SOLUTIONS MANUAL Operations Research Applications and Algorithms Second Edition Introduction to Mathematical Programming Applications and Algorithms WAYNE L. WINSTON Indiana University & PWS-KENT PUBLISHING COMPANY # BOSTONPWS-KENT Fabiing omy AO Copyright (c) 1991 by PWS-KENT Publishing Company. All rights reserved. Instructors of classes using OPERATIONS RESEARCH: Applications and Algorithms, Seeond Edition or INTRODUCTION TO MATHEMATICAL PROGRAMMING: Applications and Algorithms by Wayne L. Winston may reproduce materials for classroom use. Otherwise, no part of this Solutions Manual may be reproduced, stored in a retrieval system, or transcribed, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written consent of the publisher, PWS-KENT Publishing Company, 20 Park Pla: Boston, Massachusetts O2116. Printed in the United States of america 123456789 -- 93 92 91 90CONTENTS Guide to Teaching a Course Emphasing Model- Building from Winston's OPERATIONS RESEARCH Guide to Prerequisites for each section. PART 1 CHAPTER 2 BASIC LINEAR ALGEBRA... sean iat ee ee CHAPTER 9 INTRODUCTION TO LINEAR PROGRAMMING... . CHAPTER 4 THE SIMPLEX ALGORITHM... 2*Section 4.14 is for AN INTRODUCTION TO MATHEMATICAL PROGRAMMING only. CHAPTER 5 SENSITIVITY ANALYSIS: AN APPLIED APPROACH: ++eseeereseeeee crs CHAPTER 6 SENSITIVITY ANALYSIS AND DUALITY. ¢+esee00191 CHAPTER 7 TRANSPORTATION, ASSIGNMENT, AND TRANSSHIPMENT PROBLEMS. CHAPTER 8 NETWORK MODELS.........0.00005 CHAPTER 8 INTEGER PROGRAMMING... . CHAPTER 10 ADVANCED TOPICS IN LINEAR aaseT, PROGRAMMING. see ere CHAPTER 11 GAME THEORY. sees eerereseeee nee B87 Note: GAME THEORY is Chapter 15 in OPERATIONS RESEARCH. For solutions to REVIEW OF CALCULUS AND PROBABILITY (O.R. Chapter 11), refer to page 455. CHAPTER 12 NONLINEAR PROGRAMMING....-2.-+-+ seeeres STH ##Section 12.1 is for AN INTRODUCTION TO MATHEMATICAL PROGRAMMING only. CHAPTER 13 DETERMINISTIC DYNAMIC PROGRAMMING. ........405 Note! DETERMINISTIC DYNAMIC PROGRAMMING is Chapter 20 in OPERATIONS RESEARCH. For Solutions to DECISION MAKING UNDER UNCERTAINTY (0.R. Chapter 13), refer to page 467. **Sections 20.7 and 20.8 are for OPERATIONS RESEARCH only. aunNote: CHAPTER ‘CHAPTER CHAPTER ‘CHAPTER, ‘CHAPTER, ‘CHAPTER, ‘CHAPTER, CHAPTER, ‘CHAPTER, CHAPTER, CHAPTER CHAPTER CHAPTER CHAPTER wn 12 13 14 15 18 17 18 19 20 aL 22 23 24 PART II The Solutions in Part II pertain to OPERATIONS RESEARCH only. REVIEW OF CALCULUS AND PROBABILITY. ..++0.+455 NONLINEAR PROGRAMMING (refer to Part 1)...377 DECISION MAKING UNDER UNCERTAINTY. ..+¢+24+467 DECISION MAKING WITH MULTIPLE OBJECTIVES... . 2.005 525 GAME THEORY (refer to Part I)....+ +7387 DETERMINISTIC EOQ INVENTORY MODELS... 547 PROBABILISTIC INVENTORY MODELS... . +559 RECENT DEVELOPME! SIN INVENTORY THEORY......- Seer aoe’ 575 MARKOV CHAINS. 581 DETERMINISTIC DYNAMIC PROGRAMMING (refer to Part I)..........4..405 PROBABILISTIC DYNAMIC PROGRAMMING. +605 QUEING THEORY... SIMULATION, erereee FORECASTING MODELS...+ tvGUIDE TO TEACHING A COURSE EMPHASIZING MODEL-BUILDING FROM WINSTON'S OPERATIONS RESEARCH My goal in writing this book was to cover everything anybody might want to teach, but I have written each section to be as independent as possible of ather parts of the books. If I were teaching a course that deemphasized algorithms and emphasized model-building, here's what I would cover(as time permits). CHAPTER 2-This chapter can be completely omitted if you are emphasizing models. If you do cover this chapter, however, you may want to tell students that Lotus 1-2-3 can multiply matrices(with the \DMM command) and invert matrices (with the /DMI command). ‘CHAPTER 3-I feel strongly that sections 3.1-3.3 must be covered in their entirety, Students will have difficulty understanding computer output and sensitivity analysis unless they have solved graphical examples of unbounded LPs, infeasible LP's and LP's with alternative optimal solutions. The other sections may be covered as desired. In a course emphasizing models, I would supplement the text examples by leading the class thru some of the more complicated formulaticns, and then have the class solve some of the harder formulations with LINDO. For example, in Section 3.9 I might lead the class thru Problem 6, and then assign Review Problem 29 or 39 as a computer problem. Sections 3.4-3.12 are virtually independent of one another, but certain problems in one section may depend on other sections. CHAPTER 4-Unlike some "revisionists", I still maintain that if you are going to cover sensitivity analyis, you need to spend one or two lectures on the simplex method. Here's how I suggest proceeding. Step 1-Cover section 4.1, defining slack and excess variables. Step 2-In Section 4.2 define basic solution and basic feasible solution(bfs). Then explain how corner points correspond to basic feasible solutions, and explain that any LP with an optimal solution will have an optimal bfs. Then go through the description of the simplex on page 132. Step 3-It's not necessary to work through an entire example of the simplex, but T feel that all students can benefit from seeing how at least one algorithm works. It's sufficient to show the students an optimal tableau, and for this tableau a. show how you find the basic variable for each constraint. b. Why a non-negative row 0 indicates an optimal tableau. ‘Then I would cover sections 4.7 and 4.8. Finally, I would cover Example 6 in Section 4.12. You might want to briefly discuss degeneracy(Section 4.9). CHAPTER 5-This chapter was written precisely for a model-building course which also emphasizes interpretation of computer output. I would cover Sections 5.1-5.3 in their entirety. If time permits, I would then skip to the 100% rule in Section 6.4. Although section 5.4 is non-standard I feel it really cements the student's understanding of how changes in a problem's parameters change the problem's optimal solution;I strongly recommend covering thisGUIDE TO TEACHING A COURSE EMPHASIZING MODEL-BUILDING FROM WINSTON'S OPERATIONS RESEARCH My goal in writing this book was to cover everything anybody might want to teach, but I have written each section to be as independent as possible of other parts of the books. If I were teaching a course that deemphasized algorithms and emphasized model-building, here's what I would cover(as time permits). CHAPTER 2-This chapter can be completely omitted if you are emphasizing models. If you do cover this chapter, however, you may want to tell students that Lotus 1-2-3 can multiply matrices (with the \OMM command) and invert matrices (with the /DMI command) - CHAPTER 3-I feel strongly that sections 3.1-3.3 must be covered in their entirety. Students will have difficulty understanding computer output and sensitivity analysis unless they have solved graphical examples of unbounded LPs, infeasible LP's and LP's with alternative optimal solutions. The other sections may be covered as desired. In a course emphasizing models, I would supplement tne text examples by leading the class thru some of the more complicated formulations, and then have the class solve some of the harder formulations with LINDO. For example, in Section 3.9 I might lead the class thru Problem 6, and then assign Review Problem 29 or 39 as a computer problem. Sections 3.4-3,12 are virtually independent of one another, but certain problems in one section may depend on other sections. CHAPTER 4-Unlike some "revisionists", I still maintain that if you are going to cover sensitivity analyis, you need to spend one or two lectures on the simplex method. Here's how I suggest proceeding. Step 1-Cover section 4.1, defining slack and excess variables. Step 2-In Section 4.2 define basic solution and basic feasible solution(bfs). Then explain how corner points correspond to basic feasible solutions, and explain that any LP with an optimal solution will have an optimal bfs. Then go through the description of the simplex on page 132. Step 3-It's not necessary to work through an entire example of the simplex, but I feel that all students can benefit from seeing hov at least one algorithm works. It's sufficient to show the students an optimal tableau, and for this tableau a, show how you find the basic variable for each constraint. b. Why a non-negative row 0 indicates an optimal tableau. Then I would cover sections 4.7 and 4.8. Finally, I would cover Example 6 in Section 4.12. You might want to briefly discuss degeneracy (Section 4.5). CHAPTER 5-This chapter was written precisely for a model-building course which also emphasizes interpretation of computer output. Tf would cover Sections 5,1-5.3 in their entirety. If time permits, I would then skip to the 100% rule in section 6.4. Although Section 5.4 is non-standard I feel it really cements the student's understanding of how changes in a problem's parameters change the problem's optimal solution;I strongly recommend covering thissection. CHAPTER 6-Omit it:that's why I wrote Chapter 5!! Chapter 6 is well- suited for majors in managment science or OR, who will be going on for further study of mathematical programming, but it is too theoretical for most MBA's and non-mangement-science-major- undergraduates. CHAPTER 7-For a model-building course I suggest the following sequence of topics. 1. Do Section 7.1 in its entirety. 2. Skip Sections 7.2-7.4 3. Define an assignment problem (Section 7.5) by setting up the Machineco example and explain that a specialized computer algorithm exists for this problem. You might also want to emphasize that certain computer packages require you to balance the problem before the computer solves it (set up Problem 1 to show this). 4. Do Section 7.6 in its entirety. If you are discussing minimun cost network flow problems, however, you might want to skip this section. CHAPTER 8-I've covered most network algorithms in this chapter, but in most sections you can completely omit the discussion of algorithms. If I were covering the chapter with a model building emphasis, here's how I would procee 1. Cover the definitions (omitting chain) in Section 8.1 2. In Section 8.2, define a shortest path problem by discussing Example 1 and 2. Example 2 really is worth doing! If your students have access to a package with a specialized shortest path algorithm(such as STORM, QSB, or CMMS) you can solve Example 2 on the computer. If student's do hot have access to such a package, you may want to explain how to formulate a shortest path problem as a transsshipment or transportation problen. 3. In Section 8.3, I would skip all the Ford-Fulkerson development in a model-building course. I would explain what's involved in a maximum flow problem by discussing Examples 3, 4, and 5. I would also discuss the LP formulation for Example 3. 4. When discussing CPM and PERT in Section 8.4, I feel that students do need to see the method for determining critical paths, ET, and LT to understand what's going on. It is possible, however, to skip directly from the discussion of how to set up an AOA network to the discussion of using LP to find the critical path and crash the project. PERT may be covered or omitted, as desired, but the weaknesses of PERT should be mentioned!! 5. Given the importance of MCNFP's in today's OR and MS applications, I would cover Section 8.5 in its entirety!! I would recommend discussing one of the harder homework probelms(either 6,7, or 8) in class to give students the flavor of real-life applications. 6. Section 8.6 should be covered in its entirety. Students love the intuitive MsT algorithm. 7. Omit Section 8.7. CHAPTER 9 vi1. Cover Section 9.1 2. Cover as much of Section 9.2 as you canistudents really seem to like this section. I'd consider the material on piecewise linear functions the least important material. I'd assign one of the trickier formulations to be run on LINDO(I prefer Problem 17 in Section 9.2) 3. Skip Section 9.3-9.5 and 9.7. 4. In Section 9.6 I'd define a traveling salesperson problem and explain that computer packages such as STORM and QsB will solve small ones. If you have time the, sub-section on heuristics is a nice introduction to the important topic of heuristics. CHAPTER 10-Skip it entirely!! CHAPTER 11-How you use this chapter depends on the mathematical background of your students and the topic you cover in other chapters. CHAPTER 12-Here is the sequence of ideas I would cover in a model- building course. If you are only interested in model-building, I Yeally don't think you need to emphasize any topics requiring a knowledge of calculus 1, Cover section 12.1 in its entirety. 2. Explain that in certain situations a local extremum is guaranteed to solve an NLP. Might want to discuss convex and concave functions of one variable and say that if objective function is concave a local maximum usually solves a max NLP and if objective function is convex a local minimum usually solves a min NLP. 3. Next I would concentrate on formulating NLP'S and explaining how GINO (Section 12.12) is used to solve NLP's. Make sure you tell students that GINO (or any other NLP package) may get the wrong answer to an NLP by finding a local extremum that does not solve the NLP.!! Here are some suggested problems that can be formulated in class to give students the flavor of NLP applications. If you teach in a business school, I strongly recommend discussing the portfolio selection example (Example 22) in Section 12.9. Section Problem Numbers 12.3 1,2,4,9 12.5 3,4,8,9 22.7 4.8 izes 2,5, and Example 22 12.9 1,4,7 and Example 22 22.12 1,2,3 Review Problems 2,3,7,10,16,18,19 CHAPTER 13 Unless you are a devotee of decision theory, you probably won't want to cover the whole chapter. Here are the high points I cover 1. Do Section 13.1 in its entirety(if you don't have time just discuss minimax and and expected value) . 2. Section 13.2 is a fairly complete discussion of utility. Iproceed as follows: a. Define lottery. b. Associate a utility of 1 with most favorable outcome, and utility 0 with least favorable outcome, and explain how to choose between lotteries Li through L4. sc. Explain that we choose between lotteries by comparing their expected utilities, d. Discuss section involving Estimating a Utility Function e. By doing Problems 1-3 you can illustrate that a concave utility function yields risk-averse behavior while a convex function yields risk-seeking behavior. i 3. You can cover as much of Section 13.3 as desired. If you wish, you can cover Section 13.3 before Section 13.2, and cover the portion of Section 13.3 involving expected utility with Section 13.2. 4, Caver as much of Section 13.4 as you wish. You really don't have to cover the discussion of Bayes Theorem in Chapter 11 before discussing this Section. 5. Section 13.5 is fairly advanced, and is primarily for professors whe emphasize decision-making under uncertainty. ‘CHAPTER 14 1. In Section 14.1 [ would limit my discussion to goal programming, completely omitting the goal programming simplex. Goal programming may be discussed at the end of Chapter 4, if so desired. I would definitely explain how to use LINDO to solve goal programming problems and possibly assign Problem 15 or Review Problem 6 as a computer exercise. 2. Section 14.2 is relatively advanced and is primarily for instructors emphasizing decision-making under uncertainty. 3. I would cover Section 14.3! students love the AHP, and it is becoming more widely used each day. It is easy to implement the approximate method for determining the weights by using any spreadsheet. You can ask students to solve a decision-making problem from their own experience;they really enjoy this! Although I have used matrix notation in this section you can get around it by simply writing out the equations in (12). CHAPTER 15 Assuming that you have not covered the concepts of duality in Chapter 6, here is how I would proceed: 1, Cover Sections 15.1 and 15.2 in their entirety 2. In Section 15.3 I would derive the row and column player's problem for Stone-Paper Scissors, and explain that row and column Player problems will always have same optimal objective function value. I would also solve Morra with LINDO(you can use the FREE command and not worry about negative variables), to further illustrate the row and column player problems. If time permits I would cover Homework Problem @. 3. Sections 15.4-15.7 are all written from a model-building Perspective, so I would cover them as is. CHAPTER 16 Only sections 16.1-16.2 are really necessary. I recommend vidcovering all of Section 16.2, if possible. If time permits I would sover quantity discounts (Section 16.3), but I would omit the rest of the chapter unless you are teaching a course emphasizing inventory or operations management. In Section 16.3 you might want to omit the derivation of the quantity discount algorithm;perhaps just state it and do an example. CHAPTER 17 I recommend covering the newsperson problem(Section 17.1- 17.4). I believe the idea of marginal analysis’ is important in a model-building course, I would omit section 17.5. In Section 17.6- 17.7 you can proceed two ways. First, you may go through the derivations of safety stock formulas and then work the examples. This requires a rather mathematically sophisticated class. Second, you may begin with the definitions on pages 955-857, and then skip te an example of the safety stock formulas. If time pernits, I would discuss the two-bin and (s,S) policies. Then I would define the two types of service level(Example 6 helps here), and do an example of how to compute SIM1 and SLM2. Section 17.8 should be omitted unless your primary emphasis is inventory or operations management. Section 17.9 should definitely be coveredrit just takes a few minutes. CHAPTER 18 The three sections in this chapter are virtually independent, and may be covered as desired, MRP(Section 18.1) and JIT(Section 18.2)require an understanding of the E0Q(Section 16.2). The discussion of exchange curves requires that Sections 16.2 and 17.7 have been covered. Hopefully, this chapter will include topics helpful to instructors who emphasize inventory or operations managment. CHAPTER 19 If you like Markov chains, I believe the whole chapter is worthwhile. For a models course Section 19.7 is well worth your while. Matrix notation is used throughout the chapter, and students need to understand how to multiply and invert matrices (see Sections 2.1 and 2.5). If you want to give come of the problans as computer problems, students can multiply and invert matrices using Lotus 123(see Chapter 2 discussion). Of course, Lotus can algo be used to solve the steady state probability equations. You can also use LINDO(or any other LP package) to solve a system of linear equations. If you want to solve Ax = b,simply tell LINDO to solve the following LP: max 0x st Ax =D CHAPTERS 20 and 21 In an introductory models course, I would omit these chapters. Students usually find DP difficult, and I believe the students can Profit more if you cover other topics. For students majoring in quantitative methods, however, the chapters are important. DP is one topic (unlike LP)where I feel that students won't understand how to formate problems unless they understand the underlyingalgorithms. Therefore, you need to work thru some numerical examples(probably at least two from the network, inventory, resource allocation, or equipment replacement examples) before discussing the many tricky formulations. If you do cover Markey decision processes, however, you should tell students that value determination equations can be solved with Lotus 123 or LINDO (see Chapter 19 discussion for details). CHAPTER 22 Here's how I would approach queuing theory in a models- oriented course. 1, I would cover Sections 22.1 and 22.2 in their entirety. 2. In Section 22.3 I would begin by defining steady state probabilities(this is easy for students who have covered Markov Chains in Chapter 19). Then I would explain that when interarrival and service times are exponential, the rate at which transitions occur is all that matters. Next I vould derive the steady state equations for birth-death processes by explaining that for each state (transitions/unit time) out of state = (transitions/unit time) into state. ‘Then I would explain birth-death equations are solved for steady state probababilities. py the way, if you input the birth and death Parameters, it's easy to program Lotus to solve for the steady state probabilities. 3. From this point on you can cover any sections you desire. Certainly Sections 22.4 and 22.6 must be covered. I usually cover all Sections except for 22.9, 22.13 and 22.14. CHAPTER 23 For an intreduction to Monte carlo simulation I proceed as follows: 1. Define random numbers(Section 23.3) 2. Illustrate Monte Carlo simulation with Example 1 of Section 23.4 and Problem 1 of Section 23.4(a queuing example. Both these simulations are a snap on Lotus 123. Generate U(0,1) random numbers with the @RAND command. Then generate random variables using @IF or QVLOOKUP statements. Note that each time you recalculate a Spreadsheet, your random numbers(and lots of other stuff) will change. If you don't like this you can "freeze" the values of the random numbers with the /RV command. 3. In Section 23.5 I would explain how to simulate random variables from an exponential and normal distribution. 4. I would skip the rest of the chapter. 5. If you covered PERT in Chapter 8, you might want to simulate @ project network in an attempt to find the probability that each activity is on the critical path. CHAPTER 24 If you are going to have your students solve any real-life Problems, such as developing a queuing model for a bank or an inventory model for a store or hospital, then your students will need some knowledge of forecasting. I would cover the whole chapter if my students were working on real-life projects. If you discussextrapolation methods, I would definitely show students how to do smoothing methods on a spreadsheet. If you demonstrate graphing , then even students who have never seen a spreadsheet will be "Lotus literate" after doing smoothing methods with Lotus. The regression coverage(Sections 24.6-24.8) is probably less important than the extrapolation material, but if you cover regression you really should do Section 24.7; most functional relationships are not linear. Section 24.5 is very useful if you are forecasting demand as an input to a queuing model. My class used this section to forecast hourly customer counts at the Indiana University Credit Union and obtained MAD<8% of mean demand. To implement the methods of Section 24.5 on a spreadsheet, your students will need to use the @DAVG command to obtain monthly and daily averages and the @HLOOKUP or @VLOOKUP command to generate forecasts. Of course, you can do regressions on lotus with the \DMR command. xiPREREQUISITES FOR EACH SECTION In this section of the instructor's manual, we list the “immediate predecessors" needed to cover each section of the cexe Section Inmediate Predecessors None Section 2 Section 2 Section 2 Section 2 & 2 involving rank of matrix Section None Sectlon 3 Section 3 Section 3 Section 3 Section 3 Section 3 Section 3 Section 3 Section 3 Section 3 Section 3, 3 3 3 3 4 4 4 4 4 i 2 1 +3 and portion of Section 2.4 Fi L (some problems require 5 sone probleme require Seecion Seecion Section Section Section bue nae Section Section Section SScke essary) Section 4.4 Section 4.4 Section 4.4(1f you ignore references to Big-M method) Section 4.4 Section 4.4 Section 4.3 Section Section Section Section Sections 2.1 and 2.5 Sections 6.1 and 6.2 Section 6.3 or Section 5.2 Section 4.4 Sections 6.3 and 6.5 Sections 6.2, 6.5, and 4.10 Sections 6.1 and 6.7 perepeee Hoe Vane S 5 or Section 6.8 bene xiiImmediate Predecessors Sections 6.3 and 6.8 Sections 6,3 and 6.9 Sections 6.5 and 4.10 Seetiens 3.3 and 3.10 Section 4,4 and concept of rank from Section 2.4 Section 7.2 and Section 6.2 or Section 5.3 Sections 6.3 and 7.3 Seetion 7.1 Section 7.1 . None Sections 8.1 and 7.6 Sections 3.2 and Seetion 8.1 Sections 8.1 and Section 5.3 er Section 6.8. Also knowing how co read z-table in Section 11.7 Sections 8.1-8.4 Section 8.1 Sections 8.5 and 6.2 Section 4.3 Section 3.3 Section 4.4 Section 9.3 Section 9,3 Sections 9.3 and 7.5 Section 9.3 Section 9.2 Sections 4.4 and 6.11 Sections 4.4, 2.1, and 2.5 Section 10.1 Section 10.2 Section 10,3 Section 4.4 Sections 2.1-2.3 and Section 2.5 None Section 11,1 Section 11.2 None Section 11.4 Sections 11,2 and 11.4 Section 11.6 Sections 3.3 and 4.2 Sections 12.1, 11.1, and 2.6 Sections 11.1 and 12.1 Section 12,3 Section 12.2 Section 12.5 Section 12 Section 12 Section 12. Section 12 Section 12Section 12.12 13.2 13.2 13,3 13.8 13.5 Laid 14.2 14.3 15.1 13.2 15.3 15.4 15.5 15.6 15.7 16.1 16.2 16.3 18.4 16.5 16.6 17.1 17.2 17.3 value from Section 11.6 17.4 17.5 17.6 12.7 17.8 17.9 18.1 18.2 18.3 17.7 19.1 19.2 19.3 19.4 19.5 19.6 19.7 20.1 20.2 20.3 20.4 20.5 20.6 Innediate Predecessors Section 12.1 Section 11.6 Section 11.6 Section 1}.1(portion on expected value) Section 13.3 and possibly Section 11.5 Section 13.2 Section 4.12 Section 13.2 None, although Sections help None Section 15.1 Sections 4.3, Section 15.1 Section 15.1 Section 15.5 Section 15.6 None Section 16.1 Section 16.2 Section 16.2 Section 16.4 Section 16.2 None Section Section Q.1-2.2 will 15.2 and 6.7 17.1 17.2 and definition of expected Section 17.3 and how to read z-tables from Section 11.7 Seetion 11.2 Section 17.3, and how to 11.7 Section Section Section Section Section Section Section 16.2,Section 11.6, read z-tables from Seetion 17.6 17.6 16.2 and Section 17.6 16.2 16.2 16.2, Section 17,6, and Section None Section Section Section Section Section Section None Nene Section Section Section Section igut 19.2 and Secefon 2.1 19.3 19.4 and Section 2.2 19.5 and Section 2.5 19.5 20.2 20.2 20.2 20.2 xivSection 20.7 20.8 21.1 21.2 21.3 21.4 21.5 22.1 22.2 22.3 22.4 22.5 22.6 22.7 22.8 22.9 22.10 22.11 22.12 22.13 22,24 23.1 23.2 23.3 23.4 23.5 23.6 23.7 23.8 23.9 26.1 24.2 24.3 24.6 24.5 24.6 24.7 24.8 Immediate Predecessors Section Section Section Section Section Section Seetion None 20,3 20.3 20.4 20.3 20.2 20.2 and Section 21.2 21.2 Sections 22.1, 11,2 and 11.6 Section Series) Section Section Section Section Section Section Section Section Section Section Section None 22.2 and Section 11.1(Taylor 22.3 22.4 22.4 22.6 22a 22.6 22.6 22.6 22.6 22.12 22.6 Sections 23.1 and 22.4 Sections 23.2 and 11.6 Section Section Section Section Section Section Nene Seetion Section Section No: None Sectlon Section 23.3 23.4 23.5 23.6 23.4 23.4 26.1 24.2 20.3PART I CHAPTER 2 Basic Linear Algebra INTRODUCTION TO Mathematical Programming Applications and Algorithms and —_'”’”= Operations Research Applications and AlgorithmsChapter 2 solutions Section 2.1 <1 -2 -3 -4 -5 =6 <7 -8 -9 2 6 9 = fz 15 18 21 24 27 e le. A+2B is undefined, i 2 3 1 ij BT “102 a6 if. AB= /10 15 la. -A= Ae ab. 3, wax aa. aT od o aus 1s 24 ig. BA is undefined. 2. Ty, 750 0.10 % y,|= | -30 .70 .30 x Ys +20 .30 .60 % 3. det A = (ays), B= (Bis), and ¢ = (c,). We must show that (ac) = (apyc. “he i-j/th dfeent of a(sc) {ds given by ZF ay Ewe, be = Babi x ix’) xy®ys x oy 2 *¥oyd The i-j/th element of (aB)c is given by = ¥ Thus A(BC) = (AB)C (Bab doy = z x tx xy? yh ¥ = 5 Babe xy ex YS 4. The ivj/th element of (AB)” = element j-i of AB = Scalar product of row j of A with column i of B. Now element i-j of B"A™ = scalar product of row i of BT with column j of A” = Scalar product of row j of A with column i of B. Thus (apyT = ptat_Section 2.2 Solutions fi -v a] 4 iad z 1 =fe|ana]2 1 6 1 3] |x, 8 1 38 Section 2.3 Solutions 101/35 Liogip zoo -2 1-1 1. o1l1o4 oli of4 ik 10 4 1211/8 911 ofS o o 0 1 The “last row of the last matrix indicates that the original system has no solutien. * [Pra] Bias] & 24/3] This system has an infinite number of solutions of the form 1 x, 7k, ¥) = 2- 2k, x, 2+ ke Z apa) fa aya o 4-1 o 1f-2 2 -1} as Lo of o pt kei bon 1 i ayji Tayi 3 o-1}1 o-1fa 4 2.42 o-a} 4 This system has the unique solution x, = 2-1 1 146 naz a2 as2\3) [i -1/2 a2 1y2]3 rrr ofel [2a Cela] [oof ye alt 1-1/2 1/2 1/2| 3 1 oo 23 1/3 \10/3" 0 2 aya -173|2/3 o 1 3 -1/3 | 2/3 This system has an infinite number of solutions of the form =k, X) = 10/3 = 20/3 - k/3, x, =2/3 - o/3 + k/3. 1 wea final matrix indicates that the original system has no solution. 6.fo 2 2)4) fi 2 1/4] fi 2 1/4] fi o -1/0 2 2 afa}fo 2 2]al/fo 1 af2}ffo a aja o 1-1jo}fo 1-1 o i1-1jo/jo a -1Jo I o-21/0 o-1/07/1 0 oy. oo o 1 af2}fo a af2{fo 1 afz o1of|r o o-2--2)[o o 1Jajlo o afi aoa. Thus original system has a unique solution x, = x, = 1m [4 1 of2 1 oy27fi o 2) s)fa o ass o-1 2/3 0 1 -2/-3 | Jo 2 -2/-3//0 1 -2/-3 o 11/3 o 1 if3slo 1 ilaflo a ale10 2} 8 2 0 9/1] 0 of: a 1-2/-3| | 6 1 -2h3ffo 1 ofa © 0 af2]} loo afafio o asa Thus the original system has the unique solution x, = 1, xy = 1, % 82. a. 22 ojr 20-2 -2f2 o12ij2 o1 2 if2 000114 oo 0 J We now must pass over the third coltimn, because it has no nonzero element below row 2. We now work on column 4 and obtain 1o-10/2 o-1 oj 2 o1 21 { 2}fo21 2 0/-1 oo o1!3{(oo o1/ 3 We find that “the original system has an infinite number of solutions of the form form x, = 2+ k, x, = -1- 2k, x5 = ky x, =3. 3. In this situation there must be at least one nonbasic variable, Thus N is nonempty and Case 2 cannot apply. Thus the eriginal system cannot have a unique solution. Lor o1 Loiy fio. 1 1. t21 20 020 ol1o oO 222 22 o2 | o20 oO Row of 0's indicates that v is linearly dependent. 2. ilo 11/20 Lis2ojpfl 1/72 0 12 1/20 120 120 o3/2a o 3/20 oO 1a aaa \(3 3 1jf3 3 af/o a72af/o sven 10 ° o = o1 of foro oav2L oO The'rank of“the‘last matrix is 3,so V is linearly independent. (i | 12 22) fl ya o 12 G 2 0 3/2 {6 4. o1 Since rank of last matrix -2,V is linearly independent. “aH Bay Since rank of last matrix = 1,V is linearly dependent. This also follows from “EL Section 2.4 Solutions onl ° i o Ne ° ere17 [41 BT fe since 2h] 5]- f as v is linearly dependent. 3) Ls] fo 0 las This also follews from the fact that the matrix i 5 7 2069 has rank 2. 1 Low o 4 ° 0 1 0/ fo 10 1 o21aflfoo. = last matrix has rank 3, V is linearly independent. ° & peo Since th 7. If Ausb has a solution x= then bee Aa a2 San a. Thus if A=b has a solution, then b is a linear combination of the columns of A with x; being the weight of column i in the linear combination. Similarly, if b is a linear combination of the columns of A with weights x,,%),.-.%,, respectively then the vector aE :” |is a solution to the linear system aAx=b = 8. Let v=(v¥,,v,,-.-¥)) be any collection of m (m>3) two- dimensional* r8y vecBors, Now let Row 1 is vy A= | Row 2 is vy Row » is y, Since the Gass-Jordan Method is complete after the second column is transformed, rank A<2. Thus rank A cannot equal m{remember m>3). ‘This means that V cannot be a linearly independent set of vectors, 3. If V is linearly dependent, then c.v. + 2%, +... +e = &) where at least one c; (say c.)is nantero™ piviaing thrOudn by Sar Me Rew find that v, ‘may be dritten as a iinear Coutoattin te the other vectors in v Suppose v. may be written as a linear combination of the other vectors tn v. Then we know that et = Q%, te tow, Rearranging this equation we obtain v, 3 ¢ vo =... - evi Bg, This shovs that V is a linearly dependint s8t?og vectors? ® Section 2.5 Solutions 2 3/1 0 Tos a. | 2 slor o-1 100 1 o 2210 oa oo 1 33 lj. =s fs “43 lj. oo 2900 6 |=4 1 0 //o 1-6 1ji/2 172 172 oo 1 a2 1f2 -1sa “1 -2 3 2-172 a2 72 12-172 1 3 -2 2-1/2 1y2 1/299)(1o01/ 100 Peis 9’ 2/92 9l/o10/-110] forol-iiro6 aloo 1}( 222} 001] for ol-202 \Heo Gan V2y2 -1/2 1 0 a2 -1/2 1/2 oon] fund who12 1j100 L2ay1aedg a \fe efits 2 ojo 1 1 2 0/0 1 @& o oO 1}1-2 ® 1 6)1-1 0 0 O=-1-2 0 Oo 0 Of =2 21 * —| Again, Av} does not exist. ee 5 : 7 fal: fll tel fis hE ¥ xy 1/2 af2 -1/2Vf47 fi Thus! x,]= 2 3 fol-j2 x 12 -1/2 1/2/)2) (3 7. Suppose A is an mxa matrix .If rows of A are linearly independent, then rank A = m and we can perform ero’s on A that yield the identity matrix. This means that the Gauss-Jordan method for finding the inverse will not yield a row of o’s. Thus if rows of A are linearly independent, then A will have an inverse. If the rows of A are not linearly independent, then rank A < m, and ero’s on A cannot yield the identity matrix. Hence if rows of A are not linearly independent,then A will have no inverse. @a. Since 100B(3"1/100) = I, we have that (oop) “> 8b. Multiply each entry in column 1 of B = 8/100 by 1/2. Qe. Multiply each entry in row 1 of B 9. (BUA )aB = I and (AB) (BUA) = -I,-1 by 1/2. (apy) = 10. We know that 1 q) aa ler. By Problem 4 of Section 2.1 taking the transpose of both sides of (1) yieldswt TT a Ter Thus the inverse ot aT is (a7tyT il. For ifj, the scalar product of row i of A with row j of A must equal 0. This implies, by the vay, that tne vector aespciated with row i is orthogonal, or perpendicular, te the vector associated with row 4. Air for each row i, the sealar product of row i with itself must equal 1. This means that the “length” of each row cf A (when viewed as a vector) must equal 1 Section 2.6 Solutions 1. Expansion by row 2 yields (-292°2 (4) (6) + (-1)?*8¢5) (a2) + (-2)246 (26) = 0 Expansion by row 3 yields (ay 79 (aye (2)? ay (6) + (2534389) 3) = 0 2. By row 2 cofactors we obtain ° ° 2det |o 3 0 oo 5]. 90 o But det fo 3 o|= det I-15 oo 5 os Thus the determinant of the original natrix is 30. 3. Any upper triangular matrix a may’ be written as a- [og 34 ao ef. Expanding by row 3 cofactors we find det a= ai | = adf = product of diagonal entries of A 4a.and 4b. For any 1x1 matrix det -a = . = 7 det A. For any 202 matrix SSE Mt aaa toagg) (ng (agg) = at a. Por any aut Pee 23) (7429 aa) (rea, ag "29 "23 det (- a) = -a,, det can, wa, {* a2 eet | i 32 “833 Se Ty a, det} 22 “22 a na, a 7 31 “832'22 923 821 323 det A= ay) det = aj, det 33 31 33 21 222 ay, det 2 "32 Since det A = det (-a) holds for 2x2 matrices we find that for any 3x3 matrix det (-A) = -det A. A similar argument works for 4x4 matrices to show that det (-A) = det A. 4c. For any nxn matrix where n is even, det (-A) = det A, while if n is odd det (-A) = -det A. Solutions to Review Problems 1@2> 15 o!|2 Lo. oj2 Oo -1 j-2 o 1a |3 o 11/3 2 1/3 12145 o 1 1/3 LTS 1 0-1 Od. 8 "d |e now find that x, =k, xp <1 +m, my 9k oo o|o s wel bu o G ie a alo 4] 1 ata 0] lo 3/2 “o eal A ohne 4] 3 1) 173 “0 Thus i “ye a2 1730 e Tera ae 3 3/2 372 Le " ae PVC) (8 a8 Lee 2 2 3/2) 3/2) 7 ca yaya o F “1/2 HA] [3 2 Sak a{a 12 HA] [2 “| 0 aye LG olo Thus x, = 0 and x, = 1 is the unique solution “Bas Ten DeTe ge [32 3) =3/2 an _s ee oe ‘Thus = ee) 2 °6. [Gray 3.6 3-8 2.6 3.4 aut asus 2.7.3.1 2.9 3.6 3/14 3/14 | 12 jae 1/2 | 3/2 4 ry? [aga ls i t owe ef 0-1/2 |-103 Thus the unique solution is x) = 1, x, = 1. “G a B 2] ya ] G2 yet) 1/2 {-3/2 1 pees) 6 17 32 27 5 = Thus 5 305 “32 9: pag = -4%p and AL, = -05C, + T97A,. Using matrix notation we obtdin Ce aa _ pe *e re 605.97 fi ie 4 an Et -ajé 2 i Es 00 af 3 The row © |-1] indicates that the original equation systen has no solution. fee ad eer HS ge ore ° ul. f2 0 271 0 1 0 2/1 0 o © 1 0Jo 1 0 oasis 9 1 ifo o 1\Jo o 1fo-1 1 io oJ/2 2 0 1 ofo i 9 9 if/o-1 1 ay lo 2 1 2 -2 tus [o 1 0{ = /01 0 91a. O-1 1 10a2. 19K, + .2C, and C.,, = .1R, + .8C,. Using matrix EESA we’ obtain’ “Tt ey % =) of Bs 13. 21 tsb 0 o 20 0 0-4 1 yal fot ave The last matrix has r: a 2, thus the Sriginal set of vectors is linearly independent. 14.f. o o 6 fo 2 o 1 0 ota [3 a3 Lo oo e last aat?ix has Fank-achs ofthe eriginal set of vectors ta linearly dependent. 15, nly if a, by © and 4 are all non-zero will rank A=4. Thus exists if and only if all of a,b,c, and d are non-zero. usb. Applying the Gauss-Jordan Method we find that if a, b, c, and d are all non-zero > " cook o coke e ° 1s. ecre HonH Her ‘ Hore ORKO bore weow bEeo Hawa HUN loo or\ ooo Goon coon Hees ee HHeo HER ones coor coon ‘coon Grow Coho Hone Hobh Heer SUPP burs rude oNGD Hroe oft onHo oorrn ne a eros PROF oFHO ono cone Thus x, = c, x, = 1+ c,x,=1-¢, x, = 3 - c yields an infinite number of solutions to the original system of equations. 17. Let sestate tax paid.f=federal tax paid, and b=bonus paid to WLRust satisfy the following system of +05 (60,000 - ¢ = 5) S = .05(60,000 - b) f= .40(60,000 - b - s) 18. Expanding by row 2 cofactors we obtain 2+1 at) (72) °F" (a) det = 4 92 19. Let ab Ae ed be a 2x2 matrix that does not have an inverse. rensnt Ce i Serndpz. or then det A = 0. Now assume at least one Slement of A is nonzero (for the sake of definiteness ascune thee ne ek * Zo (hat is, it ad - bexo)then in applying ero’s fo, the matrix AlI,, A is ‘transformed into the identity and ao vill have an inverbe. However, if det A= 0, then the tecens ron Pe A will be transformed into (0 0], and A will not have on inverse. comeance Tank A = m and the system has m variables, N will be empty and the unique solution to Ax = 0 will be Hk. “x50. 20b. If rank A < m,the Gauss—Jordan Method will yield at least ore nro Of O18 on the bottom (with the right hand side for each of these equations still being 0). Thus N will ba non-empty and we will be in case 3 (an infinite number of solutions). 21. The given system may be written in the form Ax = 0 where 2 Pa “Par +++ “Pay A=] "Pig 2 Pap -e- -Ppp Pin “Pan sts 27 Pay Now use ero’s as follows:Replace row 1 of A by {row 1 of A) + (row 2 of A). Then replace the new row 1 by (new Be atea enero 2d, Continue in this fashion until you nave Feplaced the current row 1 by (current row 1) + (row n)« the resulting matrix will have the following as its first row: EPyy Pag +e Pin 1-P23-P a9” Pan 2a Pag™--“Pnn! 2the original system) has an infinite number of solution: steel cars machines steel 30045, 40 22a, A= cars 15.20 +10 machines | .40 = .10 245 Since (Total Steel Produced) = (Steel Consumed) + (Steel used to produce steel, cars and machines), s=d+ + .a5c + .4m, Similarly ve find that ema, + 158 + .20c + 10m and m= di + 405 + 106+ 145m, In matrix form we obtain the following linear system ‘3s a, 730.45 40) fs el= [a] +] 15 .20 .10}/e m a, +40 10.45 [] om = ‘3 3s 22c. Just eel as I fet and subtract A|c from a m m both sides of the answer to 22b. 22d. From our answer to 22c. we find that 8 a, “1 ce {= (1a) a, Fy al If 520,c20, and m20, then Seriland can meet the required demands. If any of s, ¢, and m are negative,then Seriland cannot. meet the required demands. 22e. Before we increase the amount of required steel by $1, s ay 1 ¢ |= (I-A) a, m a,| + After increasing amount of required steel by 1, 13a a a4 ‘s 1 =(1-a)7 i: = original | ¢ | + (r-ay74/ o a, 2 ° 8 = Original c + (First column of (1-ay~?) , cy Thus increasing steel requirements by $1 increases demand for steel by element 1-1 of (I-A) |, increases demand for eara by element 2-1 of (I-A) ", and increases demand for machines by element 3-1 of (r-ay"?. u