minimise r* = 3 w1 + 4 w2 + w3 + 6 w4 Step 2 - Write the constraints of the dual. As all the constraints in the primal are '<', the constraints in the dual will be '>'. The column COM+ efficient of the primal becomes the row coefficient of the dual. Constraints 5w1 - 2w2 + w3 - 3w4 2 6w1 + w2 - 5w3 -3w4 5 -w1 + 4w2 + 3w3 + 7w4 6 Step 3 - Therefore the dual of the primal is: Minimise Z = 3w1 + 4w2 + w3 + 6w4 Subject to: 5w1 - 2w2 + w3 - 3w4 2 6w1+ w2 - 5w3 - 3w4 5 - w1 + 4w2 + 3w3 + 7w4 6 w1, w2, w3, w4 >= 0 Solved problem 9 Obtain the dual of the following linear programming problem: Minimise Z = 5x1 -6x2 + 4x3 Subject to the constraints: 3x1 + 4x2 + 6x3 9 x1 + 3x2 + 2x3 5 7x1 - 2x2 - x3 10 x1 - 2x2 + 4x3 4 2x1 + 5x2 - 3x3 3 x1, x2, x3 0 Operations Research Unit 5 Sikkim Manipal University Page No. 96 Solution As you can see, one of the primal constraints is a " constraint while the others are all " constraints. The dual cannot be worked out unless all the constraints are in the same direction. To convert this into " constraint, multiply both the sides of the equation by "-" sign. After multiplying the constraint by "-" sign, it will become 7x1 + 2x2 + x3 -10. Now that all the constraints are in the same direction and the dual can be worked out, the dual formulation is: Maximise Z = 9w1 + 5w2 - 10w3 + 4w4 + 3w5 Subject to: 3w1 + w2 - 7w3 + w4 + 2w5 5 4w1 + 3w2 + 2w3 - 2w4 + 5w5 -6 6w1 + 2w2 + w3 + 4w4 + -3w5 4 w1, w2, w3, w4, w5 0 Self Assessment Questions 3. The coefficients of decision variables in the objective function become quantities on the right hand side of ___________________. 4. s constraints changes to ________ type in dual linear programming. 5. For every linear programming problem, there exists a unique ________ problem. 5.4 Economic Interpretation of Duality The linear programming problem is thought of as a resource allocation model, where the objective is to maximise revenue or profit subject to limited resources. The associated dual problem offers interesting economic interpretations of the linear programming resource allocation model. Consider a representation of the general primal and dual problems where primal takes the role of a resource allocation model. Table 5.1 depicts general primal and dual problems in a resource allocation model. Operations Research Unit 5 Sikkim Manipal University Page No. 97 Table 5.1: Resource Allocation Model Primal Dual Maximise
= = n j 1 z c j. x j Subject to
= s = n j 1 aij x j bi , i 1, 2, ...,m. xj 0, j = 1,2,., n Minimise
= = m i 1 w bi. yi Subject to
= > = m i 1 aij yi ci , j 1, 2, ..., n. yj 0, i = 1,2,., m In this resource allocation model, the primal problem has n economic activities and m resources. The coefficient cj in the primal represents the profit per unit of activity j. Resource i, whose maximum availability is bi, is consumed at the rate aij units per unit of activity j. 5.4.1 Economic interpretation of dual variables For any pair of feasible primal and dual solutions, (objective value in the maximisation problem) (objective value in the minimisation problem). At the optimum, the relationship holds as a strict equation. Note: Here, the sense of optimisation is very important. Hence, for any two primal and dual feasible solutions, the values of the objective functions, when finite, must satisfy the following inequality:
= = s = n j 1 m i 1 z c j x j bi y i w The strict equality, z = w, holds true when both the primal and dual solutions are optimal. Consider the optimal condition z = w. Given that the primal problem represents a resource allocation model, you can think of z as representing the profit in rupees. bi represents the number of units available of the resource i. Therefore, we can express the equation z = w as profit (Rs) = (units of resource i) x (profit per unit of resource i). This means that the dual variables yi, represent the worth per unit of resource i. Operations Research Unit 5 Sikkim Manipal University Page No. 98 Note: Variables yi are also called as dual prices, shadow prices and simplex multipliers. With the same logic, the inequality z < w associated with any two feasible primal and dual solutions is interpreted as (profit) < (worth of resources). This relationship implies that as long as the total return from all the activities is less than the worth of the resources, the corresponding primal and dual solutions are not optimal. Optimality is achieved only when the resources have been exploited fully, which can happen when the input equals the output (profit). Economically, the system is said to be disequilibrium (non optimal) when the input (worth of the resources) exceeds the output (return). Equilibrium occurs only when the two quantities are equal. Self Assessment Questions 6. Dual variables represent the worth or unit of a resource. (True/False) 7. Optimality is reached when the resources are not fully utilised. (True/False) 8. At optimum level, the relationship holds as a strict equation. (True/False) 5.5 Sensitivity Analysis The management of a company rarely restricts its interest to the numerical values of an optimal solution. It is interested in knowing the impact of changes in the input parameter values on the optimal solution. This process is called sensitivity analysis. In linear programming, all model parameters are assumed to be constant. However; in real-life situations, the decision environment is always dynamic. Therefore, it is important for the management to know how profit would be affected; by an increase or decrease in the resource level, by change in the technological process, and by change in the cost of raw materials. Such an investigation is known as sensitivity analysis or post optimality analysis. The results of sensitivity analysis establishes upper and lower bounds for input parameter values within which they can vary without causing major changes in the current optimal solution. Sensitivity analysis allows us to figure out which data offers a significant impact on the results. This in turn allows us to concentrate on getting accurate data for those items, or at least running through several scenarios Operations Research Unit 5 Sikkim Manipal University Page No. 99 with various values of the crucial data, to get an idea of the range of possible outcomes. Sensitivity analysis is important due to the following reasons: - values of linear programming parameters might change - linear programming parameters have an uncertainty factor attached to them Any change in a linear programmings parameters affects the optimality of the basic variable. This is due to the following two reasons: - When a variable (or number of variables) in row 0 has a negative coefficient, we can obtain a better basic feasible solution of larger z-value, by pivoting a non-basic variable with a negative coefficient in row 0. If this occurs, the basic variable becomes a sub-optimal basis. - A constraint (or number of constraints) may have a negative RHS. In such a case, at least one member of basic variable will become negative and will no longer yield a basic feasible solution. This is an infeasible basic. The following six types of changes in linear programmings parameters cause changes in the optimal solution: - Changing the objective function coefficient of a non-basic variable. - Changing the objective function coefficient of a basic variable. - Changing the right-hand side of a constraint. - Changing the column of a non-basic variable. - Adding a new variable or activity. - Adding a new constraint. Solved problem 10 Luminous lamps produce three types of lamps A, B and C. These lamps are processed on three machines X, Y and Z. To find out the optimal solution in order to maximise profit and minimise cost, first we find out whether a previously determined optimal solution remains optimal if the contribution rate is changed. An increase in Cj of a variable would mean that, resources from other products should be diverted to this more profitable product. The reverse is true for a minimisation problem. Operations Research Unit 5 Sikkim Manipal University Page No. 100 Table 5.2 depicts the full technology and input restrictions. Table 5.2: Lamp Production and Profit Per Unit Product Machine Profit Per Unit M1 M2 M3 A 10 7 2 12 B 2 3 4 3 C 1 2 1 1 Available time 100 77 80 Solution The linear programming model for this problem is as follows: Maximise Z = 12 x1 + 3 x2 + x3 Subject to10 x1 + 2 x2 + x3 s 100 7 x1 + 3 x2 + 2x3 s 77 2 x1 + 4 x2 + x3 s 80 x1, x2 , x3 > 0 On solving this problem by the simplex method, we get x1 = 8 73 , x2 = , 8 35 x3 = 0 And the optimal value of the objective function is 8 981 . 5.5.1 Changes in Cj of a non-basic variable A non basic variable can be brought on the basis only if its contribution rate is attractive. Therefore, we need to determine the upper limit of the profit contribution (Cj) of each non basic variable. The reverse is true for a minimisation problem. Table 5.3 depicts the final simplex table of this problem. Operations Research Unit 5 Sikkim Manipal University Page No. 101 Table 5.3: Final Simplex Table Cj 12 3 1 0 0 0 CB Basic variables B x1 x2 x3 s1 s2 s3 Solution values b (=xB) 12 x1 1 0 16 1 16 3
8 1 0 8 73 3 x2 0 1 16 13
16 7 8 5 0 8 35 0 s3 0 0 8 17
8 11 4 9 1 4 177 zj 12 3 16 27 16 15 8 3 0 zjcj 0 0 16 11 16 15 8 3 0 From this final simplex table, we can see that profit contribution for product C is Re 1, which is not greater than its zj. Therefore, to bring x3 into the basis, its profit contribution rate cj must exceed Rs. 16 27 to make zjcj value negative or zero that is zjcj s 0. Specifically, - If cj* cj>zj cj,, then a new optimal solution must be derived. - If cj* cj = zj cj, then an alternative optimal solution exists. - If cj* cj<zj cj, then the current optimal solution remains unchanged. In case of c3 = 1 and z3 c3 = 16 11 , C3* 1 > 16 11 C3* > 16 11 +1 = 16 27 x3 can be introduced in the basis, if its contribution rate c3 increases to at least Rs 16 27 . If it increases beyond that, then the current solution will no longer be optimal. Operations Research Unit 5 Sikkim Manipal University Page No. 102 5.5.2 Change in Cj of a basic variable Consider the case of product A (x1 column) and divide each zj - cj entry in the index row by the corresponding coefficient in the x1 row as shown here: - Minimum | | . |
\ | <
sA s | | . |
\ | >
;y 0 y Z C ;y 0 C Minimum y Z C ij ij j j ij i ij j j Referring to the final simplex table, we observe that corresponding to the non basic variables x3 and x5, y13, y15<0. Therefore, Minimum ( ( ( (
16 3 16 15 = 5 Hence, 5 s c1 * -12 s 3 i.e. 7 s c1 * s 15 Thus, we can say that the optimal solution is insensitive as long as the changed profit coefficient c1 * varies between Rs. 7 and Rs. 15. 5.5.3 Change in available resources If the available resources change, we need to investigate whether a previous optimal solution remains feasible or not. For long term planning, it is important to know the bounds within which each available resource (for instance machine hours) can vary, without causing drastic changes in the current optimal solution. To illustrate this, divide each quantity in the XB column by the corresponding coefficient in the X4 column of the table 5.4. Table 5.4 depicts the resources table. Operations Research Unit 5 Sikkim Manipal University Page No. 103 Table 5.4: Resources Table XB X4 X X 4 B 8 73 16 3 3 146 8 35 16 7 10 4 177 8 11 11 354 The least positive ratio | | . |
\ | 11 354 indicates the number of hours of machine M1 that can be decreased. The least negative ratio (10) indicates the increase in the number of hours of machine M1. 5.5.4 Calculating the range Lower limit = 100 11 354 = 11 746 Upper limit = 100 (10) = 110 Therefore, the range of hours for M1 is 11 746 to 110. We can calculate the range of hours for machine M2 and M3 in the same manner. Self Assessment Questions 9. Sensitivity analysis is carried out on _______ simplex table. 10. Sensitivity analysis helps us to study the effect of changes in _______ _______ in objective function. 11. The results of sensitive analysis establish ___________ and ________ ___________ for input parameters value. Operations Research Unit 5 Sikkim Manipal University Page No. 104 5.6 Summary Let us recapitulate the important concepts discussed in this unit: - For every linear programming problem there exists a dual linear programming problem. - The dual linear programming problem helps us to reduce the amount of calculation involved in original linear programming problem. - The dual linear programming problem also helps us to interpret the economic variables more effectively. 5.7 Glossary Optimality: most desirable solution subject to constraints Canonical form: standard form Primal problem: original linear programming problem 5.8 Terminal Questions Write the dual of the following linear programming problems: 1. Maximise 2 = 7x1 + 5x2 a. Subject to constraints x1 + 2x2s 6 b. 4x1 + 3x2s 12 c. x1 x2 > 0 2. Maximise z = 3x1 + 4x2 a. Subject to constrains 5x1 + 4x2s 200; b. 3x1 + 5x2s 150; c. 5x1 + 4x2> 100; d. 8x1 + 4x2> 80, e. x1> 0, x2> 0 3. Maximise z = 2x1 + x2 a. Subject to constraints 4x1 + 3x2s 12, b. 4x1 + x2s 8, c. 4x1 x2s 8, d. x1, x2> 0 Operations Research Unit 5 Sikkim Manipal University Page No. 105 5.9 Answers Self Assessment Questions 1. False 2. True 3. Dual 4. 5. Dual 6. True 7. False 8. True 9. Final 10. Resource levels 11. Upper, lower, bounce Terminal Questions 1. Minimise W = 6y1 + 12y2 Subject to constraints y1 + 4y2 7 2y1 + 3y2 5 y1 + y2 0 2. Minimise w = 200y1 + 150y2 100y3 80y4 Subject to constraints 5y1 + 3y2 5y3 8y4 3 4y1 + 5y2 4y3 4y4 4 y1, y2, y3, y4 0 3. Minimise w = 12y1 + 8y2 + 8y3 Subject to constraints 4y1 + 4y2 + 4y3 2 3y1 + y2 - y3 1 y1, y2, y3, 0 5.10 Case Study Managing a Farm The Oxman family owns and operates a 640-acre farm that has been in the family for several generations. The Oxmans always have had to work hard to make a decent living from the farm and have had to endure some occasional difficult years. Stories about earlier generations overcoming Operations Research Unit 5 Sikkim Manipal University Page No. 106 hardships due to droughts, floods, etc., are an important part of the family history. However, the Oxmans enjoy their self-reliant lifestyle, and gain considerable satisfaction from continuing the family tradition of successfully living off the land during an era, when many family farms are being abandoned or taken over by large agricultural corporations. John Oxman is the current manager of the farm while his wife Eunice runs the house and manages the farms finances. Johns father, grandpa Oxman, lives with them and still puts in many hours working on the farm. John and Eunices older children, Frank, Phyllis, and Carl, also are given heavy chores before and after school. The entire family can produce a total of 4,000 person-hours worth of labour during the winter and spring months, and 4,500 person-hours during the summer and fall. If any of these personhours are not needed, Frank, Phyllis, and Carl will use them to work on a neighbouring farm for $5 per hour during the winter and spring months and $5.50 per hour during the summer and fall. The farm supports two types of livestock: dairy cows and laying hens, as well as three crops: soybeans, corn, and wheat. (All three are cash crops, but the corn is also a feed crop for data per acre planted and wheat is also is used for chicken feed.) The crops are harvested during the late summer and fall. During the winter months, John, Eunice, and grandpa make a decision about the mix of livestock and crops for the coming year. Currently, the family has just completed a particularly successful harvest, which has provided an investment fund of $20,000 that can be used to purchase more livestock. (Other money is available for ongoing expenses, including the next planting of crops). The family currently has 30 cows valued at $35,000 and 2,000 hens valued at $5,000. They wish to keep all this livestock and perhaps purchase more. Each new cow would cost $1,500, and each new hen would cost $3. Over a years time; the value of a herd of cows will decrease by about 10 percent, and the value of a flock of hens will decrease by about 25 percent due to aging. Each cow will require 2 acres of land for grazing and 10 person-hours of work per month, while producing a net annual cash income of $850 for the family. The corresponding figures for each hen are: no significant acreage, 0.05 person-hours/month, and an annual net cash income of $4.25. The chicken house can accommodate a maximum of Operations Research Unit 5 Sikkim Manipal University Page No. 107 5,000 hens, and the size of the barn limits the herd to a maximum of 42 cows. For each acre planted in each of the three crops, table 5.5 depicts the number of person-hours of work that will be required during the first and second halves of the year, as well as a rough estimate of the crops net value (in either income or savings in purchasing feed for the livestock). Table 5.5: Corns Net Value Data per acre planted Soybeans Corn Wheat Winter and spring, person-hours 1.0 0.9 0.6 Summer and fall, person-hours 1.4 1.2 0.7 Net value Rs. 70 Rs. 60 Rs. 40 To provide much of the feed for the livestock; John wants to plant at least 1 acre of corn for each cow in the fund. Plus the value of the livestock at the end of the coming year, plus any income from working on a neighbouring farm, minus living expenses of $40,000 for the year coming years herd, in addition, at least 0.05 acre of wheat for each hen in the coming years flock. John, Eunice, and grandpa are now discussing how much acreage should be planted in each of the crops and how many cows and hens to have for the coming year. Their objective is to maximise the familys monetary worth at the end of the coming year; (the sum of the net income from the livestock for the coming year plus the net value of the crops for the coming year plus what remains from the investment). Discussion Questions: 1. Identify verbally the components of a linear programming model for this problem. 2. Formulate this model. (Either an algebraic or a spreadsheet formulation is acceptable.) 3. Obtain an optimal solution and generate the additional output provided for performing post optimality analysis (for example, the sensitivity report when using excel). What does the model predict regarding the familys monetary worth at the end of the coming year? Operations Research Unit 5 Sikkim Manipal University Page No. 108 References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E-References: - http://www.psnacet.edu.in/courses/MBA/Linearprogramming/lecture- 11.pdf - http://agecon2.tamu.edu/people/faculty/mccarl-bruce/mccspr/new04.pdf - http://www.sce.carleton.ca/faculty/chinneck/po/Chapter6.pdf. - http://www-ee.eng.hawaii.edu/~jyee/491.06/lectures/chapter06.pdf Operations Research Unit 6 Sikkim Manipal University Page No. 109 Unit 6 Transportation Problem Structure: 6.1 Introduction Objectives 6.2 Formulation of Transportation Problem (TP) 6.3 Transportation Algorithm (MODI Method) 6.4 The Initial Basic Feasible Solution North-west corner rule Matrix minimum method Vogels approximation method 6.5 Moving Towards Optimality Improving the solution Modified distribution method/MODI method/U V Method Degeneracy in transportation problem 6.6 Summary 6.7 Glossary 6.8 Terminal Questions 6.9 Answers 6.10 Case Study 6.1 Introduction In the previous unit, duality in linear programming problem, we studied about how every linear programming problem (LPP) is associated with another linear programming problem involving the same data and optimal solutions. We also learnt about formulation of dual problem, economic interpretation of duality and sensitivity analysis. In this unit, transportation problem, we will study about the formulation of transportation problem. We will learn about the initial basic feasible solution and the various methods. We will also study how to improve the solution and about the modified distribution method. To facilitate the presentation and solution, the general transportation problem is normally portrayed in a tabular form. Transportation model is an important class of linear programs. For a given supply at each source and a given demand at each destination, the model studies the minimisation of the cost of transporting a commodity from a number of sources to several destinations. Operations Research Unit 6 Sikkim Manipal University Page No. 110 To illustrate the background of a common transportation problem, a typical situation is portrayed here. Assume that petrol is imported and supplied from three ports, that is, Mumbai, Kolkata and Chennai. The destination cities where petrol needs to be supplied are Delhi, Bangalore, Hyderabad and Ahmedabad. Petrol can be transported from any of the ports to any of the destination cities; however the transportation costs vary from one place to another. The transportation problem in this unit helps in determining the quantity of petrol that needs to be supplied from each of the ports to the destination cities, such that the total transportation cost is minimised after satisfying the supply and demand conditions. Figure 6.1 depicts a transportation problem. Port Cities (Source) (Destination) Fig. 6.1: A Transportation Problem Where, Si = Quantity of petrol available at source i, (i=1, 2, 3) Dj = Quantity of petrol required at destination j, (j = 1, 2, 3, 4) Cij = The cost of shipping one barrel of petrol from source i to destination j for each route Xij =The number of barrels of petrol per route from source i to destination j Mumbai S1 Kolkata S2 Chennai S3 Delhi D1 Hyderabad D2 Bangalore D3 Ahmedabad D4 Operations Research Unit 6 Sikkim Manipal University Page No. 111 To facilitate the presentation and solution, the general transportation problem is normally portrayed in a tabular form by means of a transportation tableau as discussed later in this unit. Objectives: After studying this unit, you should be able to: - identify the transportation problem - locate the initial basic feasible solution - compare the advantages of various methods of finding initial basic feasible solution - apply the model to minimise the cost of transporting a commodity - solve the degeneracy in the transportation problem 6.2 Formulation of Transportation Problem (TP) The transportation problem involves m sources, each of which has available ai (i = 1, 2 m) units of homogeneous product and n destinations, each of which requires bj (j = 1, 2., n) units of products. Here ai and bj are positive integers. The cost cij of transporting one unit of the product from the ith source to the jth destination is given for each i and j. The objective is to develop an integral transportation schedule that meets all demands from the inventory at a minimum total transportation cost. Table 6.1 depicts a transportation problem represented in the form of a transportation table. Table 6.1 Transportation Table D1 D2 Dn Supply S1 x11 x12 x1n a1 S2 x21 x22 x2n a2 S3 x31 x32 x3n a3 Sm xm1 xm2 xmn am Demand b1 b2 bn ai = bi C11 C21 C31 Cm 1 C12 C22 C32 Cm 2 C1n C2n C3n Cm n Operations Research Unit 6 Sikkim Manipal University Page No. 112 Each source has one row and each destination has one column. The capacity of each source is shown at the end of each row, and each destination is written under its corresponding column. These capacities and demands are known as rim conditions. The unit cost to ship from each source to each destination is written in the upper-right-hand corner of the cell in the matrix. If the total demand is equal to the total supply it is a balanced transportation problem. When the demand is not equal to the supply it is an unbalanced transportation problem. It is assumed that the total supply and the total demand are equal.
= = = n j 1 m i 1 ai bj (1) The condition (1) is guaranteed by creating; either a fictitious destination with a demand equal to the surplus if total demand is less than the total supply, or a (dummy) source with a supply equal to the shortage if total demand exceeds total supply. The cost of transportation from the fictitious destination to all sources and from all destinations to the fictitious sources are assumed to be zero, so that total cost of transportation will remain the same. If the total demand is greater than total supply, then there is no feasible solution. The standard mathematical model for the transportation problem is as follows: Let xij be number of units of the homogenous product to be transported from source i to the destination j Then objective is to minimise z = ij m i 1 n j 1 Cij x = = Subject to
)
`
= = = =
= = x bj;ij 1,2,.......... .,n x a ,i 1,2,......,m m i 1 ij n j 1 ij i (2) With all xij > 0 and integers Operations Research Unit 6 Sikkim Manipal University Page No. 113 Theorem: A necessary and sufficient condition for the existence of a feasible solution to the transportation problem (2) is:
= = = n j 1 m i 1 ai bj Self Assessment Questions 1. Transportation problems are a special type of ___________. 2. The number of rows and columns need not always be ___________. 3. If total demand is greater than total supply then there is _____ ____ solution. 6.3 Transportation Algorithm (MODI Method) A feasible solution has to be found always. Rather than determining a first approximation by a direct application of the simplex method, it is more efficient to work with the transportation table. The transportation algorithm is the simplex method specialised to the format of table involving the following steps: i) Finding an initial basic feasible solution ii) Testing the solution for optimality iii) Improving the solution, when it is not optimal iv) Repeating steps (ii) and (iii) until the optimal solution is obtained The solution to transportation problem is obtained in two stages In the first stage, we find the basic feasible solution using any of the following methods: - North-west corner rule - Matrix minima method or least cost method - Vogels approximation method. In the second stage, we test the basic feasible solution for its optimality by MODI method. Self Assessment Questions 4. Basic feasible solution In transportation problem can be found using North-west corner rule. (True/False) 5. Transportation problems can also be solved by simplex method. (True/False) Operations Research Unit 6 Sikkim Manipal University Page No. 114 6.4 The Initial Basic Feasible Solution Let us consider a transportation problem involving m-origins and ndestinations. Since the sum of origin capacities equals the sum of destination requirements, a feasible solution always exists. Any feasible solution satisfying m + n 1 of the m + n constraints is a redundant one and hence, it can be deleted. This also means that a feasible solution to a transportation problem can have only m + n 1 positive component; otherwise, the solution will degenerate. It is always possible to assign an initial feasible solution to a transportation problem, satisfying all the rim requirements. This can be achieved either by inspection or by following some simple rules. We can begin by imagining that the transportation table is blank that is initial xij = 0. The simplest procedures for initial allocation are discussed in the following section. 6.4.1 North-west corner rule Step 1 - The first assignment is made in the cell occupying the upper left hand (north-west) corner of the transportation table. The maximum feasible amount is allocated here is: x11 = min (a1, b1) Either the capacity of origin O1 is used up or the requirement at destination D1 is satisfied or both. This value of x11 is entered in the upper left hand corner (small square) of cell (1, 1) in the transportation table. Step 2 - If b1> a1, the capacity of origin O is exhausted and the requirement at destination D1 is still not satisfied, then, at least one variable in the first column will have to take on a positive value. Move down vertically to the second row and make the second allocation of magnitude: x21 = min (a2, b1 x21) in the cell (2, 1) This either exhausts the capacity of origin O2 or satisfies the remaining demand at destination D1. If a1> b1, the requirement at destination D1 is satisfied, but the capacity of origin O1 is not completely exhausted. Move to the right in a horizontal position to the second column to make the second allocation of magnitude: x12 = min (a1 x11, b2) in the cell (1, 2) Operations Research Unit 6 Sikkim Manipal University Page No. 115 This either exhausts the remaining capacity of origin O1 or satisfies the demand at destination D2. If b1 = a1, the origin capacity of O1 is completely exhausted as well as the requirement at destination is completely satisfied, then there is a tie at the second allocation. An arbitrary tie breaking choice is made. Make the second allocation of magnitude: x12 = min (a1 a1, b2) = 0 in the cell (1, 2) or x21 = min (a2, b1 b2) = 0 in the cell (2, 1) Step 3 - Start from the new north-west corner of the transportation table; satisfying the destination requirements and exhausting the origin capacities one at a time, moving down towards the lower right corner of the transportation table, until all the rim requirements are satisfied. Application - It is used for transportation to satisfy such obligations where cost is not the criteria. For example, in the case of Food Corporation of India Ltd. Solved problem 1 Table 6.2 depicts a transportation problem. Determine an initial basic feasible solution using the north west corner rule: Table 6.2 Transportation Problem D1 D2 D3 D4 Supply O1 6 4 1 5 14 O2 8 9 2 7 16 O3 4 3 6 2 5 Demand 6 10 15 4 35 Where Oi and Dj represent the ith origin and the jth destination respectively. Operations Research Unit 6 Sikkim Manipal University Page No. 116 Solution 1. Start with the top most left corner 2. Allocate the maximum possible amount here. 3. Move to the right hand cell if there is still any available quantity, otherwise move to down cell. Again perform step 2 and continue until all the available quantity is exhausted. Following north west corner rule, the first allocation is made in the cell (1,1), the magnitude being x11 = min (14, 6) = 6 The second allocation is made in the 6 10 15 4 cell (1, 2) and the magnitude of allocation is given by x12= min (14 6, 10) = 8 The third allocation is made in the cell (2, 2), the magnitude being x22 = min (16, 10 8) = 2. The magnitude of fourth allocation, in the cell (2, 3) is given by x23 = min (16 2, 15) = 14. The fifth allocation is made in the cell (3, 3), the magnitude being x33 = min (5, 15 14) =1. The sixth allocation in the cell (3, 4) is given by x34 = min (5 1, 4) = 4. Now all the rim requirements have been satisfied and hence, an initial feasible solution to the transportation problem has been obtained. Table 6.3 depicts the solution. Table 6.3 Initial Feasible Solution to the Transportation Problem D1 D2 D3 D4 O1 O2 O3 6 6 4 8 1 5 14 8 9 2 2 14 7 16 4 3 6 1 2 4 5 6 10 15 4 Clearly, this feasible solution is non-degenerate basic feasible solution as the allocated cells do not form a loop. The transportation cost according to the loop is given by: Z = x11 c11 + x12 c12 + x22 c22 + x23 c23 + x33 c33 + x34 c34 = 66+84+ 29 +142 + 16+42= 128 Operations Research Unit 6 Sikkim Manipal University Page No. 117 6.4.2 Matrix minimum method Step 1 - Determine the smallest cost in the cost matrix of the transportation table. Let it be cij. Allocate xij= min ( ai, bj) in the cell ( i, j ) Step 2 - If xij= ai cross the ith row of the transportation table, decrease bj by ai and proceed to step 3. If xij= bj cross the ith column of the transportation table, decrease ai by bj and proceed to step 3. If xij= ai= bj cross either the ith row or the ith column, but not both. Step 3 - Repeat steps 1 and 2 to reduce transportation table until all the rim requirements are satisfied. Whenever the minimum cost is not unique, make an arbitrary choice among the minima. Solved problem 2 Obtain an initial basic feasible solution to the following transportation problem using the matrix minima method. Table 6.4 depicts the initial table. Table 6.4 Initial Table D1 D2 D3 D4 Capacity O1 1 2 3 4 6 O2 4 3 2 0 8 O3 0 6 8 6 10 Demand 4 6 8 6 24 Where Oi and Di denote ith origin and jth destination respectively. Solution The transportation table of the given transportation problem has 12 cells. Following the matrix minima method, The first allocation is made in the cells (3, 1), the magnitude being x31 = 4. This satisfies the requirement at destination D1 and thus we cross the first column from the table 6.5. The second allocation is made in the cell (2, 4), the magnitude being x24 = min (6, 8) =6. Cross the fourth column of the table 6.5. This yields the table 6.5 as depicted. Operations Research Unit 6 Sikkim Manipal University Page No. 118 Table 6.5: First Iteration Table 6.6: Second Iteration There is again a tie for the third allocation. Arbitrarily choose the cell (1, 2) and allocate x12 = min (6, 6) = 6. Cross the second column of the first row. Next, choose to cross off the first row of the table 6.5. The next allocation of magnitude x32 = 0 is made in the cell (3, 2), cross the second column. Table 6.7 depicts the result Table 6.7: Third Iteration Table 6.8: Final Solution Again you choose arbitrarily to make the next allocation in cell (2, 3) of magnitude x23 = min (2, 8) = 2, cross the second row in table 6.7. The last allocation of magnitude x33 = min (6, 6) = 6 is made in the cell (3, 3). Now that all the rim requirements have been satisfied, an initial feasible solution has been determined. This solution is depicted in table 6.8. Since the cells do not form a loop, the solution is basic and degenerate. The transportation cost according to the above route is given by: Z=62+22+60+40+02+62=28 6.4.3 Vogels approximation method The Vogels approximation method (VAM) takes into account not only the Operations Research Unit 6 Sikkim Manipal University Page No. 119 least cost cij, but also the cost that just exceeds cij. The steps of the method are given as follows: Step 1 - For each row of the transportation table, identify the smallest and the next to smallest costs. Determine the difference between them for each row. Display them alongside the transportation table by enclosing them in parenthesis against the respective rows. Similarly, compute the differences for each column. Step 2 - Identify the row or column with the largest difference among all the rows and columns. If a tie occurs, use any arbitrary tie breaking choice. Let the greatest difference correspond to the ith row and let Cij be the smallest cost in the ith row. Allocate the maximum feasible amount xij = min (ai, bj) in the (i, j)th cell and cross off the ith row or the jth column in the usual manner. Step 3 - Recompute the column and row differences for the reduced transportation table and go to step 2. Repeat the procedure until all the rim requirements are satisfied. Remarks - A row or column difference indicates the minimum unit penalty incurred by failing to make an allocation to the least cost cell in that row or column. - It is clear that VAM determines an initial basic feasible solution, which is very close to the optimum solution, but the number of iterations required to reach the optimal solution is small. Area of application - It is used to compute transportation routes in such a way as to minimise transportation cost for finding out location of warehouses - It is used to find out locations of transportation corporations depots where insignificant total cost difference may not matter. Operations Research Unit 6 Sikkim Manipal University Page No. 120 Solved problem 3 A company has three production facilities S1, S2 and S3 with production capacity of 7,9 and 18 units (in 100s) per week of a product respectively. These units are to be shipped to four warehouses D1, D2, D3 and D4 with requirement of 5, 8, 7 and 14 units (in 100s) per week, respectively. Table 6.9 depicts the transportation costs (in rupees) per unit between factories to warehouses. Table 6.9: Transportation Costs D1 D2 D3 D4 S1 19 30 50 10 S2 70 30 40 60 S3 40 8 70 20 Solution The differences (penalty costs) for each row and column have been calculated as depicted in table 6.10. In the first round, the maximum penalty, 22 occurs in column D2. Thus the cell (S3, D2) having the least transportation cost 8 is chosen for allocation. The maximum possible allocation in this cell is 8 and it satisfies demand in column D2. Adjust the supply of S3 from 18 to 10 (18-10). The new row and column penalties are calculated except column D2 because its demand has been satisfied. The second round allocation is made in column D1 with target penalty 21 in the same way as in the first round as depicted in cell (S1, D1) of the table 6.10.. In the third round, the maximum penalty 50 occurs at S3. The maximum possible allocation of 10 units is made in cell (S3, D4) having least transportation cost 20 as depicted in table 6.10. The process is continued with new allocations till a complete solution is obtained. The initial solution using Vogels approximation method is depicted in table 6.10. The total transportation cost associated with this method is calculated as follows: Operations Research Unit 6 Sikkim Manipal University Page No. 121 Total cost = 5 x 9 + 2 X 10 + 7 X 40 + 2 X 60 + 8 X 8 + 10 X 20 = Rs. 779 Table 6.10: Initial Solution using Vogels Approximation Method D1 D2 D3 D4 Supply row difference S1 19 (5) 30 50 10 (2) 7 9 9 40 40 S2 70 30 40 (7) 60 (2) 9 10 20 20 20 S3 40 8 (8) 70 20 (10) 18 12 20 50 X Demand 5 8 7 14 34 Column difference 21 22 10 10 21 X 10 10 X X 10 10 10 50 Self Assessment Questions 6. In matrix-minima method, you start allocating from the upper left hand (north-west) corner of the table. (True/False) 7. In Vogels approximation method, you first construct penalty and then start allocating. (True/False) 8. North-west corner rule gives optimum solution. (True/False) 9. Vogels approximation method gives solution near to the optimum solution. (True/False) 6.5 Moving Towards Optimality After evaluating an initial basic feasible solution to a transportation problem, the next question is how to get the optimum solution. The basic techniques are illustrated as follows: 1. Determine the net evaluations for the nonbasic variables (empty cells) 2. Determine the entering variable Operations Research Unit 6 Sikkim Manipal University Page No. 122 3. Determine the leaving variable 4. Compute a better basic feasible solution 5. Repeat steps (1) to (4) until an optimum solution has been obtained 6.5.1 Improving the solution Definition - A loop is the sequence of cells in the transportation table such that: - Each pair of consecutive cells lie either in the same row or same column - No three consecutive cells lie in the same row or same column - The first and the last cells of the sequence lie in the same row or column - No cell appears more than once in the sequence Consider the non-basic variable corresponding to the most negative of the quantities cij ui vj, calculated in the test for optimality; it is made the incoming variable. Construct a loop consisting exclusively of this incoming variable (cell) and current basic variables (cells). Then allocate the incoming cell to as many units as possible, after appropriate adjustments have been made to the other cells in the loop. Avoid violating the supply and demand constraints, allow all allocations to remain non-negative and reduce one of the old basic variables to zero. 6.5.2 Modified distribution method/MODI method/UV method Step 4 - You repeat steps 1 to 3 to till all allocations are over. Step 5 - For allocating all forms of equations ui+ vj = cj, set one of the dual variable ui / vj to zero and solve for others. Step 6 - Use this value to find Aij = cij - ui - vj. If all Aij> 0, then it is the optimal solution. Step 7 - If any Aijs 0 select the most negative cell and form loop. Starting point of the loop is positive and alternative corners of the loop are negative and positive. Examine the quantities allocated at negative places. Select the minimum, add it to the positive places and subtract from the negative places. Step 8 - Form a new table and repeat steps 5 to 7 till Aij> 0 Operations Research Unit 6 Sikkim Manipal University Page No. 123 Balanced transportation problem Solved problem 4 Table 6.11 depicts the cost coefficients demands and supplies for a transportation problem. Solve the transportation problem. Table 6.11 Cost Coefficients Demands and Supplies A B C Supply I II III 6 8 4 14 12 5 4 9 8 1 2 6 Demand 6 10 15 Solution Since total demand = 31 = total supply, the problem is balanced. The initial basic feasible solution is obtained by Vogels approximation method. Table 6.12 depicts the initial solution. Table 6.12 Initial Solution Supply 6 8 4 14 14 12 5 4 6 9 5 8 1 1 2 5 6 Demand 6 10 15 For allocated cells u1 + v3 = 4 u1 = -4 Set u2 = 0 u1 + v3 = 4 u3= -4 u2 + v2 = 9 v1 = 4 Operations Research Unit 6 Sikkim Manipal University Page No. 124 u2 + v3 = 8 v2 = 9 u3 + v2 = 5 v3 = 8 Note: Variable ui/vi is repeated very often for easy calculation. Here u2 is repeated often. For unallocated cells Aij = cijuivj A11 = 6 (-4) 4 = 6 A12 = 8 (-4) 9 = 3 A31 = 1 (-4) 4 = 1 A33 = 6 (-4) 8 = 2 The optimum allocations are x13 = 14, x21 = 6, x22 = 5, x23 = 1, x32 = 5 The minimum transportation cost is 144+64+59+18+52 = 143 Unbalanced transportation problem Solved problem 5 A car company is faced with an allocation problem resulting from a rental agreement, which allow cars to be returned to locations other than those from where they were originally rented. At the present time there are two locations with 15 and 13 cars respectively and 4 locations requiring 9, 6, 7 and 9 cars respectively. Table 6.13 depicts the unit transportation costs (in dollars) between the locations. Table 6.13 Unit Transportation Costs between Locations Destinations D1 D2 D3 D4 Sources S1 45 17 21 30 S2 14 18 19 31 Obtain a minimum cost schedule. Operations Research Unit 6 Sikkim Manipal University Page No. 125 Solution Since the supply and requirements are not equal it is called an unbalanced transportation problem. In general, if Eai=Ebj then it is called an unbalanced transportation problem. We introduce either a dummy row or a column with cost zero quantities and Ebi=Eaj respectively. Applying Vogels approximation method we find the basic feasible solution. Table 6.14 depicts the basic feasible solution. Table 6.14: Basic Feasible Solution using Vogels Approximation Method D1 D2 D3 D4 Supply row difference S1 45 17 (6) 21(3) 30 (6) 15/9/6 4 4 4 11 S2 14(9) 18 19 (4) 31 13/4 4 4 4 12 S3 0 0 0 0 (3) 3 0 0 0 0 Demand 9/0 6 7/3 9/6 31 Column difference 14 17 19 30 31 1 2 1 - 1 2 1 - - 2 1 X34 = Min [3, 9] = 3 x 0 = 0 X21 = Min [13, 9] = 9 x 14 = 126 X12 = Min [15, 6] = 6 x 17 = 102 X23 = Min [4, 7] = 4 x 19 = 76 X14 = Min [6, 6] = 6 x 30 = 180 X13 = Min [6, 3] = 3 x 21 = 63 Total cost 547 Testing for optimality u1 + v2 = 17 Set u1 = 0 u1 + v3 = 21 u2= -2 u1 + v4 = 30 u3 = -30 u2 + v1 = 14 v1 = 16 u2 + v3 = 19 v2 = 17 u3 + v4 = 0 v3 = 21 v4 = 30 Operations Research Unit 6 Sikkim Manipal University Page No. 126 For unallocated cells Aij = cij - ui - vj A11 = 45 0 16 = 29 A22 = 8 + 2 17 = 3 A24 = 31 +2 30 = 3 A31 = 0 + 30 26 = 14 A32 = 0 + 30 17 = 13 A33 = 0 + 30 21 = 9 For non-allocated cells, determine cij uI vi. Since all the quantities are non-negative, the current solution is optimal. The minimum transportation cost is equal to: 617+321+630+914+419+30 = 547. This is achieved by transporting x12 = 6 cars from source 1 to destination 2, x13 = 3, x14 = 6 cars from sources 1 to destinations 3 and 4 respectively; x21 = 9 and x33 = 4 cars from sources 2 to destinations 1 and 3 respectively. 6.5.3 Degeneracy in transportation problem A basic solution to an m-origin, n destination transportation problem can have at the most m+n-1 positive basic variables (non-zero), otherwise the basic solution degenerates. It follows that whenever the number of basic cells is less than m + n 1, the transportation problem is a degenerate one. The degeneracy can develop in two ways: Case 1 - The degeneracy develops while determining an initial assignment via any one of the initial assignment methods discussed earlier. To resolve degeneracy, you must augment the positive variables by as many zero-valued variables as is necessary to complete the required m + n 1 basic variable. These zero-valued variables are selected in such a manner that the resulting m + n 1 variable constitutes a basic solution. The selected zero valued variables are designated by allocating an extremely small positive value to each one of them. The cells containing these extremely small allocations are then treated like any other basic cells. The s are kept in the transportation table until temporary degeneracy is Operations Research Unit 6 Sikkim Manipal University Page No. 127 removed or until the optimum solution is attained, whichever occurs first. At that point, we set each = 0. Case 2 - The degeneracy develops at the iteration stage. This happens when the selection of the entering variable results in the simultaneous drive to zero of two or more current (pre-iteration) basic variables. To resolve degeneracy, the positive variables are augmented by as many zero-valued variables as it is necessary to complete m+n-1 basic variables. These zero-valued variables are selected from among those current basic variables, which are simultaneously driven to zero. The rest of the procedure is exactly the same as discussed in case 1. Note - The extremely small value is infinitely small and it never affects the value it is added to or subtracted from. Introduce e in unallocated minimum cost cell to avoid forming a loop. Self Assessment Questions 10. All the values of Aij = Cij - ui - vj should be __________ or _________ for the solution to be optimum. 11. In unbalanced transportation problem ai is ________ ______ to bj. 12. If the number of allocation is less than _________ then it is said to be a degenerate transportation problem. 6.6 Summary Let us recapitulate the important concepts discussed in this unit: - The transportation problem is a special type of linear programming problem in which the objective is to transport a homogeneous product manufactured at several plants (origins) to a number of different destinations at a minimum total cost. - There are several different techniques for computing an initial basic feasible solution to a transportation problem, such as North-west corner rule, Matrix minimum method and Vogels approximation method. - The degeneracy in transportation problem can develop in two ways o The degeneracy develops while determining an initial assignment via any one of the initial assignment methods discussed. o The degeneracy develops at the iteration stage. Operations Research Unit 6 Sikkim Manipal University Page No. 128 6.7 Glossary Algorithm: sequence of steps to solve complex problems Degeneracy: In transportation problem the number of occupied cells is less than m+n-1 where m is number of rows and n is number of columns 6.8 Terminal Questions 1. Table 6.15 depicts a transportation problem. Find the solution. Table 6.15 Transportation Problem 2. A company has three cement factories located in cities 1, 2, 3 which supply cement to four projects located in towns 1, 2, 3, 4. Each plant can supply daily 6, 1, 10 truck loads of cement respectively and the daily cement requirements of the projects are respectively 7, 5, 3, 2 truck loads. Table 6.16 depicts the transportation cost per truck load of cement (in hundreds of rupees) from each plant to each project site. Table 6.16: Transportation Cost Factories Determine the optimal distribution for the company so as to minimise the total transportation cost. 3. Table 6.17 depicts a transportation problem. Solve the transportation problem using all three methods to obtain the initial feasible solution. 1 2 3 4 1 2 3 11 7 2 1 0 6 1 3 5 8 15 9 Operations Research Unit 6 Sikkim Manipal University Page No. 129 Table 6.17: Transportation Problem A B C D E F supply O1 9 12 9 6 9 10 5 O2 7 3 7 7 5 5 6 O3 6 5 9 11 3 11 2 O4 6 8 11 2 2 10 9 Demand 4 4 6 2 4 2 22 6.9 Answers Self Assessment Questions 1. Linear programming problem 2. Equal 3. No feasible 4. True 5. True 6. True 7. True 8. False 9. True 10. Zero or positive 11. Not equal 12. m + n 1 Terminal Questions 1. The optimal transportation cost is Rs. 796 2 The optimal transportation cost is Rs. 10, 000 3 The minimum transportation cost is Rs. 112 as c 0 Operations Research Unit 6 Sikkim Manipal University Page No. 130 6.10 Case Study Transporting Wood to the Market Amber Wood is a lumber company that has three sources of wood and five markets to be supplied. The annual availability of wood at sources 1, 2 and 3 is 15, 20, and 15 million board feet, respectively. The amount that can be sold annually at markets 1, 2, 3, 4 and 5 is 11, 12, 9, 10 and 8 million board feet respectively. In the past, the company has shipped the wood by train. However, because shipping costs have been increasing, the alternative of using ships to make some of the deliveries is being investigated. This alternative would require the company to invest in some ships. Except for these investment costs, the shipping costs in thousands of rupees per million board feet by rail and by water would be as depicted in table 6.18 for each route: Table 6.18: Rail Costs and Shipping Costs Unit Cost by Rail (Rs. 1,000's) Unit Cost by Ship (Rs. 1,000's) Market Market Source 1 2 3 4 5 1 2 3 4 5 1 61 72 45 55 66 31 38 24 - 35 2 69 78 60 49 56 36 43 28 24 31 3 59 66 63 61 47 - 33 36 32 26 Table 6.19 depicts the capital investment in ships required for each million board feet to be transported annually by ship along each route. Table 6.19: Capital Investment for Ships Investment for Ships ($1,000's) Market Source 1 2 3 4 5 1 275 303 238 - 285 2 293 318 270 250 265 3 - 283 275 268 240 Considering the expected useful life of the ships and the time value of money, the equivalent uniform annual cost of these investments is one-tenth the amount depicted in the table 6.19. The objective is to determine the overall shipping plan that minimises the total equivalent uniform annual cost. Operations Research Unit 6 Sikkim Manipal University Page No. 131 You are the head of the operations research team that has been assigned the task of determining this shipping plan for each of the following three options: - Option 1: Continue shipping exclusively by rail - Option 2: Switch to shipping exclusively by water - Option 3: Ship by either rail or water, depending on which is less expensive for the particular route Discussion Questions: 1. Present your results for each option and compare. 2. Consider the fact that these results are based on current shipping and investment costs, so the decision on the option to adopt now should take into account, managements projection of how these costs are likely to change in the future. For each option describe the scenario of future cost changes that would justify adopting the option now. References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E-References: - http://www.rajalakshmi.org/Department/MECH/2009ES/ME1018 NOL.doc. accessed on April 28, 2010. Operations Research Unit 7 Sikkim Manipal University Page No. 132 Unit 7 Assignment Problem Structure: 7.1 Introduction Objectives 7.2 Mathematical Formulation of the Problem 7.3 Hungarian Method Algorithm 7.4 Routing Problem Unbalanced AP Infeasible assignments Maximisation in AP 7.5 Travelling Salesman Problem 7.6 Summary 7.7 Glossary 7.8 Terminal Questions 7.9 Answers 7.10 Case Study 7.1 Introduction In the previous unit we dealt with the formulation of Transportation Problem (TP), transportation algorithm (MODI method), and the initial basic feasible solution. We also discussed about how to get the optimum solution. In this unit, we will deal with assignment problem. The assignment problem is a special case of transportation problem, where the objective is to minimise the cost or time of completing a number of jobs by a number of persons, and to maximise revenue and sales efficiently. In other words, when the problem involves the allocation of n different facilities to n different tasks, it is often termed as an assignment problem. This model is mostly used for planning. The assignment model is useful in solving problems, such as assignment of machines to jobs, assignment of salesman to sales territories, travelling salesman problem, and many more similar situations. It may be noted that with n facilities and n jobs, there are n possible assignments. One way of finding an optimal assignment is to write all the n possible arrangements, evaluate their total cost, and select the assignment model Operations Research Unit 7 Sikkim Manipal University Page No. 133 offering the minimum cost. This method can be unfeasible due to involvement of computational procedures. In this unit, we will study an efficient method for solving assignment problems. Lets say there are n jobs in a factory having n machines to process the jobs. A job i (=1 n), when processed by machine j (=1 n) is assumed to incur a cost Cij. The assignment is to be made in such a way that each job can be associated with one and only one machine. You can then determine an assignment of jobs to the machines to minimise the overall cost. The cost data is given as a matrix where rows correspond to jobs and columns to machines and there are as many rows as the number of columns. The number of jobs and number of machines should be equal. Assignment becomes a problem because each job requires different skills and the capacity or efficiency of each person with respect to these jobs can be different. This gives rise to cost differences. If each person is able to do all the jobs with the same efficiency then all costs will be the same and each job can be assigned to any person. When assignment is a problem it becomes an optimisation problem. Therefore, you can compare an assignment problem to a transportation problem. The cost element is given and is a square matrix and the requirement at each destination is one and the availability at each origin is also one. Additionally, you have a number of origins, which equal the number of destinations. Therefore, the total demand is equal to the total supply. There is only one assignment in each row and each column. However, if you compare this to a transportation problem, you will find that a general transportation problem does not have the above mentioned limitations. These limitations are peculiar to assignment problems only. An assignment problem can be either balanced or unbalanced. Lets first focus on a balanced assignment problem. A balanced assignment problem is one where the number of rows = the number of columns (comparable to a balanced transportation problem where total demand =total supply). Minimisation case for an assignment problem Figure 7.1 depicts the steps that must be performed to solve the minimisation case for an assignment problem (AP). Operations Research Unit 7 Sikkim Manipal University Page No. 134 Fig. 7.1: Steps to Solve the Minimisation Case for an Assignment Problem Step 1: Determine the total opportunity cost matrix. a) Arrive at a column opportunity cost matrix by subtracting the lowest entry of each column of the given payoff matrix from all the entries in the column. b) Subtract the lowest entry of each row of the matrix obtained in (a) from all the entries in its row. The result of step 1b) gives the total opportunity cost matrix. Step 2: Determine whether an optimal assignment can be made. a) Cover all the zeros of the current total opportunity cost matrix with the minimum possible number of horizontal and vertical lines. b) If the number of lines in step 2a) equals the number of rows (or columns) of the matrix, the problem can be solved. Make a complete assignment so that the total opportunity cost involved in the assignment is zero. c) If the number of lines drawn in step 2a) is less than the number of rows (or columns) of the matrix, proceed to step 3. Step 3: Revise the total opportunity cost matrix. a) Subtract the lowest entry in the uncovered cells of the current total opportunity cost matrix from all the uncovered cells. b) Add the same lowest entry to only those cells in which the covering lines of step 2 cross. Operations Research Unit 7 Sikkim Manipal University Page No. 135 The result of steps 3a) and 3b) is a revised total opportunity cost matrix. Step 4: Repeat steps 2 and 3 until an optimal assignment having a total opportunity cost of zero can be made. Figure 7.2 depicts the various methods used to solve the assignment problem by any of the following four methods. - Enumeration method - Simplex method - Transportation problem - Hungarian method However, in this unit, we will focus on the most commonly used method, that is the Hungarian method, to solve the assignment problems. Fig. 7.2: Methods to Solve Assignment Problem Objectives: After studying this unit, you should be able to: - interpret an assignment problem mathematically - explain a routing problem - describe the travelling salesman problem - state the significance of the assignment problem - compute the problem using the Hungarian method - explain practical problems, such as routing problems and travelling salesman problems Operations Research Unit 7 Sikkim Manipal University Page No. 136 7.2 Mathematical Formulation of the Problem Let xij be a variable defined by
= if the i job is assigned to the j machine job is not assigned to the j machine x th th th ij 1 th 0 if the i Since only one job is assigned to each machine, you have
= n i 1 xij = 1 and = n j 1 xij = 1 Hence, the total assignment cost is given by z = = = n i 1 n j 1 xij cij Thus the assignment problem takes the following mathematical form: Determine xij 0 (i, j =1 n) Minimise z = = = n j 1 n i 1 xij cij Subject to the constraints
= n i 1 xij = 1 j =1, 2, n And = n j 1 xij = 1 i = 1, 2 n With xij = 0 or 1 Note: In an assignment problem, if you add a real number to or subtract a real number from each element of a row or column of the cost matrix, then the optimum assignment for the modified matrix is also optimum for the original one. Operations Research Unit 7 Sikkim Manipal University Page No. 137 Self Assessment Questions 1. In an AP, the constraints are of equality type. (True/False) 2. The number of facilities should be equal to the number of resources. (True/False) 3. A balanced assignment problem is one where the number of rows the number of columns. (True/False) 7.3 Hungarian Method Algorithm Hungarian method algorithm is based on the concept of opportunity cost and is more efficient in solving assignment problems. The following steps are adopted to solve an AP using the Hungarian method algorithm. Step 1: Prepare row ruled matrix by selecting the minimum values for each row and subtract it from the other elements of the row. Step 2: Prepare column-reduced matrix by subtracting minimum value of the column from the other values of that column. Step 3: Assign zero row-wise if there is only one zero in the row and cross (X) or cancel other zeros in that column. Step 4: Assign column wise if there is only one zero in that column and cross other zeros in that row. Step 5: Repeat steps 3 and 4 till all zeros are either assigned or crossed. If the number of assignments is equal to number of rows present, you have arrived at an optimal solution, if not, proceed to step 6. Step 6: Mark () the unassigned rows. Look for crossed zero in that row. Mark the column containing the crossed zero. Look for assigned zero in that column. Mark the row containing assigned zero. Repeat this process till all the makings are done. Step 7: Draw a straight line through unmarked rows and marked column. The number of straight line drawn will be equal to the number of assignments made. Step 8: Examine the uncovered elements. Select the minimum. - Subtract it from the uncovered elements. - Add it at the point of intersection of lines. Operations Research Unit 7 Sikkim Manipal University Page No. 138 - Leave the rest as is. - Prepare a new table. Step 9: Repeat steps 3 to 7 till optimum assignment is obtained. Step 10: Repeat steps 5 to 7 till number of allocations = number of rows. The assignment algorithm applies the concept of opportunity costs. The cost of any kind of action or decision consists of the opportunities that are sacrificed in taking that action. Example 1 Consider the problem of assigning five jobs to five persons. The assignment costs are depicted in table 7.1. Table 7.1: Assignment Costs Table Determine the optimum assignment schedule. Solution: Applying Hungarian method Table 7.2: Row Reduced Matrix (Optimum Assignment Schedule) Column reduced matrix will be the same as each columns minimum value is zero. Then you can start assigning the jobs as depicted in table 7.3. Operations Research Unit 7 Sikkim Manipal University Page No. 139 Table 7.3: Column Reduced Matrix Since the number of assignments is 5. Therefore, the optimum assignment schedule is A to 5, B to 1, C to 4, D to 3, and E to 2. Self Assessment Questions 4. In Hungarian method, you prepare row-reduced matrix. (True/False) 5. The number of assignments should be equal to the number of rows for an optimum solution. (True/False) 6. There can be more than one allocation in a row. (True/False) 7.4 Routing Problem 7.4.1 Unbalanced AP Unbalanced assignment problem is an assignment where the number of rows is not equal to the number of columns and vice versa. For example, the number of machines may be more than the number of jobs or the number of jobs may be more than the number of machines. In such a situation, you have to introduce dummy rows or columns in the matrix. The dummy rows or columns will contain all cost elements as zero. This balances the problem and then you can use Hungarian method to find the optimal assignment. Unbalanced assignment problem: No. of rows No. of columns. Operations Research Unit 7 Sikkim Manipal University Page No. 140 Solved problem 1 Solve the following AP. Table 7.4: Assignment Operations Table Solution: Introducing a dummy row and applying Hungarian method, you have: Table 7.5: Reduced Matrix Table 7.6: Optimum Assignment Solution Hungarian method leads to multiple solutions. Selecting (03, M2) arbitrarily. Therefore, the optimum assignment schedule is O1 to M1, O2 to M3, O3 to M2, and O4 to M4. Operations Research Unit 7 Sikkim Manipal University Page No. 141 7.4.2 Infeasible assignments It is sometimes possible that a particular person is incapable of doing certain work or a specific job cannot be performed on a particular machine. The solution of the assignment problem should take into account these restrictions so that the infeasible assignment can be avoided. This can be achieved by assigning a very high cost (say or M) to the cells where assignments are prohibited, thereby, restricting the entry of this pair of job machine or resource activity into the final solution. After inserting a high value o at the cell we need to apply Hungarian method to solve the problem. Solved problem 2 Solve the following A.P. Table 7.7: Assignment Table Solution: Introducing o in places having dashes and applying Hungarian method, you have: Table 7.8: Assignment Table Table 7.9: Row-reduced Matrix Operations Research Unit 7 Sikkim Manipal University Page No. 142 Table 7.10: Column Reduced Matrix Therefore, the optimum assignment schedule is T1 to P1, T2 to P4, T3 to P2, and T4 to P3. 7.4.3 Maximisation in AP Some assignment problems are phrased in terms of maximising the profit or effectiveness or payoff of an assignment of people to tasks or of jobs to machines. You cannot apply the Hungarian method to such maximisation problems. Therefore, you need to reduce it to a minimisation problem. It is easy to obtain an equivalent minimisation problem by converting every number in the table to an opportunity loss. To do so, you need to subtract every value from the highest value of the matrix and then proceed as usual. You will notice that minimising the opportunity loss produces the same assignment solution as the original maximisation problem. Operations Research Unit 7 Sikkim Manipal University Page No. 143 Solved problem 3 Solve the following maximisation assignment problem. Table 7.11: Assignment Table Solution: Since it is a maximisation problem, subtract every value from the maximum value of 90. Thus you have: Table 7.12: Opportunity Loss Table Table 7.13: Row Reduced Matrix Table 7.14: Column Reduced Matrix Operations Research Unit 7 Sikkim Manipal University Page No. 144 Therefore, the optimum assignment schedule is C1 to P2, C2 to P1, C3 to P3, and C4 to P4. Case-let A tailoring unit has four sewing machines of different makes. Each machine is capable of stitching all the required designs and patterns. However, the profit factor differs for each assignment. The unit is looking at maximising profit. To do so, the unit needs to carry out an optimal assignment exercise of assigning the right jobs to the right machines. Solved problem 4 Five different machines can do any of the required five jobs with different profits resulting from each assignment as depicted in table 7.15. Table 7.15: Five Different Machines and Their Jobs Find out the maximum profit possible through optimal assignment. Solution: To start with, convert an assignment problem of profit maximisation type into an assignment problem of cost minimisation type. To do so, form a new modified matrix by subtracting each entry of the matrix from the greatest entry 62 of the matrix. Use the resulting matrix to solve the problem of cost minimisation type. Operations Research Unit 7 Sikkim Manipal University Page No. 145 Table 7.16: Opportunity Matrix Now apply the Hungarian method to this problem and get the row and column reduced matrices as depicted in table 7.17 and 7.18. Table 7.17: Row Reduction Table 7.18: Column Reduction Then draw the minimum number of lines to suppress the zeros as depicted in table 7.19. Table 7.19: Lines Suppressing Zeros Here, number of lines drawn N = 4 < number of rows (= 5). The smallest uncovered entry is 4. Therefore, subtract 4 from all uncovered entries and add it to all the entries on the intersection of the lines as depicted in table 7.20. Operations Research Unit 7 Sikkim Manipal University Page No. 146 Table 7.20: Lines not Suppressing Zeros Again draw minimum number of lines to cover all the zeros as depicted in table 7.21. Table 7.21: Number of Lines = Number of Rows Here the number of lines drawn N = 5 = number of rows. Mark the zero entry with that occurs exactly once in a row and cancel that row or column as depicted in table 7.22. Table 7.22: Number of Lines Number of Rows After the above steps, row 2 has zero as the last entry, mark it with . Now row 1 will have the third entry zero uncovered. Mark it with . So row 4 will have only one entry in cell (4, 4) with zero. This gives you the optimal assignment as depicted in table 7.23. Operations Research Unit 7 Sikkim Manipal University Page No. 147 Table 7.23: Optimum Assignment Thus, x13 = x25 = x31 = x44 = x52 = 1 and rest all xij = 0. The optimal assignment is depicted in table 7.24. Table 7.24: Optimal Assignment Therefore, maximum profit is 214. Self Assessment Questions 7. In unbalanced AP, the number of rows ________ to number of columns. 8. Hungarian method cannot be applied directly to _________ problem. 9. If some jobs cannot be assigned to some machines, then it is called _________ assignment problem. 7.5 Travelling Salesman Problem Routing problem Network scheduling is a technique for planning and scheduling of large projects. It has successfully been applied in transportation and Operations Research Unit 7 Sikkim Manipal University Page No. 148 communication problems. A typical network problem involves finding route from one node (origin) to another (destination) between which alternative paths are available at various stages of the journey. The problem is to select the route that yields minimum cost. A number of different constraints may be placed on acceptable routes for instance, not returning to the node passed or passing through each node just once. These kinds of problems are called as routing problems. A wide variety of problems other than routing may be developed in connection with the construction and utilisation of networks. Here, you are going to consider the special type of routing problem that occurs frequently in OR the travelling salesman problem. Suppose a salesman has to visit n number of cities. He wishes to start from a particular city, visit each city once, and then return to his starting point. The objective is to select the sequence to visit the cities in such a way that his total travelling time is minimised. Starting from a given city, the salesman will have total of (n-1) different sequences. Further, since the salesman has to visit all the n number of cities; the optimal solution remains independent of selection of the starting point. The problem can be represented as a network where the nodes and arcs represent the cities and the distance between them respectively. Assume that in a five city problem, a round trip of the salesman is given by the following arcs. (3,1), (1,2), (2,4), (4,5), (5,3) These arcs in order are the first, second, third, fourth, and fifth directed arcs for the trip. Generally the kth directed arc represents the kth leg of the trip that is on leg k, the salesman travels from city i to city j. (i, j = 1, 2 n; i= j ) To formulate the problem, whose solution will yield the minimum travelling time, let the variables xijk be defined as:
= 0, otherwise directed from city i to city j th 1, if k is ijk x Where, i, j, and k are integers that vary between 1 and n. Operations Research Unit 7 Sikkim Manipal University Page No. 149 Following are the constraints of the problem. a) Only one directed arc may be assigned to a specific k, thus
j k xijk = 1 k =1, 2, 3n i= j b) Only one other city may be reached from a specific city i, thus
j k xijk =1, i = 1, 2, n c) Only one other city can initiate a direct arc to a specified city j, thus
i k xijk =1, j =1, 2... n d) Given the kth directed arc ends at some specific city j, the (k+1)th directed arc must start at the same city j; thus i xijk = r xjr ( k +1) for all j and k. i= j r = j These constraints ensure that the round trip will consist of connected arcs. The objective function is to minimise z = i
j k dij xijk i= j Where dij is the distance from city i to city j. Self Assessment Questions 10. In travelling salesman problem, the objective is to visit each cities ________ __________. 11. Salesman has ________ different sequences if n is the number of cities to be visited. 7.6 Summary Let us recapitulate the important concepts discussed in this unit: - This unit on assignment problems focuses on a special type of transportation problem, where the objective is to allocate n number of different facilities to n number of different tasks. Operations Research Unit 7 Sikkim Manipal University Page No. 150 - Although an assignment problem can be formulated as a linear programming problem, it is solved by a special method know as Hungarian method. - If the number of persons is the same as the number of jobs, the assignment problem is said to be balanced. The unit also explains the travelling salesman problem in brief. 7.7 Glossary Opportunity cost matrix: A cost matrix that depicts the lost opportunity when allocating resource to activity. 7.8 Terminal Questions 1. Four jobs are to be done on four different machines. The cost in rupees for producing ith on the j th machine is given below: Machines M1 M2 M3 M4 J1 15 11 13 15 J2 17 12 12 13 J3 14 15 10 14 J4 16 13 11 17 Assign the jobs to different machines to minimise the total cost. 2. A marketing manager has 5 salesmen and 5 sales districts. Considering the capabilities of the salesman and the nature of districts, the marketing manager estimates that the sales per month (in hundred rupees) for each salesman in each district would be as follows. Sales districts A B C D E 1 32 38 40 28 40 2 40 24 28 21 36 3 41 27 33 30 37 4 22 38 41 36 36 5 29 33 40 35 39 Find the assignment of salesman to districts that will result in maximum sales. Job s Salesman Operations Research Unit 7 Sikkim Manipal University Page No. 151 3. In a plant layout, there are five vacant places. The plant orders four machines to be installed in the vacant places. The cost of installing is as follows: M/G A B C D E M1 9 11 15 10 11 M2 12 9 - 10 9 M3 - 11 14 11 7 M4 14 8 12 7 8 Find the optimum assignment. 4. Find the assignment that maximises the total sale. Zone Sales men 1 2 3 4 M1 42 35 28 21 M2 30 25 20 15 M3 30 25 20 15 M4 24 20 16 12 7.9 Answers Self Assessment Questions 1. True 2. True 3. False 4. True 5. True 6. False 7. 8. Maximisation problem 9. Infeasible 10. Only once 11. (n-1) Terminal Questions 1. The optimum assignment policy is Job1 to machine 2, Job 2 to machine 4 Operations Research Unit 7 Sikkim Manipal University Page No. 152 Job 3 to machine1, Job 4 to machine 3 And the minimum assignment cost = Rs. (11+13+14+11) = Rs. 49. Refer 7.3 2. Optimal assignment policy is salesman 1 to district B, 2 to A, 3 to E, 4 to C, and 5 to D. Hence the maximum sales = Rs. (38+40+37+41+35) 100 = Rs. 19,100. Refer 7.4.3 3. M1 A2; M2 B; M3 E; M4 D Total 38. Refer 7.4.2 4. M1 1; M2 2; M3 3; M4 4 Total 99. Refer 7.4.3 7.10 Case Study Move-It Company The Move-It company has two plants producing fork-lift trucks that are then shipped to three distribution centres. The production costs are the same at the two plants, and the cost of shipping for each truck is shown for each combination of plant and distribution centre: Distribution Centre 1 2 3 Plant A Rs. 800 Rs. 700 Rs. 400 B Rs. 600 Rs. 800 Rs. 500 A total of 60 fork-lift trucks are produced and shipped per week. Each plant can produce and ship any amount up to a maximum of 50 trucks per week, so there is considerable flexibility on how to divide the total production between the two plants so as to reduce shipping costs. However, each distribution centre must receive exactly 20 trucks per week. Assume that distribution centres 1, 2, and 3 must receive exactly 10, 20, and 30 units per week respectively. For administrative convenience, management has decided that a single plant will supply each distribution centre totally, so that one plant will supply one distribution centre and the other plant will supply the other two distribution centres. The choice of assignments of plants to the distribution centres is to be made solely on the basis of minimising total shipping cost. Operations Research Unit 7 Sikkim Manipal University Page No. 153 Discussion Questions: 1. Formulate this case as an assignment problem by constructing the appropriate cost table, including identifying the corresponding assignee and tasks. 2. Obtain an optimal solution. References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. - Sharma S.C. Operation Research: Simulation and Replacement Theory. Operations Research Unit 8 Sikkim Manipal University Page No. 154 Unit 8 Integer Programming Problem Structure: 8.1 Introduction Objectives 8.2 Types of Integer Programming Problems 8.3 Gomorys All-IPP Method Construction of Gomorys constraints 8.4 All IPP Algorithm 8.5 Branch and Bound Technique Branch and bound algorithm 8.6 Summary 8.7 Glossary 8.8 Terminal Questions 8.9 Answers 8.10 Case Study 8.1 Introduction In the previous unit, you learnt about mathematical formulation of the problem and Hungarian method algorithm. You also learnt the routing problem and the travelling salesman problem. In this unit you will study about Integer Programming Problem (IPP). The IPP is a special case of Linear Programming Problem (LPP), where all or some variables are constrained to assume non-negative integer values. In LPP, the decision variables as well as slack or surplus variables were allowed to take any real or fractional value. However, there are certain real life problems in which the fractional value of the decision variables has no significance. For example, it does not make sense saying 1.5 men working on a project or 1.6 machines in a workshop. The integer solution to a problem can, however, be obtained by rounding off the optimum value of the variables to the nearest integer value. This approach can be easy in terms of economy of effort, time and cost that might be required to derive an integer solution but this solution may not satisfy all the given constraints. Secondly, the value of the objective function so obtained may not be an optimal value. Integer programming techniques come to our rescue during such scenarios. Operations Research Unit 8 Sikkim Manipal University Page No. 155 Integer LP problems are those in which some or all of the variables are restricted to integer (or discrete) values. An integer LP problem has important applications. Capital budgeting, construction scheduling, plant location and size, routing and shipping schedule, batch size, capacity expansion, fixed charge, etc are few problems which demonstrate the areas of application of integer programming. In this unit, you will learn about Integer Programming Problem (IPP) and the Gomorys all-IPP method. You will also learn all IPP algorithm and the branch and bound technique. Objectives: After studying this unit, you should be able to: - analyse an Integer Programming Problem (IPP) - solve an IPP using Gomorys method - describe the branch and bound technique 8.2 Types of Integer Programming Problems Linear integer programming problems can be classified into three categories: 1. Pure (all integer programming problems) - Here, all decision variables are restricted to integer values. 2. Mixed integer programming problems - Here some, but not all, of the decision variables are restricted to integer values. 3. Zero one integer programming problems - Here, all decision variables are restricted to integer values of 0 and 1. An integer programming problem(IPP) can be described as follows: Determine the value of unknowns x1, x2, , xn So as to optimise z = c1x1 + c2x2 + . . .+ cnxn Subject to the constraints ai1 x1 + ai2 x2 + . . . + ainxn = bi , i = 1,2,,m and xj > 0 j = 1, 2, ,n Where xj being an integral value for j = 1, 2, , k s n. Operations Research Unit 8 Sikkim Manipal University Page No. 156 If all the variables are forced to take only integral value that is k = n, it is called an all (or pure) integer programming problem. If some of the variables are restricted to take integral value and the remaining (n k) variables take any non-negative value, then the problem is known as a mixed integer programming problem. Self Assessment Questions 1. Integer programming is applied to problems that involve discrete variables. (True/False) 2. If some variables take on non-negative values, then it is known as pure IPP. (True/False) 8.3 Gomorys All IPP Method In the previous section, you learnt the types of integer programming problems. You will now learn the Gomorys all-IPP method. An optimum solution to an IPP is obtained by using the simplex method, ignoring the restriction of integral values. In the optimum solution, if all the variables have integer values, the current solution will be the required optimum integer solution. Otherwise, the given IPP is modified by inserting a new constraint called Gomorys constraint or secondary constraint. This constraint represents necessary conditions for integrability and eliminates some non-integer solution without losing any integral solution. On addition of the secondary constraint, the problem is solved using dual simplex method to obtain an optimum integral solution. If all the values of the variables in the solution are integers, then an optimum integer-solution is obtained, or else a new constraint is added to the modified LPP and the procedure is repeated till the optimum solution is derived. An optimum integer solution will be reached eventually after introducing enough new constraints to eliminate all the superior non-integer solutions. The construction of additional constraints, called secondary or Gomorys constraints is important and needs special attention. Operations Research Unit 8 Sikkim Manipal University Page No. 157 8.3.1 Construction of Gomorys constraints Consider a LPP or an optimum noninteger basic feasible solution. With the usual notations, let the solution be displayed as depicted in table 8.1. Table 8.1: Simplex Table yB xB Y1 y2 y3 y4 y2 y10 y11 y12 y13 y14 y3 y20 y21 y22 y23 y24 y00 y01 y02 y03 y04 The optimum basic feasible solution is given by xB = [x2, x3 ] = [y10, y20]; max z = y00 Since xB is a non-integer solution, we can assume that y10 is fractional. The constraint equation is: y10 = y11 x1 + y12 x2 + y13 x3+ y14 x4 (x1 = x4 = 0) It reduces to: y10 = y11 x1 + x2 + y14 x4 (x1 = x4 = 0)_____ (1) Because x2 and x3 are basic variables (which implies that y12 = 1 and y13 = 0). The above equation can be rewritten as: x2= y10 - y11 x1 - y14 x4 This is a linear combination of non-basic variables. Now, since y10 > 0 the fractional part of y10 must also be non-negative. You can split each of yij in equation (1) into an integral part Iij, and a non-negative fractional part, f1j for j = 0,1,2,3,4. After the breakup of equation (1), you can write it as: I10 + f10 = (I11 + f11) x2 + (I14 + f14) x4 Or f10 - f11 x2 - f14 x4 = x2 + I11 x1 + I14x4 - I10 _____ (2) If you compare equations (1) and (2), you will see that if you add an additional constraint in such a way that the left-hand side of equation (2) is an integer, then you will be forcing the non-integer y10 towards an integer. Operations Research Unit 8 Sikkim Manipal University Page No. 158 The desired Gomorys constraint is: f10 f11 x1 f11 x4 0 It is possible to have f10 f11 x1 f11 x4 = h where h > 0 is an integer. Then f10 = h + f11 x1 + f14 x4 is greater than one. This contradicts that 0 <fij< 1 for j = 0, 1, 2, 3, 4. Thus Gomorys constraint is sla 10 j 1,4 ij j j 1, 4 ij j 10 j 1, 4 ig j 10 or f x G (1) f f x f or f . x f + = > s
= = = Where Gsla (1) is a slack variable in the above first Gomorys constraint. The additional constraint to be included in the given LPP is towards obtaining an optimum all integer solution. After adding the constraint, the optimum simplex table is depicted as in table 8.2. Table 8.2: Optimum Simplex Table yB xB y1 y2 y3 y4 Gsla(1) y1 y10 y11 y12 y13 y14 0 y2 y20 y21 y22 y23 y24 0 Gsla(1) f10 f11 0 0 f14 1 y00 y01 y02 y03 y04 y05 Since f10 is negative, the optimal solution is unfeasible. Thus the dual simplex method is to be applied for obtaining an optimum feasible solution. After obtaining this solution, the above referred procedure is applied for constructing second the Gomorys constraint. The process is to be continued till all the integer solution has been obtained. Self Assessment Questions 3. An optimum solution to IPP is first obtained by using _____________. 4. With the addition of Gomorys constraint, the problem is solved by ____________________. Operations Research Unit 8 Sikkim Manipal University Page No. 159 8.4 All IPP Algorithm In the previous section, you learnt the Gomorys all-IPP method. You will now learn all the IPP algorithm. The iterative procedure for the solution of integer programming problem is as follows: Figure 8.1 depicts the iterative procedure of IPP. Fig. 8.1: Iterative Procedure of IPP Operations Research Unit 8 Sikkim Manipal University Page No. 160 Step 1: Convert the minimisation IPP into maximisation form. Ignore the integrality condition. Step 2: Introduce the slack or surplus variables, if needed to convert the inequations into equations and obtain the optimum solution of the given LPP by using simplex algorithm. Step 3: Test the integrality of the optimum solution: a) If the optimum solution contains all integer values, an optimum basic feasible integer solution has been obtained. b) If the optimum solution does not include all integer values, then proceed to the next step. Step 4: Examine the constraint equations corresponding to the current optimum solution. Let these equations be represented by: y , x b , (i 0, 1, 2, ........, m ) 1 j i n1 j 0 ij = = = Where n' denotes the number of variables and m' the number of equations. Choose the largest fractional part of bis to find { }bi i i max Let it be ( (
1 k b or write it is as o fk Step 5: Express the negative fractions if any, in the kth row of the optimum simplex table as the sum of a negative integer and a non-negative fraction. Step 6: Find the Gomorian constraint
= > n 1 j 0 kj j ko f , x f And add the equation
= = + n1 j 0 sla ko kj j G (1) f f . x To the current set of equation constraints Operations Research Unit 8 Sikkim Manipal University Page No. 161 Step 7: Start with a new set of equation constraints. Find the new optimum solution by dual simplex algorithm, that is, choose a variable to enter into the new solution having the smallest ratio: {(Cj Zj)/ yij; yij<0} so that Gsla (1) is the initial leaving basic variable. Step 8: If the new optimum solution for the modified LPP is an integer solution, it is also feasible and optimum for the given IPP. If it is not an integer solution, then return to step 4 and repeat the process until an optimum feasible integer solution is obtained. Solved Problem: 1 Find the optimum integer solution of the following all IPP. Maximise z = x1 + x2 Subject to the constraints 3x1 +2 x2 5; x2 2 x1, x2 > 0 and are integers Solution: Step 1: Introduce the slack variables s1>0 and s2 > 0, using simplex method an optimum noninteger solution is depicted in the table 8.3: Initial iteration. Non- integer optimum solution Table 8.3: Optimal Non-integer Solution cJ --> 1 1 0 0 CB Variable in Basis B Solution Values b (= xB) X1 X2 S1 S2 1 X1 1/3 1 0 1/3 -2/3 1 X2 2 0 1 0 1 Z = 7/2 cj-zj 0 0 -1/3 -1/3 Here, since all cj-zj 0, the optimal solution is : x1=1/3 and x2= 2 and max Z = 7/2 Step 2: Since x1 is the only basic variable whose value is a non-negative fraction, we consider the first row for generating the Gomory Cut. Considering x1 equation as the source row we write: 1/3 = x1+ 0 x2 + 1/3 s1 2/3 s2 (x1- source row) Operations Research Unit 8 Sikkim Manipal University Page No. 162 The factoring of solution of the x1 source row gives: (0 + 1/3) = ( 1+0 ) x1 + (0 + 1/3) s1 + (-1 + 1/3) s2 We note that each of the non-integer coefficient is factored into integer and fractional parts in such a manner that the fractional part is strictly positive. Rearranging the equation so that all of the integer coefficient appear on the left hand side. This gives: 1/3 + (s2 x1) = 1/3 s1+ 1/3 s2 Since x1 and s2 are non-negative, left hand side must satisfy: 1/3 1/3 s1+ 1/3 s2 1/3 + sg1 = 1/3 s1+ 1/3 s2 - 1/3 =sg1- 1/3 s1- 1/3 s2 Where sg1 is the new non negative (Integer) slack variable. By adding this equation (also called Gomory cut) at the bottom of the table 8.4 The new table obtained is depicted in table 8.5. Table 8.4: Optimal but Feasible Solution cJ --> 1 1 0 0 0 CB Variable in Basis B Solution Values b (= xB) X1 X2 S1 S2 Sg1 1 X1 1/3 1 0 1/3 -2/3 0 1 X2 2 0 1 0 1 0 0 Sg1 -1/3 0 0 -1/3 -1/3 1 Z = 7/2cj-zj Ratio: Min(cj-zj) / y3J(<0) 0 - 0 - -1/3 1
-1/3 1 0 - Step 4: Since solution depicted in table 8.4 is infeasible, apply the dual simplex method to find feasible as well as an optimal solution. The key row and key columns are marked in table 8.4. The new solution is obtained by applying following row operations and is depicted in table 8.5. Operations Research Unit 8 Sikkim Manipal University Page No. 163 R3(new)R3(old) X -3 ; R1(new)R1(old) X (1/3) R3(new) Table 8.5: Optimal Integer Solution cJ --> 1 1 0 0 0 CB Variable in Basis B Solution Values b (= xB) X1 X2 S1 S2 Sg1 1 X1 0 1 0 0 -1 1 1 X2 2 0 1 0 1 0 0 Sg1 1 0 0 1 1 -3 Z = 2cj-zj 0 0 0 0 -1 The solution provided in the table 8.5 is X1 = 0, X2 = 2, Sg1 = 2 and Max Z = 2. This also satisfies the integer requirement. Self Assessment Questions 5. For Gomorys constraint select the variable whose fractional value is more? (True/False) 6. Optimum values in a pure IPP can be x=2 and y=3.5. (True/False) 8.5 Branch and Bound Technique In the previous section, you learnt the IPP algorithm. You will now learn the branch and bound technique. Sometimes a few or all the variables of an IPP are constrained by their upper or lower bounds or by both. The most general technique for a solution of such constrained optimisation problems is the branch and bound technique. The technique is applicable to both all (or pure) IPP as well as mixed IPP. The technique for a maximisation problem is discussed as follows: Let the IPP be Maximise = = n j 1 z c j x j _______________________ (1) Subject to the constraints a x b i 1,2,....,m n j 1 ij j s i = = ___________________ (2) Operations Research Unit 8 Sikkim Manipal University Page No. 164 xj is integer valued, j = 1, 2, .., r (< n) ___________ (3) xj > 0 . j = r + 1, .., n ______________ (4) Further let us suppose that for each integer valued xj, we can assign lower and upper bounds for the optimum values of the variable by Lj xj Uj j = 1, 2, ., r ______________________ (5) This is the main idea behind the branch and bound technique. Consider any variable xj, and let k be some integer value satisfying LjsksUj 1. Then clearly an optimum solution (1) through (5) also satisfies either linear constraint. xj > k + 1 _________________________________ (6) Or the linear constraint xj k ______________________ (7) To explain how this partitioning helps, let us assume that there were no integer restrictions (3), and it yields an optimal solution to LPP (1), (2), (4) and (5). This indicates x1 = 1.66 (for example). Then you formulate and solve two LPPs each containing (1), (2) and (4). But (5) for j = 1 is modified to be 2 x1 U1 in one problem and L1 x1 1 in the other. Further to each of these problems, process an optimal solution satisfying integer constraint (3). Then the solution having the larger value for z is clearly the optimum for the given IPP. However, it usually happens that one (or both) of these problems have no optimal solution satisfying (3) and thus some more computations are required. Now let us discuss, stepwise, the algorithm that specifies how to apply the partitioning (6) and (7) in a systematic manner to finally arrive at an optimum solution. Lets start with an initial lower bound for z, say z(0) at the first iteration, which is less than or equal to the optimal value z*. This lower bound may be taken as the starting Lj for some xj. In addition to the lower bound z(0), you also have a list of LPPs (to be called master list) differing only in the bounds (5). To start with (the 0th iteration) the master list contains a single LPP consisting of (1), (2), (4) and (5). Let us now discuss the procedure that specifies how the partitioning (6) and (7) can Operations Research Unit 8 Sikkim Manipal University Page No. 165 be applied systematically to eventually get an optimum integer-valued solution. 8.5.1 Branch and bound algorithm At the tth iteration (t = 0, 1, 2 ), apply the following steps to get an optimum integer solution. Step 0: If the master list is not empty, choose an LPP from it. Otherwise stop the process, go to step 1. Step 1: Obtain the optimum solution to the chosen problem. If either i) It has no feasible solution or ii) The resulting optimum value of the objective function z is less than or equal to z(t), then let z(t+1) = z(t) and return to step 0; otherwise go to step 2. Step 2: If the obtained optimum solution satisfies the integer constraints (3) then record it. Let z(t+1) be the associated optimum value of z; return to step 0. Otherwise move to step 3. Step 3: Select any variable xj, j = 1, 2, ., p, that does not have an integer value in the obtained optimum solution to the LPP, chosen in step 0. Let * x j denote this optimal value of xj. Add two LPPs to the master list. These LPPs are identical to the LPPs chosen in step 0, except that, the lower bound on xj is replaced by (
(
* x j + 1. Let z(t+1) = z(t); return to step 0. Note: At the termination of the algorithm, if a feasible integer valued solution yielding z(t) has been recorded, it is optimum, or else no integer valued feasible solution exists. Solved Problem 2: Use Branch and bound method to solve the following LPP: Maximise Z = 7x1 + 9x2 subject to the constraints: -x1 + 3x2 6, 7x1 + x2 35, x2 7, x1 , x2 0 and are integers. Solution: Step 1: Ignoring the integer restriction, the optimum solution to the given LPP can easily be obtained as: x1= 9/2 , x2 = 7/2 and maximise z = 63. Operations Research Unit 8 Sikkim Manipal University Page No. 166 Step 2: Since the solution is non-integral, let us choose x1, i.e., x1* = 9/2 being the largest fractional value. Step 3: Considering the value of z as initial upper bound, i.e., z = 63; the lower bound is obtained by rounding off the values of x1 and x2 to the nearest integers, i.e., x1 = 4 and x2 = 3. Then the lower bound is z = 55. Step 4: Since [x1*] = [9/2] = 4; We have, Sub-Problem 1: Maximise Z = 7x1 + 9x2 subject to: -x1 + 3x2 6, 7x1 + x2 35, 0 x1 4, 0 x2 7. Sub-Problem 2: Maximise Z = 7x1 + 9x2 subject to: -x1 + 3x2 6, 7x1 + x2 35, x1 5, 0 x2 7. Step 5: Optimum solutions for the sub-problems are obtained as follows: Sub-Problem 1: x1= 4, x2=10/3 and maximum z = 58. Sub-Problem 2: x1= 5, x2=0 and maximum z = 35. Since the solution for sub-problem 1 is not in integers, we subdivide it into the following two sub-problems: Sub-Problem 3: Maximise Z = 7x1 + 9x2 subject to -x1 + 3x2 6, 7x1 + x2 35 , and 0 x1 4, 0 x2 3. Sub-Problem 4: Maximise Z = 7x1 + 9x2 subject to -x1 + 3x2 6, 7x1 + x2 35, and 0 x1 4, x2 4. Step 6: The optimum solutions for the sub-problem 3 and 4 are: Sub-Problem 3: x1 = 4, and x2= 3 and maximum z=55. Sub-Problem 4: No feasible solution. Step 7: Among the recorded integer valued solutions, since the largest value of z is 55; the required optimum solution is v1 = 4, x2 = 3 and maximum z = 55. Operations Research Unit 8 Sikkim Manipal University Page No. 167 The entire branch and bound procedure for the given problem is depicted in figure 8.2: Fig. 8.2: The Flow of Sub-Problems in the Branch and Bound Problems Self Assessment Questions 7. Branch and bound technique is applied when some variables have upper or lower bounds. (True/False) 8. We start the branch and bound technique with the lower bound. (True/False) 8.6 Summary Let us recapitulate the important concepts discussed in this unit: - Linear integer programming problems can be classified into pure, mixed integer and zero- one interger programming problems. - Gomorys constraint represents the necessary conditions for integrability and eliminates some non-integer solution without losing any integral solution - The most general technique for a solution of such constrained optimisation problems is the branch and bound technique. - This unit also describes two algorithms to determine the optimal solution for an integer programming problem. One is the All IPP algorithm and the other is the branch and bound algorithm. Operations Research Unit 8 Sikkim Manipal University Page No. 168 8.7 Glossary Non-basic variables: variables set to value zero in order to solve the value of basic variables. Equation constraints: A constraint written in the form of an equation containing a =. 8.8 Terminal Questions 1. Use branch and bound technique to solve the following problem: Maximise z = 3x1 + 3x2 + 13 x3 Subject to 3x1 + 6x2 + 7x3 8 6x1 3x2 + 7x3 8 0 xj 5, j = 1, 2, 3 2. What is integer programming? 3. Explain the Gomorys all integer algorithm of an IPP? 8.9 Answers Self Assessment Questions 1. True 2. False 3. Simplex method 4. Dual simplex method 5. True 6. False 7. True 8. False Terminal Questions 1. At the end of the 8th iteration we get the optional solution to the IPP is x1 = x2 = 0, x3 = 1, z* = 13. For more details refer Section 8.5 2. An integer programming problem(IPP) can be described as follows: Determine the value of unknowns x1, x2, , xn Operations Research Unit 8 Sikkim Manipal University Page No. 169 So as to optimise z = c1x1 + c2x2 + . . .+ cnxn Subject to the constraints ai1 x1 + ai2 x2 + . . . + ainxn = bi , i = 1,2,,m and xj 0 j = 1, 2, ,n Where xj being an integral value for j = 1, 2, , k n. If all the variables are forced to take only integral value that is k = n, it is called an all (or pure) integer programming problem. For more details refer to Section 8.2 3. The given IPP is modified by inserting a new constraint called Gomorys constraint or secondary constraint. For more details refer to Section 8.3 8.10 Case Study Managing Stocks Mr. Don Fredricks, an expeditor at the local warehouse for Wood World sighed as he moved boxes and boxes of inventory to the side in order to reach the shelf where the particular item he needed was located. He dropped to his hands and knees and squinted at the inventory numbers lining the bottom row of the shelf. He did not find the number he needed. He worked his way up the shelf until he found the number matching the number on the order slip. Just his luck! The item was on the top row of the shelf! Don walked back through the warehouse to find a ladder, stumbling over boxes of inventory littering his path. When he finally climbed the ladder to reach the top shelf, his face crinkled in frustration. Not again! The item he needed was not in stock! All he saw above the inventory number was an empty space covered with dust! Don trudged back through the warehouse to make the dreadful phone call. He dialled the number of Annie, the saleswoman on the kitchen showroom floor of Wood World, and informed her that the particular light fixture, the customer had requested, was not in stock. He then asked her if she wanted him to look for the rest of the items in the kitchen set. Annie told him that she would talk to the customer and call him back. Annie hung up the phone and frowned. Mr. David, her customer, would not be happy. Ordering and receiving the correct light fixture from the regional warehouse would take at least two weeks. Annie then paused to reflect Operations Research Unit 8 Sikkim Manipal University Page No. 170 upon business during the last month and realised that over 80 percent of the orders for kitchen sets could not be filled because items needed to complete the sets were not in stock at the local warehouse. She also realised that Wood World was losing customer goodwill and business because of stock outs. The furniture mega store was gaining a reputation for slow service and delayed deliveries causing customers to turn to small competitors that sold furniture directly from the showroom floor. Annie decided to investigate the inventory situation at the local warehouse. She walked the short distance to the building next door and gasped when she stepped inside the warehouse. What she saw could only be described as chaos. Spaces allocated for some items were overflowing into the aisles of the warehouse while other spaces were completely bare. She walked over to one of the spaces overflowing with inventory to discover the item that was overstocked. She could not believe her eyes! The warehouse had at least 30 rolls of pea-green wallpaper! No customer had ordered pea-green wallpaper since 1973! Annie marched over to Don demanding an explanation. Don said that the warehouse had been in such a chaotic state since his arrival one year ago. He said the inventory problems occurred because management had a policy of stocking every furniture item on the showroom floor in the local warehouse. Management only replenished inventory every three months, and when inventory was replenished, management ordered every item regardless of if it had been sold. Don also said that he had tried to make management aware of the problems with overstocking unpopular items and under stocking popular items, but that management would not listen to him because he was simply an expeditor. Annie understood that Wood World required a new inventory policy. Not only was the mega store losing money by making customers unhappy with delivery delays, but it was also losing money by wasting warehouse space. By changing the inventory policy to stock only popular items and replenish them immediately when they are sold, Wood World would ensure that the majority of customers receive their furniture immediately and that the valuable warehouse space was utilised effectively. Annie needed to sell her inventory policy to the management. Using her extensive sales experience, she decided that the most effective sales strategy would be to use her kitchen department as a model for the new inventory policy. She would identify all kitchen sets comprising 85 percent of Operations Research Unit 8 Sikkim Manipal University Page No. 171 customers orders. Given the fixed amount of warehouse space allocated to the kitchen department, she would identify the items Wood World should stock in order to satisfy the greatest number of customer orders. She would then calculate the revenue from satisfying customer orders under the new inventory policy, using the bottom line to persuade management to accept her policy. Annie analysed her records over the past three years and determined that 20 kitchen sets were responsible for 85 percent of the customer orders. These 20 kitchen sets were composed of up to eight features in a variety of styles. Annie listed each feature and its popular styles as depicted in table 8.6. Annie then created a table showing the 20 kitchen sets and the particular features comprising each set. To simplify the table, she used the codes shown in parentheses above to represent the particular feature and style. The table is given on the next page. For example, kitchen set 1 consists of floor tile T2, wallpaper W2, light fixture L4, cabinet C2, counter top O2, dishwasher D2, sink S2 and range R2. Notice that sets 14 through 20 do not contain dishwashers. Annie knew she had only a limited amount of warehouse space allocated to the kitchen department. The warehouse could hold 50 square feet of tile and 12 rolls of wallpaper in the inventory bins. The inventory shelves could hold two light fixtures, two cabinets, three countertops and two sinks. Dishwashers and ranges are similar in size, so Wood World stored them in similar locations. The warehouse floor could hold a total of four dishwashers and ranges. Every kitchen set always includes exactly 20 square feet of tile and exactly five rolls of wallpaper. Therefore, 20 square feet of a particular style of tile and five rolls of a particular style of wallpaper are required for the styles to be in stock. Floor Tile Wallpaper Light Fixtures Cabinets (T1) White textured tile (W1) Plain ivory paper (L1) One large rectangular frosted fixture (C1) Light solid wood cabinets (T2) Ivory textured tile (W2) Ivory paper with dark brown pinstripes (L2) Three small square frosted fixtures (C2) Dark solid wood cabinets Operations Research Unit 8 Sikkim Manipal University Page No. 172 (T3) White checkered tile with blue trim doors (W3) Blue paper with marble texture (L3) One large oval frosted fixture (C3) Light wood cabinets with glass (W4) Light yellow paper with marble texture (L4) Three small frosted globe fixtures (C4) Dark wood cabinets with glass Countertops Dishwashers Sinks Ranges (O1) Plain light wood countertops (D1) White energy- saving dishwasher (S1) Sink with separate hot and cold water taps (R1) White electric oven (O2) Stained light wood countertops (D2) Ivory energy- saving dishwasher (S2) Divided sink with separate hot and cold water taps and garbage disposal (R2) Ivory electric oven (O3) White lacquercoated countertops (S3) Sink with one hot and cold water tap (R3) White gas oven O4) Ivory lacquercoated countertops (S4) Divided sink with one hot and cold water tap and garbage disposal (R4) Ivory gas oven Discussion Questions: 1. Formulate and solve an IPP model to maximise the total number of kitchen sets (and thus the number of customer orders) Wood World stocks in the local warehouse. Assume that when a customer orders a kitchen set, all the particular items composing that kitchen set are replenished at the local warehouse immediately. 2. How many of each feature and style should Wood World stock in the local warehouse? How many different kitchen sets are in stock? 3. Annie convinces the management that the kitchen department should serve as a testing ground for future inventory policies. To provide adequate space for testing, management decides to allocate all the space freed by the nursery department to the kitchen department. The Operations Research Unit 8 Sikkim Manipal University Page No. 173 extra space means that the kitchen department can store not only the dishwashers and ranges from part (c), but also all sinks, all countertops, three of the four light fixtures and three of the four cabinets. How much does the additional space help? References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. Operations Research Unit 9 Sikkim Manipal University Page No. 174 Unit 9 Infinite Queuing Models Structure: 9.1 Introduction Objectives 9.2 Queuing Theory 9.3 Operating Characteristics of a Queuing System 9.4 Constituents of a Queuing System Arrival pattern Completely random arrivals 9.5 Service Facility 9.6 Queue Discipline Customer behaviour Server behaviour 9.7 Summary 9.8 Glossary 9.9 Terminal Questions 9.10 Answers 9.11 Case Study 9.1 Introduction In the previous unit, you learnt the Integer Programming Problem (IPP) and the Gomorys all-IPP method. You also learnt all IPP algorithm and the branch and bound technique. This chapter introduces the idea of queuing theory. The formation of waiting lines or queues is common and is usually formed by elements, people or events that are awaiting some form of service. Who isnt familiar with queues in our daily existence? On the way to your workplace, you have to wait at the bus stop, wait for the traffic light to turn green, wait for your turn to enter the lift. You come across cars in line at petrol pumps for service, queues to deposit cash at the bank, mobile subscriber waiting for a new connection, customers waiting at the checkout counter, goods in production/warehouse waiting to be shipped and even aircrafts waiting for a free runway to take off. The one thing common to all the above mentioned examples is that customers arrive at a service centre and wait for their turn to receive the service. Though the arrival of customers is irregular and the time taken for Operations Research Unit 9 Sikkim Manipal University Page No. 175 the service is not consistent, queues build up during hours of demand and disappear during the lull period. In commercial or industrial situations, it may not be economical to have waiting lines. On the other hand, it may not be feasible or economical to totally avoid queues. An executive dealing with the system then would like to find the optimal facilities to be provided. In this unit, you will learn about queuing theory, operating characteristics and constituents of a queuing system. You will also learn about service facility and queue discipline. Objectives: After studying this unit, you should be able to: - describe the queuing process - evaluate the Queuing Theory - differentiate the customers behaviour. 9.2 Queuing Theory Queuing theory had its beginning in the research work of a Danish engineer named A. K. Erlang. In 1909, Erlang experimented with fluctuating demand in telephone traffic. Eight years later he published a report addressing the delays in automatic dialling equipment. At the end of World War II, Erlangs early work was extended to more general problems and to business applications of waiting lines. Queuing theory is a collection of mathematical models of various queuing systems. It is based on probability concepts. It gives an indication of the capability of a given system and the possible changes in its performance with modification to the system. All the constraints of the process are not taken into account in the formulation of a queuing model. The application of queuing theory cannot be viewed as an optimisation process as there is no maximisation or minimisation of an objective function. Formation of Queues Queues or waiting lines arise when the demand for a service facility exceeds the capacity of that facility, that is, the customers do not get service immediately upon request but must wait, or the service facilities stand idle and wait for customers. Operations Research Unit 9 Sikkim Manipal University Page No. 176 For example - Supermarkets must decide how many cash register checkout positions should be opened. - Gasoline stations must decide how many pumps should be opened and how many attendants should be on duty. - Manufacturing plants must determine the optimal number of mechanics to have on duty in each shift to repair machines that break down. - Banks must decide how many teller windows to keep open to serve customers during various hours of the day. Self Assessment Questions 1. Customers arrive at a bank at regular intervals. (True/False) 2. Queuing identifies the optimal service facilities to be provided. (True/False) 3. Queuing theory is based on the deterministic model. (True/False) 9.3 Operating Characteristics of a Queuing System In the previous section, you learnt about queuing theory. You will now learn the operating characteristics of a queuing system. A queuing model has the following operating characteristics which enables us to understand and efficiently manage a queue: - Queue length: The number of customers in the waiting line reflects one of the two conditions. Short queues could mean either good customer service or too much capacity. Similarly, long queues could indicate either low server efficiency or the need to increase capacity - Number of customers in system: The number of customers in queue and also those being served in the queue relates to the service efficiency and capacity. Large values imply congestion, potential customer dissatisfaction and a need for more capacity. - Waiting time in queue: Long lines do not reflect long waiting times if the service rate is fast. However, when waiting time seems long to customers, they perceive that the quality of service is poor. Long waiting times may indicate a need to adjust the service rate of the system or change the arrival rate of customers. Operations Research Unit 9 Sikkim Manipal University Page No. 177 - Waiting time in system: The total elapsed time from entry into the system until exit from the system may indicate problems with customers, server efficiency or capacity. If some customers are spending too much time in the service system, there may be a need to change the priority discipline, increase productivity or adjust capacity in some way. - Service facility utilisation: The collective utilisation of the service facilities reflects the percentage of time the facilities are busy. Management is interested in maintaining high utilisation but this objective may adversely impact the other operating characteristic. A queuing system is said to be in transient state when its operating characteristics are dependent upon time. If the operating characteristics become independent upon time, the queuing system is said to be in a steady state. Self Assessment Questions 4. _________ represents number of customers waiting in the queue. 5. ________ ________ times may indicate a need to adjust the service rate of the system or change the arrival rate of customers 6. _____________ represents the percentage of time the facilities are busy. 9.4 Constituents of a Queuing System In the previous section, you learnt the operating characteristics of a queuing system. You will now learn the constituents of a queuing system. The constituents of a queuing system include arrival pattern, service facility and queue discipline. - Arrival pattern: It is the average rate at which the customers arrive. - Service facility: Examining the number of customers served at a time and the statistical pattern of time taken for service at the service facility. - Queue discipline: The common method of choosing a customer for service amongst those waiting for service is First Come First Serve. Operations Research Unit 9 Sikkim Manipal University Page No. 178 Figure 9.1 depicts the components of a queuing system. Fig. 9.1: Components of a Queuing System Figure 9.2 depicts an example showing the components of a queuing system. Fig. 9.2: Car Wash System Showing Components of a Queuing System 9.4.1 Arrival pattern The arrival of customers can be regular as in case of an appointment system of a doctor or flow of components on a conveyor belt. The regular pattern of arrivals is neither very common nor very easy to deal with mathematically. The following are the important arrival characteristics: Operations Research Unit 9 Sikkim Manipal University Page No. 179 1. Size of the population: Unlimited (infinite) or limited (finite) 2. Pattern of arrivals (statistical distribution) 3. Behaviour of arrivals Our primary concern is the pattern of completely random arrivals. 9.4.2 Completely random arrivals If the number of potential customers is infinitely large, then probability of an arrival in the next interval of time will not depend upon the number of customers already in the system. (The assumption is valid by and large, except for queues involving a small finite number of customers.) When the arrivals are completely random, they follow the Poisson distribution, which equals to the average number of arrivals per unit time. Sometimes it is necessary to distinguish between groups of customers, such as male and female callers, or large and small aircrafts during the arrivals. There are several other types of arrival patterns which shall not be dealt with due to their restricted applications. Self Assessment Questions 7. Every queuing process has an arrival pattern, a service facility and a queue discipline as its constituents.(True/False) 8. If the arrivals are completely random, then it follows Poisson distribution. (True/False) 9.5 Service Facility In the previous section, you learnt the constituents of a queuing system. You will now learn the service facility. Service Facility is based on three parameters Availability of service, number of service centres and duration of service. i) Availability of service It is necessary to examine if there are any constraints that reduce the number of customers to be served at a time, apart from specifying the time span of the availability of service. For example, in a waiting line for a suburban train, apart from the timings of the train services, the probability distribution of the number of passengers that can be accommodated in a train that arrives is relevant. Operations Research Unit 9 Sikkim Manipal University Page No. 180 ii) Number of service centres If only one service centre is referred to as a service channel, obviously only one customer can be served at a time. There will definitely be more than one service centre and the behaviour of the queues will vary with the number of channels available for service. Multiple service channels may be arranged in series or in parallel. Multiple service channels are arranged in series when a customer has to go through several counters one after another with each providing a different part of the service. For instance, bank counters where a customer has to go to at least two counters to withdraw is an example of arrangement in series. On the other hand ticket booths in a railway station have multiple channels with parallel arrangement. iii) Duration of service This is the length of time taken to serve a customer. This can be constant or varying. (a) Constant service time: Though not in practice, an assumption that service time is constant holds true, if the pattern of arrivals is very irregular. (b) Completely random service time: The service time can be considered completely random when: - The server does not distinguish between the various arrivals. - The server does not deliberately change the duration of service on the basis of the time taken to serve the previous arrival. - The server forgets the time for which he/she has been serving a customer. Under the above mentioned conditions, the service time follows exponential distribution which means it is equal to reciprocal of the average rate of service. (c) Service time following Erlang distribution Sometimes, the assumption of an exponential distribution for service time is not valid. Hence Erlang family of service time distributions is used. Table 9.1 depicts examples of queuing discipline in daily life. Operations Research Unit 9 Sikkim Manipal University Page No. 181 Table 9.1: Examples of Queuing Discipline in Daily Life Sr. No. Situation Arriving Customers Service Facility 1. Passage of customers through supermarket checkout Shoppers Checkout counters 2. Flow of automobile traffic through a road network Automobiles Road Network 3. Transfer of electronic messages Electronic messages Transmission lines 4. Banking transactions Bank patrons Bank tellers 5. Flow of computer programmers through a computer system Computer Programmers Central processing unit 6. Sale of theatre tickets Theatre visitors Ticket booking windows 7. Arrival of trucks to carry fruits and vegetables from a central market Trucks Loading crews and facilities 8. Registration of unemployed at employment exchange Unemployed personnel Registration assistants 9. Occurrences of fires Fires Firemen and equipment 10. Flow of ships to the seashore Ships Harbour and docking facilities 11 Calls at police control room Service calls Policemen Self Assessment Questions 9. Multiple service channels may be arranged in ______________ or in _________. 10. The service time can be __________ or _________. 9.6 Queue Discipline In the previous section, you learnt the service facility. You will now learn about queue discipline. The queue discipline is the order or manner in which customers from the queue are selected for service. There are a number of ways in which customers in the queue are served: Static queuing disciplines are based on the individual customers status in the queue like: Operations Research Unit 9 Sikkim Manipal University Page No. 182 I. First In First Out (FIFO) If the customers are served in the order of their arrival. E.g., Toll road II. Last In First Out (LIFO) Where the last in items are operated first. E.g., Cargo handling situations, getting down from the bus, etc. Dynamic queuing disciplines are based on the individual customer attributes in the queue: I. Service in Random Order (SIRO): Here customers are selected for service at random irrespective of their arrivals in the service system. II. Priority service: Under this rule, customers are grouped in priority classes on the basis of some attributes such as service time or urgency, and FIFO rule is used within each class to provide service. E.g., premium queues in temples like Tirupati. III. Pre-emptive priority (or Emergency): Under this rule, an important customer is allowed to enter into the service immediately after entering into the system even if a customer with lower priority is already in service. E.g., Emergency service in hospitals, ambulance and fire brigades in traffic signals 9.6.1 Customer behaviour a) Balking: Arriving customers are said to balk if they do not join a queue because of their reluctance to wait or some customers even before joining the queue get discouraged by seeing the number of customers already in service system or estimating the excessive waiting time for a desired service, decide to return for service at a later time. In queuing theory this is known as balking. b) Collusion: Customers may be in collusion, meaning that only one person would join the queue, but demand service on behalf of several customers. c) Reneging: Impatient customers who would not wait beyond a certain time and leave the queue are said to renege. For example, a customer who has just arrived at a grocery store and finds that the salesmen are busy in serving the customers already in the system, will either wait for service till his patience is exhausted or estimates that his waiting time may be excessive and so leaves immediately to seek service elsewhere. Operations Research Unit 9 Sikkim Manipal University Page No. 183 d) Jockeying: Some customers keep on switching over from one queue to another in a multiple service centres. This is called jockeying. 9.6.2 Server behaviour Although the service timings have been specified, the server may not be available through the entire span of time. For instance, in every one hour the server may disconnect from the service centre for 5 minutes so that it can go through its daily updates or cleanup routine. The server is characterised by: - The arrangement of service facilities - The distribution of service times - Servers behaviour - Management policies Self Assessment Questions 11. When customers keep on switching over from one queue to another then it is called ________. 12. _____ ________ ______ ______ are the types of customer behaviour. 9.7 Summary Let us recapitulate the important concepts discussed in this unit: - The waiting line theory or queuing theory analysis suggests the number of facilities required and the cost of customers waiting time and the optimum service level. - The queuing theory contributes vital information required for balancing the cost of service and cost associated with waiting time of the customer. - A queuing system is said to be in transient state when its operating characteristics are dependent upon time. 9.8 Glossary Erlang family of service time distributions: two constraint generalization of exponential distribution. Statistical pattern: set of probability distributions Operations Research Unit 9 Sikkim Manipal University Page No. 184 Probability distribution: a table containing the outcomes of an experiment with its likelihood of occurrence. 9.9 Terminal Questions 1. Explain the operating characteristics of a queuing system. 2. Which are the important arrival characteristics? 3. State the ways in which customers in a queue are served. 9.10 Answers Self Assessment Questions 1. False 2. True 3. False 4. Queue length 5. Long waiting 6. Collective utilisation 7. True 8. True 9. Series, parallel 10. Constant, varying 11. Jockeying 12. Balking, Collusion, reneging, jockeying Terminal Questions 1. The operating characteristics of a queuing system are queuing length, number of customers in system,waiting time in queue, waiting time in system and service facility utilisation. For more details, refer section 9.3 2. Size of population, pattern of arrivals and behavior of arrivals. For more details refer section 9.4.1 3. Static queuing disciplines and dynamic queuing disciplines. For more details refer section 9.5 Operations Research Unit 9 Sikkim Manipal University Page No. 185 9.11 Case Study Drive-in Photo Shop Sunder is planning to open a drive-through photo developing booth with a single service window that will be open approximately 200 hours per month in a busy commercial area. Space for a drive-through lane is available for a rental of Rs. 8000 per month per car length. Sunder needs to decide how many car lengths of space to provide for his customers. Excluding this rental cost for the drive-through lane, Sunder believes that he will average a profit of Rs 160 per customer served(nothing for a drop off of film and Rs 320 when the photographs are picked up). He also estimates that customers will arrive randomly at a mean rate of 20 per hour, although those who find the drive-through lane full will be forced to leave. Half of the customers who find the drive-through lane full wanted to drop off the film and the other half wanted to pick up the photographs. The half who wanted to drop off film will take their business elsewhere instead. The other half of the customers who find the drive-through lane full will not be lost because they will keep trying later until they can get in and pick up their photographs. Sunder assumes that the time required to serve a customer will have an exponential distribution with a mean of 2 minutes. Discussion Questions: a) Find L(expected no. of customers in queuing system) and the mean rate at which customers are lost when the number of car lengths of space provided is 2, 3, 4 and 5. b) Calculate W (waiting time in the system) and L for the cases considered in (a). c) Use the results from part(a) to calculate the decrease in the mean rate at which customers are lost when the number of car lengths of space provided is increased from 2 to 3, from 3 to 4, from 4 to 5. Then calculate the increase in expected profit per hour(excluding space rental costs) for each of these three cases. d) Compare the increases in expected profit found in part(c) with the cost per hour of renting each car length of space. What conclusion do you draw about the number of car lengths of space that Sunder should provide? Operations Research Unit 9 Sikkim Manipal University Page No. 186 References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E-References: - http://www.mrt.ac.lk/maths/TMJAC/Operational%20Research %20Techinques%20II/OR%20Lecture%203.pdf, accessed on Nov 9, 2009 Operations Research Unit 10 Sikkim Manipal University Page No. 187 Unit 10 Mathematical Analysis of Queuing Theory Structure: 10.1 Introduction Objectives 10.2 Mathematical Analysis of Queuing Process 10.3 Properties of Queuing System Arrivals Service Queuing discipline 10.4 Notations 10.5 Service System 10.6 Single Channel Models 10.7 Multiple Service Channels 10.8 Erlang Family of Distribution of Service Times 10.9 Applications of Queuing Theory 10.10 Limitations of Queuing Theory 10.11 Summary 10.12 Glossary 10.13 Terminal Questions 10.14 Answers 10.15 Case Study 10.1 Introduction In the previous unit, you learnt queuing theory, operating characteristics and constituents of a queuing system. You also learnt about service facility and queue discipline. Queuing theory is a collection of mathematical models of various queuing systems that takes in input parameters and provides quantitative parameters describing the system performance. Because of the random nature of the processes involved, the queuing theory is rather demanding and all models are based on very strong assumptions (not always satisfied in practice). Many systems (especially queuing networks) are not soluble at all, so the only technique that may be applied is simulation. Operations Research Unit 10 Sikkim Manipal University Page No. 188 Nevertheless queuing systems are practically very important because of the typical trade-off between various costs of providing service and the costs associated with waiting for the service (or leaving the system without being served). High quality fast service is expensive, but costs incurred by customers waiting in the queue are minimum. On the other hand long queues may cost a lot because customers (e.g., machines) do not work while waiting in the queue or customers leave because of long queues. So a typical problem is to find an optimum system configuration (e.g., the optimum number of servers). The solution may be found by applying queuing theory or by simulation. In this unit, you will learn about mathematical analysis of queuing theory, notations and service systems involved in queuing theory. You will also learn about Erlang family of distribution of service times and the applications and limitations of queuing theory. So, let us start with the mathematical analysis of queuing process. Objectives: After studying this unit, you should be able to: - describe the mathematical analysis of queuing theory - classify the different notations involved in queuing theory - recognise different service systems involved - explain Erlang family of distribution of service times 10.2 Mathematical Analysis of Queuing Process Statistical equilibrium While analysing the queuing process, our main interest is in developing a mathematical model which represents the system. This implies that changes occurring in the characteristics of the system are not to be considered if they are of short durations. When we specify the statistical distributions of the arrivals or service times, we want to ensure an equilibrium state. In a queuing process, at each point of time, there is a probability distribution for the length of the queue. Compare the number of customers arrived at the post office, 15 minutes after the counter is opened to that after one hour. After the initial rush, youll find the system with the same type of probabilities of arrivals. Operations Research Unit 10 Sikkim Manipal University Page No. 189 The probability distribution of the arrivals will then be different from that of the initial state. The state, in which the probability distribution remains the same, is called the steady state and the system is said to have acquired a state of statistical equilibrium. In the steady state, there will be variations in the queue from time to time but the probability distributions representing the queuing process will remain the same and are independent of the time at which the system is examined. 10.3 Properties of Queuing System In the previous section, you learnt the mathematical analysis of queuing process. You will now learn the properties of the system. While developing queuing models, we will confine our discussion to queuing systems with the following properties. 10.3.1 Arrivals - Customers are discrete entities. - Population Finite/Infinite. - No simultaneous arrivals. - Pattern of arrivals in a time period t0 follows Poisson distribution with average arrival rate . Probability (Number of arrivals = K) = t(t)x/ x ! Where x = 0, 1, 2 .. 10.3.2 Service - Single serve channel/multiple channels - Single queue/infinite capacity 10.3.3 Queuing discipline - First come first serve Note: When the number of arrivals follows Poisson distribution (discrete), the inter-arrival time, that is, the time between arrivals follows exponential distribution (continuous). Operations Research Unit 10 Sikkim Manipal University Page No. 190 10.4 Notations In the previous section, you learnt the properties of the system. You will now learn about the notations. The queuing systems with which you are concerned are denoted by M/M/1 and M/M/c where M stands for exponential inter arrival and exponential service time distributions and the third character indicates the number of channels (or servers) available (1 or c). Table 10.1 depicts a list of notations that are commonly used. Table 10.1: List of Notations Used Notation Z H p c m n Pn E(m) E(m/m > 0) E(n) W E(w) P(w = 0) E(w/w > 0) V F(v) E(v) Meaning = Average number of arrivals per time = Average number of customers served per unit time = Traffic intensity = Number of service channels = Number of customers in the queue or the length of the queue = Number of customers in the System (Number of customers in the queue + Number of customers being served ) = Steady state probability of finding n people in the system = Average length of queue = Average length of a non-empty queue = Average number of customers in the system = Steady state waiting time of a customer = Probability of not having to waiting time of a customer = Average waiting time of a customer given that there is a queue = Time spent by a customer in the system in steady state = Waiting time in the queue + Service time = Probability density function of the time spent by a customer in the system = Average time spent by a customer in the system Self Assessment Questions 1. E (m) refers to ________ _________ ________ _________. 2. Probability density function of the time spent by a customer in the system is denoted by _____________. 3. ___________ arrivals are allowed. Operations Research Unit 10 Sikkim Manipal University Page No. 191 10.5 Service System In the previous section, you learnt about notations. You will now learn about service system. The service is provided by a service facility (or facilities). This may be a person (a bank teller, a barber, a machine (elevator, gasoline pump), or a space (airport runway, parking lot, hospital bed) to mention a few. A service facility may include one person or several people operating as a team. There are two aspects of a service system (a) the configuration of the service system and (b) the speed of the service. a) Configuration of the service system The customers entry into the service system depends upon the queue conditions. If at the time of customers arrival, the server is idle, then the customer is served immediately. Otherwise the customer is asked to join the queue, which can have several configurations. By configuration of the service system we mean how the service facilities exist. Service systems are usually classified in terms of their number of channels, or numbers of servers. i. Single server Single queue The models that involve one queue one service station facility are called single server models where customer waits till the service point is ready to take him for servicing. Students arriving at a library counter is an example of a single server facility. Figure 10.1 depicts the single server single queue model. Fig. 10.1: Single Server Single Queue Model ii. Single server Several queues In this type of facility there are several queues and the customer may join any one of these but there is only one service channel. Figure 10.2 depicts the single server several queues model. Operations Research Unit 10 Sikkim Manipal University Page No. 192 Queues Service facility Customer leave Arrivals Fig. 10.2: Single Server Several Queues Model iii. Several (parallel) servers Single queue In this type of model there is more than one server and each server provides the same type of facility. The customers wait in a single queue until one of the service channels is ready to take them in for servicing. Figure 10.3 depicts the several (parallel) servers single queue model. Service stations Fig. 10.3: Several (Parallel) Servers Single Queue Model iv) Several servers Several queues This type of model consists of several servers where each of the servers has a different queue. Different cash counters in an electricity office where the customers can make payment in respect of their electricity bills provide an example of this type of model. Figure 10.4 depicts the several (parallel) servers several queues model. Operations Research Unit 10 Sikkim Manipal University Page No. 193 Fig. 10.4: Several (Parallel) Servers Several Queues Model v) Service facilities in a series In this, a customer enters the first station and gets a portion of service and then moves on to the next station, gets some service and then again moves on to the next station and finally leaves the system having received the complete service. For example, machining of a certain steel item may consist of cutting, turning, knurling, drilling, grinding and packaging operations, each of which is performed by a single server in a series. Figure 10.5 depicts multiple servers in a series. Fig. 10.5: Multiple Servers in a Series b) Speed of the service In a queuing system, the speed with which the service is provided can be expressed in either of the two ways as service rate or as service time. - The service rate describes the number of customers serviced during a particular time period. - The service time indicates the amount of time needed to service a single customer. - Service rates and times are reciprocal of each other and either of them is sufficient to indicate the capacity of the facility. Operations Research Unit 10 Sikkim Manipal University Page No. 194 Thus if a cashier can attend, on an average 5 customers in an hour, the service rate would be expressed as 5 customers/hour and service time would be equal to 12 minutes/customer. Generally, we consider the service time only. If these service times are known exactly, the problem can be easily handled. If these are different and not known with certainty, we have to consider the distribution of the service times in order to analyse the queuing system. Generally, the queuing models are based on the assumption that service times are exponentially distributed about some average service time. 10.6 Single Channel Models In the previous section, you learnt about service system. You will now learn about single channel models. The formulae are depicted in figures 10.6 and 10.7. The examples illustrate the computation of various measures of efficiency of a queuing system, but give an idea of the areas of application as well. ( ) ( ) ( ) ( ) (1 ) ( ) 1 ( ) ( ) 1 1 ( / 0) ( ) ( ) ( ) ( / 0) ( ) ( ) 1 (1 ) 2
= =
= = =
> =
=
=
> =
= = = f v e v f w e w E v P w o E w w E w E n E m m E m P P pn n n Fig. 10.6: Formulae for Poisson Arrivals, Exponential Service, Single Channel Queuing Models Infinite Population Operations Research Unit 10 Sikkim Manipal University Page No. 195 Fig. 10.7: Formulae for Poisson Arrivals, Exponential Service, Single Channel Queuing Models Number of Customers Linked to N Solved problem 1: Patrons arrive at a small post office at the rate of 30 per hour. Service by the clerk on duty takes an average of 1 minute per customer. a) Calculate the mean customer time: i) Spent waiting in line ii) Spent receiving or waiting for service b) Find the mean number of persons: i) In line ii) Receiving or waiting for service Solution: Mean arrival rate = 30 customers per hour = customer per minute Mean service rate = 1 per minute Traffic intensity (a) (i) Mean customer time spent waiting in line Operations Research Unit 10 Sikkim Manipal University Page No. 196 (ii) Mean customer time receiving or waiting for service = 2 minutes (b) (i) Mean number of persons in line = 0.5 customer (ii) Mean number of persons receiving or waiting for service Customer Solved problem 2: The Tool Companys quality control department is manned by a single clerk, who takes an average of 5 minutes in checking parts of each of the machines coming for inspection. The machines arrive once in every 8 minutes on an average. One hour of machine is valued at Rs. 15 and a clerks time is valued at Rs. 4 per hour. What is the average hourly queuing system costs associated with the quality control department? Solution: Mean arrival rate l = 1/8 per minute = per hour Mean service rate per minute = 12 per hour Average time spent by a machine in the system E (v) = = hours Average queuing cost per machine = An average arrival of machines per hour costs = Rs. 25 per hour Average hourly queuing cost = Rs. 25 Average hourly cost for the clerk = Rs. 4 Hence total hourly cost for the department = Rs. 29 Operations Research Unit 10 Sikkim Manipal University Page No. 197 Solved problem 3: A television repairman finds that the time spent on his jobs has an exponential distribution with a mean of 30 minutes. If he repairs sets in the order in which they came in and if the arrival of sets follow a Poisson distribution with an average rate of 10 per 8 hour day, what is the repairmans expected idle time each day? How many jobs are ahead of the average set just brought in? Solution: = 10/8 = 5/4 sets/hr; = (1/30) 60 = 2 sets/hr. Number of hours for which the repairman remains busy in an 8-hour day = 8* ( / ) = 8(5/8) = 5 hrs. Repairmans expected idle time in an 8 hour day = 8-5 = 3 hrs. Expected number of T.V sets in the system = (Ls) = / ( - ) = (5/4)/ [2-(5/4)] = 5/3 = 2 sets (approx.) Solved problem 4: Arrivals at a telephone booth are considered to follow Poisson distribution with an average time of 10 minutes between one arrival and the next. The length of phone call is assumed to be distributed exponentially, with a mean of 3 minutes. a) What is the probability that a person arriving at the booth will have to wait? b) The telephone department will install a second booth when convinced that an arrival would expect to wait for at least 3 minutes for a phone call. By how much should the flow of arrivals increase in order to justify a second booth? c) What is the average length of the queue that forms from time to time? d) What is the probability that it will take him/her more than 10 minutes altogether to wait for the phone and complete his call. Operations Research Unit 10 Sikkim Manipal University Page No. 198 Solution: = 1/10 person/minute; = 1/3 = 0.33 person/minute. a) Probability that a person arriving at the booth will have to wait = 1- P0 = 0.3 b) The installation of the 2nd booth will be justified only if the arrival rate is more than the waiting time. Let be the increased arrival rate. Then expected waiting time in the queue will be (Wq) = / ( - ) = / 0.33(0.33- ) Or =0.16 Hence the increase in the arrival rate is 0.16-0.10 = 0.06 arrivals /minute c) Average length of non-empty queue = / ( - ). = 0.33/0.23 = 1.42857 Customers (approx.) d) Probability of waiting for 10 minutes or more = P (t>=10)
= ( / ) ( - ) e -( - )tdt
= (0.3)(0.23)e-0.23tdt = 0.069 e -0.23t = 0.03 10 -0.23 This shows that 3 percent of the arrivals on an average will have to wait for 10 minutes or more before they can use the phone. Solved problem 5: A hospital emergency room can accommodate at most M = 5 patients. The Patients arrive at a rate of 4 per hour. The single staff physician can only treat 5 patients per hour. Any patient overflow is directed to another hospital. a) Determine the probability distribution for the number of patients either waiting for or receiving treatment at any given time b) Determine the mean values for the number of patients in the emergency room. And the number of patients waiting to see the doctor. Solution: M = Maximum number of customers in the system = 5 = 4 per hour and = 5 per hour = 8 Operations Research Unit 10 Sikkim Manipal University Page No. 199 The probability distribution for the number of patients in the system Pr = n | | . |
\ |
n P0 for 0 M The probability distribution for the number of patients either waiting for treatment or receiving treatment is as shown in table 10.2. Table 10.2: Probability Distribution N Pn 0 0.271 1 0.217 2 0.173 3 0.139 4 0.111 5 0.089 (b) (1) The mean value for the number of patients in the emergency room = 0 x (0.271) + 1 x (0.217) + 2 x (0.173) + 3 x (0.139) + 4 x (0.111) + 5 x (0.089) = 1.868332032. Solved problem 6: Cars arrive at a toll gate on a frequency according to Poisson distribution with mean 90 per hour. Average time for passing through the gate is 38 seconds. Drivers complain of long waiting time. Authorities are willing to decrease the passing time through the gate to 30 seconds by introducing new automatic devices. This can be justified, if under the old system, the number of waiting cars exceeds 5 and the percentage of gates idle time under the new system should not exceed 10. In the above scenario, can the new device be justified? Operations Research Unit 10 Sikkim Manipal University Page No. 200 Solution: In the old system, Meaning arrival rate = 90 per hour = 1.5 per minute Mean service rate = per minute Traffic intensity = = Expected number of the waiting cars = Hence the new system is justified on the basis of the expected number of waiting cars. Under the new system, the probability of device being idle = P 0 = 1 = 1 - .75 = .25 which is greater than 10% required for the new device to be introduced. As only one of the stipulated conditions is fulfilled, the new automatic device is not justified. Solved problem 7: Customers arrive at a one window drive-in bank according to a Poisson distribution with mean 10 per hour. Service time per customer is exponential with mean 5 minutes. The space in front of the window, including that for the serviced car can accommodate a maximum of 3 cars. Other cars can wait outside this space. a) What is the probability that an arriving customer can drive directly to the space in front of the window? b) What is the probability that an arriving customer will have to wait outside the indicated space? c) How long is an arriving customer expected to wait before starting service? Operations Research Unit 10 Sikkim Manipal University Page No. 201 d) How many spaces should be provided in front of the window so that all the arriving customers can wait in front of the window at least 20% of the time? Solution: (a) = 10 per hour; 10/12 = 5/6 per hour (b) p = / = 10/12 = 5/6 The probability that an arriving customer can drive directly to the space in front of the window = P0+P1+P2 = (1-p)(1 + p +p2 ) = required probability = .42 (c) Probability that an arriving customer has to wait outside the indicated space = probability that there are at least 3 customers in the space in front of the window = 0 + p1 + p2 + p3) =1 (1- p) (1 + p + p2 + p3) = .48 (d) The average waiting time of a customer in queue. (e) The existing space in front of the window can accommodate up to three cars. Probability that a customer can wait in front of the window must be at least 0.20. This is possible if the number of customers in the system is 0, 1 or 2 for at least 20% of the time. i.e., 0 + p1 + p2 0.20 It is seen from (a) that 0 + p1 + p2 = 0.42 As even with one space in front of the window, 42 percent of the times an arrival waits in the space, the number of spaces required is one or more. Self Assessment Questions 4. The expected number of customers in non-empty queue is given by ___________. Operations Research Unit 10 Sikkim Manipal University Page No. 202 5. The probability that an arriving customer has to wait for receiving service is given by ___________. 10.7 Multiple Service Channels In the previous section, you learnt about single channel models. You will now learn about multiple channel models. The analysis of systems involving several service channels is more complex. The major practical utility of these models are the ways of improving to provide additional service facilities. The list of formulae to be used is depicted in figure 10.8, followed by few examples. Fig. 10.8: Formulae for Poisson Arrivals, Exponential Service Multi channel Queuing Models Infinite Population Solved problem 8: Ships arrive at a port at a rate of one in every 3 hours, with a negative exponential distribution of interarrival times. The time a ship occupies a Operations Research Unit 10 Sikkim Manipal University Page No. 203 berth for unloading and loading has a negative exponential distribution with an average of 12 hours. If the average delay of ships waiting for berths is to be kept below 6 hours, how many berths should there be present at the port? Solution: For multichannel queues, 1/ ensure that the queue does not explode. Let us calculate the waiting time when c = 5 Substituting for n, c and p P 0 = Average waiting time of a ship E (w) = 0 rp Pe 1/(c 1)!/(c p)21 = 6.65 This is greater than 6 hours and inadequate. When P0 Moreover, average waiting time of a ship E(w) = 1.71 hours. Hence 6 berths should be provided at the port. Solved problem 9: A bank has two counters for withdrawals. Counter 1 handles withdrawals of value less than Rs. 300 and counter 2 handles withdrawals of Rs. 300 and above. Analysis of service time shows a negative exponential distribution with mean service time of 6 minutes per customer for both the counters. Operations Research Unit 10 Sikkim Manipal University Page No. 204 Arrival of customers follows Poisson distribution with mean 8 per hour for counter 1 and 5 per hour for counter 2. a) What are the average waiting times per customer of each counter? b) If each counter could handle all withdrawals irrespective of their value how would the average waiting time change? Solution: (a) For counter one 1 = 10 per hour, 1 = 8 per hour. For counter two, = 10 per hour, 2 = 5 per hour (b) When each counter can handle all withdrawals irrespective of their value with an average arrival rate of 13 per hour, mean service time of 10 per hour at each counter and number of channels = 2, = 4. 71P0 = 0.21 Average waiting time E(w) = P0 = 4.38961039 minutes Self Assessment Questions 6. Write the formula for calculating the expected number of customers in the system. 7. Write the formula for calculating the average waiting time of the customer in the queue. Operations Research Unit 10 Sikkim Manipal University Page No. 205 10.8 Erlang Family of Distribution of Service Times In the previous section, you learnt about multiple channel models. You will now learn the Erlang family of distribution of service times. The queuing processes discussed till now are mathematically simple. Assuming that the service time follows negative exponential distribution, you are also taking the standard deviation equal to its mean. There would be situations, where the mean and the standard deviation substantially differ. At these instances the models have to be made more general by using a distribution system, which coincide closely to the practical problems, but yet retains the simplicity of the properties of negative exponential distribution. A.K. Erlang first studied such a distribution system. Consider the distribution of a service time involving a fixed number of phases, k, and each phase has a negative exponential distribution. If there are k phases and the average time taken by a customer through each phase is (1/ k).units, the service time distribution f (t) is given by The mean of this distribution is (1 / ) and standard deviation is (1/ k). If k = 1, f (f) = mem1 which is the negative exponential distribution; same as the models considered earlier. For k = 2, f (f) = 4m2te2mf, etc. The mode is located at t = (k-1 / k). If k = , the variance is zero and this corresponds to a case where the service time is constant and has the value (1/). Figure 10.9 depicts the way the density functions vary as k increases. Operations Research Unit 10 Sikkim Manipal University Page No. 206 Fig. 10.9: Variation in Density Functions as k Increases The measures of efficiency should take into account the number of customers getting service, number having entered during any one or more of the phases and the number yet to join. For arrivals following Poisson distribution, with mean l and service time following the kth Erlang distribution with mean, (1/) the formulae applicable are given below. Average queue length Average number of units in the system Average waiting time Average time spent in the system For constant service time, limiting k to Operations Research Unit 10 Sikkim Manipal University Page No. 207 Solved problem 10: At the bus depots cafeteria, the customer has to pass through three counters. At the first counter, customer buys coupons, at the second they collect coffee or tea. The server at each counter takes an average two minutes although the distribution of the time of service is approximately exponential. If the arrival of customers to the cafeteria is approximately Poisson at an average rate of six per hour, what is the average time spent by customer waiting in the cafeteria? What is the average time of getting the service? What is the most probable time spent in getting the service? Solution: This is a queuing process with service time following Erlang distribution. No. of phases k = 3 persons per minute = 6 persons per hour = 1/10 person per minute (1) Average waiting time = 6 minutes (2) Average time spent in collecting coupons, snacks, etc. (3) Most probable time spent in collecting coupons, snacks. etc/ Self Assessment Questions 8. Write the formula for E (m) 9. Write the formula for E (w) Operations Research Unit 10 Sikkim Manipal University Page No. 208 10.9 Applications of Queuing Theory The applications of queuing theory are: 1. Traffic control 2. Determining the sequence of computer operations 3. Predicting computer performance 4. Health services (e.g., control of hospital bed assignments) 5. Airport traffic, airline ticket sales 6. Layout of manufacturing systems. 7. Telecommunications 10.10 Limitations of Queuing Theory In the previous section, you learnt the applications of queuing theory. You will now learn the limitations of queuing theory. Classic queuing is too mathematically restrictive to be able to model all real world situations. This restriction arises because the underlying assumptions of the theory do not always hold well in the real world. Alternative means of analysis have been devised in order to provide some insight into problems which do not fall under the scope of queuing theory, though they are often scenario-specific since they generally consist of computer simulations and/or of analysis of experimental data. 10.11 Summary Let us recapitulate the important concepts discussed in this unit: This unit discusses about the mathematical analysis of Queuing process & how the service system is useful in different process. 10.12 Glossary Queuing discipline: It refers to the decision rule by which the customers are selected from queue for service 10.13 Terminal Questions 1. Discuss about Erlang family of distribution of service times. 2. A stenographer has 5 persons for whom she performs stenographic work. Arrival rate follows Poisson distribution and service times are Operations Research Unit 10 Sikkim Manipal University Page No. 209 exponential. Average arrival rate is 4 per hour with an average service time of 10 minutes. Cost of waiting is Rs. 8 per hour while the cost of servicing is Rs. 2.50 each. Calculate: i) The average waiting time of an arrival ii) The average length of the waiting line iii) The average time which an arrival spends in the system 3. A ticket issuing office is being manned by a single server. Customers arrive to purchase tickets according to Poisson distribution with a mean rate of 30 per hour. The time required to serve a customer has an exponential distribution with a mean of 90 seconds. Find the value of pn, Ls and Ws. Denote the expected line length and waiting time in the system respectively. (Hint: =54/6 * 60, = 60/90 = 2/3. Use formulae for Pn, Ls, Ws) 4. Discuss different types of service systems? 10.14 Answers Self Assessment Questions 1. Average length of queue 2. F (v) 3. Simultaneous 4. m / m l 5. / 6. E(m/m > 0) = /( ) 7. P(w = 0) = 1-P 8. 2k ( ) k 1 2
+ 9. 2k ( ) k 1
+ Terminal Questions 1. Refer 10.8 2. i) 12.4 min ii) 0.79= one stenograph iii) 22.4 min. 3. Pn=1/4(1/4)n, Ls = 3 customers , Ws= 6 min. 4. Refer 10.5 Operations Research Unit 10 Sikkim Manipal University Page No. 210 10.15 Case Study Queuing Solution Never dull. That is how you would describe your job at the centralised records and benefits administration centre for Always First, a large company manufacturing computers and computer peripherals. The company provides a call centre at the administration centre for the 60,000 Always First employees throughout the United States. Employees contact the call centre to obtain information about dental plans and pension options, to change tax forms and personal information and to process leaves of absence and retirements. The centralised records and benefits administration centre and its call centre opened just six months ago to replace 35 small regional administration centres across the country. Since this recent opening, you and John Ducker, the director of human resources, have endured one long roller coaster ride. Receiving the go-ahead from corporate headquarters to establish the centralised records and benefits administration centre was definitely an up. Getting caught in the crossfire of angry customers (all employees of Always First) because of demand overload for the records and benefits call centre was definitely a down. Accurately forecasting the demand for the call centre provided another up. And today you are faced with another down. John approaches your desk with a not altogether attractive frown on his face. He begins complaining immediately, I just dont understand. The forecasting job you did for us two months ago really allowed us to understand the weekly demand for the centre, but we still have not been able to get a grasp on the staffing problem. We used both historical data and your forecasts to calculate the average weekly demand for the call centre. We transformed this average weekly demand into average hourly demand by dividing the weekly demand by the number of hours in the working week. We then staffed the centre to meet this average hourly demand by taking into account the average number of calls a representative is able to handle per hour. But something is horribly wrong. Operational data records show that over thirty percent of the customers wait over four minutes for a representative to answer the call! Customers are still sending me numerous complaints and executives from corporate headquarters are still breathing down my neck! I need help! You calm John down and explain to him that you think you know the problem: the number of calls received in a certain hour can be much greater (or much less) than the Operations Research Unit 10 Sikkim Manipal University Page No. 211 average because of the stochastic nature of the demand. In addition, the number of calls a representative is able to handle per hour can be much less (or much greater) than the average depending upon the types of calls received. You then tell him to have no fear; you have the problem under control. You have been reading about the successful application of queuing theory to the operation of call centres, and you decide that the queuing models you learned in school will help you determine the appropriate staffing level. Discussion Questions: (a) You ask John to describe the demand and service rate. He tells you that calls are randomly received by the call centre and that the centre receives an average of 70 calls per hour. The computer system installed to answer and hold the calls is so advanced that its capacity far exceeds the demand. Because the nature of a call is random, the time required to process a call is random, where the time frequently is small but occasionally can be much longer. On an average, however, representatives can handle 6 calls per hour. Which queuing model seems appropriate for this situation? Given that slightly more than 35 percent of customers wait over 4 minutes before a representative answers the call, use this model to estimate how many representatives John currently employs. (b) John tells you that he will not be satisfied unless 95 percent of the customers wait only 1 minute or less for a representative to answer the call. Given this customer service level and the average arrival rates and service rates from part (a), how many representatives should John employ? (c) Each representative receives an annual salary of US $30,000 and John tells you that he simply does not have the resources available to hire the number of representatives required to achieve the customer service level desired in part (b). He asks you to perform sensitivity analysis. How many representatives would he need to employ to ensure that 80 percent of customers wait 1 minute or less? How many would he need to employ to ensure that 95 percent of customers wait 90 seconds or less? How would you recommend John choose a customer service level? Would the decision criteria be different if Johns call centre were to serve Operations Research Unit 10 Sikkim Manipal University Page No. 212 external customers (not connected to the company) instead of internal customers (employees)? References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. Operations Research Unit 11 SikkimManipalUniversity Page No. 213 Unit 11 Finite Queuing Models Structure: 11.1 Introduction Objectives 11.2 Finite Queuing Models Finite queuing tables Measures of system efficiency Use of finite queuing tables 11.3 Summary 11.4 Glossary 11.5 Terminal Questions 11.6 Answers 11.7 Case Study 11.1 Introduction In the previous unit, you learnt about mathematical analysis of queuing theory, notations and service systems involved in queuing theory. You also learnt about Erlang family of distribution of service times and the applications and limitations of queuing theory. The models discussed so far relate to situations involving infinite population of customers, that is, the queue can increase indefinitely. However, you may come across scenarios where the possible number of arrivals is limited and small. In this unit, we will discuss such scenarios. In a production shop, if the machines are considered as customers requiring service from repair crews or operators, the population is restricted to the total number of machines in the shop. In a hospital ward, the probability of the doctors or nurses being called for service is governed by the number of beds in the ward. Similarly, in an aircraft the number of seats is finite and the number of stewardesses provided by the airlines will be based on the consideration of the maximum number of passengers who can demand service. Operations Research Unit 11 SikkimManipalUniversity Page No. 214 In case of a queuing system with infinite population, you can improve the efficiency of the system in terms of reduced average length of queues, average waiting time and time spent by the customer in the system. To do this, you need to increase the number of service channels. However, such increases mean additional cost and will have to be balanced with the benefits likely to accrue. If the queuing system in a machine shop is under study, the cost of providing additional maintenance crews or operators can be compared with the value of additional production due to reduced downtime of the machines. Sometimes it is not possible to quantify the benefitsthe management will have to base its decisions on the desired standards for customer service. In this unit, you will learn the finite queuing models, finite queuing tables and the use of these models and tables. Objectives: After studying this unit, you should be able to: - describe queuing models and various queuing disciplines - interpret and apply finite queuing tables to solve problems 11.2 Finite Queuing Models Basic concept of queuing models You can use a queuing model to approximate a real queuing situation or system, so that you can analyse the queuing behaviour mathematically. Queuing models allow you to determine a number of useful steady state performance measures. These measures are: - Average number in the queue, or the system - Average time spent in the queue, or the system - Statistical distribution of those numbers or times - Probability of the queue being empty or full - Probability of finding the system in a particular state Analysis of the relevant queuing models allows you to identify the cause of a queuing issue and assess the impact of the proposed changes. Typically, you need to analyse a queuing model to minimise the sum of: - Customer waiting cost - Service capacity cost Operations Research Unit 11 SikkimManipalUniversity Page No. 215 In a finite queuing model, the system processes incoming requests or messages at a certain rate. Any newly arrived request or message is added to the queue only if the number of requests or messages waiting in the queue is less than the maximum. The model is based on the following assumptions: - A queue is characterised by the maximum permissible number of customers that it can contain. This also could be finite or infinite. In most of the practical situations, it is finite. - Queuing discipline determines the manner in which the system handles calls from the customers. It defines the way in which they are served, the order in which they are served and the way in which resources are divided between the customers. A finite queuing process may follow any of the following queuing disciplines: Figure 11.1 depicts the queue discipline route in a finite queuing process. Fig. 11.1: Queue Discipline Route in a Finite Queuing Process First Come - First Served (FCFS): This principle states that customers are served one at a time and that the customer that has been waiting the longest is served first. In most queuing models, the order is usually first come-first served. Priority: Priority-discipline models give priority to rush jobs and important customers over others. For instance, machines of high cost may be given priority for maintenance while others may be kept waiting even if they had broken down earlier. Operations Research Unit 11 SikkimManipalUniversity Page No. 216 Random: This principle states that the system may randomly select a customer for service. For example, in a machine shop if a single operator is attending to several machines and several machines call for his attention at a time, he may attend first to the one nearest to him. Last Come-First Served (LCFS): This discipline is based on the stack method. Most quantitative parameters, such as the average queue length and average time spent in the systemdo not depend on priority queues. That is the reason most models either do not take the queuing discipline into account at all or assume the normal FCFS queue. The only parameter that depends on the queuing discipline is the variance (or standard deviation) of the waiting time. The analysis of finite queuing models is more complex than those with infinite population although the approach is similar. Simple waiting line model The simplest waiting line model assumes that arrivals join a queue that is of unlimited size, waiting in line until their turn for service comes on a first-cumfirst- serve basis and then enter a service facility consisting of a single channel. If, W = Average time spent in queue (i.e., sum of the expected waiting time and expected service time) = Average rate of units passing through the system per unit tim e L = Average number of units in the system. Then, L= W If = Average service rate 1/ = Expected inter arrival time 1/ = Expected service time Further, the utilisation factor, (i.e., the expected fraction of time the server is busy) is given by: = / s Where, (s) is the fraction of the system's service capacity that being utilised on the average by arriving customers (C) Operations Research Unit 11 SikkimManipalUniversity Page No. 217 The crux of the queuing theory is to 'achieve a trade-off between excessive waiting by customers (i.e., too much demand) and cost due to excessive idle time at the service facility (i.e., too little demand). Lending system in a library is an example of single facility channel waiting line. 11.2.1 Finite queuing tables Figure 11.2 depicts the different notations used in finite queuing tables. Consider a machine shop with N machines. The inter breakdown time of these machines follows a negative exponential distribution with mean U. The number of breakdowns follows Poisson distribution with mean U 1 = N = Population (machines, customer etc.) M = Service channels (repairmen, telephone linesetc.) T = Average service time (repair time, length of conversation on a telephone etc.) W = Average waiting time U = Average running time (of machines) or meantime between calls for service per unit H = Average number of units being serviced L = Average number of units waiting for services J = Average number of units in operation F = Efficiency Factor X = Service Factor D = Probability that if a unit calls for service, it will have to wait. Fig. 11.2: Notations Used in Finite Queuing Tables It is assumed that machines are kept running (or in operation) all the time, except when in repairs or waiting for repair crew to attend. If M repair crews are available, the time taken by any crew follows a negative exponential distribution with mean T. Naturally, a machine which has broken down will have to wait for repairs if all the repair crews are busy. 11.2.2 Measures of system efficiency For a given set of machines, the efficiency of the repair system may be judged by the extent to which machines have to wait for repairs. If W is the average time for which a machine has to wait, the efficiency factor F is defined as, Operations Research Unit 11 SikkimManipalUniversity Page No. 218 T U W T U F + + + = At any point of time, a machine will either be running or under repair or waiting for repairs. Therefore, the total number of machinesN= J + H + L. N L and N J , N H , correspond to the probability that a machine is being repaired, running or waiting for repairs respectively. In the finite queuing tables, service factor X is defined as T U T X + = . X is an indicator of the utilisation of repair crew. The formulae for other properties of the system are given below: T U W T U H L J H J F NF (1 X) T W U NU J N (1 F) T W U NW L FNX T W U NT H + + + = + + + = = + + = = + + = = + + = Operations Research Unit 11 SikkimManipalUniversity Page No. 219 Table 11.1: Finite queuing Tables Source: Shim, K. Jae, & Siegel, G. Joel. Operations Management . Barron's Educational Series 11.2.3 Use of finite queuing tables The above table gives values of F and D for different values of N, M and X. They are arranged in the ascending order of the values of the population. For each N, the value of X increases from .001 to .950. For a given service factor X, several values of M can be found. For each value of X and M, values of D and F are tabulated. The steps for finite queuing tables are as follows: - Find mean service time T and mean running time U. Operations Research Unit 11 SikkimManipalUniversity Page No. 220 - Compute the service factor. - Select the table corresponding to the population N. - For the given population, locate the service factor value. - Read from tables, values of D and F for the number of service crews M. - If necessary, these values may be interpolated between relevant values of X. - Calculate the other measures L, W, H and J from the formulae given. Figure 11.3 depicts the steps for finite queuing tables. Fig. 11.3: Finite Queuing Tables Steps The overall efficiency F of the system will increase with the number of service channels (M) provided. As mentioned earlier, addition of service crews involves cost, which should be justified by the increase in the efficiency of the system that is additional running time of machines. However, it will be seen from the table that as M increases the rate of increase in efficiency decreases. The practical significance is that beyond a certain value of M, it is not worthwhile increasing M as there would be no appreciable increase in the efficiency of the system. Operations Research Unit 11 SikkimManipalUniversity Page No. 221 Problem 1: In a chemical factory, there are five hoppers of identical size which feed material to grinding mills. Due to changes in the requirement of material, there are variations in the time taken for emptying the hoppers. On the basis of past experience, it was found to follow negative exponential distribution with an average of 10 hours between getting emptied. Whenever a hopper gets empty, it has to be filled by a pay loader. Although the capacity of the hopper is the same, the time taken to fill the hoppers varies due to different locations from where the material is to be loaded. The time for filling the hopper also was found to follow negative exponential distribution with an average of 2.5 hours. The company hires the pay loaders at a cost of Rs. 100 per hour irrespective of whether it is operated or not. If the mill has to be stopped due to the empty hopper, it costs Rs. 1000 per hour in terms of loss of profits. To determine the number of pay loaders which the company should engage to minimise overall cost you need to create a finite queuing table. Solved Problem 1: For the above scenario, determine the number of pay loaders the company should engage to minimise overall cost. Solution: Since T = 2.5 and U = 10 0.200 2.5 10 2.5 T U T X = + = + = For N = 5, X = 0.200 we have the following values from the tables: M D F 3 0.028 0.998 2 0.194 0.976 1 0.689 0.801 We now prepare a table as depicted in table 11.1: Operations Research Unit 11 SikkimManipalUniversity Page No. 222 Table 11.2: Number of Pay Loaders to be Engaged to Minimise Cost 1. No. of pay loaders 3 2 1 2. Overall efficiency of system (F) 0.998 0.976 0.801 3. N(1 X) 4.00 4.00 4.00 4. Average number of mills running per hour J = NF(1 X) 3.992 3.904 3.204 5. Expected profits atRs. 1000 per hour 3,992 3,904 3,204 6. Loading cost at Rs. 100 per hour per pay loader 300 200 100 7. Expected net profits per hour (5 6) 3,692 3,704 3,104 As increasing the number of pay loaders beyond two reduces the profits, the company should engage only two pay loaders. Self Assessment Questions 1. When the possible number of arrivals is limited, then we apply infinite queuing model. (True/False) 2. The queue discipline in a finite queuing process can be random. (True/False) 3. The efficiency factor for this model is T W T U F + + = . (True/False) 4. __________ principle states that customers are served one at a time and that the customer that has been waiting the longest is served first. 5. ______ discipline models give priority to rush jobs and important customers over others. 11.3 Summary Let us recapitulate the important concepts discussed in this unit: - This unit on finite queuing theory deals with situations where customers arrive, wait for the service, get the service and leave the system. - Customers may come individually or in groups from large/small population, at known/variable times, form one or more queues and move in a certain order to the service station(s)to avail of services whose speed may be fixed or variable. Operations Research Unit 11 SikkimManipalUniversity Page No. 223 - Queuing systems are analysed for determining the optimal service level, where the total cost of providing service and waiting, is minimised. - An increase in the service level increases the cost of providing service, but reduces the cost of waiting, while a decrease in the service level induces opposite changes. 11.4 Glossary Stack: Last in first out collection Variance: how far the waiting time in a queue is spread out Trade-off: compromise at the expense of one or more objectives 11.5 Terminal Questions 1. Customers arrive at the first class ticket counter of a theatre at a rate of 12 per hour. There is one clerk serving the customers at a rate of 30 per hour. i) What is the probability that there is no customer in counter (for instance, the system is idle)? ii) What is the probability that there are more than 2 customers in the counter? iii) What is the probability that there is no customer waiting to be served? iv) What is the probability that a customer is being served and nobody is waiting? 2. Assume that at a bank teller window, the customers arrive in their cars at the average rate of twenty per hour according to thePoisson distribution. Also assume that the bank teller spends an average of two minutes per customer to complete a service and the service time is exponentially distributed. Customers, who arrive from an infinite population, are served on a FCFS basis and there is no limit to the possible queue length. i) What is the expected waiting time in the system per customer? ii) What is the mean number of customers waiting in the system? iii) What is the probability of zero customers in the system? iv) What value is the utilisation factor? Operations Research Unit 11 SikkimManipalUniversity Page No. 224 11.6 Answers Self Assessment Questions 1. False 2. True 3. False 4. First Come - First Served (FCFS) 5. Priority Terminal Questions 1. i. P (the system is idle) = 1- = 1-0.4 = 0.6. ii. P (n > 2) = 1 P (n s 2) = 1 [P (0) + P (1) + P (2)] = 1- 0.936 = 0.064. iii. P (no customer waiting to be served) = P (0) + P (1) = 0.84. iv. P (a customer is being served and none is waiting) = P (1) = 0.24. For more details, refer section 11.2 2. i. Expected waiting time in the system per customer, Ws = 1/ (-) = 1/ (30-20) =1/10 hr. ii. Mean number of customers waiting in the system, Lq = 3 4 3 2 1 3 2 1 2 =
=
. iii. Probability of zero customers in the system, P (0) = 3 1 3 2 1 = 1 = . iv. Utilisation factor, 3 2 = For more details, refer section 11.2 11.7 Case Study Warehousing Corporation A single crew is provided for unloading and/or loading each truck that arrives at the loading deck of a warehouse. These trucks arrive according to Operations Research Unit 11 SikkimManipalUniversity Page No. 225 a Poisson input process at a mean rate of one per hour. The time required by a crew to unload and/or load a truck has an exponential distribution. The expected time required by a one man crew would be two hours. The cost of providing each additional member of the crew is Rs 10 per hour. The cost that is attributable to having a truck not in use is estimated to be Rs 40 per hour. Assume that the mean service rate of the crew is proportional to its size, what should the size be in order to minimise the expected total cost per hour? The general routine is that order pickers assemble orders placed by stores and load them.Oneof the company trucks wait at the single loads, on a FCFS basis. They then proceed to the store where the order is destined, unload and return for another order. Because of the many different routes and distances, and traffic problems at different times of the day, the time between arrivals of trucks at the dock is randomaveraging 30 minutes. The loading time also follows an exponential distributionaveraging 15 minutes.Truck drivers are paid Rs. 10 per hour and the crew of two loaders are each paid Rs. 4 per hour. Truckers have complained about the long waiting, so a sample was taken showing that truckers did indeed wait for an average of 32 minutes. The warehouse manager knows that a second truck dock would probably solve the problem, but the large capital expenditure plus the disruption of operations during construction are deterrents to the solution. Tests are made with different crew patterns and it is found that a crew of 3 can be used to advantage, reducing the loading time to 10 minutes. As management trainee working with the corporation, help the corporation in finding suitable answers to the following questions: Discussion Questions: a. Is the crew of 3 loaders more economical than the crew of 2? b. What is the probability that an arriving truck will find at least one truck already in the system? c. How long is the crew idle? d. If another truck dock is available with the present operating characteristics, how will it affect the total expected cost per hour? Operations Research Unit 11 SikkimManipalUniversity Page No. 226 References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. E-References: - http://www.scribd.com/doc/14160718/Operations-Management-Ch19- Waiting-Lines. accessed on Dec 20, 2011. - http://staff.um.edu.mt/jskl1/simweb/intro.htm. accessed on Dec 30, 2011. - http://docs.google.com/gview?a=v&q=cache%3AuNc0DtLpqBMJ%3Aw ww.stsc.hill.af.mil%2Fcrosstalk%2F2007%2F12%2F0712LauAppendix. pdf+finite+queuing+model&hl=en&gl=uk&sig=AFQjCNHDjm- WKl7g8LD_DztsUvEJxyr9Ig&pli=1. accessed on Dec 27, 2011. Operations Research Unit 12 Sikkim Manipal University Page No. 227 Unit 12 Simulation Structure: 12.1 Introduction Objectives 12.2 Simulation 12.3 Methodology of Simulation 12.4 Basic Concepts 12.5 Simulation Procedure Allocation of random numbers Use of random number tables 12.6 Sample Size 12.7 Summary 12.8 Glossary 12.9 Terminal Questions 12.10 Answers 12.11 Case Study 12.1 Introduction In the previous unit, you learnt the finite queuing models, finite queuing tables and the use of these models and tables. Simulation is a numerical technique for conducting experiments that involve certain types of mathematical and logical relationships necessary to describe the behaviour and structure of a complex real world system over an extended period of time. In other words, it is a quantitative technique that utilises a computerised mathematical model in order to represent actual decision making under conditions of uncertainty for evaluating alternative courses of action based upon facts and assumptions. In this unit, you will learn about simulation, methodology and basic concepts of simulation. You will also learn simulation procedure and sample size. A definition of simulation as given by Shannon: Simulation is the process of defining a model of a real system and conducting experiments with this model for the purpose of understanding the behaviour (within the limits imposed by a criterion or a set of criteria) for the operation of a system. Operations Research Unit 12 Sikkim Manipal University Page No. 228 Objectives: After studying this unit, you should be able to: - define simulation - explain the application of simulations in business problems - describe the use of random number table 12.2 Simulation Using simulation, an analyst can introduce the constants and variables related to the problem, set up the possible courses of action and establish criteria which act as measures of effectiveness. The major reasons for applying simulation technique to OR problems may be listed as below: 1. It is an appropriate tool to use in solving a problem when experimenting on the real system: a) Would be disruptive. b) Would be too expensive. c) Does not permit replication events. d) Does not permit control over key variables. 2. It is desirable tool for solving a business problem when a mathematical model is: a) Complex to solve. b) Beyond the capacity of available personnel. c) Not detailed enough to provide information on all important decision variables. 3. The major reasons for adopting simulation in place of other mathematical techniques are: a) It may be the only method available, because it is difficult to observe the actual reality. b) Without appropriate assumption, it is impossible to develop a mathematical solution. c) It may be too expensive to actually observe the system. d) There may not be sufficient time to allow the system to operate for a very long time. 4. It provides a trial-and-error movement towards the optimal solution. The decision maker selects an alternative to experience the effect of the selection and then improves the selection. In this way, the selection is adjusted until it approximates the optimal solution. Operations Research Unit 12 Sikkim Manipal University Page No. 229 12.3 Methodology of Simulation In the previous section, you learnt why simulation technique is applied. You will now learn the methodology of simulation. The methodology developed for simulation process consists of seven steps depicted in figure 12.1: Fig. 12.1: Steps in Methodology of Simulation Self Assessment Questions 1. Simulation is useful to analyse problems where analytical solution is difficult. (True/False) 2. Simulation cannot be used where mathematical techniques can be used. (True/False) 3. Simulation solutions would be identical to those using mathematical techniques. (True/False) Operations Research Unit 12 Sikkim Manipal University Page No. 230 12.4 Basic Concepts In the previous section, you learnt the methodology of simulation. You will now learn the basic concepts. Simulation is also called experimentation in the management laboratory. While dealing with business problems, simulation is often referred to as Monte Carlo Analysis. Two American mathematicians, Von Neumann and Ulan, in the late 1940s, found a problem in the field of nuclear physics too complex for analytical solution and too dangerous for actual experimentation. They arrived at an approximate solution by sampling. The method they used had resemblance to the gamblers betting systems on the roulette table, hence the name Monte Carlo has stuck. Imagine a betting game where the stakes are based on correct prediction of the number of heads, which occur when five coins are tossed. If it were only a question of one coin, most people know that there is an equal likelihood of a head or a tail occurring, that is the probability of a head is . However, without the application of probability theory, it would be difficult to predict the chances of getting various numbers of heads, when five coins are tossed. Why dont you take five coins and toss them repeatedly? Note down the outcomes of each toss after every ten tosses, approximate the probabilities of various outcomes. As you know, the values of these probabilities will initially fluctuate, but they would tend to stabilise as the numbers of tosses are increased. This approach in effect is a method of sampling, but is not very convenient. Instead of actually tossing the coins, you can conduct the experiment using random numbers. Random numbers have the property that any number is equally likely to occur, irrespective of the digit that has already occurred. Let us estimate the probability of tossing of different numbers of heads with five coins. We start with set random numbers depicted in table 12.1: Table 12.1: Random Number Set 78466 71923 78722 78870 06401 61208 04754 05003 97118 95983 Operations Research Unit 12 Sikkim Manipal University Page No. 231 By following a convention that even digits signify a head (H) and the odd digits represent a tail (T), the tossing of a coin can be simulated. The probability of occurrence of the first set of digits is and that of the other set is also a condition corresponding to the probability of the occurrence of a head and the probability of occurrence of a tail respectively. It is immaterial as to which set of five digits should signify a head. The rule could be that the digits 0, 1, 2, 3 and 4 represent a head and the digits 5, 6, 7, 8 and 9 a tail. It is only necessary to take care that the set of random numbers allotted to any event matches with its probability of occurrence. For instance, if youre interested in allotting random numbers to three events A, B and C with respective probabilities 0.24, 0.36 and 0.40, choose two digit random numbers from 00 to 99. The numbers 00 to 23 signify event A, 24 to 59 signify B and 60 to 99 signify C. The first set of five random digits in the list of random numbers implies that the outcome of the first toss of 5 coins is as depicted in table 12.2: Table 12.2: Outcome of First Toss of 5 Coins Coin 1 2 3 4 5 Random Number 7 8 4 6 6 Outcome T H H H H Hence it is 4 heads and 1 tail. Proceeding in the same way, you can tabulate the results of the first ten tosses as depicted in table 12.3. Table 12.3: Results of First Ten Tosses Toss No. Number of Heads Tails 1 2 3 4 5 6 7 8 9 10 4 3 4 3 1 1 3 4 3 1 1 2 1 2 4 4 2 1 2 4 Operations Research Unit 12 Sikkim Manipal University Page No. 232 Based on the ten tosses of the coins, the estimates of probabilities of occurrence of different numbers of heads are as depicted in figure 12.2: Fig. 12.2: Probabilities of Occurrence of Different Numbers of Heads Continue experimenting further, as these estimates come closer to the theoretical value with increasing sample size. The results for obtaining 2 heads and 3 tails for 100 throws are depicted in table 12.4: Table 12.4: Results for 100 Throws 10 throws 0 20 throws 6 30 throws 11 40 throws 14 50 throws 18 60 throws 19 70 throws 21 80 throws 22 90 throws 24 100 throws 27 Table 12.5 compares the final results at the end of 100 throws with the theoretical probabilities. Operations Research Unit 12 Sikkim Manipal University Page No. 233 Table 12.5: Final Results at the End of 100 Throws No. of heads Estimated Probability Theoretical Probability 0 1 2 3 4 5 0.03 0.21 0.27 0.33 0.12 0.04 0.03 0.16 0.31 0.31 0.16 0.03 It is observed that the results obtained with the large sample of 100, when compared are more in favour with the theoretical values than with a sample of ten sets of numbers. Self Assessment Questions 4. Simulation may be called experimentation in the _________ _______. 5. Random numbers have the property that any number is ____ to occur. 6. The totality of probability assigned to the variable should always be equal to _______. 12.5 Simulation Procedure In the previous section, you learnt the basic concepts. You will now learn the simulation procedure. The approach to solve a gambling problem can be extended to decision making in business where risk is a common feature. The probabilities associated with the variables can be estimated on the basis of availability of previous data or by entering subjective values. In any simulation problem, the variables to be studied will be given with associated probabilities. The initial conditions will also be specified. You can choose random numbers from table. However, to get uniform results, the random numbers will be specified. The first step involves coding the data that is, you assign random numbers to the variable. Then you identify the relationship between the variables and run the simulation to get the results Let us illustrate this by a simple example of a queuing process. Operations Research Unit 12 Sikkim Manipal University Page No. 234 Solved problem 1: A sample of 100 arrivals of customers at a retail sales depot is according to the following distribution: Table 12.6: Samples of 100 Arrivals of Customers Time between Arrivals (min) Frequency 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 2 6 10 25 20 14 10 7 4 2 Study of the time required to service customers by adding up the bills, receiving payment, making change and placing packages in hand trucks, yields the distribution depicted in table 12.7: Table 12.7: Study of the Time Required to Service Customers Service time (min) Frequency .5 1.0 1.5 2.0 2.5 3.0 12 21 36 19 7 5 Estimate the average percentage customer waiting time and average percentage idle time of the server by simulation for the next 10 arrivals. Solution: Step 1: Convert the frequency distributions of time between arrivals and service time to cumulative probability distributions. Step 2: Allocate random numbers 00 to 99 for each of the values of time between arrivals and service time. The range allocated to each value corresponds to the value of cumulative probability. Operations Research Unit 12 Sikkim Manipal University Page No. 235 Step 3: Using random numbers from table, sample the random time of arrival and service time for ten sets of random numbers. Step 4: Tabulate waiting time of arrivals and idle time of servers. Step 5: Estimate the percentage waiting time of arrivals and percentage idle time of servers corresponding to the ten samples. 12.5.1 Allocation of random numbers The tables 12.8, 12.9 and 12.10 depict the allocation of random numbers based on time between arrivals and service time. Table 12.8: Allocation of Random Numbers Time Between Arrivals Time Between Arrivals (1) Frequency (2) Cumulative Frequency (3) Cumulative Probability (3)/100 (4) Random Number Allocated (5) 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 2 6 10 25 20 14 10 7 4 2 2 8 18 43 63 77 87 94 98 100 0.02 0.08 0.18 0.43 0.63 0.77 0.87 0.94 0.98 1.00 00 to 01 02 to 07 08 to 17 18 to 42 43 to 62 63 to 76 77 to 86 87 to 93 94 to 97 98 and 99 Table 12.9: Allocation of Random Numbers Service Time Service Time (1) Frequency (2) Cumulative Frequency (3) Cumulative Probability (4) Random Nos. Allocated (5) 0.5 1.0 1.5 2.0 2.5 3.0 12 21 36 19 7 5 12 33 69 88 95 100 0.12 0.33 0.69 0.88 0.95 1.00 00 to 11 12 to 32 33 to 68 69 to 87 88 to 94 95 to 99 Operations Research Unit 12 Sikkim Manipal University Page No. 236 Note that the upper bound of random numbers allocated for each value of the parameter is one less than the corresponding cumulative frequency, since you have chosen a range of random numbers from 00 to 99 Table 12.10: Arrival and Service Time Arrivals Service Arrival No. Random No. Time Between Arrivals Time of Arrival Random No. Service Time Time of Start Time of Finish Waiting Time of Arrival Idle Time of server 1 2 3 4 5 6 7 8 9 10 78 78 06 04 97 71 78 61 05 95 3.5 3.5 1.0 1.0 4.5 3.0 3.5 2.5 1.0 4.5 3.5 7.0 8.0 9.0 13.5 16.5 20.0 22.5 23.5 28.0 54 24 51 45 46 84 58 58 60 24 1.5 1.0 1.5 1.5 1.5 2.0 1.5 1.5 1.5 1.0 3.5 7.0 8.0 9.5 13.5 16.5 20.0 22.5 24.0 28.0 5.0 8.0 9.5 11.0 15.0 18.5 21.5 24.0 25.5 29.0 - - - 0.5 - - - - 0.5 - 3.5 2.0 - - 2.5 1.5 1.5 1.0 - 2.5 Total 1.0 14.5 The service facility is made available at clock time zero and the server has to be idle for 3.5 minutes, when the service for first arrival starts. The service is completed at 5.0 minutes and again the server is idle for 2 minutes till the second arrival joins the system. The first three arrivals get immediate service and they dont have to wait, as the server is idle when they arrive. The fourth arrival that joins at 9.0 minutes has to wait for 0.5 minute, while the service to the third is completed. Similarly the waiting time and idle time can be computed for further arrivals. Total elapsed time = 29 minutes Waiting time of arrival = 1 minute Percentage of waiting time = ( 1 x 100) / 29 = 3.4 Idle time for server = 14.5 minutes Percentage of idle time = ( 14.5 x 100) / 29 =50% Operations Research Unit 12 Sikkim Manipal University Page No. 237 12.5.2 Use of random number tables The random numbers can be selected by any random process, similar to drawing numbered chips from a hat. However, it is convenient to use a table of random numbers prepared on the basis of some physical phenomenon. The grouping of random numbers in the tables has no significance and one should be concerned with individual digits only. The first random number could be picked at random from any point in the tables and the subsequent ones are to be selected preceding sequentially either in vertical or horizontal direction. Depending upon the number of digits required, the random numbers will be chosen in sets of single digit or two digit numbers, that is pseudorandom numbers. True random numbers cannot be produced by an algorithm and hence random numbers generated using recursive equations are referred to as pseudorandom numbers. There are several methods of generating pseudorandom numbers, but you will briefly learn only the mid square method. Operation starts with an arbitrary four digit integer called the seed. To obtain the first random number, the seed is squared and all digits except the middle four are ignored. This process is repeated each time using the previous random number as the new seed. Seed Uo = 8695 Uo 2 = 75603025 Taking the middle 4 digits, U1 = 6030 U1 2 = 36360900 U2 = 3609 Repeating the above procedure: U3=0248, U4=0615, U5= 3182, U6=3035 U7= 2112, U8 = 4805, U9 =2063 One of the disadvantages of the mid square method is that the generated numbers start cycling after a short set of random numbers is obtained. There are methods by which the seed can be chosen to obtain a fairly long sequence of numbers before cycling starts. Statistical tests check whether the generated sequence is truly random. Operations Research Unit 12 Sikkim Manipal University Page No. 238 Self Assessment Questions 7. In any simulation problem, initial conditions are stated. (True/False) 8. Random numbers are assigned for cumulative probability values. (True/False) 9. Without identifying any relationship between variables, you can solve the simulation problem. (True/False) 12.6 Sample Size In the previous section, you learnt the simulation procedure. You will now learn the sample size. As you have seen with the coin-tossing experiment, the larger the number of trials, the more confident you can be in your estimates. The question that arises is how many trials for simulation. If the experiment is as simple as tossing a coin involving only one variable, you can work out the sample size required for a given confidence level. Usually, a simulation model involves several variables making it impossible to determine the number of trials required to obtain the desired accuracy at a specified confidence level. One can only say that the accuracy associated with simulation improves as the square root of the number of trials, which results in large number of trials. This requires a great deal of computational effort and the use of computer becomes inevitable. In fact, special simulation languages such as GPSS and SIMSCRIPT have been developed to save time and effort required to structure and debug simulation models. A practical indicator of when to stop simulation trials is when the results that violently fluctuate initially tend to stabilise if the simulation is continued. If the successive cumulative results tally reasonable well, the simulation may be stopped. The degree of accuracy required varies with the problem on hand and calls for analyst judgment. Self Assessment Questions 10. A simulation model involves several variables making it impossible to determine the number of trials required to obtain the desired accuracy at a specified confidence level. (True/False) Operations Research Unit 12 Sikkim Manipal University Page No. 239 11. Large number of trails require a great deal of computational effort and the use of computer becomes optional. (True/False) 12. A practical indicator of when to stop simulation trials is when the results that violently fluctuate initially tend to stabilise if the simulation is continued. (True/False) Solved problem 2: Every morning a bread vendor buys loaves of bread at Re. 0.45 each by placing his order one day in advance (at the time of receiving his previous order) and sells them at Re. 0.70 each. Unsold bread can be sold the next day at Re. 0.20 per loaf and thereafter should be treated as of no value. The pattern of demand for bread is depicted in table 12.11: Table 12.11: Pattern of Demand for Bread Fresh Bread One Day old Bread Daily Sales Probability of demand Daily Sales Probability of demand 50 0.01 0 0.10 51 0.03 1 0.20 52 0.04 2 0.08 53 0.07 3 0.02 54 0.09 55 0.11 56 0.15 57 0.21 58 0.18 59 0.09 60 0.02 The vendor adopts the following order rule, if there is no stock with him at end of the previous day. He orders 60 units. Otherwise he orders 50 or 55 whichever is nearest the actual fresh bread sale on the previous day. Starting with zero stock and a pending order for 55 loaves simulate for 10 days and calculate the vendors profits. Operations Research Unit 12 Sikkim Manipal University Page No. 240 Solution: Table 12.12 depicts the allocation of random numbers. Table 12.12: Allocation of Random Numbers Fresh Bread One day old bread Daily Sale Prob of dema nd Cum Prob of demand Rand No. allocated Dail y Sale Prob of deman d Cum Prob of demand Rand No. Alloca -ted 50 51 52 53 54 55 56 57 58 59 60 0.01 0.03 0.04 0.07 0.09 0.11 0.15 0.21 0.18 0.09 0.02 0.01 0.04 0.08 0.15 0.24 0.35 0.50 0.71 0.89 0.98 1.00 00 01 03 04 07 08 14 15 23 24 34 35 49 50 70 71 88 89 97 98 - 99 0 1 2 3 0.70 0.20 0.08 0.02 0.70 0.90 0.98 1.00 00 to 69 70 to 89 90 to 97 98 to 99 You can now construct a table through simulation to see how the stocks and sales fluctuate. Table 12.13 depicts the results of simulation. Operations Research Unit 12 Sikkim Manipal University Page No. 241 Table 12.13: Results of Simulation FRESH BREAD ONE DAY BREAD Day Receipt of the start day Random No Sale Closing stock Order for next day Opening stock Random No. Sale 1 2 3 4 5 6 7 8 9 10 55 60 60 50 55 60 60 55 55 60 72 06 12 14 79 70 85 71 21 98 58 52 53 50 58 55 58 57 58 55 58 54 60 0 8 7 0 0 3 2 0 1 0 60 60 50 55 60 60 55 55 60 55 0 0 8 7 0 0 3 2 0 1 86 54 88 58 48 0 0 1 0 0 0 1 0 0 0 Total 570 549 21 2 Table 12.13 represents lost sales, as stock is limited as previous days closing stock is zero Estimated profit = (549 0.70 +2 20) -570 0.45 = Rs. 128.20 Solved problem 3: The maintenance manager of a chemical company is interested in determining a rational policy for maintenance of pneumatic conveying equipment. The equipment is a part of the process line and affects production. It has one bearing each on the inlet side (A) and the outlet side (B). Whenever there is a failure of any bearing, it has to be replaced immediately. The company has a good system of maintaining records on performance of the equipment and the failure records are available. Given the cost of the bearings, the cost of downtime of equipment and the time taken to replace either one bearing or both the bearings, you can use a simulation model to find out the best alternative solution for a given set of constraints. For the above scenario, the following failure data is available: Operations Research Unit 12 Sikkim Manipal University Page No. 242 Table 12.14 depicts the equipment performance data. Table 12.14: Equipment Performance Data The cost of bearing is Rs. 300 each for A and Rs. 500 each for B. The cost of downtime of equipment is Rs. 700 per hour and it takes 2 hours to replace one bearing either at inlet or outlet side and 3 hours to replace both the bearings. The three maintenance policies to be evaluated are: 1. Replace a bearing only when it fails. 2. Replace both the bearings if one fails. 3. Replace the bearing which fails plus the other one if it has been in use for more than its estimated average services life 600 hours for bearing A and 860 hours for bearing B Find the best alternative through simulation. Solution: It is assumed that the failure of a bearing is independent of the maintenance policy followed. Random numbers are allocated for different failure times: (Table 12.15 depicts the random numbers) No. of failures Age at failure (hours) Inlet side Bearing (A) Outlet side Bearing (B) 150 300 450 600 750 900 1,050 1,200 1,350 7 16 18 23 14 10 7 5 0 0 3 9 12 16 18 33 7 2 Operations Research Unit 12 Sikkim Manipal University Page No. 243 Table 12.15: Random Numbers Bearing A Bearing B Life (hrs) Cumulative Probability Random Nos. Cumulative Probability Random Nos. 150 300 450 600 750 900 1050 1200 1350 0.07 0.23 0.41 0.64 0.78 0.88 0.95 1.00 - 00 to 96 07 to 22 23 to 40 41 to 63 64 to 77 78 to87 88 to 94 95 to 99 - 0.00 0.03 0.12 0.24 0.40 058 0.91 0.98 1.00 - 00 to 02 03 to 11 12 to 23 24 to39 40 to57 58 to 90 91 to 97 98 and99 We can now select random numbers from the tables and generate a set of 12 bearings for each of the bearings. Table 12.16 depicts the set of 12 bearing for each of the bearings generated. Table 12.16: Set of 12 Bearings for Each of the Bearings Generated 1 2 3 4 5 6 7 8 9 10 11 12 10 22 24 42 37 77 99 96 89 85 28 63 300 300 450 600 450 750 1200 1200 1050 900 450 600 300 600 1050 1650 2100 2850 4050 5250 6300 7200 7650 8250 99 96 18 36 50 79 80 96 31 07 62 77 1350 1200 600 750 900 1050 1050 1200 750 450 1050 1050 1350 2550 3150 3900 4800 5850 6900 8100 8850 9300 10,350 11,400 Let us compare the costs of three policies for the first 7200 hours. Policy 1: Replace a bearing only when it fails. A requires replacement 10 times, and B 7 times during this period as seen from the lives of successive bearings. Operations Research Unit 12 Sikkim Manipal University Page No. 244 Total cost = (300 10 + 500 7) + (17 2 700) = Rs. 30, 300 Policy 2: Replace both the bearing if one fails. Table 12.17 depicts bearing failure information. Table 12.17: Bearing Failure Elapsed time (hrs) Bearing fails first Life (hrs) of newly fitted bearing A (From table 12.16) Life (hrs) of newly fitted bearing B (From table 12.16) 300 A 300 1200 600 A 450 600 1050 A 600 750 1650 A 450 900 2100 A 750 1050 2850 A 1200 1050 3900 B 1200 1200 5100 A & B 1050 750 5850 B 900 450 6300 B 450 1050 6750 A 600 1050 7350 A - - Policy 3 is the cheapest. The simulation is limited to 7200 hours of operation since the purpose is only to illustrate the method. With such small number of trials the results may turn out to be erratic. Consider policy 1 again. With a different set of random numbers and extended simulation, there may be occasions of both bearing failing at the same time, thus affecting the cost of downtime for replacement. Simulation with sufficiently large number of trials can only lead to dependable decisions. Operations Research Unit 12 Sikkim Manipal University Page No. 245 Table 12.18 depicts the bearing record. Table 12.18: Bearing Record Start of the period Life bearing in service (hrs) Be aring wh ich fail s Time elapsed (hrs) Age surviving bearing Did the surviving bearing (A/B) Complete its est. avg life? If No in col 7 balance life of surviving Replace ment policy A B (1) (2) (3) (4) (5) (6) (7) (8) (9) 0 300 1350 A 300 300 B - No 1050 Replace A 300 300 1050 A 600 600 B - No 750 Replace A 600 450 750 A 1050 1050 B - Yes -- Replace A & B 1050 600 1200 A 1650 600 B - No 600 Replace A 1650 450 600 A 2100 1050 B - Yes -- Replace A & B 2100 750 600 B 2700 600 B - Yes -- Replace A & B 2700 1200 750 B 3450 750 B - Yes -- Replace A & B 3450 1200 900 B 4350 900 B - Yes -- Replace A & B 4350 1050 1050 A& B 5400 -- -- Replace A & B 5400 900 1050 A 6300 900 B - Yes -- Replace A & B 6300 450 1200 A 6750 450 B - No 750 Replace A & B 6750 600 750 A 7350 1050 B - Yes -- Replace A & B This will be the life (from Table 12.18) if newly fitted. For survivors from previous replacement, this will be the balance life (col. 8). Operations Research Unit 12 Sikkim Manipal University Page No. 246 Solved problem 4: A factory produces 150 scooters. But the production rate varies with the distribution depicted in table 12.19. Table 12.19: Production Rate and Probability Production Rate 147 148 149 150 151 152 153 Probability 0.05 0.10 0.15 0.20 0.30 0.15 0.05 At present the track will hold 150 scooters. Using the following random numbers determine the average number of scooters waiting for shipment in the factory and average number of empty space in the truck. Random Numbers 82, 54, 50, 96, 85, 34, 30, 02, 64, 47. Solution: Table 12.20 depicts the production rate and probability. Table 12.20: Production Rate and Probability Production rate Probability Cumulative Probability Random No. Assigned 147 0.05 0.05 00 04 148 0.10 0.15 05 - 14 149 0.15 0.30 15 - 29 150 0.20 0.50 30 - 49 151 0.30 0.80 50 - 79 152 0.15 0.95 80 - 94 153 0.05 1.00 95 - 99 Table 12.21 depicts the simulation worksheet. Table 12.21: Simulation Worksheet Trial No. Random No. Simulated Production Rate Scooter Waiting in the factory Number of example spaces in the truck 1 82 152 2 - 2 54 150 2 - 3 50 150 2 - 4 96 153 5 - 5 85 152 7 - 6 34 150 7 - Operations Research Unit 12 Sikkim Manipal University Page No. 247 7 30 150 7 - 8 02 147 4 3 9 64 151 5 - 10 47 150 7 - Total 3 Therefore, average number of scooters waiting = 7/10 =0.7/day Average number of empty space = 3/10 = 0.3/day Solved problem 5: Dr. Strung is a dentist. He gives appointments to patients every half an hour. However, he does not know the nature of illness of patients arriving at his clinic. From past records, he has the following probability distribution and also knows the exact treatment timings. He starts his clinic at 8.00 A.M. Using the following information determine the average waiting time of the customers and idle time of the doctor. Random Numbers 56, 40, 26, 66, 87, 48, 17, 22, 04, 15. Table 12.22 depicts the probability distribution in past record. Table 12.22: Probability Distribution in Past Record Nature of illness Probability Time taken for treatment (min) Filling 0.10 50 Check up 0.30 15 Crowning 0.15 40 Cleaning 0.30 15 Extraction 0.15 30 Solution: Table 12.23 depicts the assigning of random number. Table 12.23: Assigning Random Number Illness Probability Cum. Prob. Random No. Assigned Filling 0.10 0.10 00 -09 Check up 0.30 0.40 10 -39 Crowning 0.15 0.55 40 -54 Cleaning 0.30 0.85 55 -84 Extraction 0.15 1.00 85 -99 Operations Research Unit 12 Sikkim Manipal University Page No. 248 Table 12.24 depicts the simulation worksheet. Table 12.24: Simulation Worksheet Trial No Random No. Nature of illness Time Taken Patients Arrival Time Treatment Doctor Idle time Patients Waiting Time Starts Finishes 1 56 Cleaning 15 8.00 8.00 8.15 - 2 40 Crowning 40 8.30 8.30 9.10 15 - 3 26 Check up 15 9.00 9.10 9.25 - 10 4 66 Cleaning 15 9.30 9.30 9.45 5 - 5 87 Extraction 30 10.00 10.00 10.30 15 - 6 48 Crowing 40 10.30 10.30 11.10 - - 7 17 Check up 15 11.00 11.10 11.25 - 10 8 22 Check up 15 11.30 11.30 11.45 5 - 9 04 Filling 50 12.00 12.00 12.50 15 - 10 15 Check up 15 12.30 12.50 13.05 - 20 Total 45 40 Doctor idle time = 45 minutes Patients average waiting time = 40/10 = 4 minutes 12.7 Summary Let us recapitulate the important concepts discussed in this unit: - Using simulation, an analyst can introduce the constants and variables related to the problem, set up the possible courses of action and establish criteria which act as measures of effectiveness. - The methodology developed for simulation process consists of seven steps. - In any simulation problem, the variables to be studied will be given with associated probabilities. - The mid square method is used for generating pseudorandom numbers. One of the disadvantages of the mid square method is that the Operations Research Unit 12 Sikkim Manipal University Page No. 249 generated numbers start cycling after a short set of random numbers is obtained. - The accuracy associated with simulation improves as the square root of the number of trials, which results in large number of trials. 12.8 Glossary Sample size: Number of trials in a simulation GPSS (General Purpose Simulation System): a discrete event simulation programming system developed in 1950s for application in queuing, manufacturing, transportation and defense systems SIMSCRIPT: a discrete event simulation programming system developed in 1962 by Harry Markowitz .It can be programmed with simple English like statements to generate the simulation model. 12.9 Terminal Questions 1. Explain the different steps involved in simulation methodologies? 2. Briefly write down the basic concepts concerned with simulation. 3. Explain the simulation procedure? 4. Briefly explain the use of random number tables. 5. Two components have to be produced in M/c A and M/c B and then finally assembled. The time taken to assemble on two machines varies with the probability distribution depicted in table 12.25. Using simulation technique and the ordered pair of random numbers, first for M/c A and second for M/c B, find the average time taken for production. Table 12.25: Assembling Time for the Two Machines M/c - A M/c - B Production Time in min Probability Production Time in min Probability 22 0.15 30 0.05 23 0.20 31 0.15 24 0.30 32 0.25 25 0.20 34 0.25 26 0.15 35 0.20 36 0.10 Operations Research Unit 12 Sikkim Manipal University Page No. 250 Random Numbers (10, 92); (25, 83); (36, 76); (44, 15); (57, 25); (62, 67); (04, 99); (72, 53); (81, 35); (94, 07) 12.10 Answers Self Assessment Questions 1. True 2. False 3. False 4. Management laboratory 5. Equally likely 6. 1 7. True 8. True 9. False 10. True 11. False 12. True Terminal Questions 1. The methodology developed for simulation process consists of seven steps. For more details refer to section 12.3 2. Concepts of simulation: simulation of complex problems, sampling allotting random numbers to events and theoretical probability. For more details refer to section 12.4 3. In any simulation problem, the variables to be studied will be given with associated probabilities. The initial conditions will also be specified. You can choose random numbers from table. For more details refer to section 12.5 4. To generate a list of number and allot it to events. For more details refer to section 12.5.2 5. 57.7 minutes. For more details refer to section 12.6 Operations Research Unit 12 Sikkim Manipal University Page No. 251 12.11 Case Study Coin-Flipping Game You are the lucky winner of a sweepstakes contest. Your prize is an allexpense- paid vacation at a major hotel in Las Vegas, including some chips for gambling in the hotel casino. Upon entering the casino, you find that, in addition to the usual games (blackjack, roulette, etc.), they are offering an interesting new game with the following rules: a) Each play of the game involves repeatedly flipping an unbiased coin until the difference between the heads tossed and the number of tails is three. b) If you decide to play the game, you are required to pay US $1 for each flip of the coin. You are not allowed to quit during a play of the game. c) You receive US $8 at the end of each play of the game. Thus, you win money if the number of flips required is fewer than 8, but you lose money if more than 8 flips are required. Here are some examples (H denotes head and T denotes Tail) HHH 3 flips You can win US $5 THTTT 5 flips You can win US $3 THHTHTHTTTT 11 flips You lose US $3 How would you decide whether to play this game? This can be simulated because you are imitating the actual play of the game without actually winning or losing any money. Discussion Question 1. Generate a sequence of random observations and give your suggestion. References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. Operations Research Unit 13 Sikkim Manipal University Page No. 252 Unit 13 Simulation Monte-Carlo Method Structure: 13.1 Introduction Objectives 13.2 Monte-Carlo Simulation Problems on Monte-Carlo method 13.3 Applications of Simulation Simulation in networks Simulation in job sequencing 13.4 Advantages of Simulation 13.5 Limitations of Simulation 13.6 Summary 13.7 Glossary 13.8 Terminal Questions 13.9 Answers 13.10 Case Study 13.1 Introduction In the previous unit, you learnt the basic concepts and methodology of simulation. You also learnt simulation procedure and sample size. Monte- Carlo technique has become such an important part of simulation models that the terms are often assumed to be synonymous. However, it is only a special technique of simulation. The technique of Monte-Carlo involves the selection of random observations within the simulation model. The technique is restricted for application involving random numbers to solve deterministic and stochastic problems. The principle of this technique is replacement of actual statistical universe by another universe described by some assumed probability distribution and then sampling from this theoretical population by means of random numbers. In fact, this process involves the generation of simulated statistics (random variables) that can be explained in simple terms as choosing a random number and substituting this value in it. In this unit, you will learn the Monte-Carlo simulation method and the application of simulation methods. You will also learn the advantages and limitations of simulation. Operations Research Unit 13 Sikkim Manipal University Page No. 253 Objectives: After studying this unit, you should be able to: - describe Monte-Carlo simulation method - analyse the applications of simulation - list the advantages and limitations of simulation 13.2 Monte-Carlo Simulation The Monte-Carlo method is a simulation technique in which statistical distribution functions are created by using a series of random numbers. This approach has the ability to develop many months or years of data in a matter of few minutes on a digital computer. The method is generally used to solve the problems that cannot be adequately represented by mathematical models or where solution of the model is not possible by analytical method. The Monte-Carlo simulation procedure can be summarised in the steps depicted in figure 13.1: Fig. 13.1: Monte-Carlo Simulation Procedure Operations Research Unit 13 Sikkim Manipal University Page No. 254 Let us now describe each step in detail. Step 1: Define the problem: a) Identify the objectives of the problem. b) Identify the main factors that have the greatest effect on the objectives of the problem. Step 2: Construct an appropriate model: a) Specify the variables and parameters of the model. b) Formulate the appropriate decision rules, i.e., state the conditions under which the experiment is to be performed. c) Identity the type of distribution that will be used. Models use either theoretical distributions or empirical distributions to state the patterns of occurrence associated with the variables. d) Specify the manner in which time will change. e) Define the relationship between the variables and parameters. Step 3: Prepare the model for experimentation: a) Define the starting conditions for the simulation. b) Specify the number of runs of simulation to be made. Step 4: Using steps 1 to 3, experiment with the model: a) Define a coding system that will correlate the factors defined in step 1 with the random numbers to be generated for the simulation. b) Select a random number generator and create the random numbers to be used in the simulation. c) Associate the generated random numbers with the factors identified in step1 and coded in step 4(a). Step 5: Summarise and examine the results obtained in step 4. Step 6: Evaluate the results of the simulation. Step 7: Formulate proposals for advice to management on the course of action to be adopted and modify the model, if necessary. Operations Research Unit 13 Sikkim Manipal University Page No. 255 13.2.1 Problems on Monte-Carlo method Solved problem 1: The occurrence of rain in a city on a day is dependent upon whether or not it rained on the previous day. If it rained on the previous day, the rain distribution is as depicted in table 13.1: Event: Table 13.1: Distribution Table Event No rain 1 cm Rain 2 cm Rain 3 cm Rain 4 cm Rain 5 cm Rain Probability 0.50 0.25 0.15 0.05 0.03 0.02 If it did not rain on the previous day, the rain distribution is as depicted in table 13.2: Table 13.2: Rain Distribution Table Event No rain 1 cm Rain 2 cm Rain 3 cm Rain Probability 0.75 0.15 0.06 0.04 Simulate the citys weather for 10 days and determine the total days without rain as well as the total rainfall during the periodby simulation. Use the following random numbers: 67 63 39 55 29 78 70 06 78 76 For simulation assume that for the first day of the simulation it had not rained the day before. Solution: We simulate the citys weather with and without rainfall in the following steps: Step 1: Previous day rain distribution is as depicted in table 13.3: Table 13.3: Rain Distribution Table Event Probability Cumulative probability RN range No rain 0.50 0 . 50 0 0 - 4 9 1 cm Rain 0.25 0 . 7 5 5 0 - 7 4 2 cm rain 0.15 0 . 9 0 7 5 - 8 9 3 cm Rain 0 . 0 5 0.95 90-94 4 cm Rain 0.03 0 . 9 8 9 5 - 9 7 5 cm Rain 0.02 1 . 0 0 9 8 - 9 9 Operations Research Unit 13 Sikkim Manipal University Page No. 256 Step 2: Previous day no rain distribution is as depicted in table 13.4: Table 13.4: Distribution Table Event Probability Cumulative probability R N r a n ge No rain 0 . 7 5 0 .75 0 0 - 7 4 1 cm Rain 0 . 1 5 0 . 15 7 5 - 8 9 2 cm rain 0 . 0 6 0.96 90-95 3 cm Rain 0 . 0 4 1.00 96-99 Step 3: Simulation for 10 days using the given random numbers is as depicted in table 13.5: Table 13.5: Random Table Day RN Events Cumulative rain 1 67 No rain 2 63 No rain 3 39 No rain 4 55 No rain 5 29 No rain 6 78 1 cm Rain 1 c m 7 70 1 cm Rain 2 c m 8 06 No rain 2 c m 9 78 1 cm Rain 3 c m 10 76 2 cm Rain 5 c m During the simulated period it did not rain on 6 out of 10 days. The total rainfall during the period is 5 cm. Solved problem 2: The number of customers at a restaurant each evening is distributed as depicted in table 13.6: Table 13.6: Distribution of Customers Number of customers Lots Average Very few Probability 0.2 0.4 0.4 The chef refuses to work on an evening when there are very few customers and walks out. He will not return to work until an evening when there are lots Operations Research Unit 13 Sikkim Manipal University Page No. 257 of customers although he always comes on Friday because he gets paid then. We are interested in fraction of evenings that the chef is at the restaurant. Solution: In this example, the trial is whether the number of customers is lots, average or very few and it is repeated for each evening when the restaurant is open. Also, whether the chef is in or out depends not only on the outcome of this trial but also on his whereabouts the previous evening and whether it is a Friday or not. Here we wish to simulate the level of attendance at the restaurant on each evening. To determine whether he is in or out on that particular evening, we use the rules governing the chefs behaviour as stated in the question. Here the sequence of ins and outs forms a simulated history of the system which helps us in determining the fraction of evenings for which the chef is in. Now we can proceed as follows: Step1: To simulate the number of customers each evening Using the random digits and the given (prescribed) probability distribution, we simulate the number of customers each evening as depicted in table 13.7: Table 13.7: Simulation of Customers Number of customers Lots Average Very few Probability 0.2 0.4 0.4 Digits 0.1 2,3,4,5 6,7,8,9 Step 2: To generate attendance level The attendance on a particular evening can be simulated by selecting a digit, from 0 to 9 inclusive, at random. Using table 13.7, we can interpret the digits as state of the restaurant. On account of the number of digits assigned to each one, the outcomes in the simulation will have the same probability of occurrence as those in the real system. Step 3: To determine whether the chef works or not After generating an attendance level, we can determine whether the chef works or not by the rules provided in the question. Table 13.8 depicts a specimen run of the simulation. Operations Research Unit 13 Sikkim Manipal University Page No. 258 The following specimen run was started with Monday and with the chef being in. Keeping the view in mind that we are trying to reproduce the long term behaviour of the restaurant, the initial conditions of the simulation will not affect the results we obtain. Table 13.8: Distribution Table Day Day of the week Random Digit Number of customers Chef in or out Number of days in to date Fraction of days in to date 1 2 3 4 5 6 7 8 9 10 Monday Tuesday Wednesday Thursday Friday Saturday Sunday Monday Tuesday Wednesday 3 5 1 8 6 6 4 4 0 Average Average Lots Very few Very few Very few Average Average Lots In In In Out In Out Out Out In 1 2 3 3 4 4 4 4 5 1.00 1.00 1.00 0.75 0.80 0.67 0.57 0.50 0.56 Step 4: When to stop the simulation run As far as determination of any average effects is concerned, the more time periods simulated the better. However, in actual practice we require a criterion for deciding when the simulation should stop. In the specimen run we could continue until the fraction at the end of each line does not change very much. At this stage there is no point in going further and we have the desired answer. In this example, it approaches 0.56 and the chef is in 56% of all days. Self Assessment Questions 1. Which of the following is not correct? a) Randomness is a key requirement of Monte-Carlo simulation. b) Monte-Carlo simulation involves modelling a deterministic system. c) Simulation is a statistical experiment as such its results are subject to statistical error. d) In Monte-Carlo simulation, a problem is solved by simulating the original data with random number generators. Operations Research Unit 13 Sikkim Manipal University Page No. 259 2. The technique of Monte-Carlo involves the selection of ____________ observations within the ____________ model. 13.3 Application of Simulation The range of application of simulation in business is extremely wide. Unlike other mathematical models, simulation can be easily understood by the users and thereby facilitates their active involvement. This makes the results more reliable and also ensures easy acceptance for implementation. The degree to which a simulation model can be made close to reality is dependent upon the ingenuity of the OR team who identifies the relevant variables as well as their behaviour. You have already seen, by means of an example, how simulation could be used in a queuing system. It can also be employed for a wide variety of problems encountered in production systemsthe policy for optimal maintenance in terms of frequency of replacement of spares or preventive maintenance, number of maintenance crews, number of equipment for handling materials, job shop scheduling, routing problems, stock control, etc. The other areas of application include dock facilities, facilities at airports to minimise congestion, hospital appointment systems and even management games. In case of other OR models, simulation helps the manager to strike a balance between opposing costs of providing facilities (usually meaning long term commitment of funds) and the opportunity costs of not providing them. Let us find some of the examples how simulation is used in different situations. 13.3.1 Simulation in networks A number of network simulation models have also been developed, e.g., simulation of probabilistic activity times in PERT network. The critical path and the project duration can be found out with a randomly selected activity times for each activity. The probability distribution of project completion time and the probability that each of given activity is on the critical path can be obtained by repeating the process a number of times. Operations Research Unit 13 Sikkim Manipal University Page No. 260 Solved problem 3: A project schedule consists of five activities and the duration of these activities is non-deterministic with the probability distribution as depicted in table 13.9: Table 13.9: Probability Distribution Table Activity Days Probability 1-2 1 4 8 0.2 0.5 0.3 1-3 2 4 7 0.3 0.5 0.2 2-4 2 4 6 0.3 0.3 0.4 3-4 3 6 8 0.3 0.4 0.3 4-5 2 3 4 0.2 0.2 0.6 Simulate the duration of the project 10 times and estimate the chances of various paths to be critical. What is the average duration of the project? Solution: From the PERT network, we observe that there are two paths from the start to end of the project, namely 1245 and 1345. Using one-digit random number corresponding to the probability distribution of various activities, we obtain the simulated duration of each path and hence the simulated duration of various activities and the simulated duration of the project. Table 13.10 depicts the results of the simulation: Operations Research Unit 13 Sikkim Manipal University Page No. 261 Table 13.10: Simulation Table Activity Simul ation 1-2 RN D 1-3 RN D 2.4 RN D 3-4 RN D 4-8 RN D Project Duration Critical Paths 1 8 8 9 2 8 6 3 6 6 4 18 1245 2 5 4 5 4 8 6 1 3 7 4 14 1245 3 5 4 6 4 7 6 4 6 7 4 14 1245 1345 4 7 8 5 4 1 2 9 8 5 4 16 1345 5 0 1 6 4 8 6 1 3 8 4 11 1245 6 8 8 3 4 0 2 4 6 3 3 13 125 7 4 4 7 4 4 4 4 0 3 3 11 145 8 5 4 8 7 2 2 6 6 6 4 17 145 9 2 4 5 4 9 6 0 3 3 3 13 145 10 6 4 1 2 3 4 7 8 0 2 12 145 RN: Random Number; D: Duration On the basis of the above simulation, we observe that there is a 70 percent chance of the path 145 being critical. Further, the average duration of the project is 13.9 days. Solved problem 4: A project consists of 7 activities. The time for performance of each of the activities is a random variable with the respective probability distribution as depicted in table 13.11: Table 13.11: Probability Distribution Table Activity Immediate predecessor Time (in days) and its probability A
3 0.20 4 0.60 5 0.20 B 4 0.10 5 0.30 6 0.30 7 0.20 8 0.10 C A 1 0.15 3 0.75 5 0.10 D B,C 4 0.80 5 0.20 E D 3 0.10 4 0.30 5 0.30 6 0.30 F D 5 0.20 7 0.80 G E,F 2 0.50 3 0.50 Operations Research Unit 13 Sikkim Manipal University Page No. 262 A) Draw the network diagram and identify the critical path using the expected activity times. B) Simulate the project using random numbers to determine the activity times. Find the critical path and the project duration. C) Repeat the simulation four more times. Is the same path critical in all the simulations? Solution: a) using the given information, the resulting network is depicted in figure 13.2. The time for each of the activities is the expected time obtained by the summation of the products of the activity time and the corresponding probability. For example, for activity A, The Expected Time = 3*0.20 + 4*0.60 + 5*0.20 = 4 days. Fig. 13.2: PERT Network Network design Observe that the critical path of the project is 123467, with an expected duration of 20.2 days. To use the simulation for obtaining the project length, we first determine the random number intervals for each activity. Random number coding is shown in the table. The five simulation runs, by taking random numbers, are as depicted in table 13.12. For each run, the critical path and the project durations are obtained and the project time is obtained as follows: Total time = Larger of times for activities A and C, and B + Time for activity D + Larger of times for activities E + Time for activity G Operations Research Unit 13 Sikkim Manipal University Page No. 263 Table 13.12: Random Number Coding Activity Time Probability Cumulative Probability Random number Interval A 3 4 5 0.20 0.60 0.20 0.20 0.80 1.00 00-19 20-79 80-99 B 4 5 6 7 8 0.10 0.30 0.30 0.20 0.10 0.10 0.40 0.70 0.90 1.00 00.09 10-39 40-69 70-89 90-99 C 1 3 5 0.15 0.75 0.10 0.15 0.90 1.00 00-14 15-89 90-99 D 4 5 3 0.80 0.20 0.10 0.80 1.00 0.10 00-79 80-99 00-09 E 4 5 6 0.30 0.30 0.30 0.10 0.70 1.00 10-39 40-69 70-99 F 5 7 0.20 0.80 0.20 1.00 00-19 20-99 G 2 3 0.50 0.50 0.50 1.00 00-49 50-99 Simulation worksheet Activity Table 13.13 depicts the simulation worksheet. Table 13.13: Simulation Worksheet Simulation A RN D B RN D C RN D D RN D E RN D F RN D G RN D 1 68 4 13 5 09 1 20 4 73 6 07 5 92 3 2 99 5 93 8 18 3 20 4 22 4 07 5 29 2 3 57 4 33 5 49 3 65 4 92 6 98 7 00 2 4 57 4 12 5 31 3 96 5 86 6 92 7 91 3 5 77 4 37 5 34 3 11 4 27 4 10 5 59 3 RN: random number; D: Duration Operations Research Unit 13 Sikkim Manipal University Page No. 264 We get the information as depicted in figure 13.3 from the above table: Fig. 13.3: Simulation of Customers It may be observed that the same path is not critical in each of the runs. 13.3.2 Simulation in job sequencing In this section, we illustrate the use of simulation in the sequencing problems. Example: A job has to be processed over two machines, machine M1 and machine M2 in that order. The distribution of inter-arrival time of the jobs at the first machine is as depicted in table 13.14: Table 13.14: Distribution of Arrival Time Time(minutes) 1 2 3 4 Probability 0.2 0.2 0.2 0.4 Operations Research Unit 13 Sikkim Manipal University Page No. 265 The processing times at the two machines are as follows: Table 13.15: Simulation Table Machine M1 Machine M2 Time (minutes) Probability Time(minutes) Probability 1 0.1 4 0.2 2 0.2 5 0.3 3 0.3 6 0.4 4 0.3 7 0.1 5 0.1 On the basis of 10 simulation runs, find out the average queue length before machine M1 and the average queue length before machine M2. Solution: The various steps for simulation are: 1. Simulate the inter-arrival time of the ith job on machine M1 with the help of a random number (i= 1, 2.). 2. Arrival time of ith job. = Arrival time of (i-1)th job (i=2,3, .) + Inter-arrival time of ith job. Arrival time of the first job= Inter-arrival time of the first job 3. Simulate the processing time on machine M1 by random numbers. 4. Departure time on machine M1 = Arrival time for processing on machine M2 = Max (Arrival time of ith job, process completion time of (i-1)thjob on machine M1) + processing time of ith job on machine M1 5. If arrival time of ith job < Process completion time of (i-1)th, ith arrival waits. If arrival time of ith job < Process completion time of (i-2)th, both ith and (i- 1)th arrivals wait. 6. Simulate the processing time of machine M2 with the help of random numbers. 7. Process completion by machine M2 of ith arrival = Max [Arrival time of ith job, process completion time of (i-1)th job] + Process time by machine M2. Operations Research Unit 13 Sikkim Manipal University Page No. 266 8. If arrival time of ith< process completion time of (i-1)th, the ith arrival waits. If arrival time of ith< process completion time of (i-2)th , both ith and (i-1)th arrivals wait, etc. Using the procedure, the simulated processing data of 10 arrivals are displayed in the simulation worksheet as depicted in figure 13.4: Fig. 13.4: Simulation of 10 Arrivals On the basis of the simulation study, the average queue length before machine M1 is 0.5 and the average length before machine M2 is 1.2. Self Assessment Questions 3. The ___________ and the project duration can be found out with a randomly selected activity times for each activity. 4. The range of application of simulation in _________ is extremely wide. 13.4 Advantages of Simulation Simulation is the process of experimenting on the model rather than on the operation which the model represents. Following are some of the advantages of simulation: Operations Research Unit 13 Sikkim Manipal University Page No. 267 - The study of very complicated systems or sub-systems can be done with the help of simulation. Simulation has been described as What to do when all else fails. - We can investigate the consequences for a system of possible changes in parameters in terms of the model. - The knowledge of a system obtained in designing and conducting the simulation is very valuable. - Simulation enables us to assess the possible risks involved in a new policy before actually implementing it. - Simulation of complicated systems helps us to locate which variables have the important influences on system performance. - Simulation methods are easier to apply than pure analytical methods. 13.5 Limitations of Simulation Notwithstanding with the above advantages, following are some of the limitations of simulation: - Simulation generates a way of evaluating solutions but it does not generate the solution techniques. - Sometimes, simulation models are expensive and take a long time to develop. For example, a corporate planning model may take a long time to develop and prove expensive also. - The simulation model does not produce answers by itself. The user has to provide all the constraints for the solutions which he/she wants to examine. - Not all situations can be evaluated using simulation. Only situations involving uncertainty are considered. - It is the trial-and-error approach that produces different solutions in repeated runs. This means it does not generate optimal solutions to problems. - Simulation is a time-consuming exercise. Self Assessment Questions 5. An advantage of simulation as opposed to optimisation is that: a) Several options of measure of performance can be examined. b) Complex reallife problems can be studied. Operations Research Unit 13 Sikkim Manipal University Page No. 268 c) It is applicable in cases where there is an element of randomness in a system. d) All the above. 6. Large complicated simulation models are appreciated because: a) Their average costs are not well-defined. b) It is difficult to create the appropriate events. c) They may be expensive to write and use as an experimental device. d) All of the above. 7. Simulation should not be applied in all the cases because it: a) Requires considerable talent for model building and extensive computer programming efforts. b) Consumes much computer time. c) Provides at best approximate solution to problem. d) All of the above. 8. Analytical results are taken into consideration before a simulation study so as to: a) Identify suitable values of the system parameters. b) Determine the optimal decision. c) Identify suitable values of decision variables for the specific choices of system parameters. d) All of the above. 13.6 Summary Let us recapitulate the important concepts discussed in this unit: - The Monte-Carlo method is a simulation technique in which statistical distribution functions are created by using a series of random numbers. - Simulation could be used many systems such as in a queuing, production, dock facilities, facilities at airports to minimise congestion, hospital appointment systems and even management games. - A number of network simulation models have also been developed, such as simulation of probabilistic activity times in PERT network. - Simulation enables us to assess the possible risks involved in a new policy before actually implementing it. Operations Research Unit 13 Sikkim Manipal University Page No. 269 - Not all situations can be evaluated using simulation. Only situations involving uncertainty are considered. 13.7 Glossary Job sequencing: the arrangement of the tasks required to be carried out. job shop: Organization of work in a manufacturing organization by job functions. 13.8 Terminal Questions 1. Explain the use of simulation in networks? 2. What are the advantages of using simulation? 3. What are the Limitations of using Simulation? 4. Write a short note on Monte-Carlo simulation? 5. For a project comprising activities A, B .. H, the following information is available: Precedence relationships: A and B are the first activities of the project. C succeeds A while B precedes D. Both C and D precede E and F. Activity G follows activity F while H is the last activity of the project and succeeds E and G. Time estimates and Probabilities: Simulate the project and determine, using random numbers, the activity times. Find the critical path and the project duration. Repeat five times. Does the critical path change? State the estimated duration of the project in each of the trials. Activity Time days 1 2 3 4 5 6 A 0.02 0.04 0.4 B 0.3 0.7 C 0.3 0.3 0.4 D 0.2 0.6 0.2 E 0.6 0.4 F 0.8 0.2 G 0.3 0.5 0.2 H 0.1 0.2 0.3 0.4 Operations Research Unit 13 Sikkim Manipal University Page No. 270 13.9 Answers Self Assessment Questions 1. (b) Monte-Carlo simulation involves modelling a deterministic system. 2. Random, Simulation. 3. Critical path 4. Business 5. (d) All of the above. 6. (c) They may be expensive to write and use as an experimental device. 7. (d) All of the above. 8. (c) Identify suitable values of decision variables for the specific choices of system parameters. Terminal Questions 1. To find the critical path and project duration. For more details refer 13.3.1 2. Advantages of simulation: to studying complicated systems, investigate the consequences for a system and assess the possible risks involved. For more details refer 13.4 3. Limitations of simulation: does not generate the solution techniques, simulation models are expensive and does not produce answers by itself. For more details refer 13.5 4. The Monte-Carlo method is a simulation technique in which statistical distribution functions are created by using a series of random numbers. This approach has the ability to develop many months or years of data in a matter of few minutes on a digital computer. For more details refer 13.2 5. Using random numbers from the first six columns for six trials: Trial Critical paths Duration 1 BDFGH 16 2 BDFGH 19 3 BDFGH 18 4 BDFGH 17 5 ACFGH BDFGH 16 6 ACFGH and BDFGH 18 Operations Research Unit 13 Sikkim Manipal University Page No. 271 13.10 Case Study Workload problems in planer department This was the first time that Carl Schilling had been summoned for a meeting with the management in the fancy executive offices upstairs and he hopes it will be the last time. Carl doesnt like the pressure. He has had enough pressure just dealing with all the problems he has been encountering as the foreman of the planer department on the factory floor. What a nightmare this last month has been! Fortunately, the meeting had gone better than Carl had feared. The management actually had been quite nice. They explained that they needed to get Carls advice on how to deal with a problem that was affecting the entire factory. The origin of the problem is that the planer department has had a difficult time keeping up with its workload. Frequently there are a number of work pieces waiting for a free planer. This waiting has seriously disrupted the production schedule for subsequent operations, thereby greatly increasing the cost of in-process inventory as well as the cost of idle equipment and resulting lost production. They understood that this problem was not Carls fault. However, they needed to get his ideas on what changes were needed in the planer department to relieve this bottleneck. Imagine this! The management, with graduate degrees from the fanciest business schools in the country, asking advice from a poor working slob like him who had barely made it through high school. He could hardly wait to tell his wife that night. The meeting had given Carl an opportunity to get two petpeeves off his chest. One peeve is that he has been telling his boss for months that he really needs another planer, but nothing ever gets done about this. His boss just keeps telling him that the planers he already has arent being used 100 percent of the time, so how can adding even more capacity be justified. Doesnt his boss understand about the big backlogs that build up during busy times? Then there is the other peeve all those peaks and valleys of work coming to his department. At times, the work just pours in, and a big backlog builds up. Then there might be a long pause when not much comes in, so the planers stand idle, part of the time. If only those departments that are feeding castings to his department could get their act together and even out the work flow, many of his backlog problems would disappear. Carl was pleased that the bigwigs were nodding their heads in seeming agreement as he described these problems. Operations Research Unit 13 Sikkim Manipal University Page No. 272 They really appeared to understand. And they seemed very sincere in thanking him for his good advice. Maybe something is actually going to get done this time. Here are the details of the situation that Carl and his management are addressing. The company has two planers for cutting flat smooth surfaces in large castings. The planers currently are being used for two purposes. One is to form the top surface of the platen for large hydraulic lifts. The other is to form the mating surface of the final drive housing for a large piece of earth-moving equipment. The time required by a planer to perform each job varies somewhat, depending largely upon the number of passes that must be made. In particular, for each platen or housing, the time required has a translated exponential distribution, where the minimum time is 10 minutes and the additional time beyond 10 minutes has an exponential distribution with a mean of 10 minutes. Castings of both types arrive one at a time to the planer department. For each type, the arrivals occur randomly with a mean rate of 2 per hour. Based on Carl Schillings advice, the management has asked an OR analyst (you) to analyse the following three proposals for relieving the bottleneck in the planer department: Proposal 1: Obtain one additional planer. The total incremental cost (including capital recovery cost) is estimated to be $30 per hour. (This estimate takes into account the fact that, even with an additional planer, the total running time for all the planers will remain the same.) Proposal 2: Eliminate the variability in the inter-arrival times of the castings, so that the castings would arrive regularly, one every 15 minutes, alternating between platen castings and housing castings. This would require making some changes in the preceding production processes, with an incremental cost of $60 per hour. Proposal 3: Make a change in the production process that would reduce the variability in the time required by a planer to perform each job. In particular, for either type of casting, the time required now would have an Erlang distribution with a mean of 20 minutes and shape parameter k _ 10. The incremental cost in this case would be $20 per hour. These proposals are not mutually exclusive, so any combination can be adopted. It is estimated that the total cost associated with castings having to wait to be processed (including processing time) is $200 per hour for each platen casting and $100 per hour for each housing casting, provided the Operations Research Unit 13 Sikkim Manipal University Page No. 273 waits are not excessive. To avoid excessive waits for either kind of casting, all the castings are processed as soon as possible on a first come, firstserve basis. Discussion Question: Managements objective is to minimise the expected total cost per hour. Use simulation to evaluate and compare all the alternatives, including the status quo and the various combinations of proposals. Then make your recommendation to the management. References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. Operations Research Unit 14 Sikkim Manipal University Page No. 274 Unit 14 Project Scheduling and PERT-CPM Structure: 14.1 Introduction Objectives 14.2 Basic Difference between PERT and CPM PERT CPM Project scheduling by PERT-CPM 14.3 PERT/CPM Network Components and Precedence Relationship Critical path calculations Determination of the critical path Determination of floats 14.4 Project Management PERT 14.5 Summary 14.6 Glossary 14.7 Terminal Questions 14.8 Answers 14.9 Case Study 14.1 Introduction In the previous unit, simulation Monte-Carlo method, we studied about the Monte-Carlo simulation. We also learnt about the various problems of this method, the applications of the simulation, the advantages and the limitations of the simulation. In this unit, project scheduling and PERT-CPM in operations research management, we will study about the basic difference between PERT and CPM. We will also learn about PERT and CPM network components, determination of floats and project management. Project management has evolved as a new field with the development of two analytic techniques for planning, scheduling and controlling projects. These are the project evaluation and review technique (PERT) and the critical path method (CPM). PERT and CPM are basically time-oriented methods in the sense that they both lead to the determination of a time schedule. Projects may be defined as a collection of interrelated activities (or tasks) which must be completed in a specified time according to a specified Operations Research Unit 14 Sikkim Manipal University Page No. 275 sequence and require resources, such as personnel, money, materials, facilities, etc. Few examples of such projects are construction of a bridge, a highway, a power plant, repair and maintenance of an oil refinery or designing of an air plane, development and marketing of a new product, and research and development. The growing complexities of todays projects have demanded more systematic and more effective planning techniques with the objective of optimising the efficiency of executing the project. Efficiency here refers to the utmost reduction in the time required to complete a project while ensuring optimum utilisation of the available resources. Objectives: After studying this unit, you should be able to: - define a project - distinguish between PERT and CPM - apply PERT-CPM method to network analysis - describe project management 14.2 Basic Difference between PERT and CPM There are no essential differences between PERT and CPM as both of them share in common the determination of a critical path. Both are based on the network representation of activities and their scheduling, which determines the most critical activities to be controlled in order to meet the completion date of the project. 14.2.1 PERT Some key points of PERT are as follows: - PERT was developed in connection with an Research and Development (R&D) work. Therefore, it had to cope with the uncertainties that are associated with R&D activities. In PERT, the total project duration is regarded as a random variable. Therefore, associated probabilities are calculated in order to characterise it. - It is an event-oriented network as in the analysis of a network, emphasis is given on the important stages of completion of a task rather than the activities required to be performed to reach a particular event or task. - PERT is normally used for projects involving activities of non-repetitive nature in which time estimates are uncertain. Operations Research Unit 14 Sikkim Manipal University Page No. 276 - It helps in pinpointing critical areas in a project, so that necessary adjustment can be made to meet the scheduled completion date of the project. 14.2.2 CPM - CPM was developed in connection with a construction project, which consisted of routine tasks whose resource requirements and duration were known with certainty. Therefore, it is basically deterministic. - CPM is suitable for establishing a trade-off for optimum balancing between schedule time and cost of the project. - CPM is used for projects involving activities of repetitive nature. 14.2.3 Project scheduling by PERT-CPM Project scheduling consists of three basic phases: planning, scheduling and controlling. Figure 14.1 depicts the three basic phases of PERT-CPM. Fig. 14.1: Phases of PERT-CPM - Project planning Figure 14.2 depicts the activities to be performed in the project planning phase. Fig. 14.2: Activities in Project Planning Phase Operations Research Unit 14 Sikkim Manipal University Page No. 277 - Identify various tasks or work elements to be performed in the project. - Determine requirement of resources, such as men, materials, and machines, for carrying out the activities listed. - Estimate costs and time for various activities. - Specify the inter-relationship among various activities. - Develop a network diagram showing the sequential inter-relationships between the various activities. - Project scheduling - Once the planning phase is over, scheduling of the project is when each of the activities required to be performed, is taken up for execution. The various steps involved during this phase are given as follows: 1. Estimate the durations of activities. Take into account the resources required for these executions in the most economic manner. 2. Based on the time estimates, prepare a time chart showing the start and finish times for each activity. Use the time chart for the following exercises: - To calculate the total project duration by applying network analysis techniques, such as forward (backward) pass and floats calculation - To identify the critical path - To carry out resource smoothing (or levelling) exercises for critical or scarce resources, including re-costing of the schedule taking into account resource constraints - Project controlling Project controlling refers to comparing the actual progress against the estimated schedule. If significant differences are observed then we need to re-schedule the project to update or revise the uncompleted part of the project. Self Assessment Questions 1. Project consists of interrelated activities. (True/False) 2. Project activities are to be completed in a specified time according to a specified sequence. (True/False) 3. PERT and CPM identifies non-critical activities. (True/False) 4. PERT is an activity-oriented network. (True/False) 5. CPM is used for projects that are repetitive in nature.(True/False) Operations Research Unit 14 Sikkim Manipal University Page No. 278 14.3 PERT/CPM Network Components and Precedence Relationship PERT/CPM networks consist of two major components which are as follows: - Events An event represents a point in time that signifies the completion of some activities and the beginning of new ones. The beginning and end points of an activity are thus described by two events usually known as the tail and the head events. Events are commonly represented by circles (nodes) in the network diagram. They do not consume time and resource. - Activities Activities of the network represent project operations or tasks to be conducted. An arrow is commonly used to represent an activity, with its head indicating the direction of progress in the project. Activities originating from a certain event cannot start until the activities terminating at the same event have been completed. They consume time and resource. Events in the network diagram are identified by numbers. Numbers are given to events such that the arrow head number is greater than the arrow tail number. Activities are identified by the numbers of their starting (tail) event and ending (head) event. Figure 14.3 depicts the network where the arrow (P.Q) extended between two events represents the activity. The tail event P represents the start of the activity and the head event Q represents the completion of the activity. Fig. 14.3: Basic PERT-CPM Network Figure 14.4 depicts an example of another PERT-CPM network with activities (1, 3), (2, 3) and (3, 4). As the figure indicates, activities (1, 3) and (2, 3) need to be completed before activity (3, 4) starts. Operations Research Unit 14 Sikkim Manipal University Page No. 279 Fig. 14.4: A PERT-CPM Network The rules for constructing the arrow diagram are as follows: 1. Each activity is represented by one and only one arrow in the network. 2. No two activities can be identified by the same head and tail events. 3. To ensure the correct precedence relationship in the arrow diagram, we need to answer the following questions as we add every activity to the network: - What activities must be completed immediately before these activities can start? - What activities must follow this activity? - What activity must occur concurrently with this activity? This rule is self-explanatory. It actually allows for checking (and rechecking) the precedence relationships as one progresses in the development of the network. Solved problem 1 Construct the arrow diagram comprising activities A, B, C .. and L such that the following relationships are satisfied: 1) A, B and C the first activities of the project, can start simultaneously. 2) A and B precede D. 3) B precedes E, F and H. 4) F and C precede G. 5) E and H precede I and J. 6) C, D, F and J precede K. 7) K precedes L. 8) I, G and L are the terminal activities of the project. Operations Research Unit 14 Sikkim Manipal University Page No. 280 Solution Figure 14.5 depicts an analysis network. Fig. 14.5: Analysis Network The dummy activities D1 and D2 are used (dotted lines) to establish correct precedence relationships. D3 is used to identify activities E and H with unique end events. The events of the project are numbered such that their ascending order indicates the direction of the progress in the project. Note: A dummy activity in a project network analysis has zero duration. 14.3.1 Critical path calculations The application of PERT/CPM should ultimately yield a schedule specifying the start and completion time of each activity. The arrow diagram is the first step towards achieving that goal. The start and completion timings are calculated directly on the arrow diagrams using simple arithmetic. The end result is to classify the activities as critical or non-critical. An activity is said to be critical if a delay in the start of the course makes a delay in the completion time of the entire project. A non-critical activity is such that the time between its earliest start and its latest completion time is longer than its actual duration. A non-critical activity is said to have a slack or float time. 14.3.2 Determination of the critical path A critical path defines a chain of critical activities that connects the start and end events of the arrow diagram. In other words, the critical path identifies all the critical activities of a project. Operations Research Unit 14 Sikkim Manipal University Page No. 281 The critical path calculations are done in two phases. The first phase is called the forward pass. In this phase, all calculations begin from the start node and move to the end node. At each node a number is computed representing the earliest occurrence time of the corresponding event. These numbers are shown in squares. Here we note the number of heads joining the event. We take the maximum earliest timing through these heads. The second phase is called the backward pass. It begins calculations from the end node and moves to the start node. The number computed at each node is shown in a triangle A near the end point, which represents the latest occurrence time of the corresponding event. In backward pass, we see the number of tails and take the minimum value through these tails. Let ESi be the earliest start time of all the activities emanating from event i. Then ESi represents the earliest occurrence time of event i. If i = 1 is the start event then conventionally for the critical path calculations, ESi = 0. Let Dij be the duration of the activity (i, j). Then the forward pass calculations for all defined (i, j) activities with ESi=0 is given by the formula: ESi = maxi {ESi+Dij} Therefore, to compute ESj for event j, we need to first compute ESi for the tail events of all the incoming activities (i, j). With the computation of all ESj, the forward pass calculations are completed. The backward pass starts from the end event. The objective of the backward pass phase is to calculate LCi, the latest completion time for all the activities coming into the event i. Thus, if i = n is the end event, LCn = ESn initiates the backward pass. In general for any node i, we can calculate the backward pass for all defined activities using the formula: LCi = min {LCj-Dij} Operations Research Unit 14 Sikkim Manipal University Page No. 282 We can now identify the critical path activities using the results of the forward and backward passes. An activity (i, j) lies on the critical path if it satisfies the following conditions: - ESi = LCi - ESj = LCj - ESj-ESi = LCj-LCi = Dij These conditions actually indicate that there is no float or slack time between the earliest start and the latest start of the activity. Thus, the activity must be critical. In the arrow diagram these are characterised by same numbers within rectangles and triangles at each of the head and tail events. The difference between the numbers in rectangles or triangles at the head event, and the number within rectangles or triangles at the tail event is equal to the duration of the activity. Thus, we will get a critical path, which is a chain of connected activities, spanning the network from start to end. Solved problem 2 Consider a network which stands from node 1 and terminates at node 6, the time required to perform each activity is indicated on the arrows. Figure 14.6 depicts an analysis network. Fig. 14.6: Analysis Network Solution Let us start with the forward pass with ESi = 0. Since there is only one incoming activity (1, 2) to event 2 with D12 = 3. ES2 = ES1+ DS2 = 0+3=3. Operations Research Unit 14 Sikkim Manipal University Page No. 283 Let us consider the end 3, since there is only one incoming activity (2, 3) to event 3, with D23 = 3. ES3 = ES2+ D23 = 3+3 = 6. To obtain ES4, since there are two activities A (3, 4) and (2, 4) to the event 4 with D24 = 2 and D34 = 0. ES4= maxi=2, 3 {ESi + De4} = max {ES2 +D24, ES3 + D34} = max {3+2, 6+0} = 6 Similarly, ES5 = 13 and ES6 = 19. This completes the forward pass. In the backward pass we have LC6= 19 = ES6 LC5 = 19-6 = 13 LC4= min J = 5, 6 {LCj D4j} = 6 LC3 = 6, LC2 = 3 and LC1 = 0 Therefore, activities (1, 2), (2, 3) (3, 4) (4, 5) (5, 6) are critical and (2, 4) (4, 6), (3, 6), are non-critical. Thus, the activities (1, 2), (2, 3), (3, 4), (4, 5) and (5, 6) define the critical path which is the shortest possible time to complete the project i.e., 19 days. 14.3.3 Determination of floats Once the determination of the critical path is done, we have to compute the floats for the non-critical activities. For the critical activities this float is zero. Before showing how floats are determined, it is necessary to define two new times that are associated with each activity. These are as follows: - Latest start (LS) time and - Earliest completion (EC) time We can define activity (i, j) for these two types of time by: LSij= LCj Dij ECij = ESi + Dij There are two important types of floats namely: - Total float (TF) - Free float (FF) Operations Research Unit 14 Sikkim Manipal University Page No. 284 The total float TFij for activity (i, j) is the difference between the maximum time available to perform the activity (= LCj ESi) and its duration (= Dij) TFij = LCj ESi Dij = LCj ECij = LSij ESi The free float is defined by assuming that all the activities start as early as possible. In this case FFij for activity (i, j) is the excess of available time (= ESi ESi) over its deviation (= Dij); i.e. FFij = ESi ESi - Dij Note: For critical activities float is zero. Therefore, the free float must be zero when the total float is zero. However, the converse is not true, that is, a non-critical activity may have zero free floats. Let us consider the example taken before the critical path calculations. The floats for the non-critical activities can be summarised as depicted in the table 14.1. Table 14.1 Floats for Non-critical Activities Activity (i j) Duration Dij Earliest Latest Total Float TFij Free Float FFij Start ESi Completion ECij Start LSij Completion LCj (1, 2) (2, 3) (2, 4) (3, 4) (3, 5) (3, 6) (4, 5) (4, 6) (5, 6) 3 3 2 0 3 2 7 5 6 0 3 3 6 6 6 6 6 13 3 6 5 6 9 8 13 11 19 0 3 4 6 10 17 6 14 13 3 6 6 6 13 19 13 19 19 0* 0* 1 0* 4 11 0* 8 0* 0 0 1 0 4 11 0 8 0 Note: Total float = ESij = LFij - ESij Free float = Total float - - Head slack * Critical activity * Operations Research Unit 14 Sikkim Manipal University Page No. 285 Solved problem 3 A project consists of a series of tasks A, B, C, D, E, F, G, H, I with the following relationships: - W < X,Y means X and Y cannot start until W is completed - X,Y < W means W cannot start until both X and Y are completed With this notation construct the network diagram having the following constraints A < D, E; B, D < F; C < G, B < H; F, G < I. Also find the minimum time of completion of the project, the critical path, and the total floats of each task, when the time (in days) of completion of each task is as follows: Task: A B C D E F G H I Time: 23 8 20 16 24 18 19 4 10 Solution Figure 14.7 depicts an analysis network. Fig. 14.7: Analysis Network ES1 = 0, ES2 = 20, ES3 = 23, ES4 = 59, ES5 = 39, ES6 = 57, ES7 = 67 Table 14.2 depicts an activity table. Table 14.2 Activity Table Activity (i, j) Duration Dij Earliest Latest Table Float TFij Free Float FFij Start ESe Finish Eeij Start Lj-Dij Finish Lj (1, 2) (1, 3) (1, 4) (2, 5) (3, 4) 20 23 8 19 16 0 0 0 20 23 20 23 8 39 39 18 0 31 38 23 38 23 39 57 39 18 0* 31 18 0* 0 0 31 0 0 Operations Research Unit 14 Sikkim Manipal University Page No. 286 (3, 7) (4, 5) (4, 6) (5, 6) (5, 7) (6, 7) 24 0 18 0 4 10 23 39 39 39 39 37 47 39 57 39 43 67 43 57 39 57 63 57 67 57 57 57 67 67 20 10 0* 18 24 0* 20 0 0 18 24 0 Critical path is 1 3 4 6 7. Self Assessment Questions 6. Events do not consume ________ and _________. 7. Arrows head number is _________ than its tail number. 8. Dummy activity is introduced in a network to keep proper _________ relationship. 9. Critical path calculation includes both _________ and _________. 14.4 Project Management PERT Probability and cost consideration in project scheduling The analysis in CPM does not take into account the cases where time estimates for the different activities are probabilistic. It also does not consider explicitly the cost of schedules. Here we will consider both probability and cost aspects in project scheduling. Probability considerations are incorporated in project scheduling by assuming that the time estimate for each activity is based on three time estimates. They are as follows: a = the optimistic time, which will be required if the execution of the project goes extremely well. b = the pessimistic time, which will be required if everything goes bad. m = the most likely time, which will be required if execution is normal. The most likely estimate m need not coincide with the mid-point 2 a + b of a and b. Then the expected duration of each activity D can be obtained as the mean of 2 a + b and 2 m. Operations Research Unit 14 Sikkim Manipal University Page No. 287 Therefore, 6 a b 4m 3 2m 2 a b D + + = + + = We can use this estimate to study the single estimate D in the critical path calculation. These formulae are derived from beta distribution. The variance of each activity denoted by V is defined by, Variance V = 2 6 b a |. | \ | The earliest expected times for the node i is denoted by E (i). For each node i, E(i) is obtained by taking the sum of the expected times of all activities leading to the node i, when more than one activity leads to a node i, then the greatest of all E(i) is chosen. Let i be the earliest occurrence time of the event i, we can consider i as a random variable. Assuming that all activities of the network are statistically independent, we can calculate the mean and the variance of i as follows: E {i } = ESi and Var{i } = k Vk Where, k defines the activities along the largest path leading to i. For the latest expected time, we consider the last node. Now for each path move backwards and substitute the D iJ for each activity (i, j). Thus we have, E (Lj) = E (a) E (i) = L (Lj) Dij if only one path events from j to i or if it is the minimum of {E [Lj) Dij] for all j for which the activities (i, j) are defined. Note: The probability distribution of times for completing an event can be approximated by the normal distribution due to central limit theorem. Since i represents the earliest occurrence time, event will meet a certain schedule time STi (specified by an analyst) with probability Operations Research Unit 14 Sikkim Manipal University Page No. 288 Pr (i s STi) = Pr | | . |
\ |
s
V( ) ST E( ) V( ) E( ) i i i i i i = Pr (Z s Ki) Where, Z ~N (01) and Ki = V( ) ST E( ) i i i
It is a common practice to compute the probability that event i will occur no later than its LCe. Such probability will represent the chance that the succeeding events will occur within the (ESe, LCe) duration. Solved problem 4 A project is represented by the network depicted in figure 14.8 and has the following data: Fig. 14.8: Analysis Network Table 14.3 depicts a data table. Table 14.3 Data Table Task A B C D E F G H I Optimistic time 5 18 26 16 15 6 7 7 3 Pessimistic time 10 22 40 20 25 12 12 9 5 Most likely time 8 20 33 18 20 9 10 8 4 Determine the following: 1. Expected task times and their variances. 2. The earliest and latest expected times to reach each event 3. The critical path 4. The probability of an event occurring at the proposed completion data if the original contract time of completing the project is 41.5 weeks. 5. The duration of the project that will have 97% channel of being completed. Operations Research Unit 14 Sikkim Manipal University Page No. 289 Solution Using the formula we can calculate expected activity times and variance as depicted in table 14.4. 2 6 b a (a b 4m) V 6 1 D | . | \ | = + + = Table 14.4 Earliest and Latest Expected Time for Events Activity A B m v 1-2 1-3 1-4 2-5 2-6 3-6 4-7 5-7 6-7 5 18 26 16 15 6 7 7 3 10 22 40 20 25 12 12 9 5 8 20 33 18 20 9 10 8 4 7-8 20-00 33-0 18-0 20-0 9-0 9-8 8-0 4-0 0.696 0.444 5.429 0.443 2.780 1.000 0.694 0.111 0.111 Forward pass E1 = 0 E2 = 7.8 E3 = 20 E4 = 33 E5 = 25.8 E6 = 29 E7 = 42.8 Backward pass L7 = 42.8 L6 = 38.8 L5 = 34.8 L4 = 33.0 L3 = 29.8 L2 = 16.8 L1 = 0. The E-values and L-values are depicted in figure 14.9. Fig. 14.9: Analysis Network - The critical path is shown by a thick line in the figure. The critical path is 1-4-7 and the earliest completion time for the project is 42.8 weeks. - The last event 7 will occur only after 42.8 weeks. For this we require only the duration of critical activities. This will help us in calculating the standard duration of the last event. Operations Research Unit 14 Sikkim Manipal University Page No. 290 Expected length of critical path = 33 + 9.8 = 42.8 Variance of article path length = 5.429 + 0.694 = 6.123 Probability of meeting the schedule time is given by Pi (Z s Ki) = Pi (Z 0.52) = 0.30 (from normal distribution table) Thus, the probability that the project can be completed in less than or equal to 41.5 weeks is 0.30. In other words, probability that the project will get delayed beyond 41.5 weeks is 0.70. - Given that P (Z s Ki) = 0.95. But Z0.9S = 1.64, from normal distribution table. Then 1.6 u = or 2.47 ST 42.8 is 1.6 u V( ) ST E( ) i i i i
=
Sji = 1.642.47+42.8 = 46.85 weeks. Self Assessment Questions 10. In a project network, a sequence of activities may form a loop. (True/False) 11. A critical activity must have its total and free floats equal to zero. (True/False) 12. A non-critical activity cannot have zero total float. (True/False) 13. The critical path of project network represents the minimum duration needed to complete the network. (True/False) 14. The analysis in CPM does not take into account the cases where time estimates for the different activities are probabilistic. (True/False) 14.5 Summary - Critical path computations are quite simple, yet they provide valuable information that simplifies the scheduling of complex projects. - The result is that PERT-CPM techniques enjoy tremendous popularity among practitioners in the field. - The usefulness of the techniques is further enhanced by the availability of specialised computer systems for executing, analysing and controlling network projects. 14.6 Glossary Uncertainty: volatility. Task: activity that has to be completed within the specified time. Operations Research Unit 14 Sikkim Manipal University Page No. 291 Latest start (LS) time: the latest time that an activity can begin without delaying the project completion time. 14.7 Terminal Questions 1. Write down the basic difference between PERT and CPM. 2. Explain project management (PERT). 3. A project has 10 activities. The following table shows the information about the activities. Table 14.5: Activities information Activity Preceding activity Duration in weeks A 6 B 3 C A 5 D A 4 E A 3 F C 3 G D 5 H B, D, E 5 I H 2 J I, G, F 3 Draw the network Find the project duration Identify the CPM Prepare the schedule 4. A small project consisting of eight activities has the following characteristics: Table 14.6: Time estimates (in weeks) Activity Preceding activity Most optimistic time (a) Most likely time (m) Most pessimistic time (b) A None 2 4 12 B None 10 12 26 C A 8 9 10 D A 10 15 20 E A 7 7.5 11 F B, C 9 9 9 G D 3 3.5 7 H E, F, G 5 5 5 Operations Research Unit 14 Sikkim Manipal University Page No. 292 1. Draw the PERT network for the project. 2. Determine the critical path. 3. If a 30-week deadline is imposed, what is the probability that the project will be finished within the time limit? 4. If the project manager wants to be 99% sure that the project is completed on this schedule date, how many weeks before that date should he start the project work? 14.8 Answer Self Assessment Questions 1. True 2. True 3. True 4. False 5. True 6. Time, resource 7. Greater than 8. Precedence 9. Forward pass & backward pass 10. False 11. True 12. True 13. True 14. True Terminal Questions 1. There are no essential differences between PERT and CPM as both of them share in common the determination of a critical path. For more details refer to section 14.2 2. Probability considerations are incorporated in PERT by assuming that the time estimate for each activity is based on three time estimates of optimistic time, pessimistic time and most likely time. For more details refer to section 14.4 3. b) 20 weeks c) A D H I - J . For more details refer section 14.3 4. b) A D G H c) 0.6591 d) 34.7 weeks. For more details refer section 14.4 Operations Research Unit 14 Sikkim Manipal University Page No. 293 14.9 Case Study Kaushik Mills production capacity building Kaushik Mills decided to increase its production capacity by building a new feed mill. The project consisted of a number of separate tasks, some of which could not be started before others were complete. The table below shows the lists of activities, along with the times expected for each. The management wanted to advance the schedule as much as possible in order to save valuable time in getting the new mill into operation. The president of the mils commented: Every week saved is worth Rs 70,000 in lost contribution if we can get going. Some of the construction activities could be sped up. For example, the firms architects could by working overtime, design the new plant in 10 weeks instead of the originally estimated 12 weeks. This advancement would cost the mills an additional Rs 25,000 per week that is advanced. The president of mills had already held discussions with mill contractors, an independent firm which was a potential contractor for one of the projects major tasks, for building the plant. Kaushik Mills intended to do the other tasks either itself or through its agents. During the talks with the mill contractors, the management had explored a number of bonus and penalty clauses. One of these was that for every week that the plant was built ahead of 10 weeks, the mills would pay contractors an additional Rs 75,000. Activity Expected Precedent Minimum Crash Activities Time(weeks) (rs/week) A 12 - 10 24,000 B 8 A 18 -11 C 6 A 11 3,000 D 13 A 13 -11 E 4 B,C 4 -11 F 101 C 8 30,000 G 6 C 4 1500 H 101 E,F 6 75,000 I 8 D,G 8 -11 J 8 H,I 8 -11 K 4 J 4 -11 Operations Research Unit 14 Sikkim Manipal University Page No. 294 1. The Management of Kaushik Mills desires to know which activities it would crash? How should they schedule their workers? References: - Kapoor V. K. (2005). Operations Research. Sultan Chand and Sons. - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Kanti Swarup & Gupta P. K., & Hira D. S., & Manmohan (2004). Operation Research. Sultan Chand and Sons. Operations Research Unit 15 Sikkim Manipal University Page No. 295 Unit 15 Game Theory Structure: 15.1 Introduction Objectives 15.2 Competitive Situations Marketing different brands of a commodity Campaigning for elections Fighting military battles 15.3 Characteristics of Competitive Games N-person game Zero-sum game Two-person zero-sum game (Rectangular game) Strategy Pure strategy Mixed strategy 15.4 Maximin Minimax Principle Saddle point Solution to a game with a saddle point 15.5 Dominance Solving games using dominance 15.6 Summary 15.7 Glossary 15.8 Terminal Questions 15.9 Answers 15.10 Case Study 15.1 Introduction In the previous unit, we learnt about project management and basic difference between PERT and CPM. We also learnt the PERT/CPM network components and precedence relationship.John Von Newman developed the Game Theory in the year 1928. The theory gained importance after 1944, when John Von Newman along with Morgenstern published their work onTheory of games and economic behaviour. This highly resourceful field of study is developing fast. Operations Research Unit 15 Sikkim Manipal University Page No. 296 In this unit, you will learn about game theory, competitive situations and characteristics of competitive games. You will also learn about Maximin Minimax principle and dominance. Objectives: After studying this unit, you should be able to: - analyse competitive situations - apply game theory to such situations - evaluate a solution using saddle point and dominance principle - define zero-sum game 15.2 Competitive Situations Competitive situations arise when two or more parties with conflicting interests operate. The competitive situations depicted in figure 15.1 may occur. Fig. 15.1: Types of Competitive Situations Let us now describe the competitive situations in detail. 15.2.1 Marketing different brands of a commodity Two or more brands of detergents/soaps trying to capture the market by adopting various methods (courses), such as advertising through electronic media, providing cash discounts to consumers or offering larger sales commission to dealers are said to be in a competitive situation. 15.2.2 Campaigning for elections Two or more candidates contesting in elections are in a competitive situation. They adopt various methods (courses), such as campaigning through TV, door to door campaigning or campaigning through public meetings to capture more votes. Operations Research Unit 15 Sikkim Manipal University Page No. 297 15.2.3 Fighting military battles Another example of a competitive situation is where two forces are fighting a war to gain supremacy over one another. They adopt various courses of action, such as direct ground attack on enemy camp, ground attack supported by aerial attack or playing defensive by not attacking. The above mentioned situations can be considered as a competitive game where the parties (players) adopt a course of action (play the game). Self Assessment Questions 1. Competitive situations arise when _______ or ________ parties with ___________ operate. 15.3 Characteristics of Competitive Games A competitive game has the following characteristics: 1. The number of players or competitors is finite. 2. Each player has finite number of courses of action or moves. 3. A game is played when each player adopts any one course of action. 4. Every time a game is played, the corresponding combination of courses of action leads to a transaction or payment to each player. The payment is called pay-off or gain. The pay-off may be monetary (money) or some benefit, such as increased sales. 5. The players do not communicate with each other. 6. The players are aware of the rules before starting the game. 15.3.1 N-person game A game having n players participating is called n-person game.A twoplayer game is called 2-person game (two person game). 15.3.2 Zero-sum game If the sum of the gains (pay-off) of the players in a gameis zero, the game is called zero-sum game. A zero-sum game with two players is called two-person zero-sum game. It is also called rectangular game. In a two-person zero-sum game, the gain of one player is the loss of the other. Operations Research Unit 15 Sikkim Manipal University Page No. 298 15.3.3 Two-person zero-sum game (Rectangular game) A two-person zero-sum game is a game where: - Two players participate. - The gain of one player is the loss of the other. Consider a two-person zero-sum game with the players A and B. Let A1 ,A2 ,...Am be the m courses of action for player A. Let B1,B2, ...Bn be the n courses of action for player B. Let aij (i=1,2,.....m;j=1,2,....n)be the pay-off (gain) of player A when he plays the course of action, i A and player B plays the course of action j B . Then, the matrix depicted in figure 15.2 is the payoff (gain) matrix of player A. Fig. 15.2: Pay-Off (Gain) Matrix of Player A This is a mn (read as m by n) game. Here, aij is As gain and Bs loss. Therefore, (-aij ) is the gain of B. To obtain the pay-off matrix of B, write (- ij a ) in the place of ij a in the above matrix and then write the transpose of the matrix. 15.3.4 Strategy In a game, the strategy of a player is predetermined. The player uses this strategy to select a course of action during the game.The strategy of a player may be pureor mixed as depicted in figure 15.3 Operations Research Unit 15 Sikkim Manipal University Page No. 299 Fig. 15.3: Types of Strategy 15.3.5 Pure strategy During the game, if a players strategy is to adopt a specific course of action, irrespective of the opponents strategy, the players strategy is called pure strategy. 15.3.6 Mixed strategy If a player chooses his course of action according to pre-assigned probabilities, then the players strategy is called mixed strategy. Thus, if player A decides to adopt courses of action 1 2 A and A with perspective probabilities 0.4 and 0.6, it is mixed strategy. Case-let 2-Finger Morra Game Two persons A and B play a game. They simultaneously raise their hand and exhibit either one finger or two fingers. If both A and B show one finger or if both show two fingers, then A should pay Rs. 10 to B. On the other hand, if one player shows one finger and the other player shows two fingers, B should pay Rs. 5 to A. The pay-off matrix of A is: Player B B1 (one finger) B2 (two fingers) Player A A1 (one finger) A2 (two fingers) -10 5 5 -10 If the player A decides to show one finger ( A1 ) , his strategy is pure strategy. On the other hand, if A decides to play A1 with probability 0.5 and A2 with probability 0.5, his strategy is mixed strategy. This means, if he is repeatedly playing, he should play both A1 and A2 randomly almost equal number of times. Operations Research Unit 15 Sikkim Manipal University Page No. 300 Self Assessment Questions 2. The number of players or competitors is finite (True/False). 3. If the sum of the gains of the players in a game is 1, the game is called zero-sum game (True/False) 4. A zero-sum game with two players is called a ______ game. 5. During the game, if a players strategy is to adopt a specific course of action, irrespective of the opponents strategy is called ______ strategy 6. If a player chooses his course of action according to pre-assigned probabilities, then the players strategy is called ______ strategy. 15.4 Maximin Minimax Principle Solving a two-person zero-sum game Player A and player B are to play a game without knowing the other players strategy. However, player A would like to maximise his profit and player B would like to minimise his loss. Also each player would expect his opponent to be calculative. Suppose player A plays 1 A . Then, his gain would be a11 ,a12 ,...a1n accordingly Bs choice would beB1 ,B2 ,...Bn .Leto1 =min{ a11 ,a12 ,...a1n } . Then, o 1 is the minimum gain of A when he plays A1 ( 1 o is the minimum pay-off in the first row.) Similarly, if Aplays A2 , his minimum gain iso 2 , the least pay-off in the second row. You will find corresponding to As play A1 ,A2 ,......Am , the minimum gains are the row minimumso1 ,o 2 ,..o m . Suppose A chooses the course of action where i o is maximum. Then the maximum of the row minimum in the pay-off matrix is called maximin. The maximin is: ) `
o = (aij ) j min i max Operations Research Unit 15 Sikkim Manipal University Page No. 301 Similarly, when B plays, he would minimise his maximum loss. The maximum loss to B is when B j is ( aij ) max | j = i . This is the maximum pay-off in the th j column. The minimum of the column maximums in the pay-off matrix is called minimax. The minimax is: Ifo =| =v( say ) , the maximin and the minimax are equal and the game is said to have a saddle point. Ifo < | , then the game does not have a saddle point. Note: o cannot be greater than| . 15.4.1 Saddle point In a two-person zero-sum game, if the maximin and the minimax are equal, the game has a saddle point. Saddle point is the position where the maximin (maximum of the row minimums) and minimax (minimum of the column maximums) coincide. If the maximin occurs in the th r row and if the minimax occurs in the th s column, the position (r, s) is the saddle point. Here, v =ars is the common value of the maximin and the minimax. It is called the value of the game. The value of a game is the expected gain of player A, when both the players adopt optimal strategy. Note 1: If a game has a saddle point, (r, s), the players strategy is pure strategy. For players A and B, the suggested solutions are Ar and Bs respectively. ) `
|= (aij ) i max j min Operations Research Unit 15 Sikkim Manipal University Page No. 302 Note 2: If a game does not have a saddle point, the solution offers a mixed strategy. Note 3: A game is said to be fair if its value is zero. 15.4.2 Solution to a game with a saddle point Consider a two-person zero-sum game with players A and B. Let A1,A2 ,...Am be the courses of action for player A. Let B1,B2 ,.....Bn be the courses of action for player B. The saddle point of the game is as follows: 1. The minimum pay-off in each row of the pay-off matrix is encircled as shown in solved problem 2. 2. The maximum pay-off in each column is written within a box as shown in solved problem 2. 3. If any pay-off is circled as well as boxed, that pay-off is the value of the game. The corresponding position is the saddle point. Let (r, s) be the saddle point. Then, the suggested pure strategy for player A is r A . The suggested pure strategy for player B isBs . The value of the game isars . Note: However, if none of the pay-offs is circled or boxed, the game does not have a saddle point. Hence, the suggested solution for the players is mixed strategy. Solved problem 1: Two persons A and B, without showing each other, place a coin each on the table. If the coins happen to be of the same denomination, player A will take both of them. If they happen to be of different denominations, player B will take both of them. Player A has a few one-rupee coins and two-rupee coins. Player B has a few one-rupee, two-rupee and five-rupee coins. i) Write down the pay-off matrix of A. Does the game have a saddle point? If so, write down the solution. ii) What happens to the game if both the players play only with one-rupee and two-rupee coins? Operations Research Unit 15 Sikkim Manipal University Page No. 303 Solution: i) The pay-off matrix of A is as follows. 1. The minimum pay-off in each row is circled. (Here, the minimums repeat.) 2. The maximum pay-off in each column is boxed. 3. The pay-off -1 is circled as well as boxed. Therefore, the game has a saddle point. It is the position (1, 3). The solution to the game is: a) Strategy for A is 1 A . b) Strategy for B is 3 B . c) Value of the game is v = -1 rupees. ii) If both the players play only with one-rupee and two-rupee coins, the pay-off matrix of A is: Here, the game does not have a saddle point. Therefore, the suggested solution is adoption of mixed strategy for the players. Operations Research Unit 15 Sikkim Manipal University Page No. 304 Solved problem 2: Verify whether the 2-finger morra game explained earlier has a saddle point. If so, write down the solution for the game. Solution: The pay-off matrix of player A is as follows: 1.. The minimum pay-off in each row is circled. 2. The maximum pay-off in each column is boxed. 3. Since none of the pay-off is circled as well as boxed, the game does not have a saddle point. Hence, solution for the game is mixed strategy for both the players. Solved problem 3: A labour union of a firm is negotiating a new five-year settlement regarding payments with the management. The union has the following options: A1 : Aggressive bargaining, A2 : Bargaining with reasoning and A3 : Conciliatory approach. The likely mode of responses from the management are: B1 : Aggressive bargaining, 2 B : Bargaining with reasoning, B3 : Legalistic approach and B4 : Conciliatory approach. The gains to the union in each case are as follows: Operations Research Unit 15 Sikkim Manipal University Page No. 305 What strategy would you suggest for the two sides? What is the value of the game? Solution: 1.. The minimum pay-off in each row is circled. 2. The maximum pay-off in each column is boxed. 3. The value 12 is circled as well as boxed. So, the game has a saddle point. It is the position (1, 3). Therefore, the solution to the game is as follows: a) Strategy for the union is A1 : Aggressive bargaining. b) Strategy for the management is B3 : Legalistic approach. c) Value of the game is v = 12. Operations Research Unit 15 Sikkim Manipal University Page No. 306 Solved problem 4: Solve the game. Is the game fair? Solution: 1. The minimum pay-off in each row is circled. 2. The maximum pay-off in each column is boxed. 3. The value 7 is circled as well as boxed. So, the game has a saddle point. It is the position (2, 1). Therefore, the solution to the game is a) Strategy for the player is 2 A (second course of action) b) Strategy for the opponent isB1 (second course of action) c) Value of the game is v = 7 Operations Research Unit 15 Sikkim Manipal University Page No. 307 Solved problem 5: A two-person zero-sum game, has the following pay-off matrix. Solve the game. Is the game fair? Solution: 1. The minimum pay-off in each row is circled. 2. The maximum pay-off in each column is boxed. 3. The value 0 is circled as well as boxed. So, the game has a saddle point at position (2, 2). Therefore, the solution to the game is: a) Strategy for the player is 2 A (second course of action). b) Strategy for the opponent is 2 B (second course of action). c) Value of the game is v = 0. d) Since the value of the game is 0, the game is fair. Self Assessment Questions 7. Saddle point occurs at row minimum and column maximum. (True/False) 8. If the value of the game is zero, then it is called zero-sum game.(True/False) 9. The pay off matrix represents the gain for top player.(True/False) Operations Research Unit 15 Sikkim Manipal University Page No. 308 15.5 Dominance In a rectangular game, the pay-off matrix of player A is pay-off in one specific row ( r row ) th exceeding the corresponding pay-off in another specific row( s row ) th . This means thatwhatever course of action is adopted by player B, for A, the course of action Ar yields greater gains than the course of action As . Therefore, Ar is a better strategy than As irrespective of Bs strategy. Hence, you can say that Ar dominates As . Alternatively, if each pay-off in a specific column ( p column ) th is less than the corresponding pay-off in another specific column( q column ) th , it means strategy Bp offers minor loss than strategy Bq irrespective of As strategy. Hence, you can say that Bp dominatesBq . Therefore, you can say that: a) In the pay-off matrix, if each pay-off in r row th is greater than (or equal to) the corresponding pay-off in thes row th , Ar dominates As . b) In the pay-off matrix, if each pay-off in p column th is less than (or equal to) the corresponding pay-off in theq column th , Bp dominatesBq . At times, a convex combination of two or more courses of action may dominate another course of action. Whenever a course of action (say As orBq ) is dominated by others, then that course of action ( As orBq ) can be deleted from the pay-off matrix. Such a deletion will not affect the choice of the solution, but it reduces the order of the pay-off matrix. Successive reduction of the order using dominance property helps in solving games. 15.5.1 Solving games using dominance Consider a two-person zero-sum game with players A and B. Let A1 ,A2 ,......Am be the courses of action for player A. Let B1 ,B2 , ...Bn be the courses of action for player B. Operations Research Unit 15 Sikkim Manipal University Page No. 309 Suppose the game has a saddle point. Use the dominance property in sequence to delete the courses of action of A as well as B till the pair comprising the saddle point remains alone. The procedure to arrive at the saddle point is as follows: a) In the pay-off matrix, if each pay-off in r row th is greater than (or equal to) the corresponding pay-off in thes row th , Ar dominates As , hence s A is deleted. b) In the pay-off matrix, if each pay-off in p column th is less than (or equal to) the corresponding pay-off in theq column th , Bp dominatesBq , henceBq is deleted. c) Repeat the above steps in succession to achieve the saddle point. Note: Sometimes, a convex combination of two or more courses of action may dominate another course of action. Solved problem 6: Solve the following game using dominance property. Solution: In the pay-off matrix, each pay-off in the first row exceeds the corresponding pay-off in the third row. Therefore, A1 dominates A3 . So, A3 is deleted. Hence the reduced matrix is as follows: Here, each pay-off in the third column is less than the corresponding pay-off in the first column. Therefore, B3 dominatesB1 . Similarly, Operations Research Unit 15 Sikkim Manipal University Page No. 310 B3 dominates B2 . Also, B3 dominates B4 . Thus, the matrix reduces to Here, since 12>8, 1 A dominates 2 A . And so, finally the matrix reduces to Thus, (1, 3) is the saddle point and the solution to the game is as follows: a) Strategy for A is A1 . b) Strategy for B isB3 . c) Value of the game is v = 12 Solved problem 7: Solve the following zero-sum game and find its value. Solution: In the pay-off matrix, each pay-off in the second column is less than (or equal to) the corresponding pay-off in the third column. And so, the course of action Q dominates R. Similarly, Q dominates S. After deleting R and S, the reduced matrix is as follows. Here, pay-off in the second row is greater than (or equal to) the corresponding pay-offs in the first, third as well as fourth rows. Therefore, Operations Research Unit 15 Sikkim Manipal University Page No. 311 B dominates A, C and D. After deleting A, C and D, the reduced matrix is as follows: Here, 1<6. Therefore, Q dominates P. After deleting P, the reduced matrix is as follows: Therefore the solution to the game is: a) Strategy for Company X is B. b) Strategy for Company Y is Q. c) Value of the game is v = 1. Solved problem 8: In a two-person zero-sum game, the pay-off matrix of A is Write down the pay-off matrix of player B. Solution: The pay-off matrix of B is as follows: Operations Research Unit 15 Sikkim Manipal University Page No. 312 Self Assessment Questions 10. In the pay-off matrix, if each pay-off in rth row is greater than (or equal to) the corresponding pay-off insth row the , Ar dominates As , hence As is __________. 11. In the pay-off matrix, if each pay-off in pthcolumnis less than (or equal to) the corresponding pay-off in qth column, Bp, dominates Bq, hence Bq is _________. 15.6 Summary Let us recapitulate the important concepts discussed in this unit: - Competitive situations arise when two or more parties with conflicting interests operate. - Competitive situations arise when two or more parties with conflicting interests operate. - In a game, the strategy of a player is predetermined. The player uses this strategy to select a course of action during the game.The strategy of a player may be pureor mixed. - The maximum of the row minimum in the pay-off matrix is called maximin. The minimum of the column maximums in the pay-off matrix is called minimax. - Saddle point is the position where the maximin (maximum of the row minimums) and minimax (minimum of the column maximums) coincide. - Whenever a course of action is dominated by others, then that course of action can be deleted from the pay-off matrix. The deletion will not affect the choice of the solution, but it reduces the order of the pay-off matrix. Successive reduction of the order using dominance property helps in solving games. 15.7 Glossary Matrix: rectangular arrangement of numbers in rows and columns Transpose of a matrix:the matrix that results from interchanging rows into and columns into rows Strategy: course of action during the game for achieving gains Operations Research Unit 15 Sikkim Manipal University Page No. 313 15.8 Terminal Questions 1. Solve the following rectangular game. 2. In a rectangular game, pay-off matrix of player A is as follows: i) Solve the game. ii) Write down the pay-off matrix of B and then, solve the game. 3. Briefly describe the characteristics of competitive game. 4. Explain maximin-minimax principle. 15.9 Answers Self Assessment Questions 1. Two, more, conflicting interest 2. True. 3. False 4. Rectangular 5. Pure 6. Mixed 7. True 8. True 9. False 10. Deleted 11. Deleted Operations Research Unit 15 Sikkim Manipal University Page No. 314 Terminal Questions 1. Value of the game is 1. As strategy is (0, 1, 0, 0), while Bs strategy is (0, 0, 1, 0, 0).For more details refer section 15.3.3. 2. Value is 5. As strategy is (1, 0), while Bs strategy is (1, 0).For more details, refer section 15.3.3. 3. Number of players is finite, each player has finite moves, player can adapt any one course of action. For more details refer to section 15.3 for answer. 4. The maximin is: ) `
o = (aij ) j min i max The minimax is: ) `
|= (aij ) i max j min For more details refer to section 15.4 for answer. 15.10 Case Study Swimming Teams The ABC swim team will soon have an important meet with the XYZ team. Each team has a star swimmer (Ram and Shyam respectively) who can swim very well in the 100-yard butterfly, backstroke and breaststroke events. However the rules prevent them from being used in more than two of these events. Therefore, their coaches now need to decide how to use them to maximum advantage. Each team will enter three swimmers per event. For each event, the following table gives the best time previously achieved by Ram and Shyam as well as the best time for each of the swimmers who will definitely enter the event.(Whichever event Ram or Shyam does not swim, his teams third entry for that event will be slower than the two shown in the table). Operations Research Unit 15 Sikkim Manipal University Page No. 315 ABC Swim Team XYZ Swim Team Entry Entry 1 2 Ram Shyam 1 2 Butterfly stroke 1:01.6 59.1 57.5 58.4 1:03.2 59.8 Backstroke 1:06.8 1:05.6 1:03.3 1:02.6 1:04.9 1:04.1 Breaststroke 1:13.9 1:12.5 1:04.7 1:06.1 1:15.3 1:11.8 The points awarded are 5 points for first place, 3 points for second place, 1 point for third place, and none for lower places. Both coaches believe that all swimmers will essentially equal their best times in this meet. Thus Ram and Shyam each will definitely be entered in two of the three events. Discussion Questions a) The coaches must submit all their entries before the meet without knowing the entries for the other team, and no changes are permitted later. The outcome of the meet is very uncertain. Formulate this as a two-person, zero-sum game. b) Eliminate dominated strategies, find the strategies for both the teams according to minimax criterion. References: - Sharma J. K. (2006). Operations Research. Macmillan India Limited. - Taha H. Operations Research. Prentice Hall. - Gupta P. K. Problems in Operations Research (Methods and Solutions). Sultan Chand and Sons Publishers. - Bronsoon, R. Theory and Problems of Operations Research. Schaums Outline Series. E-References: - http://www.springerlink.com/content/g7ph5p3k4128t608/?MUD=MP, retrieved on Dec 20, 2011.