1. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.1 Table of Contents CD Chapter 14 (Solution Concepts for Linear Programming) Some Key Facts About Optimal Solutions (Section 14.1)14.2–14.16 The Role of Corner Points in Searching for an Optimal Solution (Sec.14.2)14.17–14.21 Solution Concepts for the Simplex Method (Section 14.3)14.22–14.24 The Simplex Method with Two Decision Variables (Section 14.4)14.25–14.26 The Simplex Method with Three Decision Variables (Section 14.5)14.27–14.28 The Role of Supplementary Variables (Section 14.6)14.29–14.30 Some Algebraic Details for the Simplex Method (Section 14.7)14.31–14.40 Computer Implementation of the Simplex Method (Section 14.8)14.41 The Interior-Point Approach to Solving LP Problems (Section 14.9)14.42–14.43

2. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.2 Some Key Facts About Optimal Solutions An optimal solution must lie on the boundary of the feasible region If a linear programming problem has exactly one optimal solution, this solution must be a corner point. The simplex method is an extremely efficient procedure for solving LP problems. It only evaluate corner points. A corner point is a feasible solution that lies at the intersection of constraint boundaries.

3. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.3 Boundary of Feasible Region for the Wyndor Problem

4. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.4 Corner Points for the Wyndor Problem

5. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.5 Optimal Solution with Different Unit Profits Unit Profit DoorsWindowsObjective FunctionOptimal Solution $400 Profit = 400D + 400W(2, 6) $500$300Profit = 500D + 300W(4, 3) $300–$100Profit = 300D – 100W(4, 0) –$100$500Profit = –100D + 500W(0, 6) –$100 Profit = –100D – 100W(0, 0)

6. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.6 The Enumeration-of-Corner-Points Method Corner PointProfit = 300D + 500W (D, W) = (0, 0)Profit = 300(0) + 500(0) = $0 (D, W) = (0, 6)Profit = 300(0) + 500(6) = $3,000 Optimal  (D, W) = (2, 6)Profit = 300(2) + 500(6) = $3,600  Best (D, W) = (4, 3)Profit = 300(4) + 500(3) = $2,700 (D, W) = (4, 0)Profit = 300(4) + 500(0) = $1,200

7. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.7 More Key Facts About Optimal Solutions The only possibilities for a linear programming problem are that it has –exactly one optimal solution –an infinite number of optimal solution –no optimal solution If a linear programming problem has multiple optimal solutions, at least two of the optimal solutions must be corner points. If a linear programming problem has multiple optimal solutions, the simplex method will automatically find one of the optimal corner points and signal that there are others. If desired, it can also find the other optimal corner points. (However, Excel’s Solver does not have this feature.)

8. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.8 An Infinite Number of Optimal Solutions

9. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.9 Spreadsheet with an Infinite Number of Optimal Solutions

10. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.10 The Sensitivity Report

11. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.11 Multiple Optimal Solutions with Different Unit Profits Unit Profit DoorsWindowsObjective FunctionMultiple Optimal Solutions $300$200Profit = 300D + 200WLine segment between (2, 6) and (4, 3) $3000Profit = 300DLine segment between (4, 3) and (4, 0) 0$500Profit = 500WLine segment between (0, 6) and (2, 6) 0–$100Profit = –100WLine segment between (0, 0) and (4, 0) -$1000Profit = –100DLine segment between (0, 0) and (0, 6)

12. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.12 More Key Facts About Optimal Solutions The constraints of a linear programming problem can be so restrictive that it is impossible to satisfy all the constraints simultaneously. Thus, there are no feasible solutions and so no optimal solution. If some necessary constraints were not included, it is possible to have no limit on the best objective function value. If this occurs, then no solution can be optimal becausre there always is a better feasible solution. An optimal solution is only optimal with respect to a particular mathematical model that provides only a rough representation of the real problem. –The purpose of an LP study is to help guide decisions by providing insights into the consequences of various options under different assumptions about future conditions. –Most of the important insights are gained while conducting the analysis done after finding an optimal solution. This is often referred to as postoptimality analysis or what-if analysis.

13. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.13 No Feasible Solutions New constraint: D + W ≥ 10

14. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.14 Spreadsheet with No Feasible Solutions

15. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.15 No Bound on the Objective Function Value

16. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.16 Spreadsheet with No Bound on the Objective Function Value

17. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.17 Optimality of the Best Corner Point For any linear programming problem with an optimal solution, the best corner point must be an optimal solution. When two or more corner points tie for being the best one, all these best corner points must be optimal solutions.

18. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.18 Five Corner Points for the Wyndor Problem

19. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.19 How the Simplex Method Solves the Wyndor Problem

20. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.20 The Simplex Method Getting Started: Select some corner point as the initial corner point. If the origin is feasible, this is a convenient choice. Checking for Optimality: Check each of the adjacent corner points. If none are better, then stop because the current corner point is optimal. If one or more of the adjacent corner points are better then move on. Moving On: One of the better adjacent corner points is selected as the next current point. When more than one is better, the conventional selection method is to choose the one that provides the best rate of improvement. After making the selection, return to the Checking for Optimality step.

21. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.21 Adjacent Corner Points Corner PointIts Adjacent Corner Points (0, 0)(0, 6) and (4, 0) (0, 6)(2, 6) and (0, 0) (2, 6)(4, 3) and (0, 6) (4, 3)(4, 0) and (2, 6) (4, 0)(0, 0) and (4, 3) Two corner points are adjacent corner points if they share all but one of the same constraint boundaries. Two adjacent corner points are connected by a line segment, referred to as an edge of the feasible region.

22. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.22 Structure of Iterative Algorithms

23. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.23 Structure of Most Management Science Algorithms

24. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.24 Solution Concepts for the Simplex Method Focus solely on the corner points. For any linear programming problem with an optimal solution, the best corner point must be an optimal solution. The simplex method is an iterative algorithm. The initialization step fins an initial corner point. Each iteration moves to a new corner point. The optimality test stops when the new corner point is an optimal solution. Whenever possible, the initialization step chooses the origin to be the initial corner point. Given a corner point, it is much quicker computationally to gather information about adjacent corner points than other corner points. Each iteration only considers moving to an adjacent corner point. The simplex method uses algebra to examine each edge of the feasible region emanating form the current corner point to determine the rate of improvement. It chooses the one with the largest rate of improvement to actually move along. If none of the edges give a positive rater of improvement, then the current corner point is optimal.

25. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.25 How the Simplex Method Solves the Wyndor Problem

26. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.26 Profit & Gambit Example

27. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.27 The Simplex Method with Three Decision Variables Maximize Exposure = 1,300TV + 600M + 500SS subject to 300TV + 150M + 100SS ≤ 4,000 90TV + 30M + 40SS ≤ 1,000 TV ≤ 5 and TV ≥ 0, M ≥ 0, SS ≥ 0.

28. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.28 The Corner Points Corner PointValue of Objective Function (0, 0, 0)0 (0, 26.667, 0)16,000 (5, 16.66, 0)16,500 (5, 15, 2.5)16,750 (0, 20, 10)17,000 (0, 0, 25)10,000 (5, 0, 13.75)13, 375 (5, 0, 0)6,500

29. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.29 Slack Variables The slack variable for a ≤ constraint is a variable that equals the right-hand side minus the left-hand side. D ≤ 4 2W≤ 12 3D +2W≤ 18 s 1 = 4 – D s 2 = 12 – 2W s 3 = 18 – 3D – 2W The slack variables enable converting ≤ constraints into equations. D +s 1 ≤ 4 2W +s 2 ≤ 12 3D +2W +s 3 ≤ 18

30. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.30 The Simplex Method for the Wyndor Problem OrderCorner PointVariables = 0Other Variables Corresponding Constraint Boundary Equations 1(D, W) = (0, 0)D = 0 W = 0 s 1 = 4 s 2 = 12 s 3 = 18 D = 0 W = 0 2(D, W) = (0, 6)D = 0 s 2 = 0 s 1 = 4 W = 6 s 3 = 6 D = 0 2W = 12 3(D, W) = (2, 6)s 2 = 0 s 3 = 0 s 1 = 2 W = 6 D = 2 2W = 12 3D + 2W = 18

31. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.31 The Simplex Method for the Wyndor Problem Geometric ProgressionAlgebraic Progession IterationCorner PointCBE Nonbasic Variables Basic Variables Basic Feasible Solution (D, W, s 1, s 2, s 3 ) 1(0, 0)4, 5D, Ws 1, s 2, s 3 (0, 0, 4, 12, 18) 2(0, 6)4, 2D, s 2 s 1, W, s 3 (0, 6, 4, 0, 6) 3(2, 6)3, 2s 3, s 2 s 1, W, D(2, 6, 2, 0, 0)

32. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.32 Outline of an Iteration of the Simplex Method 1.Determine the entering basic variable. 2.Determine the leaving basic variable. 3.Solve for the new basic feasible solution.

33. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.33 Structure of the Simplex Method

34. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.34 The Initialization Step Maximize P, subject to (0)P –300D –500W= 0 (1) D+ s 1 = 4 (2)2W+ s 2 = 12 (3)3D +2W+ s 3 = 18 and D ≥ 0, W ≥ 0, s 1 ≥ 0, s 2 ≥ 0, s 3 ≥ 0. Initial Basic Feasible Solution: Nonbasic variables: D = 0, W = 0 Basic variables: s 1 = 4, s 2 = 12, s 3 = 18 Value of objective function: P = 0.

35. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.35 The Optimality Test Rule for the Optimality Test: Examine the current equation 0, which contains only P and the nonbasic variables along with a constant on the right-hand side. If none of the nonbasic variables have a negative coefficient, then the current basic feasible solution is optimal. (0)P – 300D – 500W = 0 Both D and W have a negative coefficient, so the current basic feasible solution is not optimal.

36. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.36 Determining the Entering Basic Variable Rule for Determining the Entering Basic Variable: Examine the current equation 0, which contains only P and the nonbasic variables along with a constant on the right-hand side. Among the nonbasic variables with a negative coefficient, choose the one whose coefficient has the largest absolute value to be the entering basic variable. (0)P – 300D – 500W = 0 W is chosen to be the entering basic variable.

37. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.37 Determining the Leaving Basic Variable Minimum Ratio Rule for Determining the Leaving Basic Variable: For each equation that has a strictly positive coefficient for the entering basic variable, take the ratio of the right-hand side to this coefficient. Identify the equation that has the minimum ratio, and select the basic variable in this equation to be the leaving basic variable. (0)P –300D –500W= 0 (1) D+ s 1 = 4 (2)2W+ s 2 = 12 (3)3D +2W+ s 3 = 18 Since W is the entering basic variable, only equation 2 and 3 have a strictly postive coefficient for this variable. The ratio for equation 2 (12 / 2 = 6) is smaller than the ratio for equation 3 (18/2 = 9), so s 2 (the basic variable for equation 2) is the leaving basic variable.

38. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.38 Solving for the New Basic Feasible Solution Maximize P, subject to (0)P –300D –500W= 0 (1) D+ s 1 = 4 (2)2W+ s 2 = 12 (3)3D +2W+ s 3 = 18 and D ≥ 0, W ≥ 0, s 1 ≥ 0, s 2 ≥ 0, s 3 ≥ 0. Requirements for Proper Form from Gaussian Elimination: 1.Equation 0 does not contain any basic variables. 2.Each of the other equations cotains exactly one basic variable. 3.An equation’s one basic variable has a coefficient of 1. 4.An equation’s one basic variable does not appear in any other equation.

39. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.39 Solving for the New Basic Feasible Solution 1.For the equation containing the leaving basic variable, divide that equation by the coefficient of the entering basic variable. The entering basic variable now becomes the one basic variable in this equation. 2.Subtract the appropriate multiple of this equation from each of the other equations that contain the entering basic variable. The appropriate multiple is the coefficient of the entering basic variable in the other equation. 3.The system of equations now is in proper form from Guassian elimination, so read the value of each basic variable from the right-hand side of its equation to obtain the new basic feasible solution.

40. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.40 The New Basic Feasible Solution Maximize P, subject to (0)P –300D+ 250s 2 = 3,000 (1) D+ s 1 = 4 (2)W+ 0.5s 2 = 6 (3)3D– s 2 + s 3 = 6 and D ≥ 0, W ≥ 0, s 1 ≥ 0, s 2 ≥ 0, s 3 ≥ 0. New Basic Feasible Solution: Nonbasic variables: D = 0, s 2 = 0 Basic variables: s 1 = 4, W = 6, s 3 = 6 Value of objective function: P = 3,000.

41. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.41 Computer Implementation of the Simplex Method Computer codes for the simplex method, such as the one in the Excel Solver, now are widely available on essentially all computer systems. The simplex method is used routinely to solve large linear programming problems. With large linear programming problems, it is inevitable that some mistakes will be made in formulating the model. Therefore, a thorough process of testing and refining the model (model validation) is needed. Model management encompasses a variety of activities including formulating the model, inputting the model into the computer, modifying the model, analyzing solutions from the model, and presenting results in the language of management. Packages commonly include a mathematical programming modeling language to efficiently generate the model from existing databases.

42. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.42 The Interior-Point Approach to Solving LPs Solution Concept: Interior-point algorithms shoot through the interior of the feasible region toward an optimal solution instead of taking a less direct path around the boundary of the feasible region.

43. McGraw-Hill/Irwin © The McGraw-Hill Companies, Inc., 2008 14.43 Interior-Point Algorithm in your MS Courseware