(FRSHR » SSR) —————— 5 ee a a folate agroiim 730177 S oO mR es i sFrank R. Giordano, Maurice D. Weir, and William P, Fox: A First Course in Mathematical Modeling, Third Edition (ISBN: 0-534-38428-5). Original edition copyright © 2003 Brooks/Cole, a division of Thomson Learning, Inc. First published by Brooks/Cole, an imprint of Thomson Learning, United States of America. All rights reserved. Reprinted for the People’s Republic of China by Thomson Asia Pte Ltd and China Machine Press under the authorization of Thomson Learning. No part of this book may be reproduced in any form without the express written permission of Thomson Learning Asia and China Machine Press. RAS BEC RE AI Th A AR UES BLT A EE AH A A RS BL, FABLE oly SE hth A 6 MELT A, RLF. ABMIBIAS: BF: 01-2003-2624 HS MME (CIP) ie MOPAR AL (BEIM - SBME) (98) TELA (Giordano, FR.) 44. IR: HLTA uN, 2003.5, CSR LB EE ‘BHIRX: A First Course in Mathematical Modeling, Third Edition ISBN: 7-111-12154-6 Tote Si ML See BE IV. O22 PTI FR $5 fICIP AEF (2003 ) 990360495, SUB. AH ERE tes mano er Be A20 FRAN 100037 9 eR: Hite ACSC ENB RAS SLED «BRE BIR SCR RTT BT 2003 4F5 54381 RRS YEU 787mm x 1092mm 1/16 - 3543 NM: 0001-20008 FE: 59.007 CDE ) PUA AS, SOUFFLE BTR BRT AHL eT AeDedicated to our mentors, colleagues, and students who continue to motivate mathematical modeling in our lives, our teaching, and our research, especially 10 David C. Cameron Teacher, Scholar, Soldier, Mentor, and Friend A teacher affects eternity. He knows not where his influence stops. ——AdamsPreface To facilitate an early initiation of the modeling experience, the first edition of this text was designed to be taught concurrently or immediately after an introductory business or engineering calculus course. In the second edition, we added chap- ters treating discrete dynamical systems, linear programming and numerical search ‘methods, and an introduction to probabilistic modeling. Additionally, we expanded our introduction of simulation. In this edition we have included solution methods to some simple dynamical systems to reveal their long-term behavior. We have also added basic numerical solution methods to the chapters covering modeling with differential equations. The text has been reorganized into two parts: Part One, Discrete Modeling (Chapters 1-8), and Part Two, Continuous Modeling (Chapters 9-12). This organizational structure allows for teaching an entire modeling course based on Part One and which does not require the calculus, Part Two then ad- dresses continuous models based on optimization and differential equations which can be presented concurrently with freshman calculus. The text gives students an ‘opportunity to cover all phases of the mathematical modeling process, The new CD-ROM accompanying the text contains software, additional modeling scenarios and projects, and a link to past problems from the Mathematical Contest in Mod- cling. We thank Sol Garfunkel and the COMAP staff for preparing the CD and for their support of modeling activities that we refer to under Resource Materials below. Goals and Orientation ‘The course continues to be a bridge between the study of mathematics and the ap- plications of mathematics to various fields. The course affords the student an carly opportunity to see how the pieces of an applied problem fit together. The student investigates meaningful and practical problems chosen from common experiences ‘encompassing many academic disciplines, including the mathematical sciences, op- erations research, engineering, and the management and life sciences. This text provides an introduction to the entire modeling process. The student will have occasions to practice the following facets of modeling and enhance their problem-solving capabilities 1, Creative and Empirical Model Construction: Given a real-world scenario, the student leams to identify a problem, make assumptions and collect data, propose ‘a model, test the assumptions, refine the model as necessary. fit the model to data if appropriate, and analyze the underlying mathematical structure of the mode! to appraise the sensitivity of the conclusions when the assumptions are not precisely met,Preface 2. Model Analysis: Given a model, the student learns to work backward to uncover the implicit underlying assumptions, assess critically how well those as- sumptions fit the scenario at hand, and estimate the sensitivity of the conclusions when the assumptions are not precisely met. 3. Model Research: The student investigates a specific area to gain a deeper understanding of some behavior and learns to use what has already been created or discovered. Student Background and Course Content Because our desire is to initiate the modeling experience as early as possible in the student's program, the only prerequisite for Chapters 9, 10, and 11 is a basic understanding of single-variable differential and integral calculus. Although some ‘unfamiliar mathematical ideas are taught as part of the modeling process, the em- phasis is on using mathematics already known by the students after completing high school. ‘This emphasis is especially true in Part One. The modeling course ‘will then motivate students to study the more advanced courses such as linear alge- bra, differential equations, optimization and linear programming, numerical analy- sis, probability, and statistics. The power and utility ofthese subjects are intimated throughout the text. Further, the scenarios and problems in the text are not designed for the appli- cation of a particular mathematical technique. Instead, they demand thoughtful in- genuity in using fundamental concepts to find reasonable solutions to “open-ended” problems, Certain mathematical techniques (such as Monte Carlo simulation, curve fitting, and dimensional analysis) are presented because often they are not formally covered at the undergraduate level. Instructors should find great flexibility in adapt- ing the text to meet the particular needs of students through the problem assign- ‘ments and student projects. We have used this material to teach courses to both undergraduate and graduate students, and even as a basis for faculty seminars. Organization of the Text ‘The organization of the text is best understood with the aid of Figure 1. The first eight chapters constitute Part One and require only precalculus mathematics as a prerequisite. We begin with the idea of modeling change using simple finite dif- ference equations. This approach is quite intuitive to the student and provides us with several concrete models to support our discussion of the modeling process in Chapter 2. There we classify models, analyze the modeling process, and construct several proportionality models or submodels which are then revisited in the next ‘wo chapters. In Chapter 3 the student is presented with three criteria for fitting a specific curve-type to a collected data set, with emphasis on the least-squares cri- terion. Chapter 4 addresses the problem of capturing the trend of a collected set of data. In this empirical construction process, we begin with fitting simple one- term models approximating collected data sets and progress to more sophisticated interpolating models, including polynomial smoothing models and cubic splines.Preface xi ‘Simulation models are discussed in Chapter 5. An empirical model is fit to some collected data, and then Monte Carlo simulation is used to duplicate the behavior being investigated, The presentation motivates the eventual study of probability and statistics. Chapter 6 provides an introduction to probabilistic modeling. The topics of Markov processes, reliability, and linear regression are introduced, building on sce- narios and analysis presented previously. Chapter 7 addresses the issue of finding the best-fitting model using the other two criteria presented in Chapter 3. Linear programming is the method used for finding the “best” model for one of the criteria, and numerical search techniques can be used for the other. The chapter concludes ‘with an introduction to numerical search methods including the dichotomous and golden section methods. Part One ends with Chapter 8, which is devoted to dimen- sional analysis, atopic of great importance inthe physical sciences and engineering ® © ® Experimental |_| ~~ Simulation |_| Dimensional Analysis Modeling Modeling and Models of Similtude © Probabilistic Modeling ®@ The Moding Proce, Linear Programming Proportonaliy, and and Numerical Geomeric Silay Search Methods OQ QO Part Two* Model Graphs of Functions Fiuing as Models @ ® ® Modeing witha || Modeting with || Continuous DteremilEquon Systems of Opimication [Ditereil Equations Models "Part Two requires single-variable calculus asa corequisite Figure 1 (Chapter organization and progressionxii Preface Part ‘Iwo is dedicated to the study of continuous models. Chapter 9 treats the construction of continuous graphical models and explores the sensitivity of the injodels constructed to the assumptions underlying them. In Chapters 10 and 11 we ‘model dynamic (time varying) scenarios. These chapters build on the discrete anal- ysis presented in Chapter 1 by now considering situations where time is varying continuously. Chapter 12 is devoted to the study of continuous optimization. Stu- dents get the opportunity to solve continuous optimization problems requiring only the application of elementary calculus and are introduced to constrained optimiza- tion problems as well Student Projects Student projects are an essential part of any modeling course. This text includes projects in creative and empirical model construction, model analysis, and model research. Thus we recommend a course consisting of a mixture of projects in all three facets of modeling. These projects are most instructive if they address sce- narios that have no unique solution. Some projects should include real data that, students are either given or can readily collect. A combination of individual and {Broup projects can also be valuable. Individual projects are appropriate in those Parts of the course in which the instructor wishes to emphasize the development of individual modeling skills. However, the inclusion of a group project carly in the course gives students the exhilaration of a “brainstorming” session. A variety of projects is suggested in the text, such as constructing models for various scenar- ios, completing UMAP! modules, or researching a model presented as an example in the text or class. It is valuable for each student to receive a mixture of projects requiring either model construction, model analysis, or model research for variety and confidence building throughout the course. Students might also choose to de- velop a model in a scenario of particular interest, or analyze a model presented in another course, We recommend five to eight short projects in a typical modeling course, Detailed suggestions on how student projects can be assigned and used are included in the Instructor's Manual that accompany this text. In terms of the number of scenarios covered throughout the course, as well as the number of homework problems and projects assigned, we have found it better to pursue a few that are developed carefully and completely. We have provided many ‘more problems and projects than can reasonably be assigned to allow for a wide selection covering many different application areas. The Role of Computation Although many chapters of the text do not require computing capability,” compu- tation does play an important role in any realistic modeling course. We have found " UMAP modules are developed and distributed through COMAP, Ine, 57 Bedford Steet, Suite 210, Lexington, MA 02173. 2 Chapters 2, 6, and 8-12 do not require computing capabilityPreface xii a combination of graphing calculators and computers to be advantageous through- ut the course. The use of a spreadsheet is beneficial in Chapters 1, 5, and 7, and the capability for graphical displays of data is enormously useful, even essential, whenever data is provided. Students will find computers useful, too, in tansform- ing data, least-squares curve fitting, divided difference tables and cubic splines, programming simulation models, linear programming and numerical search meth- ods, and numerical solutions to differential equations. The CD accompanying this text provides some basic technology tools that students and instructors can use as a foundation for modeling with technology. Several FORTRAN executable programs are provided to execute the methodologies presented in Chapter 4. Also included is a tutorial on the computer algebra system MAPLE and its use for this text. Resource Materials We have found material provided by the Consortium for Mathematics and Its Ap- plication (COMAP) to be outstanding and particularly well suited to the course we propose. Individual modules for the undergraduate classroom, UMAP Modules, may be used in a variety of ways. First, they may be used as instructional material to support several lessons. In this mode a student completes the self-study module by working through its exercises (the detailed solutions provided with the module can be conveniently removed before it is issued). Another option is to put together aa block of instruction using one or more UMAP modules suggested in the projects sections of the text. The modules also provide excellent sources for “model re- search,” because they cover a wide variety of applications of mathematics in many fields. In this mode, a student is given an appropriate module to research and is asked to complete and report on the module. Finally, the modules are excellent re- sources for scenarios for which students can practice model construction. In this ‘mode the instructor writes a scenario for a student project based on an application addressed in a particular module and uses the module as background material, per- hhaps having the student complete the module ata later date. The CD accompanying the text contains most of the UMAPS referenced throughout. Information on the availability of newly developed interdisciplinary projects can be obtained by writ- ing COMAP at the address given previously, calling COMAP at 1-800-772-6627, or electronically: order@comap.com A great source of student-group projects are the Mathematical Contest in Modeling (MCM) and the Interdisciplinary Mathematical Contest in Modeling (IMCM). These projects can be taken from the link provided on the CD and tailored by the instructor to meet specific goals for their class. These are also good resources to prepare teams to compete in the MCM and IMCM contests currently sponsored by the National Security Agency (NSA) and COMAP. The contest is sponsored by COMAP with funding support from the National Security Agency, the Society of Industrial and Applied Mathematics, the Institute for Operations Research and the Management Sciences, and the Mathematical Association of America. Addi- tional information concerning the contest can be obtained by contacting COMAP, ‘or visiting their website at www.comap.com.xiv Preface Acknowledgments It is always a pleasure to acknowledge individuals who have played a role in the development of a book. We are particularly grateful to B.G. (retired) Jack M. Poltin and Dr. Carroll Wilde for stimulating our interest in teaching modeling and for support and guidance in our careers. We're indebted to many colleagues for reading the first edition manuscript and suggesting modifications and problems: Rickey Kolb, John Kenelly, Robert Schmidt, Stan Leja, Bard Mansager, and especially ‘Steve Maddox and Jim McNulty. ‘We are indebted to a number of individuals who authored or co-authored UMAP materials that support the text: David Cameron, Brindell Horelick, Michael Jaye, Sinan Koont, Stan Leja, Michael Wells, and Carroll Wilde. In addition, we thank Solomon Garfunkel and the entire COMAP staff for their cooperation on this project, especially Roland Cheyney for his help with the production of the CD that accompanies the text. We also thank Tom O'Neil and his students for their contri- butions to the CD and Toms helpful suggestions in support of modeling activities. ‘The production of any mathematics text is a complex process and we have been especially fortunate in having a superb and creative production staff at Brooks/Cole. In particular, we express our thanks to Craig Barth, our editor for the first edition, Gary Ostedt, the second edition, and Gary Ostedt and Bob Pir- Ue, our editors for this edition. For this edition we are especially grateful to Tom Ziolkowski, our marketing manager; Tom Novack, our production editor; Merrill Peterson and Matrix Productions for production service; and Amy Moellering for her superb copyediting and typesetting. We are especially grateful to Wendy Fox for providing her drawing of the Cadet Chapel at West Point for the dedication age. Finally, we are grateful to our wives—Judi Giordano, Gale Weir, and Wendy Fox—for their inspiration and support. Frank R. Giordano Maurice D. Weir William P. FoxContents == __cuaerer 1 Modeling Change 1 1 12 13 14 Introduction 1 Example 1: Testing for Proportionality 2 Modeling Change with Difference Equations 4 Example 1: A Savings Certificate 5 Example 2: Mortgaging a Home 6 Approximating Change with Difference Equations 9 Example I: Growth ofa Yeast Culture 9 Example 2: Growth of a Yeast Culture Revisited 10 Example 3: Spread of a Contagious Disease. 12 Example 4: Decay of Digoxin in the Bloodstream 13 Example 5: Heating of a Cooled Object 14 Solutions to Dynamical Systems 18 Example 1: A Savings Certificate Revisited 18 Example 2: Sewage Treatment 21 Example 3: Prescription for Digoxin 25 Example 4: An Investment Annuity 26 Example 5: A Checking Account 28 Example 6: An Investment Annuity Revisited 30 Systems of Difference Equations 35 Example |; A Car Rental Company 35 Example 2: The Battle of Trafalgar 38 Example 3: Competitive Hunter Model—Spotted Owls and Hawks 41 Example 4: Voting Tendencies ofthe Political Parties 44 __carten 2 The Modeling Process, Proportionality, and Geometric Similarity 52 24 22 23 Introduction 52 Mathematical Models 54 Example 1; Vehicular Stopping Distance 59 Modeling Using Proportionality 65 Example 1: Kepler's Third Law 67 Modeling Using Geometric Similarity 75 Example 1: Raindrops from a Motionless Cloud 77 Example 2: Modeling u Bass Fishing Derby 79iv Contents 2.4 Automobile Gasoline Mileage 88 2.5 Body Weight and Height, Strength and Agility 91 ___cuaprer B Model Fitting 97 Introduction 97 3.1. Fitting Models to Data Graphically 101 3.2 Analytic Methods of Model Fitting 107 3.3. Applying the Least-Squares Criterion 114 3.4 Choosing a Best Model 119 Example 1: Vehicular Stopping Distance 122 ___cuarer 4 Experimental Modeling —_ 126 Introduction 126 4.1 Harvesting in the Chesapeake Bay and Other One-Term Models 127 Example 1: Harvesting Bluefish 130 Example 2: Harvesting Blue Crabs 131 42 High-Order Polynomial Models 138 Example 1: Elapsed Time of a Tape Recorder 140 4.3 Smoothing: Low-Order Polynomial Models 146 Example 1: Elapsed Time of a Tape Recorder Revisited 147 Example 2: Elapsed Time of a Tape Recorder Revisited Again 151 Example 3: Vehicle Stopping Distance 153 Example 4: Growth of a Yeast Culture 155 4.4 Cubic Spline Models 159 Example I: Vehicle Stopping Distance Revisited 167 __cuaeter 3 Simulation Modeling 175 Introduction 175 5.1. Simulating Deterministic Behavior: Area Under a Curve 177 5.2 Generating Random Numbers 182 5.3 Simulating Probabilistic Behavior 186 5.4 Inventory Model: Gasoline and Consumer Demand 194 5.5 Queuing Models 205 Example 1; A Harbor System 205 Example 2: Morning Rush Hour 213Contents ___cuaerer @ Discrete Probabilistic Modeling — 217 6.1 6.2 63 Introduction 217 Probabilistic Modeling with Discrete Systems 217 Example 1: Rental Car Company Revisited 217 Example 2: Voting Tendencies 219 Modeling Component and System Reliability 223 Example 1: Series Systems 224 Example 2: Parallel Systems 224 Example 3: Series and Parallel Combinations 224 Linear Regression 227 Example 1: Ponderosa Pines 229 Example 2: The Bass Fishing Derby Revisited 231 CHAPTER 7 Discrete Optimization Modeling 236 72 73 74 75 76 ___cuaprer 8 Dimensional Analysis and Si 81 8.2 ‘An Overview of Discrete Optimization Modeling 237 Example 1: Determining a Production Schedule 237 Example 2: Space Shuttle Cargo 240 Example 3: Approximation by a Piecewise Linear Function 241 Linear Programming |: Geometric Solutions 250 Example 1: The Carpenter's Problem 251 Example 2: A Data-Fitting Problem 252 Linear Programming II: Algebraic Solutions 259 Example 1: Solving the Carpenter's Problem Algebraically 261 Linear Programming Ill: The Simplex Method 263 Example 1: The Carpenter's Problem Revisited 268 Example 2: Using the Tableau Format 271 Linear Programming IV: Sensitivity Analysis 273 Numerical Search Methods 279 Example 1: Using the Dichotomous Search Method 282 Example 2: Using the Golden Section Search Method 285 Example 3: Model-Fitting Criterion Revisited 287 Example 4: Optimizing Industrial Flow 288 itude 292 Introduction 292 Dimensions as Products 295 Example 1: A Simple Pendulum 298 Example 2: Wind Force ona Van 301 ‘The Process of Dimensional Analysis 304 Example 1: Terminal Velocity of a Raindrop 309 Example 2: Automobile Gas Mileage Revisited 311vi ‘Contents 83 ADamped Pendulum 313 8.4 Examples Illustrating Dimensional Analysis 319 Example I: Explosion Analysis 319 Example 2: How Long Should You Roast a Turkey? 324 8.5 Similitude 330 Example 1: Drag Force on a Submarine 331 ___cuaerer 9 eee of Functions as Models — 336 An Arms Race 336 Example 1: Civil Defense 345 Example 2: Mobile Launching Pads 346 Example 3: Multiple Warheads 347 Example 4: MIRVs Revisited: Counting Warheads 348 9.2 Modeling an Arms Race in Stages 350 9.3 Managing Nonrenewable Resources: The Energy Crisis 355 9.4 Effects of Taxation on the Energy Crisis 359 9.5 A Gasoline Shortage and Taxation 364 _ carrer 10 Modeling with a Differential Equation 368 Introduction 368 10.1. Population Growth 371 10.2 Prescribing Drug Dosage 382 10.3 Braking Distance Revisited 391 10.4 Graphical Solutions of Autonomous Differential Equations 395 Example 1: Drawing a Phase Line and Sketching Solution Curves 396 Example 2: Cooling Soup 399 Example 3: Logistic Growth Revisited 400 10.5 Numerical Approximation Methods 404 Example 1: Using Euler's Method 406 Example 2: A Savings Certificate Revisited 407 —cuarrer 19 Modeling with Systems of Differential Equations == 412 Introduction 412 11.1. Graphical Solutions of Autonomous Systems of First-Order Differential Equations 413 Example 1: A Linear Autonomous System 414 Example 2: A Nonlinear Autonomous System 415 11.2 A Competitive Hunter Model 419Contents vii 11.3 APredator-Prey Model 427 11.4 Two Military Examples 437 Exumple 1. Lanchester Comat Models 437 Example 2: Economic Aspects of an Arms Race 444 11.5. Euler's Method for Systems of Differential Equations 450 Example 1: Using Euler's Method for Systems 451 Example 2: A Trajectory and Solution Curves 452 Continuous Optimization Modeling 458 Introduction 458 12.1. An Inventory Problem: Minimizing the Cost of Delivery and Storage 459 12.2_/A Manufacturing Problem: Maximizing Profit in Producing Competing Products 468 12.3 Constrained Continuous Optimization 474 Example 1: An Oil Transfer Company 474 Example 2: A Space Shuttle Water Container 476 12.4 Managing Renewable Resources: The Fishing Industry 480 Problems from the Mathematics Contest in Modeling, 1985-2002 490 An Elevator Simulation Algorithm 523 The Revised Simplex Method _ 532 Index 535FETT EREREEBESSE* Introduction Figure 1.1 A flow of the modeling process beginning with an ‘examination of real-world data To help us better understand our world, we often describe a particular phenomenon mathematically (by means of a function or an equation, for instance). Such a math- ematical model is an idealization of the real-world phenomenon and never a com- pletely accurate representation, Although any model has its limitations, a good one can provide valuable results and conclusions. In this chapter we direct our attention to modeling change. Mathematical Models In modeling our world, we are often interested in predicting the value of a variable a some time in the future. Perhaps it is a population, a real estate value, or the ‘number of people with a communicative disease. Often a mathematical model can help us understand a behavior better or aid us in planning for the future. Let's think of a mathematical model as a mathematical construct designed to study a particular real-world system or behavior of interest. The model allows us to reach mathematical conclusions about the behavior, as illustrated in Figure 1.1, These ‘conclusions can be interpreted to help a decision maker plan for the future. Simplification Most models simplify reality. Generally, models can only approximate real-world behavior. One very powerful simplifying relationship is proportionality. Resl-wockd — | _Simpliscation | Model ‘ta ‘Verification Analysis: Predictions Mathematical explanations Tnierreion ‘emelusionsCHAPTER 1 Modeling Change DEFINITION ‘Two variables y and x are proportional (to each other) if one is always a constant multiple of the other; that is, if yoke for some nonzero constant k. We write y ocx. ‘The definition means that the graph of y versus x lies along a straight line through the origin. This graphical observation is useful in testing whether a given ata collection reasonably assumes a proportionality relationship. If a proportion- is reasonable, a plot of one variable against the other should approximate a straight line through the origin, Here is an example. EXAMPLE 1 Testing for Proportionality Table 1.1 Consider a spring-mass system, such as the one shown in Figure 1.2. We conduct an Spring-mass experiment to measure the stretch of the spring as @ function of the mass (measured system as weight) placed on the spring. Consider the data collected for this experiment, displayed in Table 1.1. A scatterplot graph of the stretch or elongation of the spring Maus Hong versus the mass or weight placed on it reveals an approximate straight line passing 30 1.000) through the origin. (Figure 1.3). 100 1.875 ‘The data appear to follow the proportionality rule that elongation ¢ is pro- 130 2.750 portional to the mass m, or symbolically, e cx m. The straight line appears to pass a 2 through the origin. This geometric understanding allows us to look at the data to nee ‘determine if proportionality is a reasonable simplifying assumption and, if so, 10 a See estimate the slope k. In this case, the assumption appears valid, so we estimate the me. oa constant of proportionality by picking the two points (200, 3.25) and (300, 4.875) 4507250 as lying along the straight line. We calculate the slope of the line joining these 500 8.000 points as 5508750 Figure 1.2 Spring-mass systemFigure 1.3 Data from spring-mass system Figure 1.4 Data from spring-mass system with proportionality line 1.0 introduction 3 Elongation Mass ‘Thus the constant of proportionality is approximately 0.0163 and we estimate our model as ¢ = 0.0163m We then examine how close our model fits the data by plotting the line it represents superimposed on the scatterplot (Figure 1.4), The graph reveals that the simplifying proportionality model is reasonable. . 90 S60 8 100200 30000300 Modeling Change A powerful paradigm to use in modeling change is, future value = present value + change Often, we wish to predict the future on what we know now, in the present, and add the change that has been carefully observed. In such cases, we begin by studying the change itself according to the formula change = future value — present value By collecting data over a period of time and plotting that data, we often can discem patterns to model that capture the trend of the change. If the behavior is tak- ing place over discrete time periods, the preceding construct leads to a difference4 CHAPTER 1 Modeling Change equation, which we study in this chapter. If the behavior is taking place contin- uuously with respect to time, then the construct leads to a differential equation studied in Chapter 10. Both are powerful methodologies for studying change to explain and predict behavior. 11 Modeling Change with Difference Equations EFINITION Figure 1.5 ‘The first difference of a sequence is the rise in the ‘raph during one time period In this section we build mathematical models to describe change in an observed behavior. When we observe change, we are often interested in understanding why the change occurs in the way it does, perhaps to analyze the effects of different con- ditions on the behavior or to predict what will happen in the future. A mathematical ‘mode! helps us better understand a behavior while allowing us to experiment math- cematically with different conditions affecting it. Fora sequence of numbers A = {ap, a), a, a3,... } the first differences are ‘Ady = a1 — ap Aa = a2 - a ay = a3 — a ‘a3 = a4 — a3 For each positive integer n, the mth first difference is ay = Ont ~ On Note from Figure 1.5 that the first difference represents the rise or fall be- tween consecutive values of the sequence; that is, the vertical change in the graph of the sequence during one time period. Ort hay) oon . ! 84, = 8,41 ~ OyEXAMPLE 1 1.1 Modeling Change with Difference Equations 5 A Savings Certificate Consider the value of a savings certificate initially worth $1000 that accumulates interest paid each month at 1% per month. The following sequence of numbers represent the value of the certificate month by month: A = (1000, 1010, 1020.10, 1030.30, The first differences are as follows: Aap = a ~ ay = 1010 — 1000 = 10 Aay = a ~ a; = 1020.10 ~ 1010 = 10.10 ay = a3 ~ a) = 1030.30 — 1020.10 = 10.20 Note thatthe first differences represent the change in the sequence during one time period, or the interest earned in the case of the savings certificate example. ‘The first difference is useful for modeling change taking place in discrete intervals. In this example, the change in the value of the certificate from one month to the next is merely the interest paid during that month. If n is the number of ‘months and a, the value of the certificate after n months, then the change or interest growth in each month is represented by the nth difference Ady = dnt ~ dn = 0.010q This expression can be rewritten as the difference equation O41 = On + 0.010q ‘We also know the initial deposit (initial value) that then gives the dynamical system model an41 = LOlay, a = 1000 0,1,2,3,... (1.1) where ay represents the amount accrued after n months, Because n represents the nonnegative integers (0, 1,2,3,...), Equation (1.1) represents an infinite set of algebraic equations, called a dynamical system. Dynamical systems allow us to describe the change from one period to the next. The difference equation formula computes the next term knowing the immediately previous term in the sequence, but it does not compute the value of a specific term directly (e.g.. the savings after 100 periods). Because itis change we often observe, we can construct a difference equation by representing or approximating the change from one period tothe next. To modify our example, if we were to withdraw $50 from the account each month, the change during a period would be the interest earned during that period minus the monthly withdrawal, or Adin = dns ~ dy = 0.0 lay — 50 .CHAPTER 1 Modeling Change EXAMPLE 2 In most examples, mathematically describing the change is not going to be as precise a procedure as illustrated here. Often itis necessary to plot the change, observe a pattern, and then describe the change in mathematical terms. That is, We will be trying to find change = Aa, = some function f ‘The change may be a function of previous terms in the sequence (as was the cease with no monthly withdrawals), or it may also involve some external terms (such as the amount of money withdrawn in the current example or an expression involving the period n). Thus, in constructing models representing change in this chapter we will be modeling change in discrete intervals, where change = Ady = ans ~ dn = f(terms in the sequence, external terms) Modeling change in this way becomes the art of determining or approximating a function f that represents the change. Consider a second example in which a difference equation exactly models a behavior in the real world. Mortgaging a Home Six years ago your parents purchased a home by financing $80,000 for 20 years, paying monthly payments of $880.87 with a monthly interest of 1%. They have made 72 payments and wish to know how much they owe on the mortgage, which they are considering paying off with an inheritance they received. Or they could be considering refinancing the mortgage with several interest rate options, depending on the length of the payback period. The change in the amount owed each period increases by the amount of interest and decreases by the amount of the payment: Aby = bro — by = O.01b, - 880.87 Solving for b, +1 and incorporating the initial condition gives the dynamical system model Png = by + 0.015, — 880.87 by = 80000 where by represents the amount owed after n months. Thus, +1 = 80000 + 0.01(80000) — 880.87 = 7919.13, ‘by = 79919.13 + 0.01(79919.13) — 880.87 = 79837.45 yielding the sequence B = (80000, 79919.13, 79837.45, ...) ‘The sequence is graphed in Figure 1.6. .1.1 Modeing Change with Difference Equations 7 Figure 1.6 Months | Amount Owed ‘The sequence and graph a 4, a Home mongage for Example 2 0) a000.00 1 7919.13 Ld 2 | r9e3743 a) & | ones = SOM s | reser48 spoon © | wsuza 3 30000) 7 | rane 0000) 8 | 32904 9 9242.37 ston 10 | ris392 ° | 7906839 © 21 42 63 84 105 126 147 168 189 210 231 | tesna37 Months Let's summarize the important ideas introduced in Examples 1 and 2. DEFINITIONS : AA sequence is a function whose domain is the set ofall nonnegative integers and whose range is a subset of the real numbers. A dynamical system is a relationship among terms in a sequence. A numerical solution is a table of values satisfying the dynamical system. In the problems for this section we discuss other behaviors in the world that. ccan be modeled exactly by difference equations. In the next section, we use dif- ference equations to approximate observed change. After collecting data for the ‘change and discerning patterns of the behavior, we will use the concept of propor- Vionality to test and fit models that we propose. 1.1 Problems Sequences 1. Write out the first five terms ap-a4 of the following sequences: (@) angt = 3aq, ay =1 (b) Gat = 2a, +6, a9 =0 (0) ans = 2ay(Qn +3), a9 = 4 @) ans = a5, ay =1 2, Find a formula for the nth term of the sequence. (a) (3,3,3,3.3,...} (b) (1, 4, 16, 64, 256, Od bebe (@) (1, 3,7, 15,31...)CHAPTER 1 Madeling Change Difference Equations 3. By examining the following sequences, write a difference equation to represent the change during the nth interval as a function of the previous term in the sequence, (a) (2.4.6.8, 10...) tb) (2.4.16, 256...) (© (1.2.5.11.23,.0) (d) {1, 8. 29. 92. 1 4, Write out the first five terms of the sequence satisfying the following difference equations: (a) Aap Gy. ay = 1 0.015b,, by = 1000 0.001(500~ pp). po = 10 1.5100 = tq). f= 200 Dynamical Systems 5. By substituting m = 0, 1,2, 3, write out the first four algebraic equations rep- resented by the following dynamical systems: (@) any = Bay, ay = 1 (D) an41 = 2ade +6, a =0 (©) Qnt1 = 2an(an +3), a9 = 4 (@) anys 6. Name several behaviors you think can be modeled by dynamical systems. 2. ay =I Modeling Change Exactly For Problems 7-10, formulate a dynamical system that models change exactly for the described situation. 7. You currently have $5000 in a savings account that pays 0.5% interest each month, You add another $200 each month. 8. You owe $500 on a credit card that charges 1.5% interest each month. You pay $50 each month and you make no new charges. 9. Your parents are considering a 30-year, $100,000 mortgage that charges 0.5% interest each month. Formulate a model in terms of a monthly payment p that allows the mortgage (loan) to be paid off afier 360 payments. Hint: If an rep- resents the amount owed after n months, what are ay and ayo? 10. Your grandparents have an annuity. The value of the annuity increases each month by an automatic deposit of 1% interest on the previous month's balance. Your grandparents withdraw $1000 at the beginning of each month for living expenses. Currently, they have $50,000 in the annuity. Model the annuity with1.1 Project 1.2 ‘Approximating Change wi h Difference Equations EXAMPLE 1 1.2. Aporoximating Change with Difference Equations 9 a dynamical system. Will the annuity run out of money? When? Hint: What value will a, have when the annuity is depleted? 1. You wish to buy a new car and narrow your choices to a Saturn, Cavalier, and Hyundai. Each company offers you its best deal: Saturn $13,990 $1000 down 3.55 interest for up to 60 months Cavalier $1350 $1500down 4.5% interest for up to 60 months Hyundai $12,400 $500down 6.5% interest for up to-48 months You are able to spend at most $475 a month on a car payment. Use a dynamical system to determine which car to buy. In most examples, describing the change mathematically will not be as precise a Procedure as in the cases of the savings certificate and mortgage examples pre- sented in the previous section. Typically, we must plot the change, observe a pat tem, and then approximate the change in mathematical terms. In this section we approximate some observed change to complete the expression change = Ady = some function f ‘We begin by distinguishing between change that takes place continuously and that which occurs in discrete time intervals Discrete Versus Continuous Change ‘When constructing models involving change, an important distinction is that some change takes place in discrete time intervals (such as the depositing of interest in an account); in other cases, the change happens continuously (such as the change in the temperature of a cold can of soda on a warm day), Difference equations represent change in the case of discrete time intervals. Later we will see the rela- tionship between discrete change and continuous change (for which calculus was developed). For now, in the several models that follow, we approximate a conti luous change by examining data taken at discrete time intervals. Approximating a continuous change by difference equations is an example of model Growth of a Yeast Culture ‘The data in Figure 1.7 was collected from an experiment measuring the growth of 2 yeast culture. The graph represents the assumption that the change in population is proportional to the current size of the population. That is, APn = (Pav1~ Pr) = pq, where pp represents the size of the Population biomass after n hours, and k is a positive constant. The value of k depends on the time measurement,10 CHAPTER 1 Modeling Change Figure 1.7 Time Growth of a yeast na culture versus time in hours; data from R. Pearl, “The Growth of Population,” Quart Rev: Biol. 2(1927) 532-548 EXAMPLE 2 Observed | AP, yeast | Change in Change in biomass vs. biomass Monae | biomes g me | Pie E z i Although the graph of the data does not lie precisely along a straight lis passing exactly through the origin, it can be approximated by such a straight line. Placing a ruler over the data to approximate a straight line through the origin, we estimate the slope of the line to be about 0.5. Using the estimate k = 0.5 for the slope of the line, we hypothesize the proportionality model APn = Pri ~ Pn = 0.5Pn yielding the prediction p,41 = 1.5p,. This model predicts a population that is creases forever, which 1s questionable. . Model Refinement: Modeling Births, Deaths, and Resources If both births and deaths during a period are proportional to the population, then the change in population should be proportional to the population, as was illustrated in Example 1. However, certain resources (¢.g., food) can support only a maximum population level rather than one that increases indefinitely. As these maximum lev- els are approached, growth should slow. Growth of a Yeast Culture Revisited Finding a Mode! The data in Figure 1.8 show what actually happens tothe yeast culture growing in a restricted area as time increases beyond the eight observations given in Figure 1.7 From the third column of the data table in Figure 1.8 note that the change in population per hour becomes smaller as the resources become more limited or constrained. From the graph of population versus time, the population appears to be approaching a limiting value or carrying capacity. Based on our graph we es- timate the carrying capacity to be 665. (Actually, the graph does not precisely tell us the correct number is 665 and not 664 or 666, for example). Nevertheless, a8 Pp approaches 665, the change does slow considerably. Because 665 — py gets smaller 8 pr approaches 665, we propose the model APn = Pati — Pn = k(665 — Pa) PnFigure 1.8 Time ‘Yeast biomass aa approaches a limiting "™"S population evel Yeast Pr 96 183 290 412 om N91 1746 2573 350.7 410 533 5597 594.8 6294 640.8 6st 0559) 6596 661.8 biomass Changed ‘hone 7 107 182 239 480 555 B27 934 903 n3 404 351 346 ua 103 37 22 Pas iPa 1.2. Approximating Change with Difference Equations " Yeast biomass which causes the change Ap, to become increasingly small as p, approaches 665. Mathematically, this hypothesized model states that the change A, is proportional tothe product (665~p,) Pn. To test the model, plot (Pat Pn) versus (665— Pn) Pn to sce if there is a reasonable proportionality. Then estimate the proportionality constant k. Examining Figure 1.9, we see that the plot does reasonably approximate a straight line projected through the origin, We estimate the slope of the line approx mating the data to be about k ~ 0.00082, which gives the model Figure 1.9 Testing the constrained growth model 87 oa 1183461 1844400 29,160.16 4222629 65,016.69 85,623.86 104.901.21, 10,225.01, 98,788.00 7786761 3893641 41,754.96 22,406.64 15,507.36, 9050.29 5968.69 3501.84 Patt ~ Pn = 0.00082(665 — px) Pn (1.2) Growth constrained by resources 100 0 80 7) tw 3 50) 240 30] 20| 10 0 50,000 100,000 150,000 Pe(665~P,)12 CHAPTER 1 Modeling Change EXAMPLE 3 inhours | Observations Solving the Model Numerically Solving Equation (1.2) for pn gives Passi = Px + 0.00082(665 ~ Pn) Pn (1.3) ‘The right side of this equation is a quadratic in pa. Such dynamical systems are classified as nonlinear and generally cannot be solved for analytical solutions. ‘That is, usually we cannot find a formula expressing p, in terms of. However, if ‘given that po = 9.6, we can substitute in the expression to compute py P1 = Po +.0.00082(665 — po) po = 9.6 + 0.00082(665 — 9.6)9.6 = 14.76 Ina similar manner, we can substitute py = 14.76 into Equation (1.3) to compute p2 = 22.63. Iterating in this way, we compute a table of values to provide a numerical solution to the model. This numerical solution of model predictions is presented in Figure 1.10. The predictions and observations are plotted together versus time on the same graph. Note that the model captures fairly well the srend of the observed data . Pr Growth ina yeast culture 22 ‘Yeast biomasss 8 Time in hours Figure 1.10 Model predictions and observat Spread of a Contagious Disease Suppose there are 400 students in a college dormitory and that one or more students has a severe case of the flu. Let ix represent the number of infected students after rn time periods. Assume some interaction between those infected and those not infected is required to pass on the disease. If all are susceptible to the disease, then (400 — i,) represents those susceptible but not yet infected. If those infected remainEXAMPLE 4 1.2 Approximating Change with Difference Equations 3 contagious, we can model the change of those infected as a proportionality to the product of those infected by those susceptible but not yet infected, or in = inti ~ in = kin(400 = in) (1.4) In this model the product j,(400 — i,) represents the number of possible interac- tions between those infected and those not infected at time n. A fraction k of these interactions would cause additional infections, represented by Ain. Equation (1.4) has the same form as Equation (1.2), but in the absence of any data we cannot determine a value for the proportionality constant k. Nevertheless, a graph of the predictions determined by Equation (1.4) would have the same S shape as the graph of the yeast population in Figure 1.10. There are many refinements to this model. For example, we might assume that a segment of the population is not susceptible to the disease, that the infection period is limited, or that infected students are removed from the dorm to prevent interaction with the uninfected. More sophisticated models might even treat the infected and susceptible populations separately. . Decay of Digoxin in the Bloodstream Digoxin is used in the treatment of heart disease. Doctors must prescribe an amount of medicine that keeps the concentration of digoxin in the bloodstream above an effeetive level without exceeding a safe level (there is variation among patients). For an initial dosage of 0.5 mg in the bloodstream, Table 1.2 shows the amount of digoxin a, remaining in the bloodstream of a particular patient after n days, together with the change Aa, each day. Table 1.2. The change ap in digoxin in a patient's bloodstream n 0 1 2 3 4 5 6 71 8 a 0500 0345 0238 0164 0.113 0078 0058 0037 0026 Say =0.155 -0.107 -0078 -0031 -0.035 -0.024 -0017 0011 A scatterplot of Aay versus a, from Table 1.2 is shown in Figure 1.11. The graph shows that the change Aa, during a time interval is approximately propor- tional to the amount of digoxin a, present in the bloodstream at the beginning of the time interval. The slope of the proportionality line through the origin is approx- imately & * ~0.107/0.345 = —0.310. Since the graph in Figure 1.11 shows the change Ady as a linear function of a, with slope ~0.31, we have Ady = —0.31ay The Model From Figure 1.11, ‘San = -0.31a, Qn $1 ~ dy = 0.31 On 41 = 0.694y14 Figure 1.11 CHAPTER 1 Modeling Change A plot of Bag versus an from Table 1 ‘2 suggests a straight line through the origin EXAMPLE 5 ol 02030408 002 008 0.06 =008 010 0.12 018 = 016 -018 ba, A difference equation model for the decay of digoxin in the bloodstream given an initial dosage of 0.5 mg is Gn = Gy ~ 031d, = 0.6940, ag = 0.5 . Heating of a Cooled Object Now we examine a behavior that is taking place continuously. Suppose a cold can of soda is taken from a refrigerator and placed in a warm classroom and we measure the temperature periodically. The temperature of the soda is initially 40°F and the room temperature is 72°F, Temperature is a measure of energy per unit volume. Because the volume of soda is small relative to the volume of the room, we would expect the room temperature to remain constant. Furthermore, we assume the entire can of soda is the same temperature, neglecting variation within the can. We might expect the change in temperature per time period to be greater when the difference in temperatures between the soda and room is large and the change in temperature per unit time (o be less when the difference in temperatures is small, Letting ( represent the temperature of the soda after n time periods and k a positive constant of proportionality, we propose ln = tngt ~ tn = R(T2~ ty) 6 =40 Many refinements are possible for this model. Although we have assumed k is constant, it actually depends on the shape and conductivity properties of the con- tainer, the length of the time period between the measurements, and so on. Also, the temperature of the environment may not be constant in many instances, and it may be necessary to take into account that the temperature is not uniform throughout the can of soda. The temperature of an object may vary in one dimension (as in the case of a thin wire), in two dimensions (such as for a flat plate), or three dimensions (@s in the case of a space capsule reentering the earth's atmosphere). .1.2 Problems 1.2 Approximating Change with Difference Equations 15 We have presented only a glimpse of the power of difference equations to model change in the world around us. In the next section, we build numerical solutions to some of these models and observe the patterns they exhibit. Having observed certain patterns for various types of difference equations, we will then classify them by their mathematical structures. This will help to determine the long- term behavior of a dynamical system under study. 1, The following data were obtained for the growth of a sheep population intro- duced into a new environment on the island of Tasmania.' Year | 181418241834 18441854 1864 Population | 125° 275 83012001750 1650 Plot the data. Is there a wend? Plot the change in population versus years clapsed after 1814. Formulate a discrete dynamical system that reasonably ap- proximates the change you have observed. 2. The following data represent the U.S. population from 1790 to 2000: Year Population 1790 3,929,000 1800 5,308,000 1810 7,240,000 1820 9,638,000 1830 12,866,000 1840 17,069,000 1850 23,192,000 1860 31,443,000 1870 38,558,000 1880 50,156,000 1890 62,948,000 1900 75,995,000 1910 91,972,000 1920 105,711,000 1930 122,755,000 1940 131,669,000 4950, 150,697,000 1960 179,323,000 1970 203,212,000 1980 226,505,000 1990 248,710,000 2000 281,416,000 * Adapted from J. Davidson, “On the Growth of the Sheep Population in Tasmania,” Trans. R. Soc 5 Australia 62(1938); 342-346,16 CHAPTER 1 Modeling Change 3. 6, Find a dynamical system model that fits the data fairly well. Test your model by plotting the predictions of the model against the data Sociologists recognize a phenomenon called social diffusion, which is the spreading of a piece of information, a technological innovation, or a cultural fad among a population, The members of the population can be divided into two classes: those who have the information and those who do not. In a fixed population whose size is known, it is reasonable to assume that the rate of diffusion is proportional to the number who have the information times the number yet to receive it. If ay denotes the number of people who have the information in a population of N people after n days, formulate a dynamical system to approximate the change in the number of people in the population who have the information, Consider the spreading of a highly communicable disease on an isolated island with population size NV. A portion of the population travels abroad and returns to the island infected with the disease. Formulate a dynamical system to ap- proximate the change in the number of people in the population who have the disease. Assume we are considering the survival of whales and that if the number of whales falls below a minimum survival level m the species will become extinct. Assume also that the population is limited by the carrying capacity M of the environment. That is, if the whale population is above M, it will experience a decline because the environment cannot sustain that large a population level. In the following model, a, represents the whale population after n years; discuss the model: ng ~ On = K(M ~ dq) (dy ~ m) A certain drug is effective in treating a disease if the concentration remains above 100 mg/L. ‘The initial concentration is 640 mg/L. [tis known from lab- oratory experiments that the drug decays at the rate of 20% of the amount present each hour. (a) Formulate a model representing the concentration at each hour. (b) Build a table of values and determine when the concentration reaches 100 mg/L. Use the model developed in Problem 6 to prescribe an initial dosage and a maintenance dosage that keeps the concentration above the effective level of 500 ppm but below a safe level of 1000 ppm. Experiment with different values until you have results you think are satisfactory. ‘A humanoid skull is discovered near the remains of an ancient campfire. Ar- chacologists are convinced the skull is the same age as the original campfire It is determined from laboratory testing that only 1% of the original amount of carbon-14 remains in the bumed wood taken from the campfire. It is known that carbon-14 decays at a rate proportional to the amount remaining and that1.2 Project 11.2. Approximating Change with Difference Equations 7 carbon-14 decays 50% over $700 years. Formulate a model for carbon-14 dating 9. The data in the accompanying table show the speed n (in increments of 5 mph) of an automobile and the associated distance ay in feet required to stop it once the brakes are applied. For instance, n = 6 (representing 6 x 5 = 30 mph) requires a stopping distance of ag = 47 ft. (@) Calculate and plot the change Aa, versus n. Does the graph reasonably approximate a linear relationship? (b) Based on your conclusions in part a, find a difference equation model for the stopping distance data. Test your model by plotting the errors in the predicted values against n. Discuss the appropriateness of the mode!. 9 112 10 140 now 12 204 13° 241 14 282 15 325 16 376 10, Place a cold can of soda in a room. Measure the temperature of the room and periodically measure the temperature of the soda, Formulate a model to predict the change in the temperature of the soda. Estimate any constants of propor- tionality from your data. What are some of the sources of error in your model? 1, Complete the UMAP module “The Diffusion of Innovation in Family Plan- ning,” by Kathryn N. Harmon, UMAP 303. This module gives an interesting application of finite difference equations to study the process through which Public policies are diffused to understand how national governments might adopt family planning policies,18 CHAPTER 1 Modeling Change 1.2 Further Reading Fraventhal, James C. Introduction to Population Modeling. Lexington, MA: COMAP, 1979. Hutchinson, G. Evelyn. An Introduction to Population Ecology. New Haven, CT: Yale Uni- versity Press, 1978. Levins, R. “The Strategy of Model Building in Population Biology.” American Scientist 54 (1966):421-431 1.3 Solutions to Dynamical Systems EXAMPLE 1 In this section we build solutions to some dynamical systems, starting with an ini- tial value and iterating a sufficient number of subsequent values to determine the patterns involved. In some cases, we see that the behavior predicted by dynamical systems is characterized by the mathematical structure ofthe system. In other cases, we see wild variations in the behavior caused by only small changes in the initial values of the dynamical system. We also examine dynamical systems for which small changes in the proportionality constants cause wildly different predictions. The Method of Conjecture ‘The method of conjecture is a powerful mathematical technique to hypothesize the form ofa solution to a dynamical system and then to accept or reject the hypothesis It is based on exploration and observation from which we attempt to discem a pattern to the solution. Let's begin with an example. A Savings Certificate Revisited In the savings certificate example (Example 1, Section 1.1), a savings certificate ially worth $1000 accumulated interest paid each month at 1% of the balance. No deposits or withdrawals occurred in the account, determining the dynamical system n41 = 1.Oldy (15) ‘ay = 1000 Look fora Pattern We might begin by graphing the balance in the account at the end of each month (after the interest is paid). That is, we graph ay versus n. The ‘graph is displayed in Figure 1.12. From the graph it appears that the sequence {ao, a1, a2, a3, ...} grows with- out bound. Let’s examine the sequence algebraically to gain more insight into the ‘growth pattern, 4; = 1010.00 = 1.01(1000) ‘ay = 1020.10 = 1.01(1010) = .01(1.01(1000)) = 1.012(1000) 43 = 1030.30 = 1.01(1020.10) = 1.01(1.017(1000)) = 1.015(1000) a4 = 1040.60 = 1.01(1030.30) = 1.01(1.013(1000)) = 1.014(1000)Figure 1.12 Growth of the savings certificate balance An = 1.01q, a = 1000 1.3. Solutions to Dynamical Systems 19 losior 2300 1061 52 Pd or 1082.86 1500] 10m 69 ogee LIS67 500 113683 : 113809 100 1149.47 ° a 16097 117238 118430 ‘The pattern of these terms in the sequence suggests that the kth term ay is the amount 1000 multiplied by (1.01). Conjecture For k = 1,2,3,..., the term ag in the dynamical system (Equa- tion (1.5))is ay = (1.01)41000 (1.6) Test the Conjecture We test the conjecture by examining if the formula for ag satisfies the system of Equation (1.5) upon substitution, net = 10a (L.01)"*!1000 = 1.01((1.01)" 1000) = (1.01)"*!1000 Since this last equation is true for every positive integer m, we accept the conjecture in Equation (1.6). Conclusion The solution for the term ay in the dynamical system,41.5) is ay = (1.01)*1000 ay = (1.01)ap, k = 1,2,3,.. This solution allows us to compute the balance ay in the account after k months. For example, after 120 months (or 10 years), the account is worth aig9 = (4.01)29(1000) ~ $3303.90. After 30 years (k = 360), the account is worth $35,949.64. This calculation is easier to obtain than iterating the dynamical system 360 times, but a formula such as Fquation (1.6) can provide even more insight into the long-term behavior of a dynamical system, as we will soon see. .20 CHAPTER 1 ‘Modeling Change Let’s summarize the procedure followed in this example. “The Method of Conjecture 1. Observe a pattern. 2. Conjecture a form of the solution to the dynamical system. 3. Test the conjecture by substitution, 4, Accept or reject the conjecture depending on whether it does or does not satisfy the system after the substitution and algebraic manipulation. To accept, the substitution must result in an identity. Linear Dynamical Systems a,41 = ran, for r Constant ‘The dynamical system in Example 1 is of the form dyy1 = ra, where r = 1.0L Let's follow the method of conjecture for the more general casc in which r is any positive or negative constant, assuming a given initial value ag. Look for a Pattern Examining the terms in the sequence ans = ran, We see that ay = ray ay = ray = r(rag) = rag 3 = ray = r(P?a0) = rao a4 = ray = r(3ay) = ray From these terms, we observe a pattern leading to the following conjecture: Conjecture Fork = 1,2, ray is + the term ax in the dynamical system ay 4) = (1.7) Test the Conjecture We substitute the formula from Equation (1,7) into the dynamical system: nt} = Taq prt dy = r(r"an) ‘The result is an identity, and we accept the conjecture. Let's summarize our resultTHEOREM 1 1.3 Solutions to Dynamical Systems a The solution of the linear dynamical system an41 = rd, for r any nonzero ‘constant is ay = rag where ap is a given initial value. EXAMPLE 2 Sewage Treatment ‘A sewage treatment plant processes raw sewage to produce usable fertilizer and clean water by removing all other contaminants. The process is such that each hour 12% of remaining contaminants in a processing tank are removed. What percentage of the sewage would remain after 1 day? How long would it take to lower the amount of sewage by half? How long until the level of sewage is down to 10% of the original level? Solution Let the initial amount of sewage contaminants be ag and let a, denote the amount after n hours. We then build the model On 41 = Oy — 0.12dy = 0.88aq, which is a linear dynamical system. From Theorem | the solution is ay = (0.88)kag After I day, k = 24 hours and the level of sewage remaining is 4424 = (0.88)"*ag = 0.0465a9 ‘That is, the level of contaminants in the sewage has been reduced by more than 95% at the end ofthe first day. Half the original contaminants remain when ag = 0.Sag. Thus, 0.Sao = ax = (0.88)'a9 Solving for k, we find 0.5aq = (0.88)*ap (0.88) =0.5, Ittakes about 5.42 hours to lower the contaminants to half their original amount. To reduce the level of contaminants by 90% we require (0.88) = 0.14922 CHAPTER 1 Figure 1.13 Every solution of ni = an isa ‘constant solution Modeling Change so that _ bog0.1 = = 18.01 10g 0.88 It takes 18 hours before the contaminants are reduced to 10% of their original level . Long-Term Behavior of an41 = ran, for r Constant For the linear dynamical system d,1 = rq, let's consider some significant values of r. If r = 0, then all the values of the sequence (except possibly a) are zero, so there is no need for further analysis. If r = 1, then the sequence becomes 4n41 = Qq. This is an interesting case because no matter where the sequence starts, it stays there forever (as illustrated in Figure 1.13, in which the sequence is enumerated and graphed for ay = 50 and also graphed for several other start- ing values). Values for which a dynamical system remains constant at thosc valucs, once reached, are called equilibrium values of the system. We define that term more precisely later, but for now note that in Figure 1.13, any starting value would be an equilibrium value. 100 80 o 0 Ifr > 1, as in Example 1, then the sequence a, = r*ag, which solves the linear dynamical system, grows without bound. This growth was illustrated in Fig- ure 1.12. ‘What happens if r is negative? If we replace 1.01 with —1.01 in Example 1, ‘we obtain the graph in Figure 1.14. Note the oscillation between both positive and negative values, Because the negative sign causes the next term in the sequence to be of opposite sign from the previous term, we conclude that, in general, negative values of r cause oscillations in the linear form a.1 = ry, What happens if |r| < 12 We know what happens if r = 0, if r > 1, if 1 < ~I, and if r is negative in general. If 0 1 | Growth without bound Dynamical Systems of the Form an41 = ran +b, Where r and b Are Constants Now let's add a constant b to the dynamical system we previously studied. Again, we want to classify the nature of the long-term behavior for all possible cases. We begin with a definition. DEFINITION ‘A number a is called an equilibrium value or fixed point of a dynamical system n41 = f (dn) if ag = a for all k = 1,2,3,... when ag = a. That is, ag = a is a constant solution to the dynamical system.1.3 Solutions to Dynamical Systems 25. Figure 1.17 Long-term behaviors for 0 Gq) = Faq. 051 > 1 and |r| <1 ‘Grows large without bound, Grows negative without hound > 1 ay>0 PP, age € Decays a. Decays ere, ay>0 Oerel, ay<0 €.Oxcillates and grows £. Oscillates and decays ret, ay>0 1< <0, ag>0 ‘A consequence of the definition is that a is an equilibrium value for dn41 J (aq) if and only if a — f(a) when ap = a. This result is shown by simple substi- tution. Equilibrium values can be helpful in understanding the long-term behavior of a dynamical system such as a1 = ray + b. Let's consider three examples to gain insight into its behavior. EXAMPLE 3 Prescription for Digoxin Consider again the digoxin problem. Recall that digoxin is used in the treatment of heart patients. The objective of the problem is to consider the decay of digoxin in the bloodstream to prescribe a dosage that keeps the concentration between accept- able (safe and effective) levels. Suppose we prescribe a daily drug dosage of 0.1 mg and know that half the digoxin remains in the system at the end of each dosage Period. This results in the dynamical system an 41 = 0.Sdy + 0.126 Figure 1.18 ‘Taree initial doses CHAPTER 1 Modeling Change digoxin EXAMPLE 4 ‘Now consider three starting values, or initial doses: Asay = 0.1 B: ay = 0.2 C: ay = 0.3 In Figure 1.18 we compute the numerical solutions for each case. juueaCinses i 0 of 1 025 2 02s 028 3 187s | 02 0.2125 4 | 0137s | 02 | 02005 oz 5 | 0.196875 | 02 | 0203125 6 | 01984375 | 02 | 02015625 gys 7 | o19s2i875 | 02 | o2norsi2s 8 | osseasss | 02 | 020030063 9 | os9s0469 | 02 | 020019531 7 10 | 019990234 | 02 | 020009766 11 | oaswsii7 | 02 | 020008883 00s 12 | 019997559 | 02 | o-20002441 13 | 019998779 | 02 | 020001221 * 14 | o.g939 | 02 | 02000061 5 10 15 15 | 019999695 | 02 | o.20000305 Note that the value 0.2 is an equilibrium value because once reached, the system remains at 0.2 forever. Furthermore, if we start below the equilibrium (as in Case A) or above the equilibrium (as in Case C), apparently we approach the equilibrium value as a limit. In the exercises you are asked to compute solutions for starting values even closer to 0.2, lending evidence that 0.2 is a stable equilibrium value When prescribing digoxin, the concentration level must stay above an effec- tive level for a period of time without exceeding a safe level. In the exercises you are asked to find initial and subsequent doses that are both safe and effective. = An Investment Annuity Retum now to the savings account problem and consider an annuity. Annuities are often planned for retirement purposes. They are basically savings accounts that pay interest on the amount present and allow the investor to withdraw a fixed amount each month until the account is depleted. An interesting issue (posed in the ex- ercises) is to determine the amount one must save monthly to build an annuity allowing for withdrawals, beginning at a certain age with a specified amount for a desired number of years, before the account's depletion. For now, consider 1% as the monthly interest rate and a monthly withdrawal of $1000. This gives theFigure 1.19 An annuity with three initial investments 1.3 Solutions to Dynamical Systems 2 dynamical system an41 = 101d, — 1000 Now suppose we made the following initial investments: ‘A: ay = 90,000 B: ag = 100,000 C: ay = 110,000 ‘The numerical solutions for each case are graphed in Figure 1.19. alarie ‘© | 90000 | 100000110000 40,000 1 | 89900 | 100000 ] 110100 2 | 89799 | 100000 ) 110201 120,000 3. | 89607 | 100000 | 110303 -—____* 4 | 89594 | 100000 | 110406 499,000 5. | 89990 | 100000 | 110310 -—__ 6 | 89385 | 100000 | 110615, 7 | 89279 | 100000 | 110721 oom 8 | 89171 | 100000 } 110829, 9. | 39063 | 100000 | 110937 a 10 | 88954 | 100000 | 111046 11} 88843 | too000 | 111157 12 | 88732 | 100000 | 111268 13 | 88619 | 100000 | 111381 14 | 88505 | 100000 | 111495 15 | 38390 | 100000 | 111610 40,000 Notice that the value 100,000 is an equilibrium value because once reached, the system remains there for all subsequent values. But if we start above that equi- librium, there is growth without bound. (Try plotting with ap = $100,000.01.) On the other hand, if we start below $100,000, the savings are used up at an increas- ing rate. (Try $99,999.99.) Note how drastically different the long-term behaviors are even though the starting values differ by only $0.02. In this situation, we say that the equilibrium value 100,000 is unstable: If we start close to the value (even within a penny) we do not remain close. Look at the dramatic differences displayed by Figures 1.18 and 1.19. Both systems show equilibrium values, but the first is stable and the second is unstable. . In Examples 3 and 4 we considered cases in which |r| < 1 and |r| > 1. Let's see what happens if r = 1,28 CHAPTER 1 Modeling Change EXAMPLE 5 Figure 1.20 ‘A checking account to pay dorm costs A Checking Account Most students cannot keep enough cash in their checking accounts to eam any interest. Suppose you have an account that pays no interest, and that each month you pay only your dorm rent of $300, giving the dynamical system On 41 = ay ~ 300 The result is probably obvious from the application, but when compared with Figures 1.18 and 1.19 the graph is revealing. In Figure 1.20 we plot the numerical solution for a starting value of 3000. Do you see how drastically the graph of the solution differs from those of the previous examples? Can you find an equilibrium, as in Examples 3 and 4? . Now let’s collect our observations thus far, classifying the three examples and their long-term behavior according to the value of the constant r. ing and Classifying Equilibrium Values Determining if equilibrium values exist, and classifying them as stable or unstable, assists us immensely in analyzing the long-term behavior of a dynamical system. Consider again Examples 3 and 4. In Example 3, how did we know that a starting value of 0.2 would result in a constant solution or equilibrium value? Similarly, how do we know the answer to the same question for an investment of $100,000 in Example 4? For a dynamical system of the form n+] =n +b (1.8) denote by a the equilibrium value, if one exists. From the definition of an equilib- rium value, if we start at a we must remain there for all n; that is, dp41 = dq = a for all n, Substituting a for dp +1 and a, in Equation (1.8) yields a=rath