Sei sulla pagina 1di 42

OR UNIT-I Meaning of Quantitative Techniques.

Quantitative techniques are those statistical and operations research or programming techniques which help in the decision making process especially concerning business and industry. These techniques involve the introduction of the element of quantities i.e., they involve the use of numbers, symbols and other mathematical expressions. The quantitative techniques are essentially helpful supplement to judgment and intuition. Quantitative techniques may be defined as those techniques which provide the decision maker with a systematic and powerful means of analysis and help, based on quantitative data, in exploring policies for achieving pre-determined goals. These techniques are particularly relevant to problems of complex business enterprises. What is Quantitative Analysis ? Quantitative analysis is the scientific approach to managerial decision making . Whim, emotions, and guesswork are not part of quantitative analysis approach. The approach starts with data. Like raw material for a factory, these data are manipulated or processed into information that is valuable to people making decision. This processing and manipulating of raw data into meaningful information is the heart of quantitative analysis. Computers have been instrumental in the increasing use of quantitative analysis. In solving a problem, managers must consider both qualitative and quantitative factors. For example, we might consider several different investment alternatives, , including certificates of deposit at a bank, investments in the stock market, and an investment in real estate. We can use quantitative analysis to determine how much our investment will be worth in the future when deposited at a bank at a given interest rate for a certain number of years. Quantitative analysis can also be used in computing financial ratios from the balance sheets for several companies whose stock we are considering. In addition to quantitative analysis, qualitative factors should also be considered. The weather, state and federal legislation , new technological breakthroughs, the outcome of an election, and so on may all be factors that are difficult to quantify. Because of the importance of qualitative factors, the role of quantitative analysis in the decision-making process can vary. When there is a lack of qualitative factors and when the problem, model, and input data remain the same, the results of quantitative analysis can automate the decision-making process. For example, some companies use quantitative inventory models to determine automatically when to order additional new material. The Quantitative Analysis Approach The quantitative analysis approach consists of defining a problem, developing a model, acquiring input data, developing a solution, testing the solution , analyzing the results, and implementing the results. One step does not have to be finished completely before the next is started; in most cases one or more of these steps will be modified to some extent before the final results are implemented.

1. Defining the problem. The first step in the quantitative approach is to develop a clear, concise statement of the problem. This statement will give direction and meaning to the following steps. In many cases, defining the problem is the most important and the most difficult step. It is essential to go beyond the symptoms of the problem and identify the true causes.. One problem may be related to other problems; solving one problem without regard to other related problems can make the entire situation worse. Thus, it is important to analyze how the solution to one problem affects other problems or the situations in general. It is likely that an organization will have several problems. However, a quantitative analysis group usually can not deal with all of an organizations problems at one time. Thus, it is usually necessary to concentrate on only a few problems. For most companies, this means selecting those problems whose solutions will result in the greatest increase in profits or reduction in costs to the company. When the problem is difficult to quantify, it may be necessary to develop a specific, measurable Objectives. A problem might be inadequate health care delivery in a hospital. The objectives might be to increase the number of beds, reduce the average number of days a patient spends in the hospital , increase the physician-to-patient ratio, and so on. When objectives are used, however, the real problem be kept in mind. 2. Developing a Model. Once we select the problem to be analyzed , the next step is to develop a model. Simply stated, a model is a representation (usually mathematical) of a situation. You are using models most of your life. You may have developed models about peoples behavior. Your model might be that friendship is based on reciprocity, an exchange of favors. Of course, there are many other types of models. Architects sometimes make a physical model of a building that they will construct. Engineers develop scale models of chemical plants, called pilot plants. A schematic model is a picture, drawing, or chart of reality. Automobiles, lawn mowers, gears, fans, typewriters, and other devices have schematic models ( drawings and pictures ) that reveal how these devices work. A mathematical model is a set of mathematical relationship. In most cases , these relationships are expressed in equations and inequalities, as they are in spreadsheet model that computes sums, averages, or standard deviations. All models should be developed carefully. They should be solvable, realistic, and easy to understand and modify, and the required input data should be obtainable. 3. Acquiring Input Data. Once we have developed a model, we must btain the data that are used in the model ( input data) . Obtaining accurate data for a model is essential; even if the model is perfect representation of reality, improper data will result in misleading results. This situation is called garbage in, garbage out (GIGO). For a larger problem, collecting accurate data can be one of the most difficult steps in performing quantitative analysis. There are a number of sources that can be used in collecting data. In some cases , company reports, and documents can be used to obtain the necessary data. Another source is interviews with employees or other persons related to the firm. A production supervisor , for example, might be able to tell you with a great degree of accuracy the amount of time that it takes to produce a particular product. Sampling and direct measurement provide other sources of data for the model. You may need to know

how many pounds of raw material are used in producing a new photochemical product. This information can be obtained by going to the plant and actually measuring with scale the amount of raw material that is being used. In other cases , statistical sampling procedures can be used to obtain data. 4 Developing a Solution. Developing a solution involves manipulating the model to arrive at the best (optimal) solution to the problem. In some cases , this requires that an equation be solved for the best decision. In other cases, you can use a trial and error method, trying various approaches and picking the one that results in the best decision. For some problems, you may wish to try all possible values for the variables in the model to arrive at the best decision. You can also solve very difficult and complex problems by repeating a few simple steps until you find the best solution. A series of steps or procedures that are repeated is called an algorithm , named after Algorismus, an Arabic mathematician of the ninth century. The accuracy of the solution depends on the accuracy of the input data and the model. 5. Testing the Solution Before a solution can be analyzed and implemented, it needs to be tested completely. Because the solution depends on the input data and the model, both require testing. Testing the input data and the model includes determining the accuracy and completeness of the data used by the model. Inaccurate data will lead to an inaccurate solution. There are several ways to test input data. One method of testing the data is to collect additional data from a different source. If the original data were collected using interviews, perhaps some additional data can be collected by direct measurement or sampling. These additional data can then be compared with the original data, and statistical tests can be employed to determine whether there are differences between the original data and the additional data. If there are significant differences, more effort is required to obtain accurate input data. If the data is accurate but the results are inconsistent with the problem, the model may not be appropriate. The model can be checked to make sure that it is logical and represents the real situation. 6. Analyzing the Results and Sensitivity Analysis. Analyzing the results starts with determining the implications of the solution. In most cases, a solution to a problem will result in some kind of action or change in the way an organization is operating. The implications of these actions or changes must be determined and analyzed before the results are implemented. Because a model is only an approximation of reality, the sensitivity of the solution to changes in the model and input data is a very important part of analyzing the results. This type of analysis is called sensitivity analysis or post-optimality analysis. It determines how much the solution will change if there were changes in the model or the input data. When the solution is sensitive to changes in the input data and the model specification, additional testing should be performed to make sure that the model and the input data are accurate and valid. If the model or data are wrong, the solution could be wrong, resulting in financial losses or reduced profits. Input data may not always be accurate or model assumptions may not completely appropriate, sensitivity analysis can become an important part of quantitative analysis approach. 7. Implementing the Results

This is the process of incorporating the solution into the company. This can be much more difficult than you would imagine. Even if the solution is optimal and will result in millions of dollars in additional profits, if managers resist the new solution, all of the efforts of the analysis are of no value. Experience has shown that a large number of quantitative analysis teams have failed in their efforts because they have failed to implement a good, workable solution properly. After the solution has been implemented, it should be closely monitored. Over time, there may be numerous changes that call for modifications of the original solution. DEVELOPMENT OF OPERATIONS RESEARCH (i) Pre-Word War II: In 1885 Ferderick W. Taylor emphasized the application of scientific analysis to methods of production. Another man of early scientific management era was Henry L. Gantt. Most job-scheduling methods at that time were rather haphazard. Gantt mapped each job from machine to machine, minimizing every delay. In 1917, A.K. Erlang, a Danish mathematician, published his work on the problem of congestion of telephone traffic. The difficulty was that during busy periods, telephone operators were unable the calls the moment they were made, resulting in delayed calls. A few years after its appearance, his work was accepted by the British Post Office as the basis for calculating circuit facilities. The formulae he developed on waiting time are of fundamental importance to the theory of telephone traffic. The well-known economic lot size model is attributed to F.W. Harris, who published his work on the area of inventory control in 1915. During the 1930s, H.C. Levinson, an American astronomer, applied scientific analysis to the problems of merchandising. His work included scientific study of customers buying habits, response to advertising and relation of environment to the type of article sold. (ii) Word War II: During Word War II, the military management in England called on a team of scientists to study the strategic and tactical problems of air and land defense. This team was under the direction of Professor P.M.S. Blackett of University of Manchester and a former naval officer. Blacketts circus, as the group was called, included three physiologists, two mathematical physicists, one astrophysicist , one army officer, one surveyor, one general physicist and two mathematicians. Many of these problems were of executive type. The objective was to find out the most effective allocation of limited military resources to the various military operations and to the activities within each operation. The application included the effective use of newly invented radar, allocation of British Air Force Planes to missions and the determination of best patterns for searching submarines. This group of scientists formed first OR team. The name operations research (or operational research) was apparently coined in 1940 because the team was carrying out research on (military) operations. Though this science of operations research originated in England, the United States soon took the lead. In United States these OR teams helped in developing strategies for mining operations, inventing new flight patterns and planning of sea mines. (iii) Post-War War II : Immediately after the war, the success of military teams attracted the attention of industrial managers who were seeking solutions to their problems. Industrial operations

research in U.K. and U.S.A. developed along different lines. In U.K. the critical economic situation required drastic increase in production efficiency and creation of new markets. Nationalization of a few key industries further increased the potential field for OR. Consequently OR soon spread from military to government, industrial, social and economic planning. The progress of industrial operations research in U.S.A. was due to advent of Second Industrial Revolution which resulted in automation- the replacement of man by machine as a source of control. The new revolution began around 1940s when electronic computers became commercially available. In 1950, OR was introduced as a subject for academic study in American Universities. Since then this subject has been gaining ever increasing importance for the students of Mathematics, Statistics, Commerce, Economics, Management and Engineering. To increase the impact of operations research, the Operations Research Society of America was formed in 1950. In 1953, the Institute of Management Sciences (IMS) was established . Other countries followed suit and in 1959 International Federation of OR Societies was established and in many countries International Journals on OR began to appear. Some of them (in English) are :1. Operations Research, 2.Opsearch, 3.Operational Research Quarterly, 4. Management Science, 5. Transportation Science, 6. Mathematics of Operations Research, 7. International Journal of Game Theory, etc. OR activities have spread to diverse fields such as hospitals, libraries, city planning, transportation systems, crime investigation, etc. Some of the Indian organizations using OR techniques are : Indian Airlines, Railways, Defense Organizations, Fertilizer Corporation of India, Delhi Cloth Mills, Tata Iron and Steel Co., etc. Operations Research in India In India, operations research came into existence with the opening of an OR unit in 1949 at the Regional Research Laboratory in Hyderabad. At the same time another OR unit was set up at the then Defence Science Laboratory to tackle the problems of stores, purchase and weapon evaluation. An OR unit under Prof. P.C.Mahalonobis was established in 1953 in the Indian Statistical Institute, Kolkata to apply OR methods in national planning and survey. Operations Research Society of India was formed in 1957 and its first conference was held in Delhi in 1959. The first volume of journal of Opsearch came out in 1963. Prof. Mahalonobis made the first important application of OR in India in preparing the draft of the Second Five Year Plan. For academic studies, the M.Sc. course on OR was started by Delhi University in 1963. At the same time, Institutes of Management at Kolkata and Ahemdabad introduced OR in their MBA courses. At present , this subject has been introduced in almost all Institutes and Universities for the students of Mathematics, Statistics, Commerce, Economics, Management and Engineering. DEFINITION OF OPERATIONS RESEARCH Operations research, rather simply defined, is the research of operations. An operation may be called a set of acts required for the achievement of a desired outcome. Such complex, interrelated acts can be performed by four types of systems: Man, Machine, Man-Machine unit and any organization of men, machines, and man-machine units. OR is concerned with the operations of the last type of system. Some of the definitions are:

1. OR is a scientific method of providing executive departments quantitative basis for decisions regarding the operations under their control. (Morse & Kimball). 2. OR, in the most general sense, can be characterized as the application of scientific methods, tools and techniques to problems involving the operations of systems so as to provide those in control of the operations with optimum solutions to the problems.. (Churchman, Ackoff, Arnoff ). 3. Operations research is a scientific approach to problem solving for executive management. (H.M. Wagner). 4. Operations research is the art of winning wars without actually fighting them.(Auther Clark). 5. Operations research is an experimental and applied science devoted to observing, understanding and predicting the behavior of purposeful man-machine systems; and operations research workers are actively engaged in applying this knowledge to practical problems in business, government and society. (Operations Society of America). The most comprehensive and modern definition of operations research can be summarized as below: 6. O.R. is the application of modern methods of mathematical science to complex problems involving management of large systems of men, machines materials and money in industry, business, government and defence. The distinctive approach is to develop a scientific model of the system incorporating measurement of factors such as chance and risk to predict and compare the outcomes of alternative decisions, strategies or control. (Journal of Operations Research Society , U.K.). SCOPE OF OPERATIONS RESEARCH IN MANAGEMENT Operations research is a problem-solving and decision-making science. Some of the areas of management where OR techniques have been successfully applied are: 1. Allocation and Distribution (a) Optimal allocation of limited resources such as men, machines, materials, time and money. (b) Location and size of warehouses, distribution centers, retail depots, etc. (c ) Distribution policy. 2. Production and Facility Planning (a) Selection, location and design of production plants. (b) Project scheduling and allocation of resources. (c ) Preparation of forecasts for the various inventory items and computing economic order quantities and reorder levels. (d) Determination of the number and size of the items to be produced. (e) Maintenance policy. (f) Scheduling and sequencing of production runs by proper allocation of machines. 3. Procurement (a) What, how and when to purchase at the minimum procurement cost. (b) Bidding and replacement policies. 4. Marketing (a) Product selection, timing, and competitive actions. (b) Selection of advertising media (c) Demand forecasts and stock levels. (d) Consumers preference for size, color and packaging of various products.

5. Finance (a) Capital requirements, cash-flow analysis. (b) Credit policies, credit risks, etc. (c ) Determination of optimum replacement policies. 6. Personnel (a) Selection of personnel, determination of retirement age and skills. (b) Recruitment policies and assignment of jobs. 7. Research and Development (a) Determination of areas for research and development. (b) Reliability and control of development projects. (c ) Selection of projects and preparation of their budgets. APPLICATIONS OF VARIOUS OPERATIONS RESEARCH TECHNIQUES Operations Research finds extensive applications in industry, business, government, military and agriculture. Widely variety of industries namely, airlines, automobiles, transportation, petroleum, coal, chemical , mining, paper, communication computer , electronics, etc. have made extensive use of OR techniques. Some of the problems to which OR techniques have been successfully applied are: 1. Linear programming has been used to solve problems involving assignment of jobs to machines, blending, product mix, advertising media selection, least cost diet, distribution, transportation, investment portfolio selection and many others. 2. Dynamic programming has been applied to capital budgeting, selection of advertising media, employment smoothing, cargo loading and optimal routing problems. 3. Inventory Control models have been used to determine economic order quantities, safety stocks, reorder levels, minimum and maximum stock levels. 4. Queuing theory has been helpful to solve problems of traffic congestion, repair and maintenance of broken- down machines, number of service facilities, scheduling and control of air traffic, hospital operations, counters in banks and railway booking agencies. 5. Decision theory has been helpful in controlling hurricanes, water pollution, medicine, space exploration, research and development projects. 6. Network techniques of PERT and CPM have been used in planning, scheduling and controlling construction of dams, bridges, roads, highways and development and production of aircrafts, ships, computers, etc. 7. Simulation has been helpful in a wide variety of probabilistic marketing situations. 8. Replacement theory has been extensively employed to determine the optimum replacement interval for three types of replacement problems: (a) replacement of items that deteriorate with time. (b) replacement of items that do no deteriorate with time but fail suddenly. (c) staff replacement and recruitment. Other techniques extensively employed are game theory, statistical quality control, investment analysis, goal programming , etc. LIMITATIONS OF OPERATIONS RESEARCH 1. Mathematical models, which are essence of OR, do not take into account qualitative factors or emotional factors which are quite real. All influencing factors which can not be quantified find no place in mathematical models.

2. Mathematical models are applicable to only specific categories of problems. 3. OR tries to find optimal solution taking all the factors of the problem into account. 4. Being a new field, generally there is a resistance from the employees to the new proposals. 5. Management, who has to implement the advised proposals, may itself offer a lot of resistance due to conventional thinking. LINEAR PROGRAMMING Linear programming deals with the optimization of a function of variables known as objective function, subject to set of linear equalities/ inequalities known as constraints. The objective function may be profit, loss, cost, production, capacity or any other measure of effectiveness which is to be obtained in the best possible or optimum manner. The constraints may be imposed by different sources such as market demand, production processes and equipment, storage capacity, raw material availability etc. Linear programming is used for optimization problems that satisfy the following conditions: 1. There is a well-defined objective function to be optimized and which can be expressed as a linear function of decision variables. 2. There are constraints on the attainment of the objective and they capable of being expressed as linear equalities/ inequalities in terms of variables. 3. There are alternative courses of action. 4. The decision variables are inter-related and non-negative 5. Resources are in limited supply. Example: A firm produces three products. These products are processed on three different machines. The time required to manufacture one unit of each of the three products and the daily capacity of the three machines are given in the table below: Time per Unit ( Minutes) -------------------------------------------------------------Machine Product I Product 2 Product 3 Machine Capacity(minutes/day) M1 2 3 2 400 M2 4 -3 470 M3 2 5 -430 It is required to determine the daily number of units to be manufacture for each product. The profit per unit for the product 1, 2 and 3 is Rs.4, Rs.3 and Rs.6 respectively. It is assumed that all the amounts produced are consumed in the market. Formulate a linear programming model for the given problem. Solution: Let the extents (amounts) of products 1, 2 and 3 manufactured daily be x1 , x2 and x3 respectively. The objective function is the profit Z = 4 x1 + 3 x2 + 6 x3 Subject to 2 x1 + 3 x2 + 2 x3 400 4 x1 + 0 x2 + 3 x3 470 2 x1 + 5 x2 + 0 x3 430 x1 , x2 , x3 0

General Model of Linear Programming Problem. The general form of a linear programming problem with n decision variables and m constraints can be stated in the following form: Optimize (Maximize or Minimize) Z = c1 x1 + c2 x2 + +cn xn Subject to the linear constraints a11 x1 + a12 x2 + + a1n xn ( , =, ) b1 a21 x1 + a22 x2 + + a2n xn ( , =, ) b2 .. . . . . .

(*)

am1 x1 + am2 x2 + + amn xn ( , =, ) bm and x1, x2 , , xn 0 (**) where (i) x1, x2, ., xn are the variables whose values we wish to determine and are called the decision ( or structural ) variables. (ii) the linear function Z which is to be maximized or minimized is called the objective function. (iii) the inequalities (*) are called the constraints of general L.P.P. (iv). The set of inequalities (**) is known as the set of non-negative restrictions on the general LPP. (v). the constant coefficients cj (j = 1,2, , n) represents the percentage contribution (profit or cost) of decision variable xj to the value of objective function. (vi). the coefficients aij ( i = 1,2, , m ; j= 1,2, ,n) are referred to as the substitution coefficients. These represent the amount of resources consumed per unit of variable aij. (vii). bi ( i= 1,2, ,m) is the constant representing the requirement or availability of the i-th constraint. (viii). The expression ( , =, ) means that only one of the relationships in the set ( , =, ) would hold for a particular constraint. GRAPHICAL METHOD ( Two variables case) Steps: 1. Represent the given problem in the mathematical form ,i,e, formulate a L.P model for the given problem. 2. Represent the given constraints as equalities on x1 , x2 co-ordinates plane and find the convex region formed by them. 3. Plot the objective function. 4. Find the vertices of the convex region and also the value of the objective function at each vertex. The vertex that gives the optimum value of the objective function gives the optimal solution to the problem. In general , a linear programming problem may have (i) a definite and unique optimal solution (ii) an infinite number of optimal solutions (iii) an unbounded solution, and (iv) no solution. Example: Find the maximum value of

Z = 2 x1 + 3 x2 Subject to: x1 + x2 30 x2 3 x2 12 x1 - x2 0 0 x1 20 Solution: Any point in this shaded region is a feasible solution to the given problem. The coordinates of five vertices are: A (3,3) ; Z(A) = 15 B(12,12) ; Z(B) = 60 C(18,12) ; Z(C) = 72 D(20,10); Z(D) = 70 E(20,3) ; Z(E) = 49 Since the maximum value of Z is 72, which occurs at the point Z (18,12). The solution to the given problem is : x1 = 18, x2 = 12, Zmax = 72. Example: Find the minimum value of Z = 5 x1- 2 x2 Subject to 2 x1 + 3 x2 1 x1 , x2 0 Solution: Any point in the shaded region is a feasible solution to the given problem. The coordinates of the two vertices of the unbounded convex region are : A ( 0, 1/3) , B(1/2, 0) Z(A) = -2/3 Z(B) = 5/2 Since the minimum value of Z is -2/3, which occurs at the vertex A (0,1/3) . The solution to the given problem is x1 = 0 , x2 = 1/3, Zmin = -2/3 Example: Maximize Z = 5 x1 + 4 x2 Subject to x1 - 2 x2 1 x1 + 2 x2 3 x1, x2 0 Solution: The solution space satisfying the constraints x1- 2 x2 1 , x1 + 2 x2 3 and x1, x2 0 is shown. This shaded convex region is unbounded. The objective function , when Z = 0, gives the equation 5 x1 + 4 x2 = 0, which is shown by the dotted line passing through the origin O. As Z is increased from zero, this dotted line moves to the right, parallel to itself. Since we are interested in maximizing Z, we increase the value of Z till the dotted line passes through the farthest corner of the shaded region from the origin. As it is not possible to get

10

the farthest corner of the shaded convex region, the maximum value of Z can not be found as it occurs at infinity only. The problem, therefore, has an unbounded solution. Example: Maximize Z = 3 x + 2 y Subject to - 2 x + 3 y 9 3 x 2 y -20 x,y 0 Solution: The figure indicates two shaded regions, one satisfying the constraints 2 x + 3 y 9 and the other satisfying the constraints 3 x 2 y -20. These two shaded regions do not overlap with the result that there is no point (x, y) common to both the shaded regions. The solution of the problem does not exist.

Example: Maximize Z = 3 x1 + 4 x2 Subject to x1- x2 0 2.5 x1 x2 -3 x1 , x2 0 Solution: No Solution.

Example( Multiple Optimal Solution) Maximize Z = 4 x1 + 3 x2 Subject to 3 x1 + 4 x2 24 8 x1 + 6 x2 48 x1 5, x2 6, x1, x2 0 Solution.

11

The five vertices are: O(0,0) ; Z(O) = 0 A(0,6) ; Z(A) = 18 B(24/7,24/7); Z(B) = 24 C(5,4/3) ; Z(C) = 24 D(5,0) ; Z(D) = 20 Thus the maximum value of Z is 24 and it occurs at two corner points B and C . In fact, this maximum value occurs at all points on the line segment joining B and C. Thus we have an infinite number of optimal solutions. Example: An animal feed company must produce 200 kg. of a mixture consisting of ingredients X1 and X2. The ingredient X1 costs Rs.3 per kg. and X2 costs Rs. 5 per kg.. Not more than 80 kg of X1 can be used and at least 60 kg of X2 must be used. Find the minimum cost mixture. Solution: Total cost = 3 x1 + 5 x2 which is to be minimized Subject to x1 + x2 = 200 x1 80, x2 60 x1 0, x2 0 The given problem has no feasible solution space but has only one feasible point with its coordinates x1 = 80, x2 = 120. Hence the optimal solution is to mix 80 kg. of ingredient X1. and 120 kg of ingredients X2 to have a minimum cost of Rs. 840. Example: The standard weight of a special purpose brick is 5 kg. and it contains two basic ingredients B1 and B2. B1 costs Rs.5/kg. and B2 costs Rs.8/kg. Strength considerations dictate that the brick contains not more than 4 kg of B1; and a minimum of 2 kg. of B2. Since the demand for the product is likely to be related to the price of the brick. Find out graphically the minimum cost of the brick satisfying the above conditions. Solution. Let the quantities in kg. of ingredients B1 and B2 be x1 and x2 respectively. The cost of brick is Z = 5 x1 + 8 x2 , which is to be minimized subject to constraints x1 4, x2 2 x1 + x2 = 5 x1, x2 0 Since the third constraint is an equality constraint, the solution space does not consist of a feasible Region but has feasible point A(3,2). Value of the objective function Z = 5 x1 + 8 x2 at the point is Z(A) = 31, Since the minimum value of Z is Rs. 31 which occurs at the point A(3,2). The solution to the given problem is x1 = 3 kg. , x2 = 2 kg. Zmin= Rs.31. Example: A retired person wants to invest up to an amount of Rs.30,000 in fixed income securities. His broker recommends investing in two bonds: Bond A yielding 7% and bond B yielding 10%. After some consideration, he decides to invest at most Rs.12,000 in bond B and at least Rs.6000 in Bond A. He also wants the amount invested in A to be at least

12

equal to the amount invested in Bond B . What should the broker recommend if the investor wants to maximize his return on investment ? Solve graphically. Solution: Let x and y be the amount invested in Bonds A and B , respectively. Using the given data, the problem can be stated as follows: Maximize Z = 0.07 x + 0.10 y Subject to x + y 30,000 x 6,000 y 12,000 xy 0 x, y 0 The optimum solution is: invest Rs. 18,000 in Bond A and Rs. 12,000 in Bond B. It would yield a return of Rs. 2,460. Example: Maximize Z = 20 x + 30 y Subject to 2 x + y 40 4 x y 20 x 30 x, y 0 Solution: All the constraints can not be satisfied and, as such, there is no feasible solution to the given problem. Example: Maximize Subject to Z = 10 x + 20 y 2 x + 4 y 16 x + 5 y 15 x, y 0 Solution. The objective function is not bound over the feasible region and we can move the iso-profit line upward without any limit. The problem has, therefore, unbounded solution.

SIMPLEX METHOD The canonical Form of L.P. Problem. The general linear programming problem can always be put in the following form, called the canonical form: Maximize Z = j=1 cj xj Subject to j=1 aij xj bi , i = 1,2, , m xj 0 , j = 1,2, ,n The characteristics of the form are: (a) all decision variables are non-negative. (b) all constraints are of the ( ) type, and (C) Objective function is of maximization type. Any linear programming problem can be put in the canonical form by the use of elementary transformations. 1. The minimization of a function, f(x) , is equivalent to the maximization of f(x) .

13

For example, the objective function Minimize Z = c1 x1 + + cn xn is equivalent to Maximize G = - Z = -c1 x1 c2 x2 - - cn xn Where Z = - G. 2. An inequality in one direction ( or ) can be changed to an inequality in the opposite direction ( or ) by multiplying both sides by -1. For example. a1x1 + a2 x2 b is equivalent to a1 x1 a2 x2 -b 3. An equation may be replaced by two inequalities in opposite directions. For example, a1 x1 +a2 x2 = b is equivalent to the two simultaneous constraints a1 x1 + a2 x2 b and a1 x1 + a2 x2 b or a1 x1 + a2 x2 b and - a1 x1 a2 x2 - b Basic Terms involved in Simplex Procedure. 1. Standard Form. A linear program in which all the constraints are written as equalities. 2. Slack variable. A variable added to the left-hand side of a less-than or equal to constraint to convert the constraint into an equality is called a slack variable. 3. Surplus Variable. A variable subtracted from the left-hand side of the greater-than 0r equal-to constraint , to convert the constraint into an equality is called a surplus variable. 4. Basic Solution. For a system of m simultaneous linear equations in n +m variables (n > m), a solution obtained by setting n variables equals to zero and solving for the remaining m variables is called a basic solution. In addition, the solution for the m variables must be unique. The n variables set equal to zero in any solution are called non-basic variables. The other m variables whose values are obtained by solving the remaining system of equations are referred to as basic variables. 5. Basic Feasible Solution. A basic feasible solution to a linear programming problem is a basic solution for which the m variables solved for, are all greater than or equal to zero. 6. Optimal Solution. Any basic feasible solution which optimized the objective function of a general L.P. problem is called an optimum basic feasible solution to the general L.P. problem. 7. Tableau Form. The form in which a linear program must be written prior to setting up the initial simplex tableau. 8. Simplex Tableau. A table used to keep track of the calculations made at each iteration when simplex solution method is employed. 9. Zj Row. The numbers in this row under each variable represents the total contribution of outgoing profit when one unit of a non-basic variable is introduced into the basis in place of a basic variable. 10. Cj Zj (or Net Evaluation or Index) Row. The row containing the net profit (or loss) that will result from introducing one unit of the variable indicated in that column in the solution. Thus a positive ( or negative) number in the index row indicates an algebraic reduction ( or increment) in the objective function if one unit of the variable at the head of that column, were introduced in the basic (solution). 11. (a) Pivot (or key) Column. The column with the largest positive number in the net evaluation row of maximization problem, or the largest negative number in the net evaluation row in a minimization problem. It indicates which variable will enter the solution next. (b) Pivot (or Key Row. The row corresponding to the variable that will leave the basis in order to make room for the entering variable. The departing variable will correspond to

14

the smallest positive ratio found by dividing the quantity column values by the pivot column values for each row. (c ) Pivot (or key) Number. The element at the intersection of the pivot row and pivot column. The Standard Form of L.P. Problem. The characteristics of the standard form are 1. All the constraints are expressed in the form of equations, except the non-negativity constraints which remain inequalities ( 0 ). 2. The right hand side of each constraint equation is non-negative. 3. All the decision variables are non-negative. 4. The objective function is of the maximization or minimization type. The inequality constraints are changed to equality constrains by adding or subtracting a non-negative variable from the left hand sides of such constraints. These new variables are called slack variables. They are added if the constraints are the ( ) and subtracted if the constraints are ( ). Since in the case of ( ) constraints the subtracted variable is called surplus variable. Both decision variables, slack and surplus variables are called admissible variables. For example: a1 x1 + a2 x2 b , b > 0 is changed to a1 x2 + a2 x2 + s1 = b, where s1 0 Also constraint p1 x1 + p2 x2 q, q 0 is changed to p1 x1 + p2 x2 s 2 = q , s2 0 What is Simplex method ? Simplex method, also called simplex technique or simplex algorithm developed by G.B. Dantzig, an American mathematician . It has the advantage, i.e. any linear model for which the solution exists, can be solved by it. In principle, it consists of starting with a certain solution of which all that we know that it is feasible, i.e. it satisfies the non-negative conditions ( xj 0, j = 1,2, ,n). We, then, improve upon this solution at consecutive stages, until, after a certain finite number of stages, we arrive at the optimal solution. The simplex method provides an algorithm which consists in moving from one vertex of the region of feasible solutions to another in such a manner that the value of the objective function at the succeeding vertex is less than (or more as the case may be) than at the preceding vertex. This procedure of jumping from one vertex to another is then repeated. Since the number of vertices if finite, this method leads to an optimal vertex in a finite number of steps. Simplex method makes use of the following three points in achieving a systematic reduction from an infinite number of solutions to a finite number of promising solutions. 1. If there are m equality constraints and m +n is the number of variables ( m n) a start for the optimal solution is made by putting n unknowns (out of m +n unknowns) equal to zero and then solving for the m equations in remaining m unknowns, provided that the solution exists and is unique. The n zero variables are called non-basic variables and the m variables are called basic variables which form a basic solution. If the solution yields all non-negative basic variables, it is called basic feasible solution, otherwise it is infeasible.

15

This step reduces the number of alternatives for the optimal solution from infinite to a finite number , whose maximum limit is (m+n)!/ m! n! 2. The number of alternative solutions can be further reduced by eliminating all infeasible basic solutions ( solutions having variables less than zero). In the simplex method, this is achieved by starting with a basic solution which is non-negative ( 0 ) . A condition, called feasibility condition is then provided which ensures that the next basic solution to be selected from all the possible basic solutions is always feasible ( 0 ). This solution is called basic feasible solution If all the basic variables are greater than zero , the solution is called non-degenerate , if some of them are zero, the solution is called degenerate. A new basic feasible solution can be obtained from a previous one by setting one of the m basic variables equal to zero and replacing it by a new non-basic variable. The basic variable set equal to zero is called a leaving variable, while the new one is called an entering variable. 3. The entering variable can be so selected that it improves the value of the objective function so that the new solution is better than the previous one. This is achieved by the use of another condition called optimality condition which selects that entering variable which produces the largest per unit gain in the objective function. This procedure is repeated successively until no further improvement in the value of the objective function is possible. Application of Simplex Method The computation procedure requires at the most m non-zero variables (equal to number of constraints) in the solution at any time, if at any stage the number of non-zero variables becomes less than m, the problem is said to degenerate.

Example:(Maximization LP problem). Three grades of coal A,B and C contain phosphorus and ash as impurities. In a particular industrial process, fuel up to 100 tons (maximum) is required which should contain ash not more than 3% and phosphorus not more than 0.03%. It is desired to maximize the profit while satisfying these conditions. There is an unlimited supply of each grade. The percentage of impurities and profits of grades are given below: Coal A B Phosphorus % 0.02 0.04 Ash % 2.0 3.0 Profits in rupees per ton 12.00 15.00

16

C 0.03 5.0 14.00 Find the proportions in which the three grades be used. Solution. Formulation of the linear Programming Model Step 1: Key decision to be made is as to how much of each grade of coal be used. Let x1, x2 and x3 respectively be the amount in tons of grades A ,B and C used. Step 2: Feasible alternatives are set of values of x1, x2 , x3, where x1 0, x2 0, x3 0 . Step 3. Objective is to maximize the profit. i.e. Maximize Z = 12 x1 + 15 x2 + 14 x3 Step 4: Constraints are: (i) phosphorus content must not exceed 0.03%. Therefore 0.02 x1+0.04 x2+0.03 x3 0.03(x1+x2+x3 ) Or - x1 + x2 0 (ii) ash content must not exceed 3%. Therefore 2 x1+ 3x2 + 5 x3 3(x1+x2+x3 ) Or - x1 + 2x3 0 (iii) total quantity of fuel required is not more than 100 tons. Therefore x1+x2+x3 100 Solution of the problem: Step 1. If the problem is of minimization Type, convert it into maximization type of problem. It can be done because of the transformation Minimize Z = Maximize (- Z). For example, the problem Minimize Z = c1 x1 + . + cn xn can be written as Maximize Z/ (= -Z) = - c1 x1 - . - cn xn If v is the maximum value of Z/, then v will be the minimum value of Z. Step 2. Make all the bis positive. If bi in any constraint is negative, multiply both sides of that constraint by 1 to make that bi positive. Step 3. Set up the problem in the standard form: Convert the inequality constraints into equalities by adding slack variables s1,s2 and s3. Thus - x1+x2+ s1 = 0 - x1+2x3+s2 = 0 x1+x2+x3 +s3= 100 The slacks contribute zero profit to the objective function. Further, the simplex method requires that any variable that appears in one equation must appear in all the equations. This is done by proper placement of a zero coefficient. Therefore, the problem can be expressed as Maximize Z = 12x1+15x2 +14 x3 +0s1 +0s2 +0s3, Subject to - x1+x2 +0 x3 +s1 +0s2 +0s3 =0 - x1+0x2 +2 x3 +0s1 +s2 +0s3 =0 x1+x2 + x3 +0s1 +0s2 +s3 = 100 x1 ,x2, x3 , s1, s2, s3 0.

17

Step 4. Find an Initial Basic Feasible Solution. There are three equations involving six unknowns and to obtain a solution , any three of the variables must be assigned zero values. We start with a basic solution, which we shall get by assuming that the profit earned is zero. This will be so when the decision variables x 1, x2 and x3 are each set equal to zero. Substituting x1 = x2 = x3 = 0 in the above constraints we get s1 = 0, s2= 0 ,s3 = 100 as the basic solution. Since all si ( i= 1,2,3) are 0, the basic solution is feasible (but degenerate). i.e. it is basic feasible degenerate solution. Variables s1 , s2 and s3 are called basic variables and form the basis., while the variables x1 , x2 and x3are called non-basic variables. The corresponding value of the objective function is zero. The vectors associated with the slack variables form the identity matrix. The above information can be conveniently expressed in the form of a table, known as tableau. Tableau 1 Body Matrix Identity Matrix Objective function cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 0 s1 -1 1 0 1 0 0 0 0 s2 -1 0 2 0 1 0 0 coefficients 0 s3 1 1 1 0 0 1 100 In tableau 1, basis refers to the basic variables in the current basic feasible solution. They are slack variables s1 , s2 and s3. The cj- row denotes the coefficients of the variables in the objective function, while CB- column denotes the coefficients of the basic variables only in the objective function. The b- column, also known as quantity column, represents the values of the basic variables; values of all other variables not expressed in the quantity column are always zero. The coefficients under the columns of the variables and corresponding to the first, second and third row are the coefficients of these variables in the constraint equations. The body matrix contains the coefficients of the decision (non-basic) variables while the identity matrix contains the coefficients of basic variables. Although two basic variables have zero values (s1 = 0, s2 = 0), i.e. the solution is degenerate, the problem can still be solved by imaging s1 = and s2 = where (epsilon) is a very small positive number. Step 5. Perform Optimality Test By performing optimality test, we can find whether the current feasible solution can be improved or not. This can be done by computing cj Ej , where Ej = CB aij. Here aij represents matrix elements in the i-th row and j-th column. If cj Ej is positive under any column, at least one better solution is possible. Cj CB Basis 0 s1 0 s2 0 s3 Ej= CB aij Cj Ej 12 x1 -1 -1 1 0 12 15 x2 1 0 1 0 15 14 x3 0 2 1 0 14 0 s1 1 0 0 0 0 Tableau 2 0 0 s2 s3 b 0 0 1 0 0 1 100 0 0 Profit lost/ton 0 0 Net profit/ton

18

(Net evaluation row) The cj Ej row is called the index or net evaluation row and represents the per unit increase in the objective function if the variable heading the column is brought into solution. A positive value indicates the amount by which there is increase in profit, while a negative value indicates the amount by which net profit will decrease if one unit of the variable for that column were brought in the solution. Obviously, if cj Ej is positive under any column, the profit can be increased, i.e. the current basic feasible solution is not optimal and a better solution exists. When no more positive values remain in the cj- Ej row, the solution becomes optimal. As cj- Ej in tableau 2 is positive under some columns, the current feasible solution is not optimal and can be improved. Step 6. Iterate Towards an Optimal Solution. Sub-step (i): This is done by interchanging one of the non-basic variables with one of the slack s. Improvement can be made by (a) finding which variable should be made zero in the current feasible solution (basis matrix) and (b) finding which variable should enter the solution (basis matrix) by making it positive. By observing cj Ej for different columns and marking the column for maximum possible positive cj Ej suggests the variable at the top of the column to be one which should enter the solution or which should replace the slack. This variable is also termed as the incoming variable and the column in which it occurs is called key column and is marked as K. If more than one variable appears with the same maximum value , any one of these variables may be selected arbitrarily as the incoming variable. Now the elements under column b are divided by the corresponding elements of the column K and the row containing the minimum positive ratio is marked. The corresponding variable (slack) is to be made zero. This variable is called outgoing variable and the corresponding row is called key row or pivot row . The element lying at the intersection of the key column and the key row is called key element ( or pivot element) . This element must be made unity by multiplying/dividing the key row by a common multiplying factor. If all the ratios are negative or zero, the value of incoming variable, whatever it may be, can be made as large as we like without violating the feasibility condition. In such a case the problem has an unbounded solution and the iteration stops. The above procedure is represented below:

cj CB Basis 0 s1 0 s2 0 s3 Ej= CB aij cj - Ej

12 x1 -1 -1 1 0 12

15 14 x2 x3 (1) 0 0 2 1 1 0 0 15 14

Tableau 3. 0 0 0 s1 s2 s3 b 1 0 0 (key row) 0 1 0 0 0 1 100 100 0 0 0 0 0 0 0 19

K In tableau 3, maximum value of cj Ej is 15 under x2- column. This column is the key column and has been marked K; x2 is the incoming variable. To decide the outgoing variable, we divide the elements under b-column by the corresponding elements of x 2column and the ratios are represented under - column. Under this column, is the least positive ratio and the row containing it is marked as key row. s1 is, therefore, the outgoing variable and (1) is the key element. Sub-step (ii) : Now all the elements in K column are made zero except the key element which will be unity. This is done by subtracting or adding the proper multiples of key row to the other rows. This new tableau is given below: Tableau 4 cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 15 x2 -1 1 0 1 0 0 0 s2 -1 0 2 0 1 0 0 s3 2 0 1 -1 0 1 100 Ej=CBaij -15 15 0 15 0 0 0 cj Ej 27 0 14 -15 0 0 2nd feasible solution The second feasible solution gives x1 = 0, x2 = , x3 = 0 and Z = 15 Step 7. Repeat step 5 for tableau 4 which gives the 2nd feasible solution. On finding the value of cj Ej for various columns, we find that it is positive for some of them. Hence at least one better solution exists.

Step 8: Repeat step 6. Tableau 5 -------------------------------------------------------------------------cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 15 x2 -1 1 0 1 0 0 - 0 s2 -1 0 2 0 1 0 - 0 s3 (2) 0 1 -1 0 1 100 50- /2 key row Ej=CBaij -15 15 0 15 0 0 0 cj Ej 27 0 14 -15 0 0 K Tableau 6 can be written by making key element unity Tableau 6

20

cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 15 x2 -1 1 0 1 0 0 0 s2 -1 0 2 0 1 0 0 s3 (1) 0 1/2 -1/2 0 1/2 50 /2 Key element 1, entering variable x1; leaving variable s3. Replace s3 by x1. Tableau 7 cj CB Basis 15 x2 0 s2 12 x1 Ej Cj- Ej 15 14 0 0 0 x2 x3 s1 s2 s3 b 1 1/2 1/2 0 1/2 50+/2 100+ 0 (5/2) -1/2 1 1/2 50+/2 20+/5 (key row) 0 1/2 -1/2 0 1/2 50-/2 100- 15 27/2 3/2 0 27/2 0 1/2 -3/2 0 -27/2 K The third feasible solution gives x1 = 50, x2= 50, x3 = 0 and Z = Rs.1,350 Step 9: Repeat step 5. Key element is 5/2 Tableau 8 cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 15 x2 0 1 1/2 1/2 0 1/2 50+/2 0 s2 0 0 (1) -1/5 2/5 1/5 20+/5 12 x1 1 0 1/2 -1/2 0 1/2 50-/2 Entering variable x3, leaving variable s2 Replace s2 by x3. Tableau 9. cj 12 15 14 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 15 x2 0 1 0 3/5 -1/5 2/5 40+2/5 14 x3 0 0 1 -1/5 2/5 1/5 20+/5 12 x1 1 0 0 -2/5 -1/5 2/5 40- 3/5 Ej 12 15 14 7/5 1/5 68/5 cj- Ej 0 0 0 -7/5 -1/5 -68/5 As cj- Ej is not positive under any column in tableau 9, the solution given by it is an optimal solution. From this tableau, x1 = 40 tons, x2= 40 tons, x3= 20 tons and Zmax. = Rs.1,360. Thus quantities of grades A, B and C of coal to be used are 40, 40 and 20. Example: Show that there is an unbounded solution to the following L.P. problem: Maximize Z = 4 x1+x2+3 x3+ 5 x4., Subject to the constraints 21 12 x1 0 0 1 12 0

4 x1 - 6 x2 5 x3 4x4 -20 - 3x1 2 x2 +4 x3 + x4 10 - 8x1 3x2 + 3x3 + 2x4 20 x1, x2, x3, x4 0 Solution: Step 1. If the problem is of minimization type, convert it into maximization type of the problem. Step 2. Make all the bis positive. Multiplying the first constraint throughout by 1, we get -4x1 +6x2 + 5 x3 + 4 x4 20 Step 3. Set up the problem in the standard form Introducing slack variables s1, s2, s3, the problem can be expressed in standard form as Maximize Z = 4x1 + x2 +3x3 +5x4 +0s1 +0s2 +0s3, Subject to -4x1 +6x2 + 5 x3 + 4 x4 +s1 +0s2 +0s3 = 20 -3x1 2 x2 +4 x3 + x4 +0s1 +s2 +0s3 = 10 - 8x1 3x2 + 3x3 + 2x4 +0s1 +0s2 + s3 = 20 x1, x2, x3, x4, s1, s2, s3 0 Step3. Find an Initial Basic feasible Solution Setting decision variables x1, x2, x3, x4 each equal to zero, the basic (non-degenerate) feasible solution is x1= x2= x3= x4= 0 (non-basic), s1= 20 (basic), s2 = 10 (basic), s3= 20 (basic), Z = 0. The following table represents the solution. Tableau 1. cj 4 1 3 5 0 0 0 CB Basis x1 x2 x3 x4 s1 s2 s3 b 0 s1 -4 6 5 (4) 1 0 0 20 5 0 s2 -3 -2 4 1 0 1 0 10 10 0 s3 -8 -3 3 2 0 0 1 20 10 Ej 0 0 0 0 0 0 0 cj - Ej 4 1 3 5 0 0 0 Initial basic feasible solution Step 5. Perform optimality test Since cj Ej is positive under some variable columns, tableau is not optimal. Step 6. Iterate Towards an optimal Solution. In tableau 1, x4 is incoming variable, s1 is outgoing variable and (4) is the key element. This element is made unity in tableau 2. Tableau 2 cj 4 1 3 5 0 0 0 CB Basis x1 x2 x3 x4 s1 s2 s3 b 0 s1 - 1 3/2 5/4 (1) 1/4 0 0 5 0 s2 -3 -2 4 1 0 1 0 10 0 s3 -8 -3 3 2 0 0 1 20 Key element is unity In tableau 3, s1 is replaced by x4. Tableau 3

22

4 1 3 5 0 0 0 x1 x2 x3 x4 s1 s2 s3 b - 1 3/2 5/4 1 1/4 0 0 5 -5 - 2 - 7/2 11/4 0 - 1/4 1 0 5 - 5/2 - 6 - 6 1/2 0 - 1/2 0 1 10 - 5/3 - 5 15/2 25/4 5 5/4 0 0 25 9 - 13/2 -13/4 0 -5/4 0 0 Second basic feasible solution --------------------------------------------------------------------------------Step 7. Perform optimality test Since cj Ej is positive under x1- column, tableau 3 is not optimal. Step 8. Iterate towards an optimal Solution. In tableau 3, x1 is the incoming variable. However, since all replacement ratios ( in - column) are negative, the problem has an unbounded solution. CB 5 0 0

cj Basis x4 s2 s3 Ej cj - Ej

Artificial Variable Technique for Finding the Initial Basic Feasible Solution There are many L.P. problems where slack variables cannot provide such a solution. In these problems at least one of the constraints is of (=) or () type. There are two (closely related) methods available to solve such problems: 1. The big M-method or M-technique or the method of penalties due to A. Charnes. 2. The two phase method due to Dantzig, Orden and Wolfe. THE BIG M- METHOD Step 1. Express the linear programming problem in the standard form. Step 2. Add non-negative variables to the left hand side of all the constraints of (= or ) type. These variables are called artificial variables. The purpose of introducing artificial variables is just to obtain an initial basic feasible solution. However, addition of these artificial variables causes violation of the corresponding constraints. Therefore we would like to get rid of these variables and would not allow them to appear in the final solution. To achieve this, these artificial variables are assigned a large penalty ( -M for maximization problems and + M for minimization problems) in the objective function. Step 3. Continue with the regular steps of simplex method. While making iterations, using simplex method, one of the following three cases may arise: 1. If no artificial variable remains in the basis and the optimality condition is satisfied, then the solution is an optimal basic feasible solution to the given problem. 2. If at least one artificial variable appears in the basis at zero level(with zero value in replacement ratio i.e., b- column) and the optimality condition is satisfied, then the solution is an optimal basic feasible (though degenerate) solution to the given problem. 3. If at least one artificial variable appears in the basis at non-zero level(with positive value in b-column) and the optimality condition is satisfied, then the original problem has no feasible solution. The solution satisfies the constraints but does not optimize the objective function since it contains a very large penalty M and is called pseudo-optimal solution.

23

An artificial variable, once driven out of the solution, can be omitted from further consideration in the succeeding tableaux as it is never going to be reconsidered for entry into the basis. Example: Maximize Z = 3 x1 - x2 Subject to 2 x1 + x2 2, x1 + 3 x2 3, x2 4, x1, x2 0 Solution: Step 1. Set up the problem in the standard form Slack variable s1 is added to the first constraint to express it in the standard form. Thus 2 x1 + x2 +s1 = 2 The second constraint is of ( ) type. A surplus variable s2 is subtracted from it to put in the standard form. Thus x1 + 3 x2 s2 = 3 Likewise, slack variable s3 is added to the third constraint to give x2 +s3 = 4. Putting x1= x2 = 0, we get s1 = 2, s2 = -3, s3 = 4 as the first basic solution. However, this solution is not feasible since s2 is negative. Therefore, we introduce artificial variable A2 in the second constraint. Thus x1 + 3 x2 s2 +A2 = 3 Further , the simplex method requires that any variable that appears in one equation must appear in al the equations. This is done by the proper placement of a zero coefficient. Thus we an artificial system given by 2 x1 + x2 +s1 + 0s2 +0s3 +0A2 = 2 x1 + 3 x2 +0s1 s2 +0s3 +A2 = 3, 0x1 + x2 +0s1 + 0s2 + s3 +0A2 = 4, Where x1, x2 ,s1 , s2, s3, A2 0 Now artificial variable with a value greater than zero(e.g., A2 = 3 here) destroys the equality constraint . Hence it must not appear in the final solution. To achieve this, this artificial variable is assigned a large penalty ( a large negative value, -M) in the objective function, which can be written as Maximize Z = 3 x1 x2 + 0s1 + 0s2 + 0s3 MA2 Step 2. Find an Initial Basic Feasible Solution. Setting decision variables x1= x2 = 0 and also s2 = 0, the basic feasible solution to the artificial system is x1= x2 = s2 = 0, s1 = 2, A2 = 3, s3 = 4, Z=-3M The solution is represented in tableau 1.

24

3 -1 0 CB x1 x2 s1 0 2 1 1 2 -M 1 (3) 0 1 0 0 1 0 4 Ej - M -3M 0 cj - Ej 3+M -1+3M 0 Step 3. Perform Optimality test. Since cj Ej is positive under some variable columns, tableau 1 is not optimal. Step 4. Iterate towards an optimal solution. In table 1, x2 is incoming variable, A2 is outgoing variable and (3) is the key element. This element is made unity in tableau 2. Tableau 2. cj 3 -1 0 0 0 -M CB Basis x1 x2 s1 s2 s3 A2 b 0 s1 2 1 1 0 0 0 2 -M A2 1/3 (1) 0 -1/3 0 1/3 1 0 s3 0 1 0 0 1 0 4 key element unity In tableau 3, A2 is replaced by x2. Also column is omitted from the tableau . Since A2 has been replaced to zero, this tableau represents a basic feasible solution to the original problem.

cj Basis s1 A2 s3

Tableau 1 0 0 -M s2 s3 A2 b 0 0 0 2 -1 0 1 3 0 1 0 4 M 0 -M -M 0 0

Tableau 3. cj 3 -1 0 0 0 CB Basis x1 x2 s1 s2 s3 b 0 s1 (5/3) 0 1 1/3 0 1 3/5 -1 x2 1/3 1 0 -1/3 0 1 3 0 s3 -1/3 0 0 1/3 1 3 -9 Ej -1/3 -1 0 1/3 0 -1 cj -Ej 10/3 0 0 -1/3 0 Second basic feasible solution Step 5. Repeat step 3. Since cj Ej is positive under x1- column, tableau 3 is not optimal. Step 6. Repeat step 4. In tableau 3, x1 is incoming variable, s1 is outgoing variable and (5/3) is the key element. This element is made unity in tableau 4. Tableau 4 cj 3 -1 0 0 0 CB Basis x1 x2 s1 s2 s3 b 0 s1 (1) 0 3/5 1/5 0 3/5 -1 x2 1/3 1 0 -1/3 0 1 0 s3 -1/3 0 0 1/3 0 3 25

Key element unity In tableau 5, s1 is replaced by x1. Tableau 5 0 0 0 s1 s2 s3 b 3/5 1/5 0 3/5 -1/5 -2/5 0 4/5 1/5 2/5 1 16/5 2 1 0 1 -2 -1 0 Optimal basic feasible solution Step 7. Repeat step 3. Since cj- Ej is either zero or negative under all variable columns, tableau 5 is optimal and the optimal basic feasible solution is given by x1 = 3/5, x2= 4/5, Zmax = 1 CB 3 -1 0 Example: Minimize Z = 2 y1 + 3 y2 Subject to y1 + y2 5 y1 +2 y2 6 y1, y2 0 Solution: Step 1: Set up the problem into standard form Introducing surplus and artificial variables, the given problem in the standard form may be expressed as cj Basis x1 x2 s3 Ej cj- Ej 3 x1 1 0 0 3 0 -1 x2 0 1 0 -1 0

Minimize Z = 2 y1 + 3 y2 +0s1 + 0s2 + M A1 + M A2 Subject to y1 + y2 s1 + 0s2 + A1 + 0A2 = 5 y1 +2y2 + 0s1 - s2 +0 A1 + A2 = 6 y1 ,y2 , s1 , s2 ,A1 , A2 0 Step 2.: Find an Initial Basic Feasible Solution Setting variables y1 = y2 = s1 = s2 = 0, the basic feasible solution to the artificial system is A1 = 5 A2 = 6 Z = 11 M The solution is represented in tableau 1. cj 2 3 0 Basis y1 y2 s1 A1 1 1 -1 A2 1 (2) 0 Ej 2M 3M -M cj - Ej 2-2M 3-3M M Tableau 1. 0 M M s2 A1 A2 b 0 1 0 5 5 -1 0 1 6 3 -M M M M 0 0

CB M M

26

Initial solution Step 3. Perform Optimality Test. Since cj Ej is negative under variable columns, table 1 in not optimal. Step 4. Iterate Towards an Optimal Solution. In tableau 1, y2 is incoming variable, A2 is outgoing variable and (2) is the key element. This element is made unity in tableau 2. Tableau 2 cj 2 3 0 0 M M CB Basis y1 y2 s1 s2 A1 A2 b M A1 1 1 -1 0 1 0 5 M A2 1/2 (1) 0 -1/2 0 1/2 3 Key element unity In tableau 2, A2 is replaced by y2 and A2 is, therefore , omitted from further consideration. Tableau 3 cj 2 3 0 0 M Basis y1 y2 s1 s2 A1 b A1 (1/2) 0 -1 1/2 1 2 4 y2 1/2 1 0 -1/2 0 3 6 Ej (3+M)/2 3 -M (M-3)/2 M 9+2M cj- Ej (1-M)/2 0 M (3-M)/2 0 Second solution Step 5. Repeat step 3. since cj Ej is negative under some variable columns, tableau 3 is not optimal. Step 6. Repeat Step 4. In tableau 3, y1 is incoming variable, A1 is outgoing variable and (1/2) is the key element. This element is made unity in tableau 4. CB M 3 Tableau 4. cj 2 3 0 0 M CB Basis y1 y2 s1 s2 A1 b M A1 (1) 0 -2 1 2 4 3 y2 1/2 1 0 -1/2 0 3 Key element unity . In table 5, A1 is replaced by y1 and A1 is, therefore omitted from further consideration. Since A1, A2 have been reduced to zero, this tableau represents a basic feasible solution for the given problem. Tableau 5 CB 2 3 cj Basis y1 y2 Ej cj- Ej 2 y1 1 0 2 0 3 y2 0 1 3 0 0 s1 -2 1 -1 1 0 s2 1 -1 -1 1 b 4 1 11

27

Optimal feasible solution Step 7. Repeat step 3. Since cj Ej is non-negative under all variable columns, tableau 5 is optimal. The optimal basic feasible solution is y1 = 4 y2 = 1, Zmin = 11. Example: An animal feed company must produce 200 kg. of a mixture consisting of ingredients x1 and x2 respectively. x1 costs Rs.3 per kg. and x2 Rs.8 per kg. No more than 80 kg. of x1 can be used, and at least 60 kg. of x2 must be used. Find how much of each ingredient should be used if the company wants to minimize cost ? Solution. The L.P. model for the problem is Minimize Z = 3 x1 + 8 x2 Subject to x1 + x2 = 200 x1 80 x2 60 x1, x2 0 Introducing slack ,surplus and artificial variables, the given problem in the standard form is Minimize Z = 3 x1 + 8 x2 + 0s1 +0s2 + MA1 +MA3, Subject to x1 + x2 + 0s1 +0s2 + A1 +0A3= 200 x1 + 0x2 +s1 +0s2 +0A1 +0A3= 80, 0x1 + x2 +0s1- s2 +0A1 + A3= 60, x1 ,x2 , s1, s2, A1 , A3 0 Initial basic (non-degenerate) solution to the artificial system is x1= x2 = s2 = 0, s1 = 80 A1 = 200 A3 = 60 Z = 260 M The optimum basic feasible solution is given by x1 = 80, x2 = 120, Zmin = Rs. 1,200. THE TWO-PHASE METHOD The linear Programming problem is solved in two phases. Phase 1. Step 1. Assign a cost -1 to each of artificial variable and a cost 0 to all other variables and get a new objective function . Z* = - A1 A2 - .. where Ai are artificial variables. Step 2. Write down the auxiliary L.P.P. in which the new objective function is to be maximized subject to the given set of constraints. Step 3. Solve the auxiliary L.P. by simplex method until either of the following three cases arise: (i) Max Z* < 0 and at least one artificial appears in the optimum basis at positive level. (ii) Max Z* = 0 and at least one artificial variable appears in the optimum basis at zero level. (iii) Max Z* = 0 and no artificial variable appears in the optimum basis.

28

In case (i) , given L.P.P. does not possess any feasible solution, whereas in case (ii) and (iii) we go to phase II. Phase II Use the optimum basic feasible solution of phase I as a starting solution for the original L.P.P. Assign the actual costs to the variables in the objective function and a zero cost to every artificial variable in the basis at zero level. Delete the artificial variable column from the table which is eliminated from the basis in phase I. Apply simplex method to the modified simplex table obtained at the end of phase I till an optimum basic feasible solution is obtained or till there is an indication of unbounded solution. Example : Use two-phase simplex method to Maximize Z = 5x1 4 x2 + 3 x3 Subject to 2 x1 + x2 6 x3 = 20, 6 x1 + 5 x2 + 10 x3 76, 8 x1 3 x2 + 6 x3 50, x1 , x2 , x3 0. Solution. PHASE I: It consists of the following steps: Step I. Make all the bis positive Since all bis are non-negative, this step is not required in the present problem. Step 2. Set up the problem in the standard Form. The given constraints, after the introduction of slack and artificial variables take the form 2 x1 + x2 6 x3 + A1 = 20, 6 x1 + 5 x2 + 10 x3 + s2 = 76, 8 x1 3 x2 + 6 x3 + s3 = 50, x1 , x2 , x3, s2 , s3, A1 0. The new (artificial) objective function is Minimize w = A1 The new objective function is also called the infeasibility form or auxiliary (dummy) objective function. Thus the problem for phase I in standard form can be written as Max Z* = 0x1 + 0x2 +0x3 + 0s2 + 0s3 - A1, Subject to 2 x1 + x2 6 x3 +0s2 +0s3+ A1 = 20, 6 x1 + 5 x2 + 10 x3 + s2 +0s3 +0A1 = 76, 8 x1 3 x2 + 6 x3 + 0s2 + s3 +0A1= 50, x1 , x2 , x3, s2 , s3, A1 0. Step 3. Find an Initial Basic Feasible Solution Setting variables x1= x2 = x3 = 0, the basic feasible solution to the auxiliary problem is x1= x2 = x3 = 0 A1 = 20 s2 = 76 s3 = 50 Tableau 1. represents the solution. Tableau 1 cj 0 0 0 0 0 -1

29

CB Basis -1 A1 0 s2 0 s3 Ej cj Ej

x1 x2 x3 s2 s3 A1 b 2 1 -6 0 0 1 20 10 6 5 10 1 0 0 76 38/3 (8) - 3 6 0 1 0 50 25/4 -2 -1 6 0 0 -1 20 2 1 - 6 0 0 0 Initial b.f.s. for phase I problem Step 4. Perform Optimality test. Since cj- Ej is positive under some variable columns, tableau 1 is not optimal. Step 5. Iterate towards an optimal Solution. In tableau 1, x1 is the incoming variable, s3 is the outgoing variable and (8) is the key element. This element is made unity in table 2. Performing iterations to get an optimal solution results in the following tableaux. cj Basis A1 s2 s3 0 0 x1 x2 2 1 6 5 (1) - 3/8 0 x3 -6 10 3/4 Tableau 2 0 0 -1 s2 s3 A1 b 0 0 1 20 1 0 0 76 0 1/8 0 25/4 Key element Unity

CB -1 0 0

CB -1 0 0

cj 0 Basis x1 A1 0 s2 0 x1 1 Ej 0 Cj- Ej 0

0 0 x2 x3 (7/4) -15/2 29/4 11/2 - 3/8 3/4 - 7/4 15/2 7/4 - 15/2

Tableau 3 0 0 -1 s2 s3 A1 b 0 -1/4 1 15/2 30/7 1 -3/4 0 77/2 154/29 0 1/8 0 25/4 -50/3 0 1/4 - 1 15/2 0 - 1/4 0 Second b.f.s. for phase I problem Tableau 4

CB -1 0 0

cj Basis A1 s2 x1

0 0 0 0 x1 x2 x3 s2 0 (1) -30/7 0 0 29/4 11/2 1 1 - 3/8 3/4 0

cj

0 -1 s3 A1 b -1/7 4/7 30/7 -3/4 0 77/2 154/29 1/8 0 25/4 Key element unity Tableau 5 0 0 -1 30

CB 0 0 0

x3 s2 s3 A1 b -30/7 0 -1/7 4/7 30/7 256/7 1 2/7 -29/7 52/7 -6/7 0 1/14 3/14 55/7 Ej 0 0 0 0 0 cj - Ej 0 0 0 -1 Optimal b.f.s. for phase I problem Since cj Ej is negative under all columns, tableau 5 is optimal. Also ,since Max Z* = 0 and no artificial variable appears in the basis, this tableau gives a b.f.s. to the original problem. PHASE II: . Tableau 6 -4 3 0 0 CB x2 x3 s2 s3 b -4 1 -30/7 0 -1/7 30/7 0 0 256/7 1 2/7 52/7 5 0 -6/7 0 1/14 55/7 -4 90/7 0 13/14 155/7 0 -69/7 0 -13/14 Optimal b.f.s. for the original problem Since cj Ej is either negative or zero under all variable columns, tableau 6 gives optimal basic feasible solution for the original problem. The optimal solution is x1 = 55/7 , x2 = 30/7, x3 = 0 , Zmax = 155/7. Example: Use two-phase simplex method to Maximize Z = 3 x1 + 2 x2 + 2 x3, Subject to 5 x1 + 7 x2 + 4 x3 7 -4 x1 +7 x2 + 5 x3 -2 3 x1 + 4 x2 6 x3 29/7 x1, x2, x3 0 Solution: x1 = 1, x2 = 2/7, x3 = 0, Zmax = 25/7. Example: Use two-phase simplex method to solve Maximize Z = 5 x1 + 3 x2 Subject to 2 x1 + x2 1 x1 + 4x2 6 x1 , x2 0 Solution: Phase I Maximize Z* = 0 x1 + 0 x2 + 0 s1 + 0 s2 1 A1 Subject to 2 x1 + x2 + s1 + 0 s2 + 0 A1 = 1 x1 + 4 x2 + 0s1 - s2 + A1 = 6 Initial basic feasible solution is given by s1 = 1 , A1 = 6 cj Basis x2 s2 x1 Ej cj - Ej 5 x1 0 0 1 5 0 cj 0 0 0 0 -1

Basis x2 s2 x1

x1 0 0 1 0 0

x2 1 0 0 0 0

31

x2 s1 s2 A1 b (1) 1 0 0 1 1/1= 1 4 0 -1 1 6 6/4 -4 0 1 -1 4 0 -1 0 0 x2 2 1 1 0 0 1 -1 A1 -7 0 -4 -1 1 2 Ej 7 0 4 1 -1 Cj Ej -7 0 -4 -1 0 Since all Cj Ej 0, an optimum feasible solution to the auxiliary L.P.P. is obtained, x2 = 1, x1 = 0, A1 = 2. But as Max Z* 0 and an artificial variable A1 is in the basis at a positive level, the original LPP does not possess any feasible solution. Example : Use two-phase simplex method to solve Minimize Z = x1 x2 3 x3 Subject to - 2 x1 + x2 + 3 x3 = 2 2 x1 + 3x2 + 4 x3 = 1 x1, x2 , x3 0 Maximize Z = -x1 + x2 + 3 x3 Subject to - 2 x1 + x2 + 3 x3 +A1 = 2 2 x1 + 3x2 + 4 x3 +A2 = 1 x1, x2 , x3 , A1, A2 0 Phase I Max. Z* = ,0 x1 0x2 0 x3 1 A1 1 A2 Subject to - 2 x1 + x2 + 3 x3 +A1 + 0 A2 = 2 2 x1 + 3x2 + 4 x3 + 0 A1 +A2 = 1 x1, x2 , x3 , A1, A2 0 Initial basic feasible solution is A1 = 2 , A1 = 1 Cj 0 0 0 -1 -1 CB Basis x1 x2 x3 A1 A2 b -1 A1 -2 1 3 1 0 2 2/3 -1 A2 2 3 (4) 0 1 1 1/4 Ej 0 -4 -7 -1 -1 Cj - Ej 0 4 7 0 0 -1 A1 -7/2 -5/4 0 1 -3/4 5/4 0 x3 1/2 3/4 1 0 1/4 1/4 Ej 7/2 5/4 0 -1 3/4 - 5/4 Cj Ej - 7/2 - 5/4 0 0 -7/4 Since all Cj Ej 0, an optimum basic feasible solution x1 = 0 , x2 = 0, x3 = 1/4, A1 = 5/4 , A2 = 0 to the auxiliary LPP has been allowed. But since Max Z* is negative and the artificial variable A1 appears in the basis at positive level, the original problem does not possess any feasible solution. Example. Maximize Z = 5 x1 4 x2 + 3 x3 CB 0 -1 Basis s1 A1 Ej Cj Ej x1 2 1 -1 1 32

Subject to 2 x1 + x2 6 x3 = 20 6 x1 + 5 x2 +10 x3 76 8 x1 - 3 x2 +6 x3 50 x1 , x2 , x3 0 Solution. Phase I Max Z* = o x1 + 0 x2 + 0 x3 + 0 s1 + 0 s2 1 A1 Subject to 2 x1 + x2 6 x3 +0 s1 +0 s2 + A1= 20 6 x1 + 5 x2 +10 x3 + s1 + 0 s2 + 0 A1 = 76 8 x1 - 3 x2 +6 x3 + 0 s1 + s2 + 0 A1 = 50 x1 , x2 , x3 , s1 , s2 , A1 0 Initial basic feasible solution is: A1 = 20, s1 = 76, s2 = 50.

CB -1 0 0

Cj Basis A1 s1 s2 Ej Cj Ej A1 s1 x1 Ej Cj Ej x2 s1 x1 Ej

-1 0 0

0 0 0

0 x1 2 6 (8) -2 2 0 0 1 0 0 0 0 1 0 0

0 x2 1 5 -3 -1 1 (7/4) 29/4 -3/8 -7/4 7/4 1 0 0

0 x3 -6 10 6 6 -6 -15/2 11/2 3/4 15/2 -15/2 -30/7 256/7 -6/7 0

0 s1 0 1 0 0 0 0 1 0 0 0 0 1 0 0

0 s2 0 0 1 0 0 -1/4 -3/4 1/8 1/4 -1/4 -1/7 2/7 1/14 0

-1 A1 1 0 0 -1 0 1 0 0 -1 0 4/7 -29/7 3/4 0

b 20 76 50

10 76/6 50/8

15/2 30/7 77/2 154/29 50/8 -50/3

30/7 52/7 55/7

33

Cj Ej 0 0 0 0 0 -1 Since all Cj Ej 0, an optimum solution to the auxiliary L.P.P. has been obtained. Also Max Z* = 0 with no artificial variable in the basis. We go to phase II. Phase II. Consider the actual cost associated with the original variables. Delete the artificial variable A1 column from the table. Cj 5 -4 3 0 0 CB Basis x1 x2 x3 s1 s2 b -4 x2 0 1 -30/7 0 -1/7 30/7 0 s1 0 0 256/7 1 2/7 52/7 5 x1 1 0 -6/7 0 1/14 55/7 Ej 5 -4 90/7 0 13/14 Cj Ej 0 0 - 69/7 0 -13/14 Since all Cj Ej 0 and optimum basic feasible solution has been reached. Hence x1 = 55/7, x2 =30/7, x3 = 0, Max Z = 155/7 Degeneracy in Simplex Method A basic feasible solution is said to be degenerate if at least one of the basic variables is zero and this phenomenon of getting a degenerate basic feasible solution is called degeneracy. Degeneracy in L.P. may arise: (i) at the initial stage when at least one basic variable is zero in the initial basic feasible solution. (ii) at any subsequent stage when the minimum positive replacement ratio ( under column) is same for two or more rows. In such a case arbitrary selection of the outgoing variable may result in one or more basic variables becoming zero in the next iteration. There is no assurance that the value of the objective function will improve, since new solutions may remain degenerate . This seriously reduces the efficiency of the simplex method. More serious, however, is the condition of cycling or circling in which the same sequence of simplex iterations are repeatedly endlessly without improving the solution. Nevertheless, if such a situation occurs, this difficulty may be overcome by applying the following procedure, called perturbation method by A. Charnes: 1. Divide each element in the tied rows by the positive coefficients of the key-column in that row. 2. Compare the resulting ratios, column by column, first in the identity and then in the body, from left to right. 3. The row which first contains the smallest algebraic ratio contains the outgoing slack variable. Example: Maximize Z = 2x1 + 3 x2 + 10 x3 Subject to x1 + 2 x3 = 0 x2 + x3 = 1, x1 , x2, x3 0 Solution: Step 1. Set up the problem in the standard form The given problem can be expressed as Maximize Z = 2x1 + 3 x2 + 10 x3 34

Subject to x1 + 0 x2 + 2 x3 = 0 0 x1 + x2 + x3 = 1, x1 , x2, x3 0 Note that there is no need to introduce artificial variables as x1 and x2 are themselves forming identity matrix [ 1 0 ] and can be 0 1 treated as basic variables . Step 2. Find an initial Basic Feasible Solution Setting x3 = 0, the basic feasible (degenerate) solution to the problem is x1 = 0 (basic), x2 = 1 (basic), x3 = 0 (non-basic) Z=3 Table 1. Cj 2 3 10 CB Basis x1 x2 x3 b 2 x1 1 0 (2) 0 0 3 x2 0 1 1 1 1 Ej 2 3 7 3 Cj- Ej 0 0 3 Initial basic feasible solution Step 3. Perform Optimality test Since Cj Ej is positive under x3- column, table 1 is not optimal. Step 4. Iterate Towards an optimal solution. Table 2 Cj 2 3 10 CB Basis x1 x2 x3 b 10 x3 1/2 0 1 0 3 x2 - 1/2 1 0 1 Ej 7/2 3 10 3 Cj Ej - 3/2 0 0 Optimal b.f.s. Table 2 is optimal and the optimal b.f.s. is x1 = 0 (non-basic) x2 = 1 (basic) x3 = 0 (basic) Zmax = 3. The above solution is, however, degenerate because basic variable x3 has zero value. Note that the value of Z in table 1 as well as table 2 is same. Thus we have obtained the optimal degenerate solution from a degenerate solution without improving the value of Z . In other words, it is possible to move from on table to another without any improvement in the objective function.

35

Example : Maximize Z = 2 x1 + x2 Subject to 4 x1 + 3 x2 12 4 x1 + x2 8 4 x1 - x2 8 x1 , x2 0 Solution: The problem , in the standard form , can be expressed as Maximize Z = 2 x1 + x2 +0s1 +0s2 + 0s3 Subject to 4 x1 + 3 x2 + s1 +0s2 + 0s3 = 12 4 x1 + x2 + 0s1 +s2 + 0s3 = 8 4 x1 - x2 + 0s1 + 0s2 +s3 = 8 x1 , x2 , s1, s2, s3 0 The initial basic feasible solution is x1 = x2 = 0 (non-basic), s1 = 12 (basic), s2 = 8 (basic), s3= 8 (basic) , Z = 0 The Table 1 represent this solution Table 1 Cj 2 1 0 0 0 CB Basis x1 x2 s1 s2 s3 b 0 s1 4 3 1 0 0 12 3 0 s2 4 1 0 1 0 8 2 0 s3 (4) - 1 0 0 1 8 2 Ej 0 0 0 0 0 0 Cj Ej 2 1 0 0 0 Initial b.f.s Perform Optimality test Since Cj Ej is positive under some variable columns, table 1 is not optimal. Iterate Towards an Optimal Solution In table 1, x1 is the incoming variable. As s2 and s3 are tied rows, perturbation method by A. Charnes is applied to determine the outgoing variable. The first column of the identity has elements 0 and 0 in the tied rows. Dividing them by the corresponding elements of key column, the resulting ratios are 0 and 0. Hence the first column of the identity fails to identify the outgoing variable. The second column of the identity has elements 1 and 0 in the tied rows. Dividing them by the corresponding elements of the key column, the resulting ratios are 1/4, 0 . As s3-row yields the smaller ratio, it is marked s key row and (4) is the key element. Performing iterations to get an optimal solution results in the following tables. Table 2 Cj 2 1 0 0 0 CB Basis x1 x2 s1 s2 s3 b 0 s1 0 4 1 0 -1 4 1 0 s2 0 (2) 0 1 -1 0 0 2 x1 1 -1/4 0 0 1/4 2 -8 Ej 2 - 1/2 0 0 1/2 4

36

Cj Ej

3/2 1 x2 0 1 0 1 0

- 1/2 second b.f.s

CB 0 1 2

Cj 2 Basis x1 s1 0 x2 0 x1 1 Ej 2 Cj Ej 0

Table 3 0 0 0 s1 s2 s3 1 -2 (1) 0 1/2 - 1/2 0 1/8 1/8 0 3/4 - 1/4 0 - 3/4 1/4 Table 4 0 s1 1 1/2 - 1/8 1/4 - 1/4 0 s2 -2 - 1/2 3/8 1/4 - 1/4

b 4 0 2 4

4 -0 16

third b.f.s. 0

CB 0 1 2

Cj Basis s3 x2 x1 Ej Cj- Ej

2 x1 0 0 1 2 0

1 x2 0 1 0 1 0

s3 b 1 4 0 2 0 3/2 0 5 0 Optimal b.f.s.

Table 3 gives the following optimal b.f.s.: x1 = 3/2 , x2 = 2, Zmax = 5. Example : Maximize Z = 5 x1 - 2 x2 + 3 x3 Subject to 2 x1 + 2 x2 x3 2, 3 x1 - 4 x2 3, x2 + 3 x3 5 x1, x2, x3 0 Solution: Step 1. Setup the problem in the standard form Maximize Z = 5 x1 - 2 x2 + 3 x3 +0s1 + 0s2 + 0s3 MA1, Subject to 2 x1 + 2 x2 x3 s1 +0s2 +0s3 +A1= 2, 3 x1 - 4 x2 +0x3 +0s1 +s2 + 0s3 +0A1= 3, 0x1 + x2 + 3 x3 +0s1 + 0s2 +s3 +0A1 = 5 x1, x2, x3 , s1, s2, s3, A1 0 Step 2. Find an initial basic feasible solution The i.b.f.s. is x1= 0, x2 = 0, x3 = 0 , s1 = 0 (non-basic) A1 = 2, s2 = 3, s3 = 5, Z = - 2M Table 1 Cj 5 -2 3 0 0 0 -M CB Basis x1 x2 x3 s1 s2 s3 A1 b -M A1 (2) 2 -1 -1 0 0 1 2 1

37

0 0

s2 3 -4 0 s3 0 1 3 Ej -2M -2M M Cj Ej 5+2M -2+2M 3-M

0 0 M -M

1 0 0 3 1 0 1 0 5 0 0 -M -2M 0 0 0 Initial solution

Step 3. Perform Optimality test Since Cj Ej is positive under some variable columns, table 1 is not optimal. Step 4. Iterate Towards an Optimal Solution In table 1, x1 is the incoming variable. A1 and s2- rows are tied . But since A1 is an artificial variable, it is chosen as the outgoing variable and (2) is the key element. Making this element unity and performing iterations to get an optimal solution results in the following tables: Table 2 Cj 5 -2 3 0 0 0 CB Basis x1 x2 x3 s1 s2 s3 b 5 x1 1 1 -1/2 -1/2 0 0 1 -2 0 s2 0 -7 (3/2) 3/2 1 0 0 0 0 s3 0 1 3 0 0 1 5 5/3 Ej 5 5 -5/2 -5/2 0 0 5 Cj Ej 0 -7 11/2 5/2 0 0 Second b.f.s. Cj 5 CB Basis x1 5 x1 1 3 x3 0 0 s3 0 Ej 5 Cj Ej 0 Table 3 -2 3 0 x2 x3 s1 -4/3 0 0 -14/3 1 1 (15) 0 -3 -62/3 3 3 56/3 0 -3 -2 x2 0 0 1 -2 0 Table 4 3 x3 0 1 0 3 0 0 0 s2 s3 1/3 0 2/3 0 -2 1 11/3 0 -11/3 0 Third

b 1 0 5 5 b.f.s.

-3/4 -0 1/3

Cj CB Basis 5 x1 3 x3 -2 x2 Ej Cj Ej

5 x1 1 0 0 5 0

0 0 0 s1 s2 s3 b -4/15 7/45 4/45 13/9 -65/12 (1/15) 2/45 14/45 14/9 70/3 -1/5 -2/5 1/15 1/3 -5/3 -11/15 53/45 56/45 101/9 11/15 -53/45 -56/45 Fourth b.f.s.

38

Cj CB Basis 5 x1 0 s1 -2 x2 Ej Cj Ej

5 x1 1 0 0 5 0

Table 5 -2 3 x2 x3 0 4 0 15 1 3 -2 14 0 -11

0 s1 0 1 0 0 0

0 0 s2 s3 1/3 4/3 2/3 14/3 0 1 5/3 14/3 -5/3 -14/3 Optimal b.f.s,

b 23/3 70/3 5 85/3

Optimal solution is: x1 = 23/3, x2=5, x3=0, Zmax = 85/3. DUAL PROGRAMMING PROBLEM For every L.P. problem there is a related unique L.P. problem involving the same data which also describes the original problem. The given original problem is called the primal program (P) . This program can be solved by transposing (reversing) the rows and columns of the algebraic statement of the problem. Inverting the program in this way results in dual program (D). A solution to the dual program may be found in a manner similar to that used for the primal. The two programs have very closely related properties so that optimal solution of the dual gives complete information about the optimal solution of the primal and vise-versa. Useful Aspects of Duality Property (i) If a primal contains a large number of rows ( constraints) and a smaller number of columns (variables), the computational procedure can be considered reduced by converting it into dual and then solving it. (ii) It gives additional information as to how the optimal solution changes as a result of the changes in the coefficients and the formulation of the problem. This is termed as post optimality or sensitivity analysis . (iii) Duality in linear programming has certain far reaching consequences of economic nature. (iv) Calculation of the dual checks the accuracy of the primal solution. Dual Problem When Primal is in the Canonical Form. The general linear programming problem in canonical form is written as Maximize Z= c1 x1 + c2 x2 + . + cn xn Subject to a11 x1 + a12 x2 + + a1n xn b1 a21 x1 + a22 x2 + + a2n xn b2 .. (1) am1 x1 + am2 x2 + + amn xn bm x1, x2, .. xn 0 If the above problem is referred to as primal, then its associated dual will be Minimize W = b1 y1 + b2 y2 + + bm ym Subject to a11 y1 + a21 y2 + + am1 ym c1 a12 y1 + a22 y2 + + am2 ym c2 (2) a1n y1 + a2n y2 + + amn ym cn

39

where the associated dual variables y1, y2, , ym 0 Relations (1) and (2) are called symmetric primal dual pairs. The above pair of programs can be written as Primal Dual n m Maximize Z = j=1 cj xj Minimize W = i=1 bi yi n m Subject to j=1 aij xj bi, i=1,2, ..m Subject to i=1 aij yi cj , i = 1,2, ,m j= 1,2, ..,n Where xj 0, j= 1,2, ,n where yi 0, i= 1,2, ,m Points to remember (1) If the primal contains n variables and m constraints, the dual will contain m variables and n constraints. (2) The maximization problem in the primal becomes the minimization problem in the dual and vise-versa. (3) The maximization problem has ( ) constraints while the minimization problem has ( ) constraints. (4) The constants c1, c2 , , cn in the objective function of the primal appear in the constraints of the dual. (5) The constraints b1, b2 , , bm in the primal appear as constants in the objective function of the dual. (6) The variables in both problems are non-negative. Example: Construct the dual to the primal problem Maximize Z = 3 x1 + 5 x2 Subject to 2 x1 + 6 x2 50 3 x1 + 2 x2 35 5 x1 3 x2 10 x2 20 where x1 0, x2 0. Solution: Let y1, y2, y3 & y4 be the corresponding dual variables, then the dual problem is given by Minimize W = 50 y1 + 35 y2 + 10 y3 + 20 y4 Subject to 2 y1 + 3 y2 + 5 y3 +0 y4 3 6 y1 + 2 y2 3 y3 + y4 5 y1, y2, y3, y4 0 Dual Problem when Primal is in the Standard Form Consider the problem Maximize Z = c1 x1 + c2 x2 Subject to a11 x1 + a12 x2 = b1 a21 x1 + a22 x2 b2 x1 , x2 0 The first constraint can be expressed as a11 x1 + a12 x2 b1 and a11 x1 + a12 x2 b1 or - a11 x1 - a12 x2 - - b1

40

Let y1 , y1 and y2 be the dual variables associated with the first, second and third constraints. Then the dual problem is Minimize W= b1 ( y1/- y1// ) + b2 y2 Subject to a11 ( y1/ y1// ) + a21 y2 c1 a12 (y1/ y1// ) + a22 y2 c2 y1/, y1//, y2 0 Note that the term ( y1/- y1//) occurs in the objective function as well as all the constraints of the dual. If we put y1/ y1// = y1, then the new variable y1 , which is the difference between two nonnegative variables , becomes unrestricted in sign and the dual problem becomes Minimize W = b1 y1 + b2 y2 Subject to a11 y1 + a21 y2 c1 a12 y1 + a22 y2 c2 y2 0, y1 unrestricted in sign. Example: Find the maximum of Z = 6 x + 8y Subject to 5 x + 2 y 20 x + 2 y 10 x, y 0 by solving its dual problem. Solution: The dual of the given primal problem is given below. As there are two constraints in the primal, we have two dual variables namely w1 , w2 . Min Z* = 20 w1 + 10 w2 Subject to 5 w1 + w2 6 2 w1 + 2 w2 8 w1 , w2 0 The problem is reformulated as Max Z* = - 20 w1 10 w2 + 0 s1 + 0s2 - MA1 MA2 Subject to 5 w1 + w2 s1 + A1 = 6 2 w1 + 2 w2 s2 + A2 = 8 W1, w2 , s1, s2, A1, A2 0 Cj -20 -10 0 0 -M -M CB Basis w1 w2 s1 s2 A1 A2 b -M A1 (5) 1 -1 0 1 0 6 6/5 -M A2 2 2 0 -1 0 1 8 8/2 Ej -7M -3M M M -M -M Cj Ej -20+7M -10+3M -M -M 0 0 -20 w1 1 1/5 -1/5 0 1/5 0 6/5 6 -M A2 0 (8/5) 2/5 -1 1 28/5 7/2 Ej - 20 -4-8/5M 4-2/5M M -M Cj- Ej 0 8/5M-6 -4+2/5M -M 0 -20 w1 1 0 -1/4 1/8 1/2 -10 w2 0 1 1/4 -5/8 7/2 Ej -20 -10 5/2 15/4 -45 41

Cj Ej 0 0 -5/2 -15/4 Since all Cj Ej 0, the solution is optimum. Therefore, the optimal solution of dual is w1 = 1/2, w2 = 7/2. Min.Z* = 45 The optimal solution of the primal problem is given by the value of Ej - Cj in the optimal table corresponding to the column surplus variables s1 and s2. Therefore x = 5/2 , y = 15/4 Max Z = 6 x5/2 + 8x 15/4 = 45.

42