Gord.3896.FM.

pgs 5/2/03 10:34 AM Page i

Functioning
in the
Real World
A Precalculus Experience
Second Edition

Sheldon P. Gordon
Farmingdale State University of
New York

Florence S. Gordon
New York Institute of Technology

Alan C. Tucker
SUNY at Stony Brook

Martha J. Siegel
Towson State University

Boston San Francisco New York

London Toronto Sydney Tokyo Singapore Madrid
Mexico City Munich Paris Cape Town Hong Kong Montreal
Gord.3896.FM.pgs 11/21/03 9:54 AM Page ii

Publisher: Greg Tobin

Managing Editor: Karen Guardino
Sponsoring Editor: Anne Kelly
Project Editor: Rachel Reeve
Production Supervisor: Cindy Cody
Marketing Manager: Becky Anderson
Marketing Coordinator: Julia Coen
Media Producer: Sharon Smith
Prepress Supervisor: Caroline Fell
Senior Designer: Barbara Atkinson
Cover and Interior Designer: Andrea Menza
Production Services: Jenny Bagdigian
Compositor: WestWords Inc.
Art Creation: Techsetters
Cover Photograph: Eyewire/Getty

Dedicated to:
Craig Eric Gordon and Kenneth Scott Gordon,
Our love, our pride, our hope, our joy.

Library of Congress Cataloging-in-Publication Data

Gordon, Sheldon P.
Functioning in the real world : a precalculus experience.— 2nd ed. / Sheldon P.
Gordon, Florence S. Gordon, Alan C. Tucker.
p. cm.
Rev. ed. of: Functioning in the real world / Sheldon P. Gordon . . . [et al.]. c1997.
Includes bibliographical references and index.
ISBN 0-201-38389-6
1. Functions. I. Gordon, Florence S. II. Tucker, Alan, 1943 July 6-
III. Functioning in the real world. IV. Title.

QA331.3.G67 2004
515—dc21
2001056076

This book is based on a portion of the materials developed with the support of the Divi-
sion of Undergraduate Education of the National Science Foundation under grants #USE-
91-50440 and #DUE-9254085 for the Math Modeling/PreCalculus Reform Project.
However, any views expressed are not necessarily those of the Foundation.

Copyright © 2004 Pearson Education, Inc. All rights reserved. No part of this publication may be repro-
duced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical,
photocopying, recording, or otherwise, without the prior written permission of the publisher. Printed in
the United States of America.

1 2 3 4 5 6 7 8 9 10 CRW 06 05 04 03
Gord.3896.FM.pgs 5/2/03 10:34 AM Page iii

Contents

Preface vii

CHAPTER 1 Functions in the Real World

1.1 Functions Are All Around Us 1
1.2 Describing the Behavior of Functions 8
1.3 Representing Functions Symbolically 18
1.4 Connecting Geometric and Symbolic Representations 25
1.5 Mathematical Models 34
Chapter Summary and Review Problems 39

CHAPTER 2 Families of Functions

2.1 Introduction 43
2.2 Linear Functions 44
2.3 Linear Functions and Data 59
2.4 Exponential Growth Functions 70
2.5 Exponential Decay Functions 87
2.6 Logarithmic Functions 99
2.7 Power Functions 113
2.8 Comparing Rates of Growth and Decay 131
2.9 Inverse Functions 140
Chapter Summary and Review Problems 154

CHAPTER 3 Fitting Functions to Data

3.1 Introduction to Data Analysis 161
3.2 Linear Regression Analysis 164
3.3 Fitting Nonlinear Functions to Data 181
3.4 How to Fit Exponential and Logarithmic Functions to Data 196
3.5 How to Fit Power Functions to Data 208
3.6 How Good is the Fit? 221
3.7 Linear Models with Several Variables 237
Chapter Summary and Review Problems 243

CHAPTER 4 Extended Families of Functions

4.1 Introduction to Polynomial Functions 249
4.2 The Behavior of Polynomial Functions 255
4.3 Modeling with Polynomial Functions 273

iii
Gord.3896.FM.pgs 5/2/03 10:34 AM Page iv

iv Contents

4.4 The Roots of Polynomial Equations: Real or Complex? 285

4.5 Finding Polynomial Patterns 292
4.6 Building New Functions from Old: Operations on Functions 304
4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 320
4.8 Using Shifting and Stretching to Analyze Data 331
4.9 The Logistic and Surge Functions 346
Chapter Summary and Review Problems 352

CHAPTER 5 Modeling with Difference Equations

5.1 Eliminating Drugs from the Body 357
5.2 Modeling with Difference Equations 371
5.3 The Logistic or Inhibited Growth Model 385
5.4 Newton’s Laws of Cooling and Heating 407
5.5 Geometric Sequences and their Sums 416
Chapter Summary and Review Problems 425

CHAPTER 6 Introduction to Trigonometry

6.1 The Tangent of an Angle 429
6.2 The Sine and Cosine of an Angle 440
6.3 The Sine, Cosine and Tangent in General 456
6.4 Relationships Among Trigonometric Functions 463
6.5 The Law of Sines and the Law of Cosines 469
Chapter Summary and Review Problems 479

CHAPTER 7 Modeling Periodic Behavior

7.1 Introduction to the Sine and Cosine Functions 483
7.2 Modeling Periodic Behavior with Sine and Cosine 494
7.3 Solving Equations with Sine and Cosine: The Inverse Functions 515
7.4 The Tangent Function 527
Chapter Summary and Review Problems 533

CHAPTER 8 More About the Trigonometric Functions

8.1 Relationships Among Trigonometric Functions 539
8.2 Approximating Sine and Cosine with Polynomials 550
8.3 Properties of Complex Numbers 566
8.4 The Road to Chaos 574
Chapter Summary and Review Problems 581
Gord.3896.FM.pgs 5/2/03 10:34 AM Page v

Contents v

CHAPTER 9 Geometric Models

9.1 Introduction to Coordinate Systems 583
9.2 Analytic Geometry 585
9.3 Conic Sections: The Ellipse 595
9.4 Conic Sections: The Hyperbola and the Parabola 606
9.5 Parametric Curves 619
9.6 The Polar Coordinate System 630
9.7 Families of Curves in Polar Coordinates 636
Chapter Summary and Review Problems 644

CHAPTER 10 Matrix Algebra and Its Applications

10.1 Geometric Vectors 649
10.2 Linear Models 659
10.3 Scalar Products 670
10.4 Matrix Multiplication 686
10.5 Gaussian Elimination 700
Chapter Summary and Review Problems 718

CHAPTER 11 Probability Models (Available on website only)

11.1 Introduction to Probability Models
11.2 Binomial Probability and the Binomial Formula
11.3 Using Probability to Investigate Polynomials
11.4 Geometric Probability
11.5 Estimating Areas of Plane Regions
11.6 The Normal Distribution Function
11.7 Waiting Time at a Red Light
11.8 Random Patterns in Chaos
Chapter Summary and Review Problems

CHAPTER 12 More About Difference Equations

(Available on website only)
12.1 Solutions of Difference Equations
12.2 Constructing Solutions of First Order Difference Equations
12.3 Modeling with First Order Non-Homogeneous, Difference Equations
12.4 Financing a Car or a Home
12.5 Solving the Fibonacci and Other Second Order Difference Equations
12.6 The Predator-Prey Model
Gord.3896.FM.pgs 5/2/03 10:34 AM Page vi

vi Contents

12.7 Competing Species

12.8 Iteration and Chaos
Chapter Summary and Review Problems

APPENDIX A Some Mathematical Moments to Remember 721

APPENDIX B Solving Systems of Linear Equations Algebraically 727

APPENDIX C Solving Systems of Linear Equations using Matrices 731

APPENDIX D Symmetry 732

APPENDIX E Arithmetic of Complex Numbers 734

APPENDIX F Introduction to Data Analysis 737

APPENDIX G 2002 World Population Data 740

ANSWERS Selected Answers 746

Index 775
Gord.3896.FM.pgs 5/2/03 10:34 AM Page vii

Preface

To the Student
Picture yourself as a homeowner whose only tools are a set of screwdrivers. You are
perfectly capable of driving screws into or out of wood. But what about other jobs
around the house? A screwdriver is not useful for banging a nail into the wall or cutting
a board in two. Other tools are needed, and for larger jobs, power tools are essential.
In many ways this situation is analogous to standard mathematics courses in
which the emphasis has been almost completely on algebraic methods. These meth-
ods give you a powerful set of tools—you can collect like terms, factor various ex-
pressions, cancel common factors, expand powers of binomial terms, multiply out
polynomials, and so forth. But there are many jobs requiring mathematics that can-
not be solved at all using only algebraic methods. For such problems other mathe-
matical tools, including graphical and numerical methods, are far more useful.
For instance, suppose a doctor wants to study a patient’s heartbeat using an
EKG or a patient’s brainwaves using an EEG. There are no known formulas to ex-
press these quantities algebraically, but the doctor certainly can get critical infor-
mation about a patient by interpreting the graphs produced by these devices.
Suppose an engineer develops a new tread design for automobile tires and wants to
test its braking effectiveness for a car going 20, 30, 40, 50, and 60 miles per hour.
There is no exact formula for either the braking distance or the time until the car
comes to rest—too many unpredictable factors are involved. All the engineer has
is a set of measurements from the experimental runs, and he or she must make
decisions based on an understanding of what information the data provides.
Both cases illustrate themes that run through this book. We focus on the appli-
cations of mathematics to situations all around us and on the function concept that
allows us to study these phenomena. That’s why this book is titled Functioning in the
Real World. In this course you will learn to use a combination of algebraic, graphical,
and numerical methods, depending on which is the most helpful tool in any given
context. You will develop an understanding of the mathematical concepts and learn
how to apply them to realistic problems, and not merely perform operations me-
chanically. You will learn to interpret results, not just obtain answers. You will use
technology as a tool for answering the kinds of questions that arise naturally. But this
tool is not intended merely to give you answers; technology will help you learn the
mathematics. Above all, you will increase your ability to think mathematically so that
you can apply mathematics in many other arenas—in other math courses, in courses
in other disciplines, in your eventual careers, and in all other aspects of your life.
To do all this, you will look at a much wider variety of topics than you probably
have seen in previous math courses. You also will face much more varied types of
problems than you might have encountered before. Many of these problems will re-
quire you to think about the mathematics, not just redo a worked-out example from
the text with the numbers changed slightly. To do such problems, you will have to und-
erstand the mathematical ideas, not merely memorize solutions. To gain that under-
standing, you will have to pay careful attention in class and read the text thoroughly.
Real-world problems, such as the realistic applications in this book, involve far
more than solving an equation that someone hands you. Instead, you face a situation

vii
Gord.3896.FM.pgs 5/2/03 10:34 AM Page viii

viii Preface

from which you have to identify and extract the mathematical component (which
usually means creating an equation), you then have to solve that equation (perhaps
using pencil and paper or more likely some kind of technology), you have to inter-
pret the results to make sure that they make sense and are realistic, and finally you
have to communicate your answers to other people.
In some ways, this course will be more challenging than others you have taken,
but it certainly will be much more rewarding because you will see the value of
mathematics all around you. We’d like to give you some suggestions that will make
things easier for you in this course:
◆ Read the book. It was written for you and is very readable.
◆ Work in teams outside of class. Students are very good at explaining mathe-
matical ideas to one another in terms that they understand.
◆ Attacking problems in teams is not only a good learning strategy but also
the way people in science, engineering, business, and other fields function in
the real world.
◆ Ask questions in class. Many of the problems are ideal for class discussions,
but it is very hard for a teacher to answer questions no one asks.
◆ Feel free to suggest your own interpretations. Many of the problems can be
approached in very different ways, and different students (and instructors)
are likely to come up with different solutions depending on their viewpoint.
◆ Talk to your instructor during office hours if you need help. Your professor
understands that this is a demanding course for some students and will be
happy to work with you. But you must seek out the help.
◆ If you use a graphing calculator, carry it to class and use it often to add a
graphical dimension to whatever you are studying.
◆ Some of you already may use more sophisticated computer programs, the
mathematical power tools that are widely available today. These might in-
clude software packages, such as DeriveTM, MathematicaTM, or MapleTM,
that can perform virtually any algebraic operation or a spreadsheet, such as
ExcelTM, that organizes data and produces sophisticated graphs quickly and
easily. Become comfortable with this software as soon as you can so that
they are familiar tools throughout the course.
◆ Realize that the most powerful and effective tool you have is your mind. It
does things no machine is capable of doing—thinking, understanding, cre-
ating, and interpreting.
Remember, a carpenter equipped with all the right tools is able to build almost
anything. Likewise, a student equipped with all the right mathematical tools and
the knowledge and judgment to select the right one is prepared for almost any-
thing. Above all, we hope that you will have a very exciting and rewarding experi-
ence as you are Functioning in the Real World.

To the Instructor
The mathematics curriculum is in the process of change to establish a better bal-
ance among geometric, numerical, symbolic, and verbal approaches. There is a
much greater emphasis on understanding fundamental mathematical concepts, on
Gord.3896.FM.pgs 5/2/03 10:34 AM Page ix

Preface ix

realistic applications, on the use of technology, on student projects, and on more

active learning environments. These approaches tend to make greater intellectual
demands on the students compared to traditional courses that place heavy empha-
sis on rote memorization and manipulation of formulas.
With support from the National Science Foundation, the Math Modeling/
PreCalculus Reform Project developed a new precalculus or college algebra/
trigonometry experience with the following goals:
◆ Extend the common themes in most of the calculus reform projects.
◆ Reflect the common themes for new curricula and pedagogy as called for in
the MAA/CUPM recommendations, the AMATYC Crossroads standards
recommendations, and the NCTM Standards recommendations.
◆ Focus more on mathematical concepts and mathematical thinking by
achieving a balance among geometric, numerical, symbolic, and verbal ap-
proaches rather than focusing almost exclusively on developing algebraic
skills.
◆ Provide students with an appreciation of the importance of mathematics in
a scientifically oriented society by emphasizing mathematical applications
and models.
◆ Introduce some modern mathematical ideas and applications that usually
are not encountered in traditional courses at this level.
◆ Provide students with the skills and knowledge they will need for subse-
quent mathematics courses.
◆ Reflect the major changes in the mathematical needs of other disciplines
and so provide students with the skills and knowledge that they will need
for courses in the disciplines.
◆ Make appropriate use of technology without becoming dominated by the
technology to the exclusion of the mathematics.

Philosophy of the Project

To accomplish these goals, we have adopted several basic principles advocated by
most leading mathematics educators.

Students should see the power of mathematics.

Most students take precalculus or college algebra/trigonometry courses at the col-
lege level because they are prerequisites for future math courses or for courses in
one of the client disciplines or as general education requirement, not because they
are turned on by the elegance of mathematics. When students see interesting and
significant applications, they see the power of mathematics and so are willing to
put in the effort needed to learn the subject.
Such applications are the primary focus of the Functioning in the Real World
course. For instance, in Chapter 2, we build mathematical models for the growth of
populations and for the decay in the level of a medication in the blood. In Chapter 3
we analyze the data related to the Challenger disaster showing how mathematics could
have been used to decide against the launch that day. In Chapter 4 we analyze the
spread of AIDS to determine what kind of mathematical model best describes its
Gord.3896.FM.pgs 5/2/03 10:34 AM Page x

x Preface

growth. In Chapter 7 we see how periodic phenomena in nature, such as temperatures,

tides, or hours of daylight over the course of a year, can be modeled mathematically.

Students should focus on mathematical ideas,

not mathematical calculations.
Our goal is to achieve mathematical understanding, which too often is lost when
students concentrate on routine computations that can, should, and in practice
will be done by machine. Computer Algebra Systems (CAS) now perform virtually
any type of manipulation we could ever expect a student to do. Even if such sys-
tems are not necessarily incorporated into the new calculus courses, their existence
has major implications about what is important to teach. Further, graphing calcu-
lators allow us to study considerably more complicated problems from a geometric
or numeric perspective than conventional problems that require factoring artifi-
cially constructed functions. Consequently, there is less need to develop as broad a
level of manipulative skills in precalculus courses as in the past. However, the alge-
bra that arises in contexts and modeling throughout the book is very substantial,
especially the properties of exponential and logarithmic functions. Rather than fo-
cusing on producing students who are little better than imperfect organic clones of
CAS systems, we emphasize the power of the human mind to inquire, explore, an-
alyze, and interpret.
Technology certainly has a role in the Functioning course. We use graphing
technology to compare the growth behavior of different members within a family
of functions or to compare the behavior among different families of functions.
This theme comes up repeatedly throughout the book. Another technological
theme is to find the equation of the function (linear, exponential, power, logarith-
mic, trigonometric, logistic, and so on) that best fits real sets of data. In Chapter 5
we use technology to generate and display solutions of difference equations and in-
vestigate their dependence on initial conditions.
Whether you choose to have your students use graphing calculators, CAS,
spreadsheets, or other computer packages, technology should be used to motivate
and explain the mathematical ideas, not just to produce answers. Whatever tech-
nology you and your students use, we firmly believe that the overriding focus
should remain on the mathematics and not on the machinery.

Students should DO mathematics, not just passively

watch mathematics.
The typical mathematics course involves classroom lectures and homework. But
mathematics is all around us, and the best way to appreciate its power and useful-
ness is to apply it directly. To achieve this, we suggest having students do several
mathematical projects either individually or in small groups. Such projects give the
students the opportunity to “get their hands dirty” by:
◆ formulating mathematical questions
◆ collecting appropriate sample data
◆ analyzing that data
◆ drawing conclusions based on the analysis
◆ preparing reports, which promote the organization of ideas as well as their
communication skills
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xi

Preface xi

For instance, when we look at linear functions early in Chapter 2, students can be
asked to select a set of data of interest to them and estimate, by eye, the equation of
the line that best fits the data. In Chapter 3, students can be asked to fit a variety of
functions (linear, exponential, power, and logarithmic) to a set of data and make
predictions based on the results. (See the Instructor’s Resource Manual for addition-
al suggestions related to each chapter.) The results of doing projects increase the
students’ level of enthusiasm for the subject matter and their understanding of the
mathematical ideas.

Students should be exposed to a broad view of mathematics.

Traditional precalculus courses focus exclusively on preparing students for calculus,
particularly in terms of the algebraic skills they may need. The Functioning course is
intended to prepare students for mathematics in a more general sense. We have in-
corporated a variety of nontraditional topics in the course that simultaneously in-
troduce new mathematical ideas while advancing the students toward a calculus
experience.
◆ The laboratory sciences routinely use mathematics to analyze laboratory
data, but rarely explain the underlying mathematical foundations—students
may use semi–log paper or log–log paper to produce the appropriate results
without understanding why. Throughout the book, we emphasize the con-
cept of fitting functions to data using real-life situations, which motivates
the mathematical ideas as well as providing contexts in which to develop
important algebraic skills at a high level. For example, in Chapter 3 we ana-
lyze the growth of the U.S. population via an exponential function; in Chap-
ter 5 we return to the U.S. population and attempt to fit it with a logistic, or
inhibited growth, model.
◆ In most calculus reform courses the emphasis on differential equations has
greatly increased. We extensively discuss difference equations in Chapter 5 to
introduce most of the comparable ideas and models in a discrete setting.
This discussion reinforces critical ideas on functional behavior and the mod-
eling of real-world phenomena while providing opportunities for honing
important algebraic skills. We return to the study of difference equations in
supplementary Chapter 12 (available on the web at www.aw.com/ggts).
◆ Probabilistic reasoning has become increasingly important in recent decades,
and we use it in supplementary Chapter 11 (also available on the web) and
we use it in the service of reinforcing other previous ideas and methods
from the study of functions.

What’s New in the Second Edition

In the second edition, we have tried to build on the strengths of the first edition. The
new edition contains a wealth of new applications, examples and problems, and all
real-world data sets have been updated. All concepts and methods are approached
using the Rule of Three: graphically, symbolically, and numerically.
The new edition has been reorganized and completely rewritten to provide a
slower pace through topics that some students find challenging. It also contains a
more prominent role for algebraic topics, where the algebraic steps involved in der-
ivations are now highlighted to assist students who may have forgotten some of the
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xii

xii Preface

algebra they learned in prior courses. Many of the problem sets now include col-
lections of problems, called Exercising Your Algebra Skills, to give those students
who need it some practice with routine algebra. The book also includes a consider-
ably expanded treatment of trigonometry and the use of the trig functions as mod-
els of periodic behavior; there are now three chapters devoted to these ideas and
methods.
In particular, some of the major changes in the second edition are:

Chapter 2: Families of Functions The long section on linear function has been
split into two shorter sections to slow the pace. Similarly, the treatment of expo-
nential functions has been slowed by presenting the material in two sections, one
on exponential growth functions where the motivating illustration remains popu-
lation growth, and the other on exponential decay functions where the unifying
application is the level of a medication in the bloodstream. Both of these applica-
tions then serve as the key themes in later chapters on difference equation models.

Chapter 3: Fitting Functions to Data Many new examples, particularly relating

to power functions, have been added and considerably more emphasis has been
placed on the judgmental issues and mathematical reasoning involved in deciding
which function is the best fit to a set of data. A new optional section on multivari-
able linear regression has been added.

Chapter 4: Extended Families of Functions The material has been reorganized

to bring together all the discussions related to polynomial functions. New sections
have been added that relate the ideas on shifting and stretching functions to oper-
ations on tables of data. A new section has been added on the logistic and surge
functions as applications of the material on building new functions from old.

Chapter 5: Modeling with Difference Equations The introduction to difference

equation models has been totally rewritten and reorganized. The two unifying
themes are models for the level of a medication in the blood and population growth.
The more challenging material, particularly the heavy manipulative topics involving
methods for determining closed form solutions, has been moved to Chapter 12.

Chapter 6: Introduction to Trigonometry A new chapter on right angle tri-

gonometry has been written for those students who need an exposure to this mate-
rial. The development starts with the tangent ratio, because it makes more sense in
most applications to consider the adjacent and opposite sides. The chapter in-
cludes the law of sines and the law of cosines.

Chapter 7: Modeling Periodic Behavior This chapter presents the use of the
trigonometric functions as models for periodic behavior.

Chapter 8: More About the Trigonometric Functions This chapter presents

more advanced ideas on the trigonometric functions, especially trigonometric
identities, complex numbers and DeMoivre’s theorem, and the use of DeMoivre’s
theorem in understanding some chaotic phenomena.

Chapter 9: Geometric Models The material on the conic sections has been split
into several sections, one on the ellipse and another on the hyperbola and parabo-
la. Additional applications of the hyperbola have been added.
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xiii

Preface xiii

Chapter 10: Matrix Algebra and its Applications A new section introducing
geometric and physical vectors has been added for those instructors who only have
time to develop these ideas. Additional examples and problems have been added
that link matrix methods more directly to previous topics in the book, including
functions, trigonometry, and finding equations of conic sections.
Chapter 11: Probability Models In the first edition, we had integrated some
ideas on probability throughout the text. However, all reports from users that we re-
ceived indicated that, unfortunately, very few people were able to take advantage of
this material because of time pressures. Accordingly, we collected all of this material
into a chapter on probability models. We have also added a new section introducing
the normal distribution and its uses. Because relatively few instructors have the time
to get to this chapter, it is being posted on the web at www.aw.com/ggts, where it is
available for downloading. Wherever appropriate, we refer to this chapter, or indi-
vidual sections, as a supplementary chapter or supplementary sections.
Chapter 12: More About Difference Equations As mentioned, the more so-
phisticated ideas and methods on difference equations have been combined into the
new Chapter 12. Moreover, additional sections have been added introducing sever-
al models based on systems of difference equations; these include the predator-prey
model and a model for competitive species. This supplementary chapter is also
available for downloading from the web at www.aw.com/ggts.

The Intended Audiences

The materials in this book were developed with different courses in mind:
1. An alternative to the usual one- or two-semester precalculus course de-
signed to prepare students for calculus, one that is in the spirit of the
MAA’s CUPM recommendations and AMATYC’s Crossroads Standards.
2. An alternative (perhaps as a course in mathematical modeling) to a one-
or two-semester course in college algebra and trigonometry.
3. An alternative to traditional high-school precalculus courses, one that is in
the spirit of the NCTM Standards and the recommendations of the Pace-
setter curriculum project.
4. An alternative to related courses that often are used as a terminal or cap-
stone mathematics course.
5. An alternative to a precalculus-level course for education majors.
We presume that the students taking this course have had a reasonably good mathe-
matical background at the level of intermediate algebra and a previous exposure to
some right-angle trigonometry and logarithms (but likely remember very little of it).
As an alternative to traditional precalculus and college algebra/trig courses, the
Functioning course certainly provides a strong preparation for a standard calculus
course. However, the primary emphases we have adopted make the approach par-
ticularly well-suited as preparation for virtually any calculus course. We put a
strong emphasis on:
◆ the applications of the mathematics
◆ mathematical reasoning and understanding of mathematical concepts, not
just symbol manipulation
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xiv

xiv Preface

◆ the use of the Rule of Three: Topics are approached geometrically, numeri-
cally, and symbolically wherever possible
◆ the use of verbal reasoning and communication skills

Overall, we believe that it is very important to emphasize repeatedly why you

are teaching a course with a very different approach. Students need to be reminded
of the reasons why you are expecting more of them and asking them to do different
things. Pointing out the limitations of purely algebraic methods or the power of
modern technology to solve problems that could not be touched just a few years
ago helps. Pointing out the type of traditional manipulative operations that now
can be done easily by machine also helps. Most importantly, remind students that
they now need to develop the thinking skills to know the questions to ask, to decide
which tools to use for answering those questions, and to develop the ability to in-
terpret and communicate these solutions.

Suggested Time Frame

The suggested pacing below is for a one-semester course that meets four hours per
week. By including the optional sections and Chapters 10–12, this text can give a
two-semester sequence that provides a strong foundation in precalculus ideas and
in applied mathematics, particularly discrete mathematics. The text lends itself to
many other possible courses and the suggested timetable can provide instructors
with some guidance of the time needed for various topics.

Chapter 1 Functions in the Real World Much of this chapter can be assigned as independent
1–1.5 Weeks reading. However, we suggest spending some of the
first week talking about and developing the critical
ideas on the behavior of functions and introducing stu-
dents to the function concept from geometric, numeri-
cal, symbolic, and verbal points of view.

Chapter 2 Families of Functions This chapter is critical in helping students to reach the
3 weeks same plateau of mathematical background and to make
Section: 2.2 and 2.3 Class hours: 3 the transition to a new way of looking at and thinking
2.4 and 2.5 3 about mathematics. You should give students time to
2.6 1 reorient themselves.
2.7 2
2.8 (optional) 0.5–1
2.9 1

Chapter 3 Fitting Functions to Data This chapter provides the link between mathematics
2–2.5 Weeks and the real world. It shows where functions come
Section: 3.1 and 3.2 Class Hours: 2 from, reinforces ideas about the behavior of families of
3.3 1.5–2 functions, and provides the opportunities to develop
3.4 1.5–2 important algebraic skills. Section 3.6 may be assigned
3.5 1.5–2 as reading.
3.6 (optional) 1
3.7 (optional) 1
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xv

Chapter 4 Extended Families of Functions This chapter extends the idea of families of functions to
2.5–3 Weeks polynomials. It also introduces the idea of constructing
Section: 4.1 Class hours: 1.5 new functions from old, including operations on func-
4.2 1.5 tions and shifting and stretching functions.
4.3 0.5–1
4.4 (optional) 0.5–1
4.5 (optional) 1
4.6 1.5
4.7 1.5
4.8 (optional) 1
4.9 (optional) 1

Chapter 5 Modeling with Difference Equations

1.5–2.5 Weeks This chapter develops and analyzes models for describ-
Section: 5.1 Class hours: 1.5 ing: • population growth • logistic growth • eliminating
5.2 1.5–2 drugs from the body • radioactive decay • Newton’s
5.3 1.5–2 laws of heating and cooling • geometric sequences and
5.4 1.5–2 their sums. If pressed for time, you may wish to select
5.5 1.5 some, but not all, of the models discussed.

Chapter 6 Introduction to Trigonometry

1.5 Weeks This chapter presents a brief introduction to right
Section: 6.1 Class hours: 1 angle trigonometry, including some of the simplest and
6.2 1–1.5 most common trig identities.
6.3 1
6.4 1.5
6.5 1.5

Chapter 7 Modeling Periodic Behavior

2–2.5 Weeks This chapter uses trigonometric functions to model pe-
Section: 7.1 Class hours: 2 riodic phenomena. The four parameters in sinusoidal
7.2 3–4 functions are introduced and developed in contexts in-
7.3 1.5 volving periodic phenomena.
7.4 1–1.5

Chapter 8 More About the Trigonometric Functions

2 Weeks This chapter • examines the relationship between the
Section: 8.1 Class hours: 2–2.5 trigonometric functions • approximates sine and cosine
8.2 2 functions with polynomials • examines the properties
8.3 1.5 of complex numbers • explains chaotic phenomena.
8.4 (optional) 1.5 Coverage of Section 8.4 may take several class hours,
particularly if you wish to include live computer graph-
ics demonstrations.

Chapter 9 Geometric Models This chapter includes analytic geometry, the conic sec-
2.5–3 Weeks tions, parametric curves, and curves in the polar coor-
Section: 9.1 Class hours: 0.5 dinate system. The two parts of this chapter, Sections
9.2 1.5 9.1–9.4 (analytic geometry) and Sections 9.5–9.7 (polar
9.3 2 coordinates and parametric curves) can be considered
9.4 2 as minichapters that could be covered independently, if
9.5 (optional) 1.5 so desired.
9.6 (optional) 1
9.7 (optional) 1.5

Detailed suggestions can be found in the Instructors Resource Manual.

xv
Gord.3896.FM.pgs 11/21/03 9:54 AM Page xvi

xvi Preface

Supplements

◆ The Instructor’s Solutions Manual (available through http://suppscentral.

aw.com) contains complete solutions to all problems for instructors.
◆ The Student Solutions Manual (ISBN 0-201-61137-6) contains complete
solutions to selected problems, particularly nonroutine problems.
◆ The Instructor’s Resource Manual (available through http://suppscentral.
aw.com) contains
— detailed suggestions for topic selection and pacing for the course,
— additional problems and examples,
— ideas for classroom activities,
— suggestions for student projects related to each chapter,
— suggestions for computer laboratory exercises and assignments, and
— samples of tests, exams, and project assignments.

Acknowledgments
We gratefully acknowledge the contributions of many people whose advice and as-
sistance immeasurably helped the project and the development of these materials.
Judy Broadwin (Jericho High School, NY) brought the secondary school per-
spective into the project, contributed many innovative suggestions for graphing
calculator usage, and assisted in project dissemination activities; we greatly ap-
preciate her ongoing efforts. We also want to thank Ben Fusaro (Florida State
University), Joe Fiedler (California State University-Bakersfield), Ignacio Alar-
cón (Santa Barbara College), Jim Sandefur (Georgetown University), Walter
Meyer (Adelphi University), Arlene Kleinstein (Farmingdale State University),
Yajun Yang (Farmingdale State University), Ray Bigliani (Farmingdale State
University), Anna Silverstein (New York Institute of Technology), Bob Feldman
(New York Institute of Technology), Walter Yurek (WorWic College), Tony Per-
essini (University of Illinois at Champaign-Urbana), Paula Maida (Western
Connecticut State University), I-Lok Chang (American University), Harry
Hauser (Suffolk Community College), William Abrams (Longwood College),
Anne Landry (Dutchess Community College), William Steger (Essex Commu-
nity College), Sylvia Sorkin (Essex Community College), and Chuck Laufman
(Westbury High School) for their valuable suggestions and contributions to the
project.
We are indebted to Elka Block and Frank Purcell for the lovely and thorough
job they did in producing both the Instructor’s Solutions Manual and the Student’s
Solutions Manual that accompany the book. We are also indebted to Kathy Bavelas
(Manchester Community College) for the very creative job she did in putting to-
gether the Instructor’s Resource Manual.We also want to thank Paula Maida (West-
ern Connecticut State University) for her creative contributions to the Instructor’s
Resource Manual. We also greatly appreciate the thorough job that Erica O’Leary
(Towson State University) did in preparing the short answers to all the problems.
We also acknowledge the careful and intensive work that Paul Lorczak (MathSoft,
Inc.) has done in checking the entire manuscript.
Gord.3896.FM.pgs 11/21/03 9:54 AM Page xvii

Preface xvii

We want to thank the National Science Foundation for their support and for
making this project possible. In particular, we are indebted to James Lightbourne,
William Haver, Elizabeth Teles, and Lee Zia at the NSF Division of Undergraduate
Education for their ongoing encouragement and assistance. It is truly appreciated.
We want to express our appreciation to the reviewers of the first edition for
their many contributions to the development of the book. For this edition, it is our
pleasure to thank the following reviewers, whose comments and suggestions re-
sulted in significant improvements.
Vic Akatsa Chicago State University
Kathleen Bavelas Manchester Community College
Therese Bennett Southern Connecticut State University
Laurie Burton Western Oregon University
Bob Dobrow Carleton College
William Fox Francis Marion University
Debra L. Hydorn Mary Washington College
Matthew Isom Arizona State University
Sandra Lofstock California Lutheran University
Mohammad Moazzam Salisbury State University
Bernd Rossa Xavier University
Marvin Stick University of Massachusetts–Lowell
Richard West Francis Marion University
We are indebted to the wonderful team at Addison-Wesley, who share our vision
for the course and helped bring this book to fruition. We appreciate all of the
outstanding contributions of Anne Kelly, Rachel S. Reeve, Cindy Cody, Jenny
Bagdigian, Becky Anderson, Karen Guardino, Greg Tobin, Julia Coen, and
Sharon Smith. They are a great team, and it has been a true pleasure working
with each and every one of these professionals. We also appreciate the wonderful
efforts of Beverly Fusfield at Techsetters and Pat McCutcheon at WestWords.

Sheldon P. Gordon Florence S. Gordon

E. Northport, NY E. Northport, NY

Alan C. Tucker Martha J. Siegel

Cold Spring Harbor, NY Baltimore, MD
Gord.3896.FM.pgs 5/2/03 10:34 AM Page xviii
Gord.3896.Front.EP.pgs 4/29/03 12:46 PM Page B

Function Equation Behavior Graph

Linear y  mx  b Strictly increasing when m  0. y
The more positive the slope, the faster the rate of in-
crease. y = mx + b
m>0
Strictly decreasing when m  0.
The more negative the slope, the greater the rate of x
decrease. y

y = mx + b
m<0

Exponential y  cx Strictly increasing when the growth factor c  1. y

(for c  0) The larger c is, the faster the function grows.
y = cx
Strictly decreasing when the decay factor 0  c  1. c>1
The smaller c is, the faster the function decays toward 1
zero. x
y
Exponential graphs are always concave up.
y = cx
c<1
1
x

y
Power y  xp Strictly increasing when p  0.
The larger p is, the faster the function grows beyond
x  1.
y = xp
If p  1, the graph is concave up—it grows more 0<p<1
and more rapidly. y = xp
If 0  p  1, the graph is concave down—it grows p>1
more and more slowly. x

Strictly decreasing for x  0 when p  0. y

The more negative p is, the faster the function decays
toward zero. y = xp
If p  0, the graph is concave up, for x  0. p<0

y
Logarithmic y  log x Strictly increasing.
x0 Logarithmic graphs are always concave down.
y = log x

x
1
Gord.3896.Front.EP.pgs 4/29/03 12:46 PM Page C

Function Equation Behavior Graph

Quadratic y  ax  bx  c
2
One turning point (the vertex) y
(polynomial Concave up when a  0
of degree 2) Concave down when a  0 a>0

a<0

Cubic y  ax 3  bx 2  cx  d Typically, two turning points y

(polynomial One inflection point

of degree 3) Rises toward  as x S 
when a  0 a<0

Falls toward  as x S 
when a  0
x

a>0

Quartic y  ax 4  bx 3  cx 2  dx  e Typically, three turning points y

(polynomial Typically, two inflection points

of degree 4) Rises toward  as x S 
when a  0
x
Falls toward  as x S 
when a  0

a>0

x
Gord.3896.01.pgs 4/24/03 9:23 AM Page 1

1
Functions in the
Real World
1.1 Functions Are All Around Us
The notion of function is a fundamental idea in mathematics. Functions are the basis
of most mathematical applications in nearly all areas of human endeavor. To see how
functions can arise in unexpected places, look at the graph shown in Figure 1.1.

1000
Elephant

Bull
Horse
100 Boar Cow

Sow
Metabolic rate (watts)

Man
Woman
Chimpanzee Sheep
Goat
10 Dog Cassowary
Goose Condor
Wild birds
Hen Cat
Rabbit
Guinea pig Marmot
1 Giant rats
Pigeon Rat
Small
birds

Mouse
0.1
0.01 0.1 1 10 100 1000 5000
(10 g) (100 g)
FIGURE 1.1 Mass (kilograms)

This graph appears in many introductory biology textbooks. It shows the re-
sults of a study comparing the masses of various mammals and birds with their
metabolic rates. The biologist who conducted the study first plotted the data—the

1
Gord.3896.01.pgs 4/24/03 9:23 AM Page 2

2 CHAPTER 1 Functions in the Real World

raw measurements on body mass (measured in kilograms) and metabolic rate

(measured in watts)—on a graph and then drew a line that passes very close to most
of the points. What does this graph show? It is clear from the pattern of the data
points that there must be some relationship between the body mass and the meta-
bolic rate of mammals and birds. If there were no relationship, the points would not
fall into such a clear pattern. Thus we conclude that, in some way, the metabolic rate
of an organism depends on the mass of that organism. Such a relationship is a
function, and we say that metabolic rate R is a function of body mass W.
Informally, a function is a rule that associates a set of values of one quantity
with a set of values of another quantity. Functions are usually represented in four
different ways:
1. by formulas or equations,
2. by graphs,
3. by tables, and
4. in words.
For instance, a function might be expressed as a mathematical formula such as
A  s2, which gives the area A of a square in terms of its side s. The equation might
be D  50t, which gives the distance D you travel at a constant rate of 50 mph in
terms of the time t that you drive. A function might be given as a graph, as in the
relationship between metabolic rate R as a function of body mass W of various or-
ganisms illustrated in Figure 1.1. A function might be given as a table of data. For
instance, you compute your income tax for the Internal Revenue Service by using a
table—for each level of taxable income, there is a corresponding tax levied, as
shown in Figure 1.2. The rule for a function might be expressed in words, as in

If line 37
(taxable And you are—
income) is—

At But Single Married Married

least less filing filing
than jointly sepa-
rately
Your tax is—
29,000
29,000 29,050 5,092 4,354 5,592
29,050 29,100 5,106 4,361 5,606
29,100 29,150 5,120 4,369 5,620
29,150 29,200 5,134 4,376 5,634
29,200 29,250 5,148 4,384 5,648
29,250 29,300 5,162 4,391 5,662
29,300 29,350 5,176 4,399 5,676
29,350 29,400 5,190 4,406 5,690
29,400 29,450 5,204 4,414 5,704
29,450 29,500 5,218 4,421 5,718
29,500 29,550 5,232 4,429 5,732
29,550 29,600 5,246 4,436 5,746
29,600 29,650 5,260 4,444 5,760
29,650 29,700 5,274 4,451 5,774
29,700 29,750 5,288 4,459 5,788
29,750 29,800 5,302 4,466 5,802
29,800 29,850 5,316 4,474 5,816
29,850 29,900 5,330 4,481 5,830
29,900 29,950 5,344 4,489 5,844
29,950 30,000 5,358 4,496 5,858

FIGURE 1.2
Gord.3896.01.pgs 4/24/03 9:23 AM Page 3

1.1 Functions Are All Around Us 3

“The cost of postage is 37 cents for the first ounce and 23 cents for each additional
ounce.”
Typically, when a functional relationship exists between two quantities, the
values of one of the quantities depends on the values of the other quantity. That is:
A function is a rule that assigns to each value of one quantity precisely one
related value of another quantity.
But, if one value of a quantity leads to two or more values of the other quantity, the
relationship between them is not a function. For instance, consider the relationship
between the number of home runs that a batter has hit by the end of the baseball
season and the number of runs he has batted in (RBIs). How many RBIs are asso-
ciated with 10 home runs? Many different players hit 10 home runs say, but each
likely had a different number of RBIs, so this relationship is not a function.

Representing Functions with Formulas and Equations

When you think of functions, the first thing you probably think of is a relation-
ship between two quantities that is given by a formula, such as A  pr2, which
gives the area A of a circle in terms of its radius r. Similarly, the ideal gas law from
chemistry, which says that P  kT>V, expresses the pressure P of a gas as a func-
tion of its temperature T, where V is the volume of the container that holds the
gas and k is a constant. The conversion between Fahrenheit and Celsius tempera-
ture readings, F  95 C  32, expresses the functional relationship between the
two temperature scales.
Frequently, when we observe that one quantity is a function of another, we
would like to determine an appropriate formula that expresses this relationship. For
example, throughout most of human history, people believed that objects fall at a
constant speed. Then, in about 1590, Galileo realized that this belief might not nec-
essarily be true. He also had the insight to realize that this conjecture could be tested
experimentally. Galileo conducted his now-famous experiments of dropping objects
from the top of the Tower of Pisa and found that they fell at ever-increasing rates and
that the weight of the objects didn’t affect how fast they fell. Galileo’s study of the re-
lationship between the distance that an object falls and the time it takes to fall was the
key connection for Newton that enabled him to develop his theories of motion that
transformed the physical sciences. Based on either Newton’s laws of motion or the
analysis of data from such an experiment, a formula for the height y of an object
dropped from the top of the 180-foot-high Tower of Pisa is y  180  16t 2, where t
is the number of seconds since the object was dropped. We show where this formula
comes from later in the book.

Representing Functions with Graphs

Many effective ways are used to display functions graphically in everyday life—in
newspapers, magazines, and scientific, business, and government reports. The
graph of a function is valuable because it displays accurate information about a
quantity while simultaneously giving an overview of the behavior of that quantity.
In particular, a graph can show any trends or patterns in the process being studied.
The graph shown in Figure 1.3 shows the increase in life expectancy in the Unit-
ed States in years since the beginning of the twentieth century. This graph is a func-
tion of time t because, for any given year, there is a single value for the life
Gord.3896.01.pgs 4/24/03 9:23 AM Page 4

4 CHAPTER 1 Functions in the Real World

90

75

Life expectancy (years)

60

45

30

15

0
1900 1915 1930 1945 1960 1975 1990 2000
Year
FIGURE 1.3

expectancy of a child born in that year. From the graph, for example, we can esti-
mate that a child born in 1900 would have had, on average, a life expectancy of
about 47 years, and that a child born in 1990 would have a life expectancy of about
75 years. The rise in life expectancy is a remarkable achievement due to advances in
science and medicine and improvements in lifestyles. However, there are also some
unfortunate aspects connected with living longer. Can you think of any?
From this graph, not only can you observe the rising trend, but you can also
look ahead to predict life expectancies in the not-too-distant future. Note that life
expectancy is not merely increasing, but it is actually increasing more slowly as
time goes by.

Think About This What is the significance of this growth pattern for life expectancy if it continues? ❐

Functions are displayed graphically in many ways in newspapers and maga-

zines. Keep an eye out for them in your daily activities.

Representing Functions with Tables

Consider the following table of values, which shows the acceleration of a Pontiac
Trans Am. The table gives the time in seconds needed to reach different speeds.

Final speed, v (mph) 30 40 50 60 70

Time, t (sec) 3.00 4.29 5.52 7.38 9.81

Although you may not have thought of something such as this as being a func-
tion, the time t needed for a Trans Am to achieve a certain speed v is a function of
the speed. There may not be an explicit formula for this time as a function of the
final speed, but it nevertheless satisfies the definition of a function: For each final
speed v, there is a unique time t needed for a Trans Am to accelerate to that speed.
We can plot these points and connect them with a series of straight line seg-
ments or even by a smooth curve, as shown in Figure 1.4. Note that the times de-
pend on the speeds. Thus we plot the speeds on the horizontal axis and the times
needed to achieve those speeds on the vertical axis. Also, you should realize that the
values in the table represent only the actual measured points. Drawing a smooth
curve through the points requires making assumptions about what happens be-
Gord.3896.01.pgs 4/24/03 9:23 AM Page 5

1.1 Functions Are All Around Us 5

v
FIGURE 1.4 0 10 20 30 40 50 60 70

tween those points. The curve is just an artist’s rendition of what the pattern could
be; the actual pattern might have some minor variations. Note that we have now
represented the same function by both a table and a graph.

Think About This Estimate the time needed for a Trans Am to accelerate from 0 to 45 mph and from
0 to 75 mph. Which estimate do you think is more accurate? Why? ❐

Now consider the following daily high temperatures in Phoenix during a se-
vere heat wave in June 1990.

Date 19 20 21 22 23 24 25 26 27 28 29

Temperature (F) 109 113 114 113 113 113 120 122 118 118 108

Note that a single high-temperature reading is associated with each day, so high tem-
perature is a function of the day. This function makes sense only for the 11 days—
June 19 through June 29—and its values consist of the high-temperature readings
108, 109, 113, 114, 118, 120, and 122.
However, the date is not a function of the high temperature because a given
temperature (say, 113°2 was reached on more than one date (in this case the 20th,
the 22nd, the 23rd, and the 24th).
The function that associates the high temperature in Phoenix with the corre-
sponding day of the month can be depicted graphically by plotting the individual
points, as shown in Figure 1.5, so again we have represented the same function by
both a table and a graph. The points in the figure can be joined by a series of line
segments or by a smooth curve to give a sense of an overall trend or pattern, as
shown in Figure 1.6. However, doing so requires some careful thought. When we
connect the points, we are not indicating that this graph represents temperature as
a function of time; we are just connecting the maximum temperatures recorded
each day, and the curve shown gives absolutely no information about the tempera-
ture at any intermediate time. In fact, the actual graph of temperature versus time
would typically show the type of oscillatory effect depicted in Figure 1.7.

Representing Functions with Words

Functions are expressed verbally in many different ways. The maximum load that a
jet plane can lift is related to its wingspan. The number of different species that can
live on an island depends on the size of the island. The population of the world over
Gord.3896.01.pgs 4/24/03 9:23 AM Page 6

6 CHAPTER 1 Functions in the Real World

T T

125 125

120 120

Temperature (°F)
Temperature (°F)

115 115

110 110

105 105

100 100

t t
19 20 21 22 23 24 25 26 27 28 29 19 20 21 22 23 24 25 26 27 28 29
Day in June 1990 Day in June 1990

FIGURE 1.5 FIGURE 1.6

125

120
Temperature (°F)

115

110

105

100

t
19 20 21 22 23 24 25 26 27 28 29
FIGURE 1.7 Day in June 1990

time is a function of time. If money is borrowed at simple interest (no compound-

ing), the amount of interest earned is a multiple of the amount borrowed.

Why Study Functions?

Any situation involving two quantities usually raises several questions:
1. Is there a functional relationship between the two quantities?
2. If there is a relationship, can we find a formula for it?
3. Can we construct a table or graph relating the two quantities, especially if
we can’t find a formula?
4. If we can find a formula, or if we have a graph of the relationship, or if we
have a table of values relating the two quantities, how do we use it? That is,
how can knowledge of the function aid in understanding the relationship
between the two quantities or allow us to make predictions or informed
decisions about one of the variables based on the other?
Figure 1.1 clearly suggests a relationship between metabolic rates R in mammals
and birds and their body mass W. This same relationship can then be used to predict
the metabolic rate of other species—say, lions or Kodiak bears—based on knowledge
of their mass. This relationship could even be used to predict the metabolic rate of an
extinct pterodactyl from estimates of its body mass made from its skeletal remains.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 7

1.1 Functions Are All Around Us 7

However, it could not be used to predict the metabolic rate for a crocodile because a
crocodile is neither a mammal nor a bird; the relationship observed in Figure 1.1 for
mammals and birds may not apply to reptiles.

Think About This Based on the graph shown in Figure 1.1, would you use the relationship to predict
the metabolic rate for extinct mammoths, which were slightly larger than today’s
elephant? ❐

Problems
1. Which of the following relationships are functions
and which are not? Explain your reasoning. For
those that are functions, identify which of the two
quantities depends on the other. Again, explain
your reasoning.
a. The number of miles driven in a car versus the
(iii) (iv)
number of gallons of gas used.
b. The price of a diamond versus the number of
carats.
c. The major league baseball player who has a certain
number of home runs at the end of the season.
d. The student who has a specific score on the SAT
in a particular year.
e. The amount of rain that falls on any particular (v) (vi)
day of the year in Seattle.
f. The day of the year on which given amounts of 3. The following graphs show the noise level of a
snow, in inches, fall in Buffalo. crowd of college students watching their school’s
basketball team playing at home in the champi-
2. Match each of the following functions with a corre-
onship finals for the league title. Match the three
sponding graph. Explain your reasoning.
graphs with the corresponding scenarios (reactions)
a. The population of a country as a function of and then draw a graph for the remaining scenario.
time.
a. Our team started slowly but eventually began to
b. The path of a thrown football as a function of
pull away.
time.
b. It was a disaster from start to finish.
c. The distance driven at a constant speed as a
function of time.
d. The daily high temperature in a city as a func-
tion of time over several years.
e. The number of cases of a disease as a function of
time.
f. The percentage of families owning VCRs as a
function of time. (i) (ii)

(i) (ii) (iii)

Gord.3896.01.pgs 4/24/03 9:23 AM Page 8

8 CHAPTER 1 Functions in the Real World

c. The score kept seesawing back and forth, but we life expectancy over the past century means. For ex-
finally won on a three-point shot at the buzzer. ample, you might consider it in terms of your own
d. Our team started well, then the opposition took expected life span compared to those of your chil-
the lead, but we finally won. dren and grandchildren. Alternatively, you might
4. Consider the scenario: “You left home to run to the consider the effects on the overall distribution of
local gym. You started at a constant rate of speed people of different ages in the population at large,
but sped up when you realized how energetic you or you might discuss the question of whether there
felt. About halfway there, you began to tire, so you is a natural limit to how long the human life span can
started slowing down.” Sketch a graph of your dis- be extended in the future. Compare the values for life
tance from home as a function of time. expectancies in the United States in Figure 1.3 with
the values for life expectancies of other nations
5. Sketch a graph of your distance from home as a
given in Appendix G.
function of time for each situation.
8. Which table of values represents a function and
a. You drove steadily across town, speeding up as
which doesn’t? Explain your reasoning.
traffic diminished until the road turned into a
highway. a.
x 0 3 6 1 5 2 4
b. You drove steadily toward town but slowed down
as the traffic increased. Eventually you inched y 8 6 2 2 4 5 3
forward around a car that had broken down be-
fore you could resume normal speed. b. x 0 2 3 4 1 3 5
c. You drove steadily but realized you had left
y 8 4 7 2 6 10 9
something behind, so you returned home and
then drove all the way to school without any
9. The Dow-Jones average of 30 industrial stocks is
further trouble.
probably the most closely watched measure of stock
d. You drove steadily across town but then had a
market performance. Below are the Dow values at
flat tire; after changing it, you drove much faster
the beginning of each year from 1980 to 2000.
so that you wouldn’t be too late for class.
Write a short paragraph describing the behavior of
6. For each of the scenarios in Problem 5, sketch a
the stock market over this period of time. When did
graph of the total distance you’ve traveled as a func-
it rise? When did it fall? Which years would have
tion of time.
been the best times to buy stocks? Which would
7. Consider again the graph in Figure 1.3. Write a have been the worst times to do so?
paragraph or two interpreting what the increase in

Year 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989

Dow 839 964 875 1047 1259 1212 1547 1896 1939 2169

Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Dow 2753 2634 3169 3301 3758 3834 5177 6447 7965 9184 11358
Source: Wall Street Journal.

1.2 Describing the Behavior of Functions

Functions are used to represent quantities in the real world. Because most of these
quantities change over time or depend on some other quantity, we need some ter-
minology to describe the behavior of the function—that is, how the function
changes. We can describe the behavior of a function in two different ways.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 9

1.2 Describing the Behavior of Functions 9

Increasing and Decreasing Functions

The first and most immediate aspect of behavior is whether the function is in-
creasing or decreasing. The graph of a function is increasing if, as you look from left
to right, the vertical values get larger; that is, the graph rises. We also describe this
behavior as growth. Similarly, the graph of a function is decreasing if, as you look
from left to right, the vertical values get smaller; that is, the graph falls. We also de-
scribe this behavior as decay. Figure 1.8 illustrates these characteristics.

Increasing function Decreasing function

FIGURE 1.8

For instance, the world’s population is growing. Therefore the function that
expresses the population over time is an increasing function. Also, the heavier a car
is, the lower its gas mileage will be. Therefore the function that relates gas mileage
to the weight of a car is a decreasing function.
Of course, not every quantity merely increases or decreases. Often, a quantity
will rise some of the time and fall some of the time, such as the height of a bouncing
ball, the value of the Dow-Jones average, or the high temperature recorded in a par-
ticular location each day of the year. Thus a function whose graph looks like the one
shown in Figure 1.9 increases for some values of the variable and decreases for oth-
ers. Here the function rises (increases) to a maximum or largest value compared to
nearby points, then falls (decreases) to a minimum value compared to nearby
points, and then rises again. We call any point where the behavior of the function
changes from increasing to decreasing or from decreasing to increasing a turning
point of the function. Turning points occur at points where a function reaches a
local maximum (the value where the function is larger than any nearby value) or a
local minimum (the value where the function is smaller than any nearby value).
y

Increasing
Turning points
Turning point
Decreasing

Turning point
Increasing 0 t
FIGURE 1.9 FIGURE 1.10 t=0 t=4 t=6 t = 10

Note that a function increases or decreases over an interval of values on the hori-
zontal axis; it has a turning point at a particular point corresponding to a single value
along the horizontal axis. Figure 1.10 shows a function of t decreasing from t  0 to
t  4, increasing from t  4 to t  6, decreasing from t  6 to t  10, and then in-
creasing after t  10. This function has turning points at t  4, t  6, and t  10.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 10

10 CHAPTER 1 Functions in the Real World

Think About This Sketch the graph of a different function having the same behavior. ❐

We describe a function that is always increasing or always decreasing (it has no

turning points) as being strictly increasing or strictly decreasing.

Concavity: How A Function Bends

There is a second aspect to a function’s behavior. Figure 1.11 shows two increasing
functions. How do they differ? In Figure 1.11(a), the function isn’t merely increasing;
it is actually increasing faster and faster as time goes by. Think of the curve as bending
upward. For instance, population growth typically follows this type of growth pattern.
The function shown in Figure 1.11(b) is also increasing, but it is increasing more and
more slowly as time goes by. Think of the curve as bending downward. For instance,
the increasing human lifespan previously depicted in Figure 1.3 grows in this way.

Bending downward

Bending upward

FIGURE 1.11 (a) (b)

Now look at the two decreasing functions in Figure 1.12. The function shown
in Figure 1.12(a) decreases very rapidly at first and then more slowly as time pass-
es—it is decreasing at a decreasing rate. For instance, if a pollutant is released into a
lake, the level of pollution in the lake will decrease ever more slowly as time goes by.
Like the function shown in Figure 1.11(a), this curve is also bending upward. The
graph in Figure 1.12(b) also decreases, but it is decreasing slowly at first and then
more and more rapidly—it is decreasing at an increasing rate. For instance, if an ob-
ject is tossed off the roof of a tall building, its height above ground will decrease in
this manner as it speeds up in its descent because of the effects of gravity. Note that
this curve is bending downward, as is the curve shown in Figure 1.11(b).

Bending upward

Bending downward

FIGURE 1.12 (a) (b)

We use the term concavity to describe the way a function bends. Curves that
bend upward, such as those shown in Figures 1.11(a) and 1.12(a), are concave up.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 11

1.2 Describing the Behavior of Functions 11

Concave up Concave down

FIGURE 1.13 (a) (b)

Note that one curve is increasing and that the other is decreasing, so concavity is a
completely different concept from increasing/decreasing. Similarly, curves that
bend downward, such as those shown in Figures 1.11(b) and 1.12(b), are concave
down. Again, note that one is increasing and the other is decreasing. Figure 1.13(a)
illustrates the two types of concave up behavior, and Figure 1.13(b) illustrates the
two types of concave down behavior.

Think About This Imagine a ball bouncing up and down across the floor in front of you. Is the path
of the ball concave up or concave down? ❐

Just as a function can be increasing over one interval and decreasing over an-
other, a function can be concave up over one interval and concave down over an-
other. For instance, think of the behavior of the Dow-Jones average. This function
is increasing during some time intervals and is decreasing during other time inter-
vals, as shown in Figure 1.14. It is also concave up over some time intervals and is
concave down over other time intervals.

Closing Shares for the DJIA for Selected Dates

12,000
10,000
Number of Shares

8,000
6,000
4,000
2,000

1987 1989 1991 1993 1995 1997 1999 2001

FIGURE 1.14 Year

A point on a graph where the concavity changes from concave up to concave

down or vice versa is called a point of inflection or an inflection point. In Figure 1.15,
we show two curves, one having a point of inflection where the curve changes
from concave up to concave down and the other where the curve changes from
concave down to concave up. Observe that neither point of inflection occurs at
the turning points where the curve reaches a local maximum or a local mini-
mum, so turning points are not the same as inflection points.
Note that the function on the left in Figure 1.15 grows faster and faster to the left
of the inflection point and then grows slower and slower to the right of the inflection
point. As a result, the function is growing most rapidly at the inflection point.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 12

12 CHAPTER 1 Functions in the Real World

Concave down Concave down

Point of inflection Point of inflection

Concave up Concave up
FIGURE 1.15

Think About This What happens at the inflection point of the function on the right in Figure 1.15? ❐

We summarize the preceding information about functions as follows:

A function of x is increasing if the values of the function increase as x

increases.
A function of x is decreasing if the values of the function decrease as x
increases.
The graph of an increasing function rises from left to right.
The graph of a decreasing function falls from left to right.
The points where a function changes from increasing to decreasing or
from decreasing to increasing are the turning points.
The graph of a function is concave up if it bends upward.
The graph of a function is concave down if it bends downward.
The points where the concavity changes from concave up to concave
down or from concave down to concave up are the points of inflection
or inflection points.

In addition, the rate of change and concavity of the graph of a function are re-
lated in the following ways.

If the graph of a function is increasing and concave up, it is increasing at

an increasing rate.
If the graph of a function is increasing and concave down, it is increasing
at a decreasing rate.
If the graph of a function is decreasing and concave up, it is decreasing at
a decreasing rate.
If the graph of a function is decreasing and concave down, it is decreasing
at an increasing rate.
A function grows fastest or decays fastest at a point of inflection.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 13

1.2 Describing the Behavior of Functions 13

E XAMPLE 1
Identify all intervals where the function f shown in Figure 1.16 is
a. increasing;
b. decreasing;
c. concave up;
d. concave down.
Then indicate all points where the function has a
e. turning point;
f. local maximum;
FIGURE 1.16
g. local minimum;
h. point of inflection.
Solution For a–d, the task is to find the intervals of x-values where the different types of
behavior occur. We begin by redrawing the graph and introducing all the points x1 ,
x2 , . . . , x9 where the behavior of the function changes, as shown in Figure 1.17.
a. The function is increasing for values of x between x3 and x5 and again between x7 and
x9 , as shown in Figure 1.17(a).
b. The function is decreasing between x1 and x3 and again between x5 and x7 , as shown
in Figure 1.17(a).
c. The curve is concave up between x2 and x4 and again between x6 and x8 , as shown
in Figure 1.17(b).
d. The function is concave down between x1 and x2 , between x4 and x6 , and again from
x8 to x9 , as shown in Figure 1.17(b).

y y

x x
x1 x2 x3 x4 x5 x6 x7 x8 x9 x1 x2 x3 x4 x5 x6 x7 x8 x9
FIGURE 1.17 (a) (b)

Next, we look for particular points on the curve.

e. The turning points for this function are at x  x3 , at x  x5 , and at x  x7 .
f. The function has a local maximum at x  x1 (when compared to other nearby
points); the function also has a local maximum at x  x5 and again at x  x9 .
g. Similarly, the function reaches a local minimum at x  x3 (when compared to other
nearby points) and again at x  x7 .
h. The points of inflection occur where the concavity changes, which happens at
x  x2 , at x  x4 , at x  x6 , and at x  x8 .
◆
Gord.3896.01.pgs 4/24/03 9:23 AM Page 14

14 CHAPTER 1 Functions in the Real World

E XAMPLE 2
Two functions are defined in the accompanying tables of values. Describe the behav-
x 1 2 3 4 ior of each function.
y 6 11 15 18 Solution Both functions are obviously increasing as x increases. Note that the first
function grows first by 5 (from 6 to 11), then by 4 (from 11 to 15), and then by 3 (from
x 1 2 3 4 15 to 18), so it is growing at a decreasing rate. Therefore it is concave down. The second
function, however, grows by larger and larger amounts—first by 5 (from 6 to 11), then
y 6 11 17 24 by 6 (from 11 to 17), and then by 7 (from 17 to 24)—so the function is increasing at an
increasing rate and thus is concave up. Plot the points to verify both behaviors.
◆
Periodic Behavior
Another behavior pattern for functions is extremely common in real life. Many
natural processes have the property of being periodic—that is, the pattern repeats
over and over. We see this in the height of tides that rise and fall in the same pat-
tern roughly every 12 hours in most coastal locations. It also occurs in the pattern
of temperature readings in any location from one year to the next. Spotting a pe-
riodic function from its graph is easy: The identical pattern appears repeatedly.
For instance, consider the following data based on historical records giving the av-
erage number of tornados reported in the United States, per month, in a typical
year.

Month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Tornados 16 24 60 111 191 179 96 66 41 26 31 22

Source: National Oceanic and Atmospheric Administration.

200
Number of tornados reported

175
150
125
100
75
50
25
0
FIGURE 1.18 Jan Mar May Jul Sep Nov

Figure 1.18 shows a graph of these points. Note, either from the table or the
graph, how the values increase from a minimum level of tornado sightings in Janu-
ary to a maximum number in May and then decrease toward the minimum as the
year ends. Because these values are based on historical averages, this cycle will like-
ly repeat yearly with little change from one year to the next. It is therefore a rough-
ly periodic phenomenon. Figure 1.19 shows a smooth curve that captures the
longer term behavior of this roughly periodic function.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 15

1.2 Describing the Behavior of Functions 15

Number of tornados
January, Year 1 January, Year 2 January, Year 3

FIGURE 1.19

Problems
1. Which of the functions are strictly increasing, function is increasing. With the other, trace all parts
strictly decreasing, or neither? of the curve where the function is decreasing. Then
a. The cost of first-class postage on January first of mark all turning points on the curve.
each year. 3. Consider the function shown in the accompanying
b. The time of sunrise associated with each day of the graph. Use two different colored pens or pencils.
year. With one, trace all parts of the curve where the func-
c. The high temperature associated with each day tion is concave up. With the other, trace all parts of
of the year. the curve where the function is concave down. Then
d. The closing price of one share of IBM stock for mark all points of inflection on the curve.
each trading day on the stock exchange.
e. The area of an equilateral triangle in terms of its
base b.
f. The height of a bungee jumper t seconds after
leaping off a bridge.
g. The height of liquid in a 55-gallon tank h hours
after a leak develops.
h. The daily cost of heating a home as a function of
the day’s average temperature.
i. The world record times for running the 100 meter
dash.
2. Consider the function shown in the accompanying 4. Sketch the graph of a single smooth curve that is
graph. Use two different colored pens or pencils. first increasing and concave up, then increasing
With one, trace all parts of the curve where the and concave down, and finally decreasing and con-
cave down. Mark all turning points and points of
inflection on your curve.
5. Sketch the graph of a single smooth curve that is
first decreasing and concave up, then increasing
and concave up, and finally increasing and con-
cave down. Mark all turning points and points of
inflection on your curve.
6. Sketch a possible graph of the temperature in your
hometown over an entire week as a function of time.
On the graph indicate all the turning points. Where
Gord.3896.01.pgs 4/24/03 9:23 AM Page 16

16 CHAPTER 1 Functions in the Real World

is the temperature function increasing? Where is it 12. Each part of the table of values below defines a func-
decreasing? Where is the temperature function con- tion. Determine the concavity of each function.
cave down? Where is it concave up?
x y x y x y
7. Each year the world’s annual consumption of water
rises, as does the amount of increase in water con- 1 36 10 160 3 84
sumption. Sketch a graph of the annual world con-
2 31 15 172 7 74
sumption of water as a function of time.
8. A human fetus grows rapidly at first and then grows 3 27 20 189 11 61
with decreasing rapidity. Draw a graph showing the 4 24 25 209 15 45
size of a fetus as a function of time.
9. Sales of microwave ovens grew slowly when they 5 22 30 243 19 22
were first introduced and then increased dramati-
13. The Apollo-12 mission involved a flight to the
cally as more people appreciated their usefulness.
moon (250,000 miles from Earth), five circular or-
Eventually, sales began to slow as most households
bits about the moon, and a return to Earth.
already owned one. Sketch the graph of microwave
oven sales as a function of time. Indicate the loca- a. Assume (incorrectly) that the spacecraft traveled
tion of the point of inflection. at a constant speed between the Earth and the
moon. Sketch a rough graph of the distance from
10. Sales of VCRs grew slowly at first and then in-
Earth as a function of time.
creased tremendously as people came to accept
b. Assume (correctly) that the spacecraft’s speed
them widely. Eventually new sales began to level off
diminished the farther it got from Earth’s gravity
as market saturation neared. Sketch a possible
until it neared the moon and then increased due
graph of the percentage of U.S. homes owning a
to the moon’s gravitational force. The behavior
VCR as a function of time, paying careful attention
of the spacecraft’s speed reversed on the return
to the behavior of the function. Indicate any turn-
trip. Sketch a rough graph of the spacecraft’s dis-
ing points and points of inflection.
tance from the Earth as a function of time.
11. The Environmental Protection Agency (EPA) mon- (Think concavity!)
itors the levels of industrial pollutants in many
14. Water is being poured, at a constant rate, into vases
lakes and rivers. The following graphs show the
having the shapes shown. Sketch a graph showing
level of pollutants L in four different lakes as a func-
tion of time t. For each, write a short paragraph ei- a.
ther from the point of view of the EPA bringing
charges against a company for polluting or from the
point of view of a company defending itself against
such charges.
a. L b. L

b.
t t

c. L d. L

t t
Gord.3896.01.pgs 4/24/03 9:23 AM Page 17

1.2 Describing the Behavior of Functions 17

the level of the water as a function of time, paying d.

careful attention to concavity. What is the signifi-
cance of each of the points of inflection, if any?

c. —3 —2 —1 1 2

e.

—3 —2 —1 1 2

f.

d.

—3 —2 —1 1 2

g.

—2 —1 1 2
15. Decide which functions in a–j are periodic. (As-
h.
sume that the graphs continue indefinitely to the
left and right in the same pattern.) 1

a. —1 1 2

i.
—3 —2 —1 1 2 3

1 2 3
b.

j.

—3 —1 1 2 3
10 30

c.

16. Janis trims her fingernails every Saturday morning.

Sketch the graph of the length of her nails as a func-
—3 —2 —1 1 2 3 tion of time. Is this process periodic?
Gord.3896.01.pgs 4/24/03 9:23 AM Page 18

18 CHAPTER 1 Functions in the Real World

17. Craig is a perfectly normal individual with a pulse radio and TV signals, occur in periodic cycles. The
rate of 60 beats per minute and a blood pressure of accompanying figure is a graph of the number of
120 over 80. Thus his heart is beating 60 times> sunspots observed each year.
minute and his blood pressure is oscillating between a a. Estimate the period of the sunspot cycle.
low (diastolic) reading of 80 and a high (systolic) b. Estimate when the next two peaks will occur in
reading of 120. Sketch the graph of his blood pressure the cycle.
as a function of time. Be sure to indicate appropriate c. Suppose that you were required to come up with a
scales on each axis. reasonable estimate for the maximum number of
18. a. The thermostat in Sylvia’s home in Baltimore is set at sunspots that will occur during the next peak in
66°F during the winter. Whenever the temperature the cycle. How might you create such an estimate
drops to 66° (roughly every half-hour), the furnace based on the information given in the figure?
comes on and stays on until the temperature reaches
70°. Sketch the graph of the temperature in her house
as a function of time. Be sure to indicate appropriate
scales on each axis.
200
b. Gary, who lives in upstate New York, also has his
thermostat set to come on at 66°F. How will a
sketch of the temperature in his house differ from 150

Number of sunspots
the one you drew in part (a) for Sylvia’s house?
c. Jodi, who lives in central Florida, likewise has her
100
thermostat set to come on at 66°. How will a
sketch of the temperature in her house differ
from the other two? 50
19. Astronomers have been observing sunspots on the
face of the sun for centuries. These dark spots on
0
the sun, which are accompanied by the release of 1940 1950 1960 1970 1980 1990 2000
bursts of electromagnetic radiation that disrupt Year

1.3 Representing Functions Symbolically

A function is a rule that associates one and only one value of a quantity (say, y) with
each value of another quantity (say, x). The quantities x and y are variables. We use a
single letter, such as f, g, or h, as the name of a function. The particular formula for
the function, if it is known, is usually written as y  f 1x 2. It is read as “y equals f of
x” or possibly “y is a function of x.” For instance, the function f that takes any real
number x and squares it can be written as
y  f 1x 2  x2.
Some particular values of this function are
f 132  32  9; f A 13 B  A 13 B 2  19 ;
f 15 2  152 2  25; f 10.02 2  10.02 2 2  0.0004;
f 10.012  10.01 2 2  0.0001; f 1p2  p2  9.8696.
In each case we replaced the variable x in f 1x 2  x2 with the indicated value,
x  3, x  13 , and so on, and then evaluated the expression 32, A 13 B 2, and so on.
Similarly, the function g that takes the square root of any nonnegative real
number x can be written as

y  g 1x 2  2x .
Gord.3896.01.pgs 4/24/03 9:23 AM Page 19

1.3 Representing Functions Symbolically 19

For instance, some values of this function g are

g 116 2  216  4;
g 10.042  20.04  0.2;
g A 14 B  3 14  12 .
But g 125 2  125 does not make sense because it isn’t possible to take the square
root of a negative number in the real number system. That is, the function g is not de-
fined for x  25.
The function h that gives the reciprocal of any nonzero number x can be
written as
y  h1x 2  1x .
Some values of this function h are
h15 2  15  0.2;
h1200 2  200
1
 0.005;
h10.1252   0.125
1
 8.
But h10 2 does not make sense because division by 0 is not possible. That is, the
function h is not defined at x  0.
To work with functions requires some terminology. In the form y  f 1x 2, we
call x the independent variable because it can take on any appropriate value. We
call y the dependent variable because its value depends on the choice of x.
You can use letters other than f, g, or h to represent functions; other common
choices are F, G, or f 1 , f 2 , f 3 , and so on. You can use letters other than x to repre-
sent the independent variable; other common choices are t (for time), u (for an
angle), and r (for radius). Similarly, you can use letters other than y to represent the
dependent variable; for instance, you can use A for area, D for distance, P for pop-
ulation, and C for cost.
The area A of a circle is a function of its radius r—for each radius r, there is one
and only one area A. We write this function as A  f 1r2  pr 2. Here r is the inde-
pendent variable, A is the dependent variable because the area depends on the
choice of r, and f is the function. The distance D that a car moves in t hours at a
steady speed of 50 miles per hour (mph) is given by D  g 1t2  50t. Here t is the
independent variable, D is the dependent variable, and g is the function.
Suppose that you toss a ball straight up with an initial velocity of 64 feet per
second. The function
y  f 1t2  64t  16t2
gives the height in feet of the ball above ground level after t seconds, until the instant
that the ball hits the ground. Picture what happens. As the ball rises, it slows due to
the effect of gravity. Eventually it reaches a maximum height and then begins to fall
back to the ground. As the ball falls, its speed increases, again because of gravity.
Now let’s see how the function f gives the height of the ball above ground at any
time t. After half a second, when t  12 , the ball is 28 feet above the ground because
y  f A 12 B  64 A 12 B  16 A 12 B 2  32  4  28 feet.
After 1 second, it is at a height of
y  f 11 2  6411 2  1611 2 2  48 feet.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 20

20 CHAPTER 1 Functions in the Real World

After 2 seconds, it is at a height of

y  f 12 2  6412 2  1612 2 2  64 feet,
which happens to be the maximum height the ball reaches. After 3 seconds, the
height is
y  f 13 2  6413 2  1613 2 2  48 feet,
and the ball is on its way down. After 4 seconds, the ball is back at ground level because
f 14 2  6414 2  1614 2 2  0.

Domain and Range of a Function

In each of the preceding functions, there were some natural limitations on the pos-
sible values for both the independent variable and the dependent variable. The ball
is released at time t  0 and returns to the ground at t  4 seconds. It therefore
makes no sense in this problem to think about what happens before time t  0 or
after time t  4. Thus the permissible values for t are between 0 and 4 seconds.
Furthermore, the ball rises to its maximum height of 64 feet and then falls back to
the ground. Therefore the only meaningful values for the height of the ball are be-
tween y  0 and y  64 feet. (Of course, it is more realistic to think of throwing a
ball upward from about 4 or 5 feet above the ground rather from ground level—
we used ground level here just to simplify the mathematics.)
Similarly, the function y  g 1x 2  1x makes sense only if the independent
variable x is not negative. The possible corresponding values for y must be positive or
zero. The function y  h1x 2  1>x makes sense only if x is not zero. The possible
corresponding y-values of this function can be any number other than 0 because there
is no value of x such that y  1>x  0. Finally, for the function y  f 1x 2  x2, there
is no limitation on the possible values of x, but there certainly is a limitation on the
corresponding values for y  x2 because they can never be negative.
For any function f, the set of all possible values for the independent variable is
called the domain of f; the set of all possible values for the dependent variable is
called the range of f.
Typically, the domain and range consist of intervals of values for the inde-
pendent variable and the dependent variable, respectively. For instance, with the
function representing the height of the ball, the domain consists of the interval
from t  0 to t  4 and the range consists of the interval from y  0 to y  64.
We can also use inequalities to write these intervals expressing the domain as
0  t  4 and the range as 0  y  64.
Because each of these intervals contains endpoints (t  0 and t  4 for the
domain and y  0 and y  64 for the range), they are called closed intervals. We
can also write these intervals using interval notation, so that the domain is the
closed interval [0, 4] and the range is the closed interval [0, 64]. We use square
brackets to indicate that the endpoint value is included in the interval.
If one or both endpoint values is not included in an interval, we use paren-
theses instead of square brackets. For instance, if an interval is 3  x  6, where
both endpoints are not included, we write it in interval notation as the open in-
terval (3, 6). (Caution: Don’t misinterpret this notation as the coordinates of the
point with x  3 and y  6. The symbols are identical, but the meaning should
be clear from the context.)
Gord.3896.01.pgs 4/24/03 9:23 AM Page 21

1.3 Representing Functions Symbolically 21

An interval can also contain one endpoint, but not the other, as in 2  x  8.
We write this in interval notation as 12, 8 4 , where the square bracket on the right
indicates that x  8 is included and the left parenthesis indicates that x  2 is not
included. For instance, the domain of the function y  g 1x 2  1x is the interval
3 0,  2 , which indicates that x  0; it is closed on the left, and extends toward  , but
never reaches  , and so is open on the right.

E XAMPLE 1
Find the values of the function y  g 1x 2  1x at x  0, 14 , 1, 2, p, and 4. What is the
domain and range of this function?
Solution For each of the given values of x in the domain of this function, the corre-
sponding y-values in its range are
y  g 10 2  20  0;
y  g 1 14 2  3 14  12;
y  g 11 2  21  1;
y  g 12 2  22  1.41421 . . . ;
y  g 1p2  2p  23.14159 . . .  1.77245;
y  g 14 2  24  2.
Note that you can take the square root of any positive value of x or of 0, so the domain of
the function g consists of all nonnegative numbers. Similarly, the square root of any such
number is positive or 0, so the range of g also consists of all nonnegative numbers.
We can use inequality notation to write x  0 for the domain and y  0 for the
range. Alternatively, using interval notation, we have 30,  2 for the domain and 30,  2
for the range.
◆

E XAMPLE 2
Discuss the range of the function

y  F1x 2  x 
1
x
when the domain for F is restricted to the set of all positive numbers.
Solution We start by looking at the graph of the function F, as shown in Figure 1.20.
Note that the function is decreasing rapidly to the right of x  0. It has a turning point at

10

0 x
FIGURE 1.20 0 1 2 3 4 5
Gord.3896.01.pgs 4/24/03 9:23 AM Page 22

22 CHAPTER 1 Functions in the Real World

about x  1, and then it increases slowly thereafter. Try different values for x with your
calculator to verify that this result is indeed the case numerically. Also, extend the graph
farther to the right with your function grapher to see that this pattern continues indefi-
nitely. It turns out that the smallest possible value for y, which is y  2 exactly, corre-
sponds to x  1 at the turning point. For any other value of x, the value for y is larger.
Therefore the range of F is all values y  2.
◆
You can visualize a function f as an operation that transforms each value x from
its domain into the corresponding value y in its range. Figure 1.21 illustrates how
each point x in the domain is transformed into a single point in the range. Thus x1 is
transformed into y1 . We also can say that x1 is carried into y1 or that x1 is mapped into
y1 . Similarly, x2 is carried into y2 and x3 is mapped into y3 . Note that x4 and x5 are
both transformed into y4 , which is perfectly legitimate for a function. Each x-value
must be mapped into a single y-value, although it is certainly possible for several dif-
ferent x’s to be mapped into the same y. Think about the function y  f 1x 2  x2,
where both x  2 and x  2 are transformed into y  4.

y1
x1
x2 y2

y3
x3

x4 y4
x5

FIGURE 1.21 Domain Range

We now summarize the preceding ideas in a formal definition of a function.

Definition of a Function
A function f is a rule that assigns to each permissible value of the inde-
pendent variable x one and only one value of the dependent variable y.
The domain of f is the set of all possible values for the independent variable.
The range of f is the set of all possible values for the dependent variable.

E XAMPLE 3
Discuss the domain and range for the function relating acceleration time t to final speed
v for a Trans Am, based on the following set of data.

Final speed, v (mph) 30 40 50 60 70

Time, t (sec) 3.00 4.29 5.52 7.38 9.81

Gord.3896.01.pgs 4/24/03 9:23 AM Page 23

1.3 Representing Functions Symbolically 23

Solution The independent variable is the final speed v, so the domain of this function
consists of all possible speeds. We therefore might conclude that the domain would be 0
to 70 mph; however, if we want to use the function to predict the time needed to reach a
higher speed, we would need a somewhat larger domain—say, 0 to 100 mph. It probably
isn’t reasonable to think of speeds any faster than that. The dependent variable is the time
t needed for a Trans Am to accelerate to a given speed, v. If we use only speeds between 0
and 70 mph, the associated range would be 0 to 9.81 seconds. If we use the extended do-
main of v of 0 to 100 mph, however, the associated range might be more like 0 to 20 sec-
onds. It takes about 2.5 seconds to accelerate from 60 to 70 mph. The pattern suggests
that it will take even more time to accelerate from 80 to 90 mph and still more time to go
from 90 to 100 mph. Thus an estimate of 20 seconds to accelerate from 0 to 100 mph is
reasonable.
◆
Consider the relationship between people and their telephone numbers. Is this
relationship a function? If there is even one person who has two different telephone
numbers, the relationship does not satisfy the definition and so is not a function.
But a person’s height is a function of the person—each individual has one and only
one height at any particular time.
Often a verbal description of a function includes the idea of proportionality
from elementary algebra. Recall that y is proportional to x means that y  k . x,
for some constant of proportionality k. For instance, the area of a circle is pro-
portional to the square of the radius because A  pr 2 and p is a constant (it is
the constant of proportionality). Similarly, y is inversely proportional to x if
y  k . 1>x  k>x, where k is a constant of proportionality.
Throughout this book, unless some restriction is indicated, we assume that all
functions discussed are defined (either mathematically or practically) on the
largest possible domain that makes sense.

Problems
1. Which of the relationships are functions and which h. The area of an equilateral triangle whose base is b.
are not? For those that are not functions, explain i. The height of a bungee jumper t seconds after
why. For those that are functions, identify the inde- leaping off a bridge.
pendent and dependent variables and give a reason- j. The time it takes the bungee jumper to reach a
able domain. height H above the ground.
k. The number of baseball players who have n
a. The cost of first-class postage on January first of
home runs in a full season.
each year since 1900.
l. The height of liquid in a 55-gallon tank h hours
b. The weight of letters you can mail with n  1, 2,
after a leak develops.
3, . . . postage stamps.
m. The daily cost to a family of heating their home
c. The time of sunrise associated with each day of the
versus the average temperature that day.
year.
d. The time of high tide associated with each day of 2. The balance B, in thousands of dollars, in a CD ac-
the year. count at a bank is a function of time t, in years, since
e. The high temperature associated with each day you opened the account, so B  f 1t2.
of the year. a. What does f 14 2  2 tell you? What are appro-
f. The closing price of one share of IBM stock each priate units?
trading day on the stock exchange. b. Is f an increasing or decreasing function of t?
g. The area of a rectangle whose base is b. c. Discuss the concavity of f.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 24

24 CHAPTER 1 Functions in the Real World

3. The height H in inches of a child is a function of the b. Suppose that Adam is on a diet and wants to
child’s age a, so H  f 1a2. limit his calorie intake from a peanut butter and
a. What does f 1102  50 tell you? What are appro- jelly sandwich to a maximum of 300 calories.
priate units? Find two reasonable combinations of amounts
b. Is f an increasing or decreasing function of a? of peanut butter and jelly that produce a sand-
c. Discuss the concavity of f. wich with exactly 300 calories.
c. Which is more caloric, a gram of peanut butter
4. The surface area S of a sphere of radius r is 4p times
or a gram of jelly? Explain how you know.
the square of the radius. Write a formula for S as a
function of r. 12. A car rental company charges a fixed daily rate for a
midsize car plus a charge for each mile more than
5. The pressure P of a gas in a container of fixed size is
100 miles that the car is driven per day. A formula
proportional to the temperature T of the gas. Write a
for the cost of a rental car driven more than 100
miles is c  f 1m2  35  0.251m  100 2, where
formula for the pressure as a function of temperature.
6. The pressure P of a gas held at a constant tempera- m is the number of miles that the car is driven.
a. Find f 1100 2. What does it mean?
ture in a container is inversely proportional to the
b. Find f 1150 2, f 1200 2, and f 1500 2.
volume V of the container. Write a formula for the
pressure as a function of volume.
c. What is a reasonable domain and range for this
7. The force of gravity F between two objects is in- function?
versely proportional to the square of the distance d
13. Suppose that you throw a ball upward, with an initial
between the objects. Write a formula for F as a
velocity of 60 ft>sec, from the roof of a 120-ft-high
function of d.
building.
8. When a cup of hot coffee is left to cool on the table
a. Sketch a possible graph of the height of the ball
where the air temperature is 70°F, the change T in
as a function of time, as you visualize it.
the temperature T of the coffee is proportional to
b. Suppose that the height of the ball as a function
the difference between the temperature of the coffee
of time is given by
and the room temperature. Write a formula for T
as a function of T. H1t2  120  60t  16t2.
9. Kim has a peanut butter sandwich on white bread Find the height of the ball when t  1; when
each day. The number of calories C in the sand- t  4.
wich, as a function of the number of grams P of c. Find H12 2 and H13 2. What do they represent?
peanut butter, is C  f 1P2  150  6P. d. Use your function grapher to estimate how long
a. What is f 112? What does it mean?
it takes for the ball to reach its maximum height.
b. What is f 1102? f 115.52? f 120 2? f 130 2 ?
What is the maximum height?
e. How long does it take until the ball first hits the
c. How many calories come from the bread alone?
street below?
d. Explain why using P  1 makes no sense.
f. What are the domain and range for this function?
e. What is a reasonable domain and range for this
function? 14. For the function f 1t2  t2  5, find the values cor-
responding to t  2, 4, 6, 10.
10. Suppose that Jim wants his peanut butter sandwich
on rye bread instead of white bread. Rye bread con- 15. For the function1
F1x 2  2
tains 85 calories per slice. What would be the corre- 1
,
sponding formula for the number of calories in x 4
Jim’s sandwich? find F10 2, F11 2 , F132 , F14 2, F15 2 . Why did we skip
11. The number of calories in a peanut butter and jelly x  2? Are there any other values of x that should
sandwich on white bread is C  150  6P  2.7J, be skipped? What is the domain of this function?
where P and J are the number of grams of peanut 1
butter and jelly, respectively. Note that when you enter this expression in a calculator or most
computer programs, you must key the expression in as 1> 1x^2  42.
a. How many calories are in a sandwich with 24 g Pay careful attention to when you need to use parentheses in any
of peanut butter and 20 g of jelly? such expression.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 25

1.4 Connecting Geometric and Symbolic Representations 25

16. For the function 19. For the function f 1x 2  x3  8x2  15x  1, find
x 4
2
g 1x2 
three different values of x between 1 and 8 for
x2  9
, which f 1x 2  0. Then find at least two noninteger
values of x for which f 1x 2  0.
find g 10 2, g 11 2, g 122, g 142, g 11 2. Why did we skip
x  3? Are there any other values of x that should 20. A simple substitution code in which each letter is re-
be skipped? What is the domain of this function? placed by a different letter can be thought of as a
17. For the function g 1s2  s  1s , find the values
function f whose domain is the letters of the alphabet
A, B, . . . , Z. Suppose that f 1A2  M, f 1B2  D,
f 1C2  K, f 1D2  V, f 1E 2  X, f 1F2  B,
corresponding to s  4, 16, 25, 100. Are there any
f 1G2  P, f 1H2  T, f 1I2  J, f 1J2  S,
values for s that will make the function come out
f 1K 2  Z, f 1L2  Q, f 1M 2  H, f 1N2  O,
negative? What does that tell you about the range of
f 1O2  A, f 1P2  L, f 1Q2  W, f 1R2  C,
g? What is its domain?
18. For the function z  f 1q2  q3  5, find the value f 1S2  F, f 1T2  Y, f 1U 2  R, f 1V2  G,
of the dependent variable that corresponds to a f 1W 2  I, f 1X2  U, and f 1Y2  N.
a. What is f 1Z2?
value of the independent variable of 4. Find the
b. What is the solution to the equation f 1x2  R?
value of the independent variable that corresponds
to a value of the dependent variable of 6.

1.4 Connecting Geometric and Symbolic Representations

y
One of the most significant advances in mathematics is based on the idea of con-
necting the geometric and symbolic representations of functions. It allows you to
think of functions from a visual rather than an exclusively symbolic perspective.
II I Begin by drawing two perpendicular axes, as shown in Figure 1.22, whose
point of intersection O is the origin. The horizontal axis represents values of the
x
O(0, 0) independent variable (in this case, x); by convention, these values increase from left
to right, as indicated by the arrow. The vertical axis represents values of the
III IV dependent variable (in this case, y); by convention these values increase upward, as
indicated by the vertical arrow. The two axes divide the plane into four quadrants:
the first quadrant (I), the second quadrant (II), the third quadrant (III), and the
FIGURE 1.22 fourth quadrant (IV). Whenever appropriate, indicate the units used for each vari-
able and label the axes accordingly.
This representation is called a rectangular or Cartesian coordinate system, and it
is a way of associating points in the plane with ordered pairs of numbers. Every
point P in the plane can be represented by an ordered pair of numbers, 1x, y2 . Al-
ternatively, every ordered pair, such as 12, 5 2, 127, 12, or 123.84, 21.022, represents a
point in the plane. We call 1x, y2 the coordinates of the point P. In mathematics,
the letters x and y generally represent the independent and dependent variables, re-
spectively. However, in any given context, you should use letters that suggest the
quantities being studied.
Consider again the function
y  f 1t2  64t  16t2,
which represents the height, at any time t, of a ball tossed vertically upward with
initial velocity of 64 feet per second. The formula for the function f gives the verti-
cal height y of the ball at any instant t. When t  0, y  0 and the corresponding
point 10, 02 is the origin. Further, as we calculated before, f 1 12 2  28, f 112  48,
Gord.3896.01.pgs 4/24/03 9:23 AM Page 26

26 CHAPTER 1 Functions in the Real World

f 122  64, f 13 2  48, and f 14 2  0, which give rise to the points 1 12 , 28 2, 11, 48 2,
12, 642, 13, 48 2 , and 14, 02 in the 1t, y2 coordinate system. These six points are plot-
ted in Figure 1.23.

y y

64 (2, 64) 64

Height (feet)
Height (feet)

48 (1, 48) (3, 48) 48

32 32
( 12 , 28)
16 16
(4, 0)
t t
(0, 0) 1 2 3 4 O 1 2 3 4
FIGURE 1.23 Time (seconds) FIGURE 1.24 Time (seconds)

We can determine many other ordered pairs 1t, y2 satisfying the equation
y  64t  16t2. (Simply pick any other value for t between 0 and 4 and calculate
the associated value of y by using the equation.) Each such ordered pair can be
plotted as a point in the coordinate system. When all possible points are plotted,
they form the curve shown in Figure 1.24. This curve is the graph of the function f.
It consists of all points in the plane whose coordinates 1t, y2 satisfy the given equa-
tion. Thus we have a direct connection between the graph of a function and its al-
gebraic equation. The graph of a function is therefore another representation of
the same function. Note that the graph shown in Figure 1.24 represents the height
of the ball at any time t; it doesn’t show the path of the ball, which goes straight up
and then down.

The graph of a function y  f 1x2 consists of all points 1x, y2 in the plane
whose coordinates satisfy the equation of the function.

A table of values for a function is also useful when you’re creating a hand-drawn
graph of the function f from the formula y  f 1x 2 . It provides a simple method of
organizing the values of the independent variable x and the associated values of the
dependent variable y that produce each point to be plotted. The number of points
that you need to calculate for a table to draw a reasonable graph of a function de-
pends on how complicated the behavior of the function is. For a line, all you need is
two points because two points completely determine a line. We used six points to
produce the graph of the height of the thrown ball shown in Figure 1.24. For com-
parison, a graphing calculator uses about 100 points to construct a curve.
When drawing the graph of a function, you should determine several key points.
One point is where the graph crosses the vertical axis. You can easily find this point if
you have a formula for the function: Just set the independent variable x equal to zero
in the algebraic formula for the function and calculate the corresponding y-value. Al-
though often desirable, finding the point(s) where the curve crosses the horizontal
axis is usually more complicated. To find them, set the dependent variable y equal to
zero and then solve the resulting equation. For the function representing the height
of the ball, we can factor the expression for y and then set y  0:
y  64t  16t2  16t14  t2  0.
Gord.3896.01.pgs 4/24/03 9:23 AM Page 27

1.4 Connecting Geometric and Symbolic Representations 27

When you solve this equation for t, you get either t  0 or t  4. The time t  0 is
the instant when the ball is first released, so y  0. At the instant when t  4, the
corresponding value for y, which represents the height of the ball, is also zero. That
is, at time t  4, the ball has come back to the ground. You can see the pattern for
the values of this function (and thus the pattern for the height of the ball) in the
following table.

Time t (sec) 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Height y (ft) 0 28 48 60 64 60 48 28 0

E XAMPLE 1
Determine the domain and range of the function shown in Figure 1.25.

x
–2 O 3
–4
FIGURE 1.25

Solution Note that the axes shown are labeled x and y; x is the independent variable,
and y is the dependent variable. Further, observe that the graph extends from x  2 at
the left to x  3 at the right, so the domain of this function is from 2 to 3. We can
write this domain in terms of inequalities as 2  x  3. Similarly, the graph extends
vertically from a low of y  4 to a high of y  8, so the range is 4  y  8.
◆
In many situations, we typically start with a set of data collected from some exper-
iment or from measurements taken on some process. We then graph the data to get a
feel for the behavior of the quantity. Often, we try to connect the points on the graph
with a smooth curve to get a better indication of the behavior of the quantity. Finally,
we would like to obtain an equation for a function that fits these data points because
many questions can be answered far more easily and accurately when an equation is
available. We illustrate this methodology in Examples 2–4.

E XAMPLE 2
The snow tree cricket, which lives in the Colorado Rockies, has been studied by field
biologists who have gathered the following measurements on how the chirp rate de-
pends on the air temperature.

Temperature T (F) 50 55 60 65 70 75 80

Rate R (chirps>min) 40 60 80 100 120 140 160

Plot the points to determine the kind of trend in the data.

Gord.3896.01.pgs 4/24/03 9:24 AM Page 28

28 CHAPTER 1 Functions in the Real World

Solution Plotting these data points gives a visual dimension, as shown in Figure 1.26. Note
that the chirp rate is growing at a constant rate as the temperature increases. Moreover, the
corresponding points in the figure seem to fall into a straight line pattern, as indicated by
the line drawn through them. In Chapter 2, we discuss how to find the equation of this line
and how to predict the chirp rate R of the cricket based on the temperature T, or vice versa.
R

160

(chirps/minute)
120

Chirp rate
80

40

T
50 60 70 80
FIGURE 1.26 Temperature (°F)

◆
E XAMPLE 3

The following table of values gives the population, in millions, of the state of Florida
since 1990.
a. Plot the data points and describe the behavior pattern.
b. If this trend continues, estimate the population in the year 2000.

Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Population 12.94 13.32 13.70 14.10 14.51 14.93 15.36 15.81 16.27 16.74 ?

Solution
a. The graph of this set of data is shown in Figure 1.27. The growth pattern clearly is not
a straight line pattern; rather, the population grows ever faster. The function is both in-
creasing and concave up.
P

18

17
Population (millions)

16

15

14

13

12

Y
1990 1992 1994 1996 1998 2000
FIGURE 1.27 Year

b. The increase from 1997 to 1998 was 16.27  15.81  0.46 million, and the increase
from 1998 to 1999 was 16.74  16.27  0.47 million. As a result, we could estimate
Gord.3896.01.pgs 4/24/03 9:24 AM Page 29

1.4 Connecting Geometric and Symbolic Representations 29

that the increase from 1999 to 2000 might be about 0.48 million, so our prediction for
the year 2000 is about 16.74  0.48  17.22 million people. We determine a formula
for this function in Chapter 2 so that we can make such a prediction in a much simpler
and more confident way.

◆
E XAMPLE 4
The following table of values shows measurements, at different times, of the height of an
object dropped from the top of the 1250-foot-high Empire State Building. Construct a
graph of the height as a function of time and describe its behavior.

Time (sec) 0 1 2 3 4 5 6 7 8

Height (ft) 1250 1234 1186 1106 994 850 674 466 226

Solution The graph of the height of the object versus time in Figure 1.28 shows that
the object is falling ever faster as time goes by. The function is decreasing and concave
down. Again note that the graph represents the height of the object, not its path, which
is straight down.

1200
Height (feet)

100
t
0 1 2 3 4 5 6 7 8 9
FIGURE 1.28 Time (seconds)

◆
Although we could estimate from either the table or the graph how long it
takes the object to hit the ground or to pass, say, the 30th floor, we could answer
such questions more precisely if we knew the formula for the function.
In Examples 2–4, we simply connected the points to construct a smooth curve
that seemed to fit the pattern. Doing so, however, can sometimes lead to serious er-
rors. Suppose that we had some data on the turkey population of the United States
taken on January 1 each year. It would likely show a growth trend similar to that in
Example 3 on the population of Florida. However, a little thought will convince
you that this population will change quite drastically about the middle of Novem-
ber each year. The smooth curve drawn using the January 1 turkey census data
would therefore be a rather poor description of the actual population over all in-
termediate times.
Gord.3896.01.pgs 4/24/03 9:24 AM Page 30

30 CHAPTER 1 Functions in the Real World

Nevertheless, the idea of connecting a series of points to form a curve is precisely

how a computer or graphing calculator produces the graph of a function. We strong-
ly urge you to become comfortable with using a graphing calculator or a computer
program to investigate the graph of any desired function given by a formula. The vi-
sual dimension invariably provides a wealth of information about the behavior of the
function, and we continually turn to graphical images throughout this book.

Connections Between Geometric,

Numerical, and Symbolic Representations
We have shown that a function can be represented in a variety of ways—as a formula
giving a symbolic representation, as a graph giving a geometric representation, as a
table giving a numerical representation, or in words giving a verbal representation.
The first three ways (formula, graph, or table) are the most useful, but each approach
provides a very different perspective.
The problem is: Can you always move back and forth between these different
Symbolic representations? Figure 1.29 illustrates schematically the interrelationships between
the three most useful representations—symbolic, geometric, and numeric—by ar-
rows. Ideally, you should be able to start with any one of these representations for a
function and shift to the other two representations. Some of these shifts are very sim-
ple. If you know a formula for a function, you can create a table of numerical values.
Similarly, if you know a formula, you can create its graph at the push of a button,
Geometric Numeric using a graphing calculator or computer graphics program or even by hand, as the
graph of a function consists of all points 1x, y2 that satisfy the equation of the func-
tion. If you have the graph of a function, you can easily read off a set of points on the
FIGURE 1.29 curve and so produce a table of values. If you have a table of values, you can easily
plot the points and connect them with an appropriate, usually smooth, curve to gen-
erate a graphical representation.
Unfortunately, the two remaining shifts in perspective are considerably more
complicated. If you start with a table of values, how do you produce a formula for
the function? Similarly, if you start with the graph of a function, how do you con-
struct a formula for it? Both shifts can be extremely difficult, but fortunately mod-
ern technology provides the tools by which you can create reasonably accurate
formulas. That often is the key step in most real-life applications of mathematics.

Does Every Curve Represent a Function?

Let’s consider one last question: Is every curve the graph of some function
y  f 1x 2? Consider the five curves shown in Figure 1.30. Are they all the graphs of
functions? That is, does each value of the independent variable x correspond to one
and only one value of the dependent variable y? We can test a curve in the follow-
ing way. Imagine a vertical line moving across the curve from left to right so that it
passes through every possible value of x in the domain. If, for each x, the line cross-
es the curve at only one point, there is exactly one y-value for that x and so the
curve represents a function. If the vertical line crosses the curve at more than one
point for any value of x, the curve does not represent a function. This criterion,
called the vertical line test, shows that the curves (a), (b), and (c) are all graphs of
functions. However, when the vertical line test is applied to curves (d) and (e), the
line crosses both curves at more than one y-value and thus neither are graphs of
functions.
Gord.3896.01.pgs 4/24/03 9:24 AM Page 31

1.4 Connecting Geometric and Symbolic Representations 31

y y y

x x x

(a) (b) (c)

y y

x x

FIGURE 1.30 (d) (e)

Problems
y
1. Which of the following graphs are functions? For
each function, (a) give its domain and range, 9
(b) identify where it is increasing or decreasing, and
(c) identify where it is concave up or concave down. f (x)

y y
R
P
2 2
x
1 1 0 8
Q
x t
–4 5 5
–1
a. What are the coordinates of the points P, Q, and R?
Is the point 12, 52 on the curve?
(i) (ii )
b.
Is the point 15, 22 on the curve?
P
Q c.
2 d. What is f 152?
Find x if f 1x 2  1.
2
e.
Is f 14 2  1 true or false?
1 1
r T
f.
–1
–1
5 –1
–1
5 g. Find y when x  2.
h. Find x when y  2.
Solve f 1x 2  0 for x.
(iii ) (iv)
i.
w
j. Solve f 10 2  y for y.
2
1 t 1970 1980 1990 1991 1992 1993 1994 1995
z
–1 5
–1 f (t) 400 300 210 190 175 162 150 135
(v)
3. The accompanying graphs are based on the set of
2. The following questions all relate to the accompa- data above, but something is wrong with each graph.
nying graph of a function y  f 1x 2 . What was done incorrectly in each instance?
Gord.3896.01.pgs 4/24/03 9:24 AM Page 32

32 CHAPTER 1 Functions in the Real World

y 4. Refer to the function f relating the snow cricket’s

chirp rate to the air temperature in Example 2.
a. What is f 160 2?
1995

1990
b. Solve f 1x 2  120.
1985 c. In a complete sentence, tell what the equation
1980 f 1622  88 means.
1975 5. For the function f 1x 2  x 2  3x  2, find the val-
1970
ues of y corresponding to x  3, 2, 1, . . . , 4, 5.
Plot the corresponding points and connect them
x
0 50 100 150 200 250 300 350 400 with a smooth curve. Then use your function gra-
(a) pher to graph the function. How do the two graphs
compare? Find the values of the function corre-
y sponding to x  12 , x  32 , x   52 and indicate the
location of the corresponding points on the curve
400
you drew.
6. For the function g 1t2  9  t 2, use an appropriate
350
300
250
set of values for t and the corresponding y values to
200
get a feel for the behavior of the curve when you
draw and connect the points. How does your sketch
150
compare to what you see when you use your func-
100
tion grapher?
50
0 x
7. Repeat Problem 6 for h1s2  s3  7s  5.
’70 ’80 ’90 ’91 ’92 ’93 ’94 ’95
(b)
8. One of the functions f or g in the following table of
values is concave up and the other is concave down.
Which is which? Explain how you know.
y

400
300 x 5 10 15 20
210
f (x) 80 70 62 56
190
175 g(x) 80 70 58 43
162
150
9. A function f 1x 2 whose values are given in the fol-
135
lowing table is increasing and concave up. Give a
possible value for f 15 2.
0 x
1970 1975 1980 1985 1990 1995
(c)

y x 4 5 6
50 f (x) 10 ?? 20
100

10. A function f 1x 2 whose values are given in the fol-

150
200
lowing table is increasing and concave down. Give a
possible value for f 150 2.
250
300
350
400
0 x
x 30 40 50
1970 1975 1980 1985 1990 1995
(d) f (x) 12 20 ??
Gord.3896.01.pgs 11/21/03 10:20 AM Page 33

1.4 Connecting Geometric and Symbolic Representations 33

11. For the function shown in the accompanying fig- d. Near what x-values is the function at a local
ure, indicate minimum?
a. the intervals of x-values where the function is e. Between what pair of successive x-values is the
increasing. function increasing most rapidly?
b. the intervals where the function is decreasing. f. Between what pair of successive x-values is the
c. all points x where the function has a turning function decreasing most rapidly?
point. g. Over what intervals is the function concave up?
d. all points x where the function has a local h. Over what intervals is the function concave down?
maximum. i. Near what x-values does the function have points
e. all points x where the function has a local of inflection?
minimum. j. Estimate the location of any zeros of the function.
f. all points x where the function has points of 13. Functions f, g, and h in the following table are in-
inflection. creasing functions of x, but each function in-
g. the intervals of x-values where the function is creases according to a different behavior pattern.
concave up. Which of the accompanying graphs best fits each
h. the intervals where the function is concave down. function?
i. approximately where the function is increasing
most rapidly.
j. approximately where the function is decreasing x f (x) g(x) h(x)
most rapidly.
k. the location of any zeros of the function (points 1 11 30 5.4
where the curve crosses the x-axis). 2 12 40 5.8
l. For any of parts (a)–(k) that asks for intervals,
write the interval both in terms of inequalities 3 14 49 6.2
and interval notation. 4 17 57 6.6
y
5 21 64 7.0

6 26 70 7.4

x
x1 x4 x5 x6 x7 x8 x9 x11 x14 (a) (b) (c)

12. Consider the data below. Assume that these values 14. Functions f, g, and h in the following table are de-
represent a sample of values for a smooth, or con- creasing functions of t, but each function de-
tinuous, function. creases according to a different behavior pattern.
a. Over what intervals of x-values is the function Which of the accompanying graphs best fits each
increasing? function?
b. Over what intervals is the function decreasing?
c. Near what x-values is the function at a local
maximum?

x 2.5 2.0 1.5 1.0 0.5 0 0.5 1.0 1.5 2.0 2.5 3.0
f (x) 62.3 28.4 6.8 4.3 11.9 33.2 14.7 2.3 12.5 38.8 5.2 11.7
Gord.3896.01.pgs 4/24/03 9:24 AM Page 34

34 CHAPTER 1 Functions in the Real World

c. f has a turning point at x  3.

t f (t) g(t) h(t) d. f is decreasing from 3 to 5.
1 200 30 5.4 e. f is increasing for x
5.
f. f is concave down from 0 to 4.
2 180 27.6 5.2 g. f has a point of inflection at x  7.
3 164 25.2 4.8 17. Consider the function f 1x 2  x 2  4 for x  3.
4 151 22.8 4.1 a. Is f increasing or decreasing?
b. Is f concave up or concave down?
5 139 20.4 3.1 18. Consider the function f 1x 2  x 3  7 for x  3.
6 129 18.0 1.8 a. Is f increasing or decreasing?
b. Is f concave up or concave down?
19. For the function f 1x 2  1x , find the values of y cor-
responding to x  0, 1, 2, . . . , 6. Plot the correspon-
ding points and connect them with a smooth curve.
Then use your function grapher to graph the func-
tion. How do the two graphs compare? Find the val-
(a) (b) (c) ues of the function corresponding to x  12 , x  32 ,
x  52 and indicate the location of the corresponding
15. Sketch the graph of a function that passes through points on the curve you drew. What is the domain?
the point 10, 12 and is
20. For the function
a. increasing and concave up for x  0 and in-
f 1x 2 
x
creasing and concave down for x
0. x1
b. increasing and concave up for x  0 and de-
calculate f 10 2, f 11 2 , f 12 2, . . . , f 18 2. Plot the corre-
creasing and concave up for x
0.
sponding points and connect them with a smooth
c. decreasing and concave up for x  0 and in-
curve. Then use your function grapher to graph the
creasing and concave up for x
0.
function. How do the two graphs compare? Find the
d. decreasing and concave up for x  0 and de-
creasing and concave down for x
0. values of the function corresponding to x  12 ,
x  32 , x  52 and indicate the location of the corre-
16. Sketch the graph of a single smooth curve that sat- sponding points on the curve you drew. What is the
isfies all the following conditions. domain?
a. f 102  4 b. f 152  2

1.5 Mathematical Models

A model is an image or representation of an object or process. A diagram of the
human circulatory system is a model of the veins and arteries in the human body;
the picture can be used to help us understand how blood circulates throughout the
body. Similarly, an architect’s sketch of a proposed shopping center is a model of
the actual center; a wooden model built to scale is a still more realistic representa-
tion for that shopping center.

A model is a representation that highlights the most important characteris-

tics of an object or process.

Models can be found everywhere: the tide tables used by fishermen; a comput-
er scientist’s flowchart for a new program; plastic replicas of jet fighters; and many,
many more. Because our focus here is on mathematics, the models we present are
Gord.3896.01.pgs 4/24/03 9:24 AM Page 35

1.5 Mathematical Models 35

mathematical models. A mathematical model is a representation of a process ex-

pressed by a formula, an equation, a graph, a sequence of numbers, or a table of
values. Once we have developed any such mathematical representation, we can use
it to examine the behavior of the actual process.
For example, the equation for the motion of the ball thrown vertically upward
from ground level with an initial velocity of 64 ft>sec,
y  f 1t2  64t  16t 2,
is a mathematical model that describes one important aspect of the motion of that
ball—its height at any time t. There may be other aspects of the motion that may not
be captured in this mathematical model (such as releasing it from some initial height
above the ground or the effects of air resistance). These other factors may likely re-
quire development of a more sophisticated model, but often a simple model gives a
reasonably accurate first approximation.
How can we use mathematics to describe the real world via a mathematical
model? We begin by looking at some process in the real world, such as the motion
of a ball thrown upward, the growth of a population, or a person’s reaction to a
drug. Typically, the process of trying to explain what is happening requires some
simplifying assumptions. For instance, in modeling the motion of a ball, we as-
sumed that the only force acting on the ball is the force of gravity and ignored the
negligible effects of air resistance. (Of course, if the ball were replaced with a bal-
loon, a feather, or a piece of paper, this assumption would be invalid.)
After making reasonable assumptions, we then express the process in a mathe-
matical form, which leads to a mathematical representation of the process in terms
of a formula, an equation, a graph, or a table. This mathematical model then has to
be interpreted. Does it truly seem to reflect what happens in the real world? Does
the behavior of the function mirror the behavior of the process? If so, we can use
the mathematical model to describe of the process under study and as the basis for
predictions about the process. If the model doesn’t adequately reflect the actual
process, we may have done something wrong—overlooked some important aspect
of the situation or ignored some critical factor; made some erroneous assump-
tions; or made an error in our work. We illustrate this interplay between mathe-
matics and the real world via mathematical modeling schematically in Figure 1.31.

Formulation

Real Math
World Model

FIGURE 1.31 Interpretation

The concept of function is closely connected to the idea of a mathematical

model, and most mathematical models are expressed as functions. Let’s look at an
example of this process. Researchers have studied the relationship between the
level of animal fat in women’s diets and the death rate from breast cancer in differ-
ent countries. Some of their data are shown in the following table, which gives the
average daily intake of animal fats in grams per day and the age-adjusted death rate
from breast cancer per 100,000 women. We begin by looking at these data, first as
a set of numerical values in the table and then visually on a graph, as shown in
Gord.3896.01.pgs 4/24/03 9:24 AM Page 36

36 CHAPTER 1 Functions in the Real World

Figure 1.32. There is obviously a relationship between the death rate and the daily
fat intake. Clearly the death rate D increases as the daily fat intake F increases, so D
is an increasing function of F. Moreover, the points fall into a straight line pattern.

Country Japan Spain Austria U.S. U.K.

Daily fat intake(grams) 20 40 90 100 120

Death rate per 100,000 3 7 17 19 23

Death rate
versus
Daily fat intake
Number of deaths per 100,000 30
25 U. K.
20 U. S.

15 Austria
10
5 Spain
Japan
0 20 40 60 80 100 120
FIGURE 1.32 Daily fat intake (grams)

It turns out that the equation for this line is

D  f 1F2  0.2F  1,
and we use this equation as our mathematical model in the following example.

E XAMPLE
The average daily animal fat intake in Mexico is F  23 grams and the average daily an-
imal fat intake in Denmark is F  135 grams. Predict the death rate from breast cancer
per hundred thousand women in Mexico and Denmark.
Solution We use the mathematical model to predict that the death rate from breast
cancer in Mexico for the average daily fat intake F  23 will be
D  0.2123 2  1  3.6 per 100,00 Mexican women.
Similarly, for the average daily fat intake F  135 grams in Denmark, the equation
predicts a death rate of
D  0.21135 2  1  26 per 100,000 Danish women.
◆
The type of prediction for the death rate for breast cancer in Mexico is
called interpolation because we are predicting the value of a quantity using a
measurement within the set of data. The type of prediction for the death rate in
Denmark is called extrapolation because we are predicting the value of a quanti-
ty beyond the set of data.
In Section 2.2, we show how to find an equation such as the one relating the
death rate to the daily fat intake. Once we have such an equation as a mathemat-
ical model, we can base some informed judgments on it. This model is based on
the average daily intake in each country, which can vary tremendously among
Gord.3896.01.pgs 4/24/03 9:24 AM Page 37

1.5 Mathematical Models 37

individuals. Even so, there is obviously a link between consumption of animal

fat and the incidence of breast cancer. Thus the mathematical relationship indi-
cates that women should drastically reduce their daily animal fat intake to re-
duce their chance of breast cancer. Furthermore, knowing that such a link exists,
researchers have since been conducting follow-up studies to determine why the
link exists. They have also found links to other items in the diet as well as in the
environment. Thus, the incidence of breast cancer depends not just on a single
variable but on a number of different variables, so it is a function of several in-
dependent variables. Many situations that you encounter in real life are exam-
ples of functions of more than one variable. Although the study of such
functions is somewhat beyond the scope of this book, we introduce and briefly
discuss functions of several variables in Section 3.8.

Parameters and Mathematical Models

Consider again the formula for the height y at any time t of an object dropped from
the top of the 180-foot-tall Tower of Pisa: y  180  16t 2. In comparison, the com-
parable formula for the height of an object dropped from the top of the 555-foot-high
Washington Monument at any time t is
y  555  16t 2,
and the height of an object dropped from the top of the 1821-foot-high CN tower
in Toronto at any time t is
y  1821  16t 2.
Each of these functions has the same structure mathematically; what differs among
them is the leading number that represents the initial height from which the object
is dropped. Based on these specific functions, we can hypothesize a general formu-
la for the height y at any time t of an object dropped from any initial height y0 :
y  y0  16t 2.
In this formula the height y clearly is a function of time t— they are the depend-
ent and independent variables, respectively. The quantity y0 can also take on different
values, but it isn’t a variable in the same sense that t and y are. Although y0 can take
on different values, in any particular situation it has just one value—in this case the
initial height of the object. That value doesn’t change even though the variables t and
y change during the event. A quantity such as y0 is called a parameter. Note that each
value of y0 yields a different function, although each function has the same form.
Now suppose that you’re driving steadily at a rate of 40 mph; the relationship
between the distance D you travel and the time you drive is D  40t. If you drive at
a steady 50 mph, the relationship is D  50t, and if you drive at a steady 65 mph, the
relationship is D  65t. Obviously, distance is a function of time. The independent
variable is time t, the dependent variable is distance D, and we write the function as
D  r . t. This relationship holds for any choice of speed r and gives a slightly differ-
ent function, but one having the identical form, for each possible value of r. The
quantity r is a parameter in the formula for this distance function.

How Accurate is a Mathematical Model?

Recall that a mathematical model is only a mathematical description of a process,
not the process itself. So, every model carries with it some degree of inaccuracy. For
instance, we used the formula y  64t  16t 2 as a mathematical model for the
Gord.3896.01.pgs 4/24/03 9:24 AM Page 38

38 CHAPTER 1 Functions in the Real World

height at any time t of an object thrown vertically upward with an initial velocity of
64 ft>sec. With this model, we can find how long it takes for the object to come
back to the ground. That occurs when y  0, so we solve the equation
y  64t  16t 2  0,
We factor out the common factor of 16t to get
16t 14  t2  0
so that the object is at ground level when t  0 (when the object is initially re-
leased) or t  4 (when it has come back to the ground).
However, according to the laws of physics, the coefficient of t 2 is actually one
half of the Earth’s gravitational constant. Its value is not precisely 16, but rather
more like 16.1, so the solution t  4 is not quite accurate. Instead, we really
should say that t is about 4 or that it is approximately equal to 4, which we write
symbolically as t  4. We can improve on this estimate by using the more accurate
value of 16.1 for the coefficient of t 2 and then solving the equation
64t  16.1t 2  0.
The only common factor now is t, so that
t 164  16.1t2  0,
giving either t  0 or t  64>16.1  3.975, which is correct to three decimal places.
How many decimal places are reasonable for this answer? We could use more decimal
places when we divide out the fraction—say t  3.97516 or even t  3.97515528—
but when we are measuring time in seconds, both results are unrealistic levels of accu-
racy and should be avoided. Even using the three decimal places in t  3.975 may be
too many, both from a practical point of view—think about timing in Olympic events
where time is usually measured to the hundredth of a second—and from a mathe-
matical point of view—we used only one decimal place in the coefficient, 16.1. In
any context, you should determine a reasonable number of decimal places for your
final answer, both practically and in terms of the number of digits used.
In fact, rarely in applied situations do you get an “exact” answer such as x  5 or
x  18 . Even when you do get an exact answer involving a radical or a fraction,
you should usually convert it to a decimal, which automatically introduces another
level of inaccuracy. Thus 18  2.828 or 18  2.82843 or 18  2.82842712. But
18 is an irrational number and its decimal equivalent is an infinite, nonrepeating
decimal. Just because your calculator displays 10 or 12 decimal places does not nec-
essarily mean that the result is exactly that number.

Problems
1. An uncooked chicken (temperature of 70° F) is Sketch a graph of the temperature T of the soda as
placed in a hot oven at a temperature of 350° to a function of time t. Identify appropriate intervals
cook. The chicken is removed when its internal for the domain and range of this temperature
temperature reaches 180°. Sketch a possible graph function. Describe the behavior (increasing/de-
for the temperature T of the chicken as a function creasing, concavity) of this function.
of time t. What would be appropriate values for the 3. An Olympic diver dives off the 10 meter platform,
domain and range of this function? Describe the enters the water cleanly, and rises slowly to the sur-
behavior (increasing/decreasing, concavity) for the face. Sketch a possible graph for the height of the
graph. diver above water level as a function of time. What
2. A warm can of soda 180° F2 is placed in a refrigera- might be appropriate values for the domain and
tor at a temperature of 36° and left there to cool. range of this function? (Estimate how long it will
Gord.3896.01.pgs 4/24/03 9:24 AM Page 39

Chapter Summary 39

probably take the diver to reach the water from the b. The rim of the basket is 10 feet above the floor.
platform.) Describe the behavior of this function. Use your calculator to estimate all times t when
4. Repeat Problem 3 by sketching the graph of the the ball is at the height of the rim.
diver’s height above the diving platform as a func- 7. At the beginning of this section, we gave the equa-
tion of time. How does the shape of this graph com- tion for the height y of a ball thrown vertically up-
pare to the one you drew in Problem 3? ward from ground level with initial velocity 64 feet
5. Police sometimes use the formula s  f 1d2  per second,
124d as a model to estimate the speed s in miles y  f 1t2  64t  16t 2,
per hour that a car was going on dry concrete pave- as a function of time t.
ment if it left a set of skid marks d feet long. Using a. Suppose that the initial velocity of the ball is
this model, estimate the speed of a car whose skid 80 feet per second, Write a comparable formula
marks stretched for the height y as a function of time t.
a. 60 ft. b. 100 ft. b. If you think of the initial velocity v0 of the ball as
c. 140 ft. d. 200 ft. a parameter, write a formula for the height y as a
e. Suppose that you’re driving at 60 mph on dry con- function of time t with any initial velocity v0 .
crete pavement and slam on your brakes. How long 8. In each expression for a function, identify which
will your skid marks be, according to this model? letters represent variables, which letters represent
6. When a basketball player takes a long shot, the functions, and which letters therefore represent
height H of the ball above the floor can be modeled parameters.
by the equation H1t2  16t 2  24t  7, where t a. y  f 1x 2  ax 3  bx 2  cx  d
is the number of seconds since the ball was released. b. z  g1t2  at b
a. Use your calculator to estimate the maximum c. z  h1t2  abt
height that the ball reaches, correct to two deci- am
d. Q  k1m2 
mal places. bm2  c

Chapter Summary

In this chapter we introduced you to functions, their importance, and some of

their uses. Specifically we showed you how to work with functions in the following
ways.
◆ How functions arise in the real world.
◆ How functions can be represented in different ways—by graphs, by tables,
by formulas, and in words—and how to move from one representation to
another.
◆ How to identify whether a relationship between two variables given by an
equation, a graph, or a table is or is not a function.
◆ How to decide which is the independent variable and which is the dependent
variable.
◆ What the domain and range of a function are.
◆ The important characteristics about the behavior of functions—where they
increase and decrease, where their turning points are, where they are con-
cave up and concave down, where their points of inflection are, and whether
they are periodic.
◆ How to interpret concavity—whether the growth (increase) or decay (de-
crease) in a function is speeding up or slowing down.
◆ How mathematics is used to model phenomena in the real world.
Gord.3896.01.pgs 4/24/03 9:24 AM Page 40

40 CHAPTER 1 Functions in the Real World

Review Problems
1. In determining the amount of radiation to apply
Budget ($ millions) 10.0 3.4 27.0 6.2 9.7
to a tumor site, doctors take into account the
depth of the tumor within the body. What is the Attendance (millions) 1.0 0.5 2.0 0.6 1.3
independent variable and what is the dependent
Budget ($ millions) 7.0 4.8 18.0 6.5 13.0
variable in such a relationship? Give reasons for
your answer. Attendance (millions) 1.0 1.1 4.0 0.6 3.0
2. The accompanying graph describes the loudness
Budget ($ millions) 9.0 15.7 7.0 3.2 14.7
of a crowd watching a baseball game during the
ninth inning. Write a scenario that might explain Attendance (millions) 0.5 1.3 1.0 0.5 2.7
what was happening on the field as the inning
progressed. 7. The table of values shows the number, in millions,
of prerecorded cassette tapes sold in the United
Loudness States in various years between 1982 and 1998.
a. Draw a graph of the number of cassettes sold as
a function of the year since 1982.
b. In approximately what year did the sales of cas-
settes reach its maximum?
c. During which year, approximately, did the sale
Outs of cassettes change most rapidly? Most slowly?
1 2 3 4 5 6

3. An experimental form of insulin is being adminis- Year 1982 1985 1990

tered every 4 hours to a person with diabetes. The
Cassettes sold 182.3 339.1 442.2
body uses or excretes about 40% of the drug over
the 4-hour period. Draw a graph that shows the Year 1993 1994 1995
amount of the drug in the body as a function of
Cassettes sold 339.5 345.4 272.6
time over a 24-hour period.
4. Populations tend to grow steadily until there are too Year 1996 1997 1998
many members for the space and resources avail- Cassettes sold 225.3 172.6 158.5
able. Then the population size levels off. Sketch a
Source: 2000 Statistical Abstract of the United States
function that gives population size as a function of
time.
8. For the function f 1x2  3x 2  2x  1, find f 10 2,
f 11 2, f 11.1 2 , f 11.01 2, f 13 2, and f 1a2 .
5. Determine whether each table of values could rep-
resent a function. If not, explain why not.
a. 9. During the 1990s, the average cost of a new car
x 1 2 3 4 5 6 bought in the United States can be approximated by
the function C  f 1t2  659.7t  15598, where C
f (x) 10 10 12 14 18 25
is the cost of the car and t is the number of years
since 1990. (Source: 2000 Statistical Abstract of the
b. United States)
x 11 15 9 20 15 8
a. Determine the average cost of a car purchased in
g(x) 12 13 13 15 16 17 1995.
b. If this trend continues, estimate the average cost
of a car in 2003.
6. The table of values shows the budget and attendance c. According to this function, the average cost of a
at 15 U.S. zoological parks. Write a short description new car was increasing, on average, $659.70 each
of how attendance and budget are related. year after 1990. Use this information to calculate
Gord.3896.01.pgs 4/24/03 9:24 AM Page 41

Review Problems 41

how many years it would take for the average b. Sketch a graph of this function F1w 2, showing
price of a new car to reach $20,000. the cost of sending in 2003 a first-class letter as a
10. A study of the relationship between the average function of the weight w of the letter in ounces.
longevity (in years) and the gestation period (in 15. Consider the roller coaster shown in the accom-
days) for a sample of animals shows that the ani- panying figure. The points A, B, C, D, E, and F di-
mals’ average longevity L can be predicted reason- vide the curve representing the track into por-
ably well as a function of the gestation period t by tions that are increasing and concave up, increas-
the function L  f 1t2  1.04t 0.49, where t is the ges- ing and concave down, decreasing and concave
tation period in days. up, and decreasing and concave down. For each of
a. Estimate the lifetime of a chipmunk whose ges- the five portions of the track, A to B, B to C, and
tation period is 31 days. so on, (a) identify the mathematical behavior of
b. The gestation times in the study extend from 15 the curve, and (b) describe whether the speed of
days (opossum) to 645 days (elephant). What the cars is increasing at an increasing rate, increas-
would the range of the average longevity be if ing at a decreasing rate, decreasing at a decreasing
the given function were a good predictor? rate or decreasing at an increasing rate. In each
c. Use your function grapher to graph the func- case, explain your answer.
tion. Is it increasing or decreasing? Is it concave
up or concave down? C
d. Use the graph to estimate the gestation period of
B
an animal whose average longevity is 15 years.
(Hint: Find the point on the graph where L  15.)
A
e. The gestation time for humans is 9 months, or D
about 270 days. What does the formula predict F
for the average longevity of human beings? Can
you think of any reasons why the value you ob- E
tained is so inaccurate?
11. The domain of the function f 1x 2  4x 2  3x  5
is all real numbers.
a. Use your function grapher to estimate the coor-
dinates of the turning point.
b. What is the range of this function?
12. The domain of the function f 1x 2  x 3  9 is all 16. Picture a water slide at an amusement park. The
real numbers. slide starts at an initial height H0 above the pool and
a. Use your function grapher to estimate the coor- smoothly drops to water level. The slide is first con-
dinates of the point of inflection. cave down, then concave up, then concave down,
b. What is the range of this function? and finally concave up.
13. Give the domain of each of the following functions. a. Sketch a graph of the slide’s height H above
a. f 1x2  2x  5 . b. f 1x 2  2x 2  16 . water level as a function of horizontal distance x.
b. Suppose that you go down the slide in a sitting
x2  4 x2  4
2
position. Sketch a graph of the height of your
c. g1x2  . d. h1x  .
x2  9 x2  9 eye above the water as a function of horizontal
14. The U.S. Postal Service rates for first-class mail in distance x.
2003 were 37¢ for the first ounce and 23¢ for every c. Sketch the graph of the height of your eye above
additional ounce. the water as a function of time t.
d. Sketch the graph of your speed as a function of
a. Construct a table showing the cost of postage to
time t.
mail a first-class letter weighing 0 to 1 oz, 1 to 2 oz,
2 to 3 oz, 3 to 4 oz, and 4 to 5 oz. 17. Pacific coast salmon hatch in rivers and then mi-
grate to the ocean where they live most of their
Gord.3896.01.pgs 4/24/03 9:24 AM Page 42

42 CHAPTER 1 Functions in the Real World

lives. Eventually, they migrate back up the same 600,000

river to spawn and then die. The accompanying
500,000
graph shows the number of chinook salmon, in
various years, that pass a particular point on one

Number of salmon
400,000
waterway near Seattle as they wend their way up-
stream to spawn. (The U.S. Department of Fish- 300,000
eries keeps accurate counts in its efforts to
maintain a healthy salmon population.) Based on 200,000
this graph, write a paragraph explaining why the
number of salmon who swim upstream here is a 100,000

roughly periodic function of time. What is the pe-

0
riod? What does this graph suggest about the life 1992 1993 1994 1995 1996 1997 1998 1999 2000
span of the chinook salmon? Source: Hiram M. Chittenden Locks data, U.S. Army Corps of Engineers,
Seattle, Washington.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 43

2
Families of Functions

2.1 Introduction
Functions are fundamental to mathematics and its applications. Although there
are many different types of functions, most of our work focuses on just a few: func-
tions that are simple and yet sufficiently powerful to meet our needs. These types
of functions can be thought of as families of functions because the members of each
family are closely related to one another in terms of their essential properties. We
have described several distinct behavior patterns already as phenomena that
1. increase at a fixed rate and so go up by the same amount each fixed time
period;
2. decrease at a fixed rate and so go down by the same amount each fixed time
period;
3. increase at an increasing rate and so are concave up;
4. increase at a decreasing rate and so are concave down;
5. decrease at an increasing rate and so are concave down;
6. decrease at a decreasing rate and so are concave up.
Figure 2.1 illustrates these behavior patterns.

x x x

(a) (b) (c)

y y y

x x x
FIGURE 2.1 (d) (e) (f)

43
Gord.3896.02.pgs 4/24/03 9:25 AM Page 44

44 CHAPTER 2 Families of Functions

To model such phenomena and make predictions based on the models, you need
to know families of functions that behave in each of these six ways. In this chapter, we
present the families of linear functions, exponential functions, and power functions, as
well as several other useful families, that possess these behavior patterns. In later
chapters, we consider other families of functions, including polynomial functions and
trigonometric functions, that exhibit more complex behavior patterns.
As discussed in Section 1.3, we use the letters x and y generically for the inde-
pendent and dependent variables, respectively. However, in any specific context, we
use letters that directly suggest the quantities under discussion.

2.2 Linear Functions

The simplest and probably most useful family of functions are the linear func-
tions. These functions model any quantity that increases steadily or decreases
steadily—that is, it goes up or down by a fixed amount for any fixed change in the
independent variable. The graph of such a function is always a straight line.

Linear Functions That Pass Through the Origin

The simplest type of linear function is of the form y  mx, which can be inter-
preted as y is proportional to x. For example, suppose that you go to a deli to
buy some roast beef that is selling at $5.99 per pound. If you purchase 1 pound,
the roast beef costs C  5.99  1  $5.99; if you purchase 2 pounds, it costs
C  5.99  2  $11.98. If you buy N pounds, the cost is C  5.99N, so the cost C
of the roast beef is proportional to the number of pounds N that you buy. The mul-
tiple 5.99 is the constant of proportionality. We also say that the cost of the roast beef
is a linear function of the number of pounds of roast beef purchased.
Similarly, the distance D that a car travels at a constant speed of 50 mph is pro-
portional to the number of hours t driven, so the number of miles traveled is
D  50t.
As another illustration, it is reasonable to assume that the quantity G of garbage
produced in a city is proportional to the number of people P living there, so that
G  kP, for some constant multiple k. In fact, the average amount of garbage
produced annually in the United States is about 1500 pounds per person, so
k  1500. A mathematical model for the quantity G of garbage produced annu-
ally in a city whose population is P is therefore G  1500P pounds.
E XAMPLE 1
Suppose that gas costs $1.50 per gallon. (a) Create a function, as a table, as a graph, and as
a formula, to represent the cost C of G gallons of gas. (b) Create comparable functions if
the price of gas rises to $1.75>gal and to $2.00>gal and compare them to one another.
Solution
a. If gas costs $1.50>gal, the cost for 1 gallon is $1.50, the cost for 2 gallons is $3.00, the
cost for 3 gallons is 3  1.50  $4.50, and so on. We therefore get the following table
of values for this function.

G (gal) 1 2 3 4 5 ... 10 ... 20 ...

C ($) 1.50 3.00 4.50 6.00 7.50 ... 15.00 ... 30.00 ...
Gord.3896.02.pgs 4/24/03 9:25 AM Page 45

2.2 Linear Functions 45

When we plot these points, as shown in Figure 2.2, they all fall onto a line that passes
through the origin (the cost of 0 gallons of gas is $0). To find an equation for this line, we
note that, because each gallon of gas costs $1.50, the cost for buying G gallons must be
C  f 1G2  1.50  G  1.50G.
C

16
14
12
$1.50/gal
10
8
6
4
2
G
FIGURE 2.2 0 2 4 6 8 10

b. If the cost of gas rises to $1.75>gal, the corresponding function would be

C  g 1G2  1.75  G  1.75G.
Similarly, if the cost of gas rises to $2.00>gal, the function would be
C  h1G2  2.00  G  2.00G.
The graphs of all three of these linear functions are shown in Figure 2.3. Note that all
three lines pass through the origin, but that each is inclined at a slightly different
angle. In particular, the greater the cost for a gallon of gas, the steeper the line, which
makes sense because filling a tank costs more.
C

20 $2.00/gal

16

12 $1.75/gal

8
$1.50/gal

FIGURE 2.3 G
0 2 4 6 8 10

◆
The Graph of a Linear Function
That Passes Through the Origin
The graph of any linear function of the form y  mx is a line that passes through
the origin, as shown in Figure 2.4. What distinguishes one line from another is the
constant m, which represents how much y changes for a given change in x. A large
value for m, either positive or negative, means that y changes by a large amount for
a fixed change in the variable x. A small m means that y changes relatively little for
Gord.3896.02.pgs 4/24/03 9:25 AM Page 46

46 CHAPTER 2 Families of Functions

a fixed change in x. A positive value for m means that y gets larger as x gets larger. A
negative value for m means that y gets smaller as x gets larger.

y m large and positive

∆y
large
m small and
positive
∆y small
x
∆x
∆y negative

m negative
FIGURE 2.4

We use the Greek letter  (delta) to represent a change in any quantity. Hence
 x means the change in x, and y means the change in y. The quantity
change in y y rise
m  
change in x x run
is called the slope of the line. More generally, the slope of a line is
change in dependent variable
m .
change in independent variable
The letters used for the independent and dependent variables reflect what
those quantities are and often are different from x and y. For instance, based on the
data in the preceding table of values for the cost of gasoline, the independent vari-
able G is the number of gallons of gas purchased and the dependent variable C is
the cost of the gas. The slope of the line, based on the first two points, is
C 3.00  1.50 1.50
m    1.50.
G 21 1
We get the identical value for the slope if we use any two of the points.

The Meaning of Slope

The slope of a line indicates how fast the linear function is changing. For the gaso-
line example C  1.50G, the slope of 1.50 is the cost, $1.50, of each additional gal-
lon of gas. For the roast beef example C  5.99N, the slope, 5.99, is the cost per
pound, so each additional pound of roast beef costs an additional $5.99. If the roast
beef is on sale for $3.99, the cost of the roast beef goes up more slowly as the weight
of the purchase increases, and the slope is smaller. Similarly, if the price of roast beef
goes up to $6.99>lb, the cost goes up more rapidly, and the slope is steeper.
In general, whenever a linear function (or a line) arises in some context, the
slope of that line should be given in terms of units. For instance, if we use a linear
function to model the growth in the U.S. prison population over time, the units for
the slope of the line might be the number of new prisoners per year.
Suppose that a car gets 25 mpg. Then
Number of miles driven  25 miles per gallon  number of gallons used,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 47

2.2 Linear Functions 47

or
number of miles driven
25 miles per gallon  .
number of gallons used
Therefore the expression “25 miles per gallon” actually describes the slope of a lin-
ear function. The units of the slope are always a ratio: the units of the dependent
variable divided by the units of the independent variable.
Because the slope indicates how fast a line rises or falls, it is also known as the
average rate of change, or simply the rate of change. For a linear function, the rate
of change is always constant and is equal to the slope. For a nonlinear function,
which may be concave up or concave down, the rate of change is not constant, as
we demonstrate later in this chapter.

Lines That Don’t Pass Through the Origin

Next let’s consider lines that don’t pass through the origin. The equation of any
such line has the form
y  mx  b,
where
m is the slope of the line and
b is the vertical intercept.
The vertical intercept b represents the value of y when x is zero, because
y  m . 0  b  b. The vertical intercept is sometimes called the y-intercept. A
vertical intercept of zero 1b  0 2 corresponds to the special case y  mx, which is
the equation of a line that passes through the origin, as we discussed previously.

E XAMPLE 2
Graph the line y  3x  4 and describe it.
Solution The line has a slope of 3, so it rises 3 units for each increase of 1 unit to the
right. It has a vertical intercept of 4, so the line crosses the y-axis 4 units below the ori-
gin. The graph of this line is shown in Figure 2.5.
y

10
y = 3x − 4
8

4
3
2
1
x
–2 –1 1 2 3 4 5
–2

FIGURE 2.5 –4

◆
The graphs associated with the equations
y  f 1x 2  2x  1, y  g 1x 2  2x  1, and y  h1x2  2x  2
Gord.3896.02.pgs 4/24/03 9:25 AM Page 48

48 CHAPTER 2 Families of Functions

are shown in Figure 2.6. The three lines are parallel because they all have the same
slope, m  2, and so all rise 2 units for each 1 unit increase to the right. But their
vertical intercepts are different: b  1, b  1, and b  2 respectively.
y
y = 2x + 1

y = 2x + 2 y = 2x − 1

b=2

b=1

b = –1

FIGURE 2.6

The graphs associated with the equations

y  f 1x 2  2x  1, y  g 1x 2  3x  1, and y  h1x 2  2x  1
are shown in Figure 2.7. Note that, because all three lines cross the y-axis at the
point y  1, all have the same y-intercept, b  1. However, the three lines have dif-
ferent slopes and so behave differently. The functions f and g are increasing as you
move from left to right (as x increases), whereas the function h is decreasing as x
increases. Moreover, the line y  g 1x 2  3x  1 is increasing more rapidly than
the line y  f 1x 2  2x  1 because it has a larger slope. Again, the slope of the
line determines whether a line rises or falls and how rapidly it does so.
y
m=3 m=2
y = 3x + 1
y = 2x + 1

y = −2x + 1
FIGURE 2.7 m = −2

In summary, when the slope m is positive, the line rises as x increases from left
to right and the linear function is increasing; the larger m is, the faster the line rises.
When the slope m is negative, the line falls as x increases from left to right and the
linear function is decreasing; the more negative the slope, the faster the line drops.
(However, because a line doesn’t bend, either up or down, it is neither concave up
nor concave down.)
We can express the slope of a line,
change in y y rise
m   .
change in x x run
Gord.3896.02.pgs 4/24/03 9:25 AM Page 49

2.2 Linear Functions 49

in another way. If 1x1 , y1 2 and 1x 2 , y2 2 are two points on the line, then
x  x2  x1 and y  y2  y1 ,
where the order of the coordinates must be the same in both x and y. Therefore
the equation for the slope becomes
y y2  y1
m  .
x x2  x1
Now let’s explore these ideas in terms of the real world.
E XAMPLE 3
A wholesale supplier quoted the following costs, C, in dollars, for graphing calculators,
depending on the number n of units ordered.

n 1 2 3 4 5 ...

C 87 167 247 327 407 ...

Find a linear function that models the costs and discuss the meaning of the slope and
vertical intercept.
Solution Note that the cost for each additional calculator after the first is $80. The $87
charged for the first calculator consists of the $80 for the calculator and shipping, plus an ad-
ditional $7 that covers the fixed cost for processing the order. This amount remains fixed no
matter how many units are purchased. Therefore the cost C of buying n calculators can be
written as the linear function C  f 1n2  80n  7. The slope, 80, represents the increase in
the total cost for each additional unit ordered—every time n increases by 1, C increases by
80. That is, the rate at which the cost is increasing is $80 per calculator sold. The vertical in-
tercept, 7, is the fixed cost for any size order. Figure 2.8 depicts the slope as the ratio
C rise 167  87
m    80,
n run 21
or simply m  $80 per calculator.
C

∆C
m= = 80
∆n

∆C

7
∆n

FIGURE 2.8 n

◆
The value m  80 was based on the two points 11, 872 and 12, 1672. If you cal-
culate the slope by using any two points on the line, you get the same value. It is this
fact—that the slope, or rate of change, is the same at every point—that makes a line
straight. If the rate of change varies from one point to another, then the function is
not linear.
The form for the equation of the line C  80n  7 that we found in Exam-
ple 3 is known as the slope–intercept form and usually is written as y  mx  b
Gord.3896.02.pgs 4/24/03 9:25 AM Page 50

50 CHAPTER 2 Families of Functions

because it highlights the slope m  80 of the line and the vertical intercept
b  7.
The slope–intercept form is very useful for displaying the equation of a line.
However, it is usually a poor choice for finding the equation of a line because the
vertical intercept is often difficult to determine. Even when we can find the vertical
intercept, it may have little to do with the situation we’re studying. Example 4
demonstrates an easy way to apply the slope–intercept form.

E XAMPLE 4
A plumber charges $50 for a service call to come to the job and $70 per hour for labor.
(a) Find a linear function for the plumber’s charges for a job taking t hours (disregard-
ing the costs for any parts). (b) What is the meaning of the slope in this function?
Solution

a. The plumber charges $50 just for coming. For each hour on the job, the charge is an
additional $70, so a job lasting t hours costs an additional 70t dollars. Therefore the
total cost is
C  50  70t.

b. The slope of this line, 70, is the charge for each hour of labor and its units are dollars
per hour.
◆
Usually, a much better method for determining an equation of a line is the
point–slope form. It is based on the idea that a line is determined by its slope m and
one point 1x0 , y0 2 on the line. Suppose that 1x, y2 is any other point on the line, as
shown in Figure 2.9. Because
y y  y0
m  ,
x x  x0

∆y y − y0
m= = Q(x, y)
∆x x − x0

∆y

P(x 0, y0 ) ∆x

FIGURE 2.9 x

we can multiply both sides by 1x  x0 2 to get

y  y0  m1x  x0 2 .
In summary we have the following formula.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 51

2.2 Linear Functions 51

Point–Slope Formula for the Equation of a Line

The equation of the line with slope m that passes through the point 1x0 , y0 2 is
y  y0  m1x  x0 2

For instance, the line through the point 15, 2 2 with slope 4 is
y  2  41x  52 .
You are almost always better off using the point–slope form rather than the
slope–intercept form to find the equation of a line. In Example 5 we revisit the
snow tree cricket from Example 2 in Section 1.4.
E XAMPLE 5
The following set of measurements relate the snow tree cricket’s rate of chirping, in
chirps per minute, to the temperature, in Fahrenheit.

Temperature, T (F) 50 55 60 65 70 75 80

Rate, R (chirps>min) 40 60 80 100 120 140 160

a. Find a function that models the chirp-rate as a function of temperature.

b. Discuss the reasonableness of the model and give reasonable values for the domain
and range.
Solution

a. The chirp rate is increasing steadily, so it is an increasing function of the temperature T.

In particular, the chirp rate goes up 20 chirps>min for every 5°F increase in tempera-
ture. Equivalently, the chirp rate goes up 4 chirps>min for every 1°F increase in tem-
perature. Figure 2.10 shows that the corresponding points clearly fall in a linear pattern.
In other words the chirp rate R as a function of temperature T is a linear function.

160
Chirps/minute

120

80

40

T
50 60 70 80
FIGURE 2.10 Temperature (°F)

Because two points determine a line, we can use any two of the given points—
say, 155, 602 and 175, 1402 —to find the equation for this line. Using these two points,
as shown in Figure 2.11, we find that the slope of the line is
R 140  60 80
m    4.
T 75  55 20
Gord.3896.02.pgs 4/24/03 9:25 AM Page 52

52 CHAPTER 2 Families of Functions

160

(chirps/minute)
120

Chirp rate
Rise = 140 − 60 = 80
80

40 Run = 75 − 55 = 20

T
0 20 40 60 80
FIGURE 2.11 Temperature (°F)

This value for the slope means that, for each 1°F increase in temperature, the cricket
chirps 4 more times per minute. Thus, if the temperature goes up 5°F, the cricket
chirps 20 more times per minute; if it goes up 10°F, the cricket chirps 40 more times
per minute, and so on.
Next we apply the point–slope formula to find the equation of the line, using
any point on the line. If we pick the point 155, 602 used earlier, we obtain
R  60  41T  55 2
 4T  220.
Adding 60 to both sides of this equation, we get
R  f 1T2  4T  160.
This equation tells us that the vertical intercept is R  160 (when T  0). Of
course, a chirp rate of R  160 is meaningless! Was the formula wrong? No. But it
makes sense to describe the snow tree cricket’s chirp rate only for temperatures be-
tween, or possibly near, the given set of readings—that is, from 50°F to 80°F. It does
not make real-world sense to use this linear relationship far outside of this interval,
such as at 0°. The formula doesn’t predict sensible chirp rates for temperatures less
than 40°F, when R becomes negative. It doesn’t hold at temperatures high enough to
cook the cricket either. Because temperatures in the Colorado Rockies aren’t likely to
rise above 100°F, there is a natural domain for this function:
Domain of f  all values between 40°F and 100°F or
40  T  100 or 3 0, 100 4.
This function is strictly increasing, so we can find the corresponding range:
f 1402  4 . 40  160  0 and f 1100 2  4 . 100  160  240.
Thus
Range of f  all values R from 0 to 240 or 0  R  240 or 30, 240 4.
b. How reasonable are these results? At 100°F, the equation predicts that a snow tree
cricket will chirp 240 times per minute, or 4 times per second, which we might de-
cide is a bit unreasonable. Thus, even though the linear model predicts this value, we
might want to rethink whether extending the linear model as far as T  100°F makes
sense when the upper limit of the data values is T  80°F. As we’ve said previously, it
is often misleading to extrapolate too far beyond the actual data values.
◆
So far we have used the temperature to predict the chirp rate, and we thought
of the temperature as the independent variable and the chirp rate as the dependent
Gord.3896.02.pgs 4/24/03 9:25 AM Page 53

2.2 Linear Functions 53

variable. However, we could reverse the role of the variables and think of temper-
ature as a function of chirp rate. How we view a relationship determines which
variable is dependent and which is independent. Thinking of temperature as a
function of chirp rate would enable us to approximate temperature for given
chirp rates. To do so, we again start with the formula
R  f 1T2  4T  160
and solve it algebraically for T as a function of R. We add 160 to both sides to obtain
4T  R  160
and then divide both sides by 4 to get

T  1R  160 2  R  40  g 1R2.
1 1
4 4
This linear function has slope 14 and vertical intercept 40, except now the inde-
pendent variable is R. So, if you ever encounter a snow tree cricket who is chirping
merrily away, knowing this equation can help you determine the local temperature
just by using your watch. Count the number of chirps in a one-minute interval and
apply the formula to calculate the temperature.
We summarize the important information about linear functions as follows.

The slope-intercept form for the equation of a line is

y  mx  b,
where m is the slope, or rate of change of y with respect to x,
y rise
m  ,
x run
and b is the vertical intercept, or value of y when x  0.
The point–slope form for the equation of a line with slope m that passes
through the point 1x0 , y0 2 is
y  y0  m1x  x0 2 .

Note that in the slope-intercept form for the equation of a line, there are
two parameters, the slope m and the vertical intercept b. So linear functions are
a two-parameter family of functions. We determine the equation of a particular
line by finding the values of the two parameters.
E XAMPLE 6
During the early years of the Indianapolis 500 race held annually on Memorial Day, the
average winning speed increased as shown in the following table. Find a formula to
model these values.

Year 1919 1922 1925

Average speed (mph) 88 94.5 101

Source: World Almanac and Book of Facts.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 54

54 CHAPTER 2 Families of Functions

Solution The average winning speed starts at 88 mph and increases at the rate of 6.5 mph
each three years. Because the average winning speed S increases consistently by 6.5 mph
every three years, S is a linear function of time over the period 1919 to 1925, as shown in
Figure 2.12. The slope of this line is
rise S 6.5
Slope     2.17,
run t 3

∆S
m= = 2.17
101 ∆t
rise = 6.5
94.5
run = 3
88

t
FIGURE 2.12 0 3 6

which shows the rate at which the winning speed increased each year. Let t be the
number of years since 1919. Using the initial point 10, 88 2 gives the equation of this
line as
S  88  2.171t  0 2  2.17t,
or
S  f 1t2  2.17t  88.
In this formula, t represents the number of years since 1919 and S is the speed in miles
per hour. Note that we would get the same result if we used the slope–intercept form.
◆
You may wonder whether this linear trend continued beyond 1925. Let’s
compare what it predicts with what actually happened. The fastest average win-
ning speed in the Indy 500 was 186 mph in 1990, when t  71 years after 1919.
Using the linear equation S  2.17t  88, we predict an average speed of 242 mph
in 1990. Clearly, although speeds have increased dramatically, they haven’t kept
up with the linear function we constructed based on just a few early data points.
Further, this model again illustrates the danger of extrapolating too far from the
given data.

Think About This What does this information indicate about how long the 500-mile race takes?
How much longer did it take the winning car to drive the 500 miles in 1919 than
it took in 1990? ❐

Because the data in the table are given only at specific points (every 3 years),
we say that the data are discrete. However, because the function S  2.17t  88
makes sense for all possible values of t, we treat the variable t as though it were
continuous (or defined for all points). The graph shown in Figure 2.12 is of a con-
tinuous function because it is a solid line including infinitely many points, not
Gord.3896.02.pgs 4/24/03 9:25 AM Page 55

2.2 Linear Functions 55

just the three distinct points representing the winning speeds in the race in three
particular years.

E XAMPLE 7
Search and Rescue teams are often called on to find lost hikers in remote areas in the
Southwest. Members of the search team walk through the search area parallel to each
other at a fixed distance d between searchers. Experience has shown that the team’s
chance of finding those who are lost is related to the distance of separation d. The clos-
er together the searchers are, the better are their chances of success. Based on a num-
ber of simulated missions, the percentage of lost people who were found was used to
assess the probability of finding someone based on various separation distances, as
shown in the following table of values. Find a formula to model these probabilities.

Distance d (ft) 20 40 60 80 100

Probability of success P (%) 90 80 70 60 50

(These values correspond to searches conducted in the relatively open terrain of the
Southwest; searchers in other regions where there is dense forest or undergrowth
would have to use much narrower separation distances to achieve comparable levels of
success.)
Solution Because the value for the probability of success P decreases as distance d
(the independent variable) increases, the function P  f 1d2 is a decreasing function
of d. The data indicate that each 20 foot increase in distance causes the probability of
success P to decrease by 10%. Because this fact holds for any successive pair of
points, P is a linear function of d, and the graph of the probability of success versus
distance is a line, as shown in Figure 2.13. Based on the two data points (20, 90) and
(100, 50), say, the slope of this line is
P 50  90 40 1
m    .
d 100  20 80 2

100
Probability of success

–40

50
80 1
P = 100 − d
2

d
0 20 40 60 80 100 120 140 160 180 200
FIGURE 2.13 Distance

The negative sign reinforces the fact that P decreases as d increases. The slope is the rate
at which P is decreasing as d increases.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 56

56 CHAPTER 2 Families of Functions

To find the equation of the line, we use the point–slope formula. We choose any one
of the given points—say, 120, 902 —and obtain

P  90   1d  20 2
1
2
1
  d  10.
2
Adding 90 to both sides of this expression, we get

P  f 1d2   d  100.
1
2
◆

Think About This Pick any one of the other points in Example 7 and show that you get the same
equation for P. ❐

What is the meaning of the vertical intercept, P  100? Suppose that d  0 so

that the searchers are walking shoulder to shoulder; we would expect everyone to
be found, or P  100. What is the horizontal intercept? When P  0, we have
0   12 d  100, or d  200. According to the model, the value d  200 repre-
sents the separation distance at which no one is found. This outcome is unreason-
able because, even when the searchers are far apart, the search will sometimes be
successful. What this situation suggests is that, somewhere outside the data given,
the linear relationship ceases to hold. As in the Indy 500 example, extrapolating too
far beyond the given data may not make sense.

Some Useful Facts

Several facts about lines are useful to remember.
1. Parallel lines have the same slope. That is, the quantities they represent are
growing at the same rate.

Think About This The lines y  4x  3, y  4x  11, and 4x  y  15  0 are all parallel. What is
their common slope? ❐

2. Perpendicular lines have slopes that are negative reciprocals.

The lines y  2x  9 and y   12 x  3, having slopes of 2 and  12 , respec-
tively, are perpendicular to each other. Sketch their graphs to convince yourself of
this fact. Similarly, the lines y  0.162x  7.4 and y  6.173x  1.03, which
have slopes of 0.162 and 6.173  1>0.162, respectively, are perpendicular to
each other.

Think About This Write the equation of a line that is perpendicular to y  54 x  7. (Of course, your
answer will likely be different from your classmates’ choices.) ❐

3. The point where any two lines cross is known as their point of intersection.
The x- and y-coordinates of this point must satisfy both equations simul-
taneously. You find the point of intersection by solving the system of si-
multaneous equations either algebraically or graphically. (See Appendix B
and C for a discussion of ways to solve such systems of equations.)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 57

2.2 Linear Functions 57

Problems
1. Match each equation with its graph. (Note that the 4. The graph of Fahrenheit temperature F versus Cel-
scales of the graphs are different.) sius temperature C is a line. Water boils at 212°F
and 100°C and freezes at 32°F and 0°C.
y y y
a. Sketch the graph of the line.
b. Find the slope of the line relating the two tem-
perature scales.
x x x c. Find the equation of this line.
d. Use the equation to find the Fahrenheit temper-
ature that corresponds to 30°C.
(i) (ii) (iii) e. Use the equation to find the Celsius temperature
that corresponds to 98.6°F.
y y y f. When is the Fahrenheit temperature the same
numerical value as the Celsius temperature?
5. In 1990, 442.2 million prerecorded cassette tapes and
x x x 865.7 million CDs were sold in the United States. In
1998, 158.5 million cassettes tapes and 1,124.3 mil-
lion CDs were sold. Assume (incorrectly) that the
(iv) (v) (vi)
pattern of sales for both items is linear.
a. Find the equation for the number of cassette
a. y  x  2 b. y  x  3 tapes sold as a linear function of time.
c. y  2x  4 d. y  3x  4 b. Find the equation for the number of CDs sold as
e. y  12 x f. y  3 a linear function of time.
c. What is the practical significance of the slopes in
2. Estimate the slope of each line. Then use the slope
parts (a) and (b)?
to find an equation of the line.
d. If the trends in sales of both items were indeed
a. y linear, find when the number of CDs sold over-
took the number of cassette tapes sold.
3 e. Use the data given to find the total number of both
2
CDs and cassette tapes sold in 1990 and 1998 and
use these values to find the equation for the total
1 number of sales of both items combined as a lin-
x
ear function of time.
0 1 2 3 4 5 f. Use the fact that, in 1995, 272.6 million cassette
tapes and 272.6 million CDs were sold to explain
b. P
why assuming that the sales trends were linear is
300 incorrect.
6. The charges for a taxi ride are an initial charge of
200
$1.80 and $0.75 for each mile driven.
100 a. Write a formula for the charge for a taxi ride as a
t linear function of the distance traveled.
0 10 20 30 40 50 b. What is the meaning of the slope of this linear
function?
3. Find the equation of the line passing through each c. What is the cost of a 12-mile trip?
pair of points. d. Suppose that you have only $15. How far can
a. 11, 2 2, 12, 5 2 b. 11, 22 , 13, 2 2 you go in the taxi? (Assume that you will give a
c. 13.52, 4.962, 11.91, 8.362 $2 tip out of the $15 you have.)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 58

58 CHAPTER 2 Families of Functions

7. A long-distance telephone company charges 40¢ to b. What is the practical meaning of the value you
place a call from Los Angeles to London and 30¢ for get for the slope?
each minute. c. Sketch a single graph showing both tax formu-
a. Write the equation of a linear function that mod- las. Is there any discrepancy?
els this situation. 12. When filing income tax returns, many people can
b. What is the practical significance of the slope? claim deductions for depreciation on items such as cars
Of the vertical intercept? and computers used for business purposes. The idea is
c. What is the cost of a 26-minute call? that the value of such an asset decreases, or depreciates,
d. Suppose that there is a 30% discount on the over time. The simplest method used to find the depre-
rates for calls made in off-peak hours. Repeat ciated value is called straight-line depreciation, which
parts (a)–(c). assumes that the item’s value decreases as a linear func-
8. (Continuation of Problem 7) A competing long- tion of time. If an $1800 computer system depreciates
distance company claims that it is cheaper because completely in five years, find a formula for its value as a
its rates on the Los Angeles to London call are 15¢ function of time. What is it worth after three years?
to place the call and 36¢ for each minute. 13. The Athabasca glacier in southern Alberta, Canada, is
a. For the 26-minute call in Problem 7(c), which part of the largest mass of ice in the Rocky Mountains.
carrier is actually cheaper? (Tourists who visit the Jasper and Banff National
b. Graph both lines. What does the point where Parks can take a side trip out onto the actual glacier.)
they intersect signify? Over the past 120 years, the glacier has been steadily
c. Find the length of call at which the second com- “withdrawing” at a rate of about 15 meters per year, as
pany becomes more expensive than the first. it slowly melts.
a. Express the approximate position of the south-
9. A disk jockey (DJ) charges a flat fee of $120 per
ernmost extent of Athabasca as a function of
party plus $60 for each hour of the party. A sec-
time, measured in years from 1900. Measure its
ond DJ charges $100 per party plus $75 for each
position northward from the U.S.–Canada bor-
hour.
der, which was about 300 kilometers south of the
a. For each DJ find a formula that gives the cost of glacier in 1900.
hiring the DJ as a function of the number of b. If the current rate of withdrawal has been in ef-
hours the party lasts. fect indefinitely, how long ago did the toe of the
b. Sketch the graphs of both functions on the same glacier extend over the border?
set of axes. c. Can the function in part (a) continue to apply
c. How do you decide which DJ costs less? for the next million years? Why or why not?
10. The net income of the Apex Company was $240 mil- 14. Jen is typing her term paper for Psych 101. She types
lion in 1980 and has been increasing by $30 million the body of the paper at the rate of 35 words
per year since. Over the same period, the net income per minute for 30 minutes, then takes a 5-minute
of its chief competitor, the Best Corporation, has break, and comes back to do the references at a rate of
been growing by $20 million per year, starting with 20 words per minute for 12 minutes.
$300 million in 1980. Which company earned more a. Sketch the graph of Jen’s typing rate as a func-
in 1990? When did Apex surpass Best? tion of time.
11. According to the IRS, the formula T  0.15I, which b. Sketch the graph of the total number of words
gives income tax as a function of taxable income, ap- she types as a function of time.
plies only for single taxpayers with taxable incomes c. Find the equations of the different line segments
up to $21,450. The IRS tax table states: “If the taxable you drew in part (b).
income is over $21,450, But not over $51,900, Enter 15. A bicyclist pedals at the rate of 1000 ft>min for 20 min-
on Form 1040: $3,217.50  28% of the amount utes, then slows to 500 ft>min for 6 minutes, then
over 21,450.” races at 1200 ft>min for 4 minutes, and cools down at
a. Rewrite this statement as an equation that can be 500 ft>min for 5 minutes.
used to calculate your taxes. What are the do- a. Sketch the graph of the bicyclist’s rate as a func-
main and range of the resulting function? tion of time.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 59

2.3 Linear Functions and Data 59

b. Use the graph from part (a) to determine the b. Solve the two equations algebraically.
total distance biked. c. Add two times Equation (2) to Equation (1) to
c. Sketch the graph of the distance traveled as a get a new linear equation. Plot that line on the
function of time. same graph you created in part (a). What do you
d. Find the equations of the different line segments observe about the three lines?
you drew in part (c). d. Add three times Equation (2) to Equation (1)
16. The points P, Q, and R lie in order from left to right and plot that line on the same graph. What do
on the graph of a function f that is increasing. If the you observe about the four lines?
slope of line segment PQ is less than that of line e. Add four times Equation (2) to Equation (1) and
segment QR, is the curve concave up or concave plot that line on the same graph. What do you
down? Explain your reasoning. conclude from this result?
f. Find an appropriate multiple of Equation (2)
17. Find the equation of the line that passes through
the point 16, 42 and is
that, when added to Equation (1), will eliminate
the x-term. What will the graph of the resulting
a. parallel to the line y  5x  3. line look like when x has been eliminated?
20. The point 13, 4 2 is on the circle x 2  y2  25.
b. perpendicular to this line.
18. Find the equation of the line that passes through
the point 16, 42 and also passes through the point of
a. Find the equation of the line that is tangent to
the circle at this point.
intersection of y  2x  1 and y  3x  6.
b. Find the points where the line intersects the x and
19. The algebraic method of elimination for solving a y axes. (Hint: The line tangent to a circle at a point
system of linear equations involves adding a multiple is always perpendicular to the radius at that
of one equation to another equation to eliminate one point.)
of the variables. Consider the system of two equations 21. a. Of the following three linear functions, which
in two unknowns: two represent perpendicular lines?
3x  4y  1 (1) i. 3x  4y  12 ii. 2x  5y  10
2x  y  8. (2) iii. 8x  6y  7
a. Plot the two lines carefully on a sheet of graph b. For the two lines that are perpendicular, find the
paper and determine the point of intersection. point of intersection.

Exercising Your Algebra Skills

Solve each equation for the appropriate variable. 13. 312x  5 2  4 14. 214  3w2  7
1. 5x  7  12 2. 3x  8  7 15. 3.21t  19802  1700 16. 1.351t  752  8
3. 4x  3  5 4. 8x  7  15 Find (a) the slope and (b) the x and y intercepts of each
5. 18y  7  22 6. 5.4x  7.2  0.8 line in Problems 17–20.
7. 9  3x  6 8. 5  4p  7 17. 2x  3y  8 18. 2x  3y  8
9. 4.7q  5.1  24.5 10. 1.3w  12.8  22.7 19. 4x  7y  5  0 20. 3y  2x  4  0
11. 4k  7  9k  8 12. 6z  5  4z  11

2.3 Linear Functions and Data

Determining Whether a Set of Data Is Linear
When you encounter tables of data relating two quantities, you will often need to
determine whether a linear relationship exists between the two variables. You could
plot the data points to see if the points fall into a linear pattern, but this approach is
imprecise. Alternatively, you could decide whether a function y  f 1x 2 given by a
table of values is linear by examining the data. If the data fall into a linear pattern,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 60

60 CHAPTER 2 Families of Functions

you should get the same slope no matter which pair of points you use. This reason-
ing gives a simple criterion for determining linearity: See whether the differences in
y-values are constant for equally spaced x-values.

If the x-values are uniformly spaced and there is a constant difference

among the y-values, the data fall into a linear pattern.

You can visualize this principle by thinking of a long plank of wood and a
flight of stairs. If the steps all have the same height—say, 8 inches, and the same
depth, you can lay the plank on the stairs and it will touch the edge of each one, as
illustrated in Figure 2.14. The plank plays the role of a line. But, if the stairs have
different heights or depths, the plank won’t touch every one of the edges—those
edges do not fall in a linear pattern.

FIGURE 2.14

E XAMPLE 1
The following two sets of data represent values for a linear function and a nonlinear
function. Identify which is the linear function. Find the equation of the line and the con-
cavity of the nonlinear function.

x f (x) x g(x)
1.0 7.0 1 2
1.2 7.8 2 3

1.4 8.6 3 6

1.6 9.4 4 11

1.8 10.2 5 18

2.0 11.0 6 27

2.2 11.8 7 38

Solution In both sets of values, the x-values are evenly spaced, so we can proceed to
examine the successive differences in the values of the two functions, which we write
as f 1x2 and g 1x 2. For instance, for the function f the difference between the first
two values is 7.8  7.0  0.8. Continuing in this manner, we obtain the data on the
next page.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 61

2.3 Linear Functions and Data 61

x f (x) f (x)
1.0 7.0
0.8  7.8  7.0
1.2 7.8
0.8  8.6  7.8
1.4 8.6
0.8
1.6 9.4
0.8
1.8 10.2
0.8
2.0 11.0
0.8
2.2 11.8

Because the difference is a constant 0.8 between the values of the function f, we conclude
that this set of data is indeed linear. The slope of the line through these points is
y f 1x 2 0.8
m    4.
x x 0.2
Further, using the first point 11, 72 and the point–slope form for the equation of a line,
we find that the equation of the line is
y  7  41x  1 2
 4x  4.
When we add 7 to both sides of this expression, we get
y  4x  3  f 1x 2.
Suppose that we try the same analysis on the values for the function g.

x g(x) g(x)
1 2
132
2 3
363
3 6
5
4 11
7
5 18
9
6 27
11
7 38

The differences are not constant, so we conclude that these points don’t fall into a
linear pattern and hence no line passes through them. Consequently, the function g can-
not be a linear function. In fact, because the differences are successively larger, the func-
tion is growing faster than a linear function grows. Because the function g is increasing
at an increasing rate, it is concave up.
◆
So far, we have given you information on some process or quantity that clearly
is a linear function. In practice, however, you may face a situation in which you
Gord.3896.02.pgs 4/24/03 9:25 AM Page 62

62 CHAPTER 2 Families of Functions

simply assume that one quantity grows or decays in a linear manner. Or you may
even encounter a set of data that appears to be roughly linear in nature, but the
particular data points do not precisely fall on a line. We illustrate both situations in
Examples 2 through 4.
E XAMPLE 2
In 1990, the United States imported $495 billion worth of goods. In 1998, the United
States imported $912 billion worth of goods. (Source: 2000 Statistical Abstract of the
United States.)
a. Assuming that the growth in imports followed a linear pattern, find an equation of
the linear function that models U.S. imports.
b. What is an appropriate domain for this model?
c. Use the model to predict the amount of imports in the year 2005.
d. Predict when the United States will import $1 trillion worth of foreign goods ac-
cording to this model.
Solution
a. For convenience, we take the independent variable t to be the number of years since
1990 and measure imports I in billions of dollars. We therefore have two points
10, 4952 and 18, 912 2 for our linear model, as shown in Figure 2.15. The slope of the
line through these points is
912  495
m  52.1;
80
I

912

912 – 495

495
8

t
FIGURE 2.15 0 t=8

that is, imports have been growing at a rate of $52.1 billion per year. Using the point–
slope form for a line and the point 10, 495 2 yields
I  495  52.11t  0 2,
or equivalently the slope–intercept form
I  52.1t  495.
b. The data extend from 1990 when t  0 to 1998 when t  8. A reasonable domain
might be from t  5 to t  13, allowing us to predict 5 years before and after the
data points.
c. Assuming that this linear trend continues, we predict that the value of foreign goods
that will be imported in 2003, when t  13 years after 1990, is
I  52.1113 2  495  $1172.3 billion.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 63

2.3 Linear Functions and Data 63

d. Using this linear model, we have to solve for the value of t when
I  52.1t  495  1000 billion 11 trillion2 .
If we subtract 495 from both sides of this equation, we obtain
52.1t  505
so that
t  505>52.1  9.7,
or sometime late in 1999.
◆
In Example 2 we arbitrarily chose to count years from 1990. Let’s see what hap-
pens if we choose a different baseline.

E XAMPLE 3
As in Example 2, the United States imported $495 billion worth of goods from abroad in
1990 and $912 billion in goods in 1998. Assuming that the growth in imports followed a lin-
ear pattern, find an equation of the linear function that models U.S. imports based on using
the independent variable t to represent (a) the number of years since 1900 and (b) the
number of years since year 0. For each model, state an appropriate domain and compare
each model to the one constructed in Example 2. (c) Use each model to predict the
amount of imports in 2003.
Solution In Example 2, we took the independent variable t to be the number of years
since 1990, or equivalently used t  0 in 1990, and constructed the linear model
I  52.1t  495.
a. Now suppose that the independent variable t is the number of years since 1900. We
therefore have the two points 190, 4952 and 198, 9122. The slope of the line through
these points is
912  495
m  52.1,
98  90
which is the same value obtained before. Using the point–slope formula and the
point 190, 4952 gives the equation of this linear function as
I  495  52.11t  90 2
or
I  52.1t  4194.
Although the slope remained the same, the vertical intercept changed dramatically.
The reason is that we now think of the line as “starting” in 1900, not 1990, so it has
been climbing for 90 years at the rate of $52.1 billion per year. A reasonable domain
for this linear model might be from t  85 to t  103.
b. Now suppose that the independent variable t is the number of years since the year 0.
Our two points are now 11990, 495 2 and 11998, 9122 , and the slope of the line
through these points is
912  495
m  52.1,
1998  1990
Gord.3896.02.pgs 4/24/03 9:25 AM Page 64

64 CHAPTER 2 Families of Functions

which again is the same value. Using the point–slope formula and the point 11990, 4952,
the equation of this linear function is
I  495  52.11t  19902
or
I  52.1t  103,184.
There has been a huge change in the vertical intercept because this line has been
climbing at a rate of $52.1 billion per year for almost 2000 years! An appropriate do-
main for this linear function might be from t  1985 to t  2003.
c. Using any one of the three linear models, the first with t  13, the second with
t  103, and the third with t  2003, we obtain the identical prediction for the total
value of imports into the United States of about I  $1172 billion in 2003.
◆
So, which of these three models is correct? In one sense, all three are correct be-
cause they give the same predictions. In another sense, they are all wrong, because the
equation by itself, without reference to what the variable t stands for, is incomplete—
if we don’t specify the meaning of the variable or its “starting” point, someone using
the equation to make predictions may well use a different interpretation. We get very
different answers from the first model,
I  52.1t  495,
if we use t  13, t  103, and t  2003. Therefore we should write the three
models as
I  52.1t  495, where t is the number of years since 1990,
or
I  52.1t  4194, where t is the number of years since 1900,
or
I  52.1t  103,184, where t is the number of years since the year 0.

Capturing a Linear Pattern in Data

In most applications of linear functions in the real world, you will typically have far
more than two points; in fact, you will often have a relatively large set of points that
fall into a roughly linear pattern. In Example 4 we illustrate how to deal with this
important situation.

E XAMPLE 4
The following table of values gives some measurements for the rate of chirping (in
chirps>sec) of the striped ground cricket as a function of the temperature.

T (F) 89 72 93 84 81 75 70 82 69 83 80 83 81 84 76
Chirps>sec 20 16 20 18 17 16 15 17 15 16 15 17 16 17 14
Source: Adapted from George W. Pierce, The Songs of Insects. Boston: Harvard University Press, 1948.

Even though the measurements presented for the snow tree cricket in Chapter 1 fell
exactly onto a straight line, Figure 2.16(a) shows that comparable measurements for
Gord.3896.02.pgs 4/24/03 9:25 AM Page 65

2.3 Linear Functions and Data 65

the striped ground cricket clearly do not. The difference may be due to errors in
measurement; it may be that the striped ground cricket is less sensitive to tempera-
ture; or perhaps the snow tree cricket has more mathematical aptitude to get the sit-
uation right. Even though the points for the striped ground cricket do not fall
precisely on a line, they do fall in a roughly linear pattern. Find an equation that cap-
tures this linear pattern.

Solution Suppose that we take a piece of black thread (or a clear plastic ruler), hold it
taut, and move it back and forth over the points in Figure 2.16(a). Each possible orienta-
tion for the thread represents a different line. We can then select an orientation that
seems, by eye, to give the best match or fit to the linear pattern in the data. Usually we
want roughly half the points to be above the thread and half below it, so that the line
passes “midway” between the points and follows the overall trend. Such a line superim-
posed over the data points is shown in Figure 2.16(b). (Obviously, different people will
come up with slightly different lines.) We now estimate the equation of this line that
captures the overall trend of the chirp rate function.

20 20
19 19
Chirps per second

Chirps per second

18 18
17 17
16 16
15 15
14 14
13 13

65 70 75 80 85 90 95 65 70 75 80 85 90 95
Temperature (°F) Temperature (°F)

(a) (b)

FIGURE 2.16

In Figure 2.16(b) this line seems to pass through the points 168, 142 and 190, 192.
(Note that these points are on the line; we chose them for convenience. The points used
are not necessarily actual data points. In fact, unless the line drawn happens to pass
through a data point, you should not use any of the data points to estimate the slope.)
Using these two points, we find that the slope of the line is approximately
19  14 5
m   0.23.
90  68 22
This means that the chirp rate increases about 0.23 chirp>sec for each 1°F increase in
temperature. Further, because the line apparently passes through the point 168, 142 , we
conclude that the equation of the line is
C  14  0.231T  682 ,
or, when simplified,
C  0.23T  1.64.
◆
Gord.3896.02.pgs 4/24/03 9:25 AM Page 66

66 CHAPTER 2 Families of Functions

In applying this “black thread method” to find the equation of the line, you must
use two points that are on the line you draw. Do not use data points that are not on the
line and do not force a line by drawing one that must pass through any of the data
points. Note that the result you get is just an estimate for the equation of the line that
visually best fits the linear trend in the data. In Chapter 3, we introduce methods for
finding the equation of the one line that is the best fit to a set of data in a certain sense.

Implicit Linear Functions

We now consider a somewhat different type of situation, in which two quantities are
related but in such a way that we can’t necessarily identify which variable is independ-
ent and which is dependent. An ongoing debate at all levels of government concerns
the allocation of money among different programs. Because typically only a fixed
amount of money is available, the more that is spent on one program, the less there is
to spend on other programs. Let’s look at a simple case involving just two competing
programs, funding road and highway repairs versus funding day-care centers.
E XAMPLE 5
Suppose that we have a total of $1,000,000 available to divide between day-care centers,
which cost $200,000>center, and road repaving, which costs $50,000>mile. Find an
equation of a linear function relating the number of day-care centers and the number of
miles of road to be repaved. What are the domain and range of this function?
Solution Let c be the number of day-care centers and r be the number of miles of road to
be repaved. Then the amount of money spent on road repaving is $50,000r (it costs $50,000
to repave each mile), and the amount spent on day-care centers is $200,000c (it costs
$200,000 for each day-care center). Assuming that all the available money is spent, we get
amount spent on centers  amount spent on repaving  1,000,000
200,000c  50,000r  1,000,000
or equivalently, when we divide both sides by 50,000,
r  4c  20.
This equation is called the budget constraint. To graph this equation, we first find the points
at which the graph crosses the axes (the intercepts), as shown in Figure 2.17. If c  0, then
r  4102  20, so that r  20. At the other extreme, if r  0, we have 0  4c  20, so
that c  5.
c

r + 4c = 20

r
0 10 20
FIGURE 2.17

Because all the money not spent on road work is used for day-care centers, the num-
ber of centers funded is a function of the number of miles of roads repaved. That is, c is a
Gord.3896.02.pgs 4/24/03 9:25 AM Page 67

2.3 Linear Functions and Data 67

function of r, and we can solve the budget constraint equation r  4c  20 for c. We

subtract r from both sides of the equation and get
4c  20  r.
We then divide both sides by 4 and obtain

c  f 1r2  120  r2  5  r.
1 1
4 4
Similarly, the number of miles repaved is a function of the number of centers funded,
so r is a function of c. We can solve the budget constraint equation for r by subtracting 4c
from both sides to get
r  g 1c2  20  4c.
To determine the applicable domain and range, we recognize that r makes sense only
for values between 0 and 20, whereas c makes sense only for values between 0 and 5.
Which of these is the domain and which is the range depends on which variable we think
of as the independent variable and which as the dependent variable.
◆
Note that the budget constraint equation
r  4c  20
is an example of a third way of writing the equation of a line, called the normal form
of a line. In general, an equation of the form ax  by  c is the normal form of a
line. It is algebraically equivalent to either the point–slope form or the slope–intercept
form. For instance, if
3x  5y  15,
then
5y  3x  15
and when we divide both sides by 5, we get

y  a b 13x  15 2  a b x  3,
1 3
5 5
a line that has slope 35 and vertical intercept 3.
To graph a line given in normal form, the easiest way is to find and plot both
the vertical and the horizontal intercepts and connect them with a straight line.
Thus, to find the vertical intercept of 3x  5y  15, we set x  0 and solve
5y  15 to get y  3. To find the horizontal intercept, we set y  0 and solve
3x  15 to get x  5. The resulting graph is shown in Figure 2.18.
y
3

x
–1 1 2 3 4 5 6
–1

–2 3x − 5y = 15

–3

FIGURE 2.18 –4
Gord.3896.02.pgs 4/24/03 9:25 AM Page 68

68 CHAPTER 2 Families of Functions

Incidentally, as a function, an equation such as ax  by  c is called an

implicit function. It is a function, but neither variable, x nor y, is given explicitly in
terms of the other.

Problems
y
1. Determine which of the functions are linear. For
any linear function, find the equation of the line
and use it to predict the next entry to extend the 600
table of values.

a. 400
x 5 6 7 8 9

y 77 71 65 59 53 200

b.
t 50 60 70 80 90 x
0 50 60 70 80
L(t) 23.2 23.9 24.6 25.2 25.9
b. Find the correct equation of the line.
c. 5. In 1980 (when t  0), $26.5 billion were spent on
t 75 80 85 90 95
water pollution prevention and cleanup in the
Q(t) 125.1 127.5 129.9 132.3 134.7 United States. In 1990, $33.1 billion were spent.
a. Construct the linear function giving the amount
2. The data in each table of values lie along a line. spent on water pollution as a function of time t,
a. For each set of data, carefully plot the points on where t  0 in 1980.
graph paper, estimate by eye the slope and verti- b. Use the linear function to estimate the amount
cal intercept, and use these values to approxi- spent in 2002.
mate the equation of the line. c. Repeat parts (a) and (b) if t  0 in 1900.
b. Then find the equation of the line algebraically. 6. Inspector Clueless, while investigating the murder
How close was your estimate? of Mr. Jones, found the murderer’s size 11 12 footprint
in a flower bed. The inspector mutters something
i.
x 1 2 3 4 about the killer being “a man who is %#$#$&&#
tall.” If the equation of the best-fit line relating shoe
y 1.81 3.34 4.87 6.40 size to height in inches is S  0.51H  25.2, deci-
pher Clueless’s muttering.
ii.
x 1 2 3 4 7. The table of values shows the total value, in billions
of dollars, of electronics and electronic components
y 1.08 0.69 0.30 0.09 produced in the United States during the 1990s.

3. Find the equation of a linear function that fits this

set of values. t 1990 1993 1994 1995 1996 1997 1998
E 43.0 52.5 58.2 62.5 68.3 72.4 76.0

x 4 5 6 7 Source: 2000 Statistical Abstract of the United States.

y 1.557 1.614 1.671 1.728 a. Use graph paper to plot these data and use the
black thread method to sketch the best-fit line.
4. a. Explain why the equation of the line shown in b. Estimate the slope of this line and tell what it
the accompanying figure is not y  10x  200. means.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 69

2.3 Linear Functions and Data 69

c. What is your best estimate for the equation of c. Use your answer to part (b) to estimate a car’s
this line? gas mileage if it weighs 2350 pounds, 3100 pounds,
d. Use this line to estimate the total value of elec- 1950 pounds.
tronics and electronic components produced in d. What is your best estimate of the weight of a car
2004. that gets 32 mpg?
8. The table shows the height H in feet and the num- 10. A student who works as a waiter in a restaurant
ber of stories n in some notable buildings. records the cost C of meals and the tip T left by cou-
ples. His data for one evening are as follows.

H n C ($) 28.55 31.04 32.76 33.38 36.10 38.54

Empire State Building (New York) 1250 102 T ($) 4.25 4.50 5.00 5.00 5.50 6.00
John Hancock Tower (Boston) 788 61 Source: Student project.

Sears Tower (Chicago) 1450 110 a. Plot these points on a sheet of graph paper and
draw the best line you can to fit the points. Explain
NationsBank Plaza (Dallas) 921 72 your choices of the independent and the depend-
TCBY Tower (Little Rock) 546 40 ent variable.
b. Suppose that the equation for this function is
Peachtree Center (Atlanta) 374 31 T  0.18C  0.93. In terms of this mathemati-
Place Ville Marie (Montreal) 620 45 cal model, what is the increment in the tip for
each $1 increment in the cost of the meal?
Source: World Almanac and Book of Facts.
c. What does the slope of the line in part (b) rep-
resent? What significance does the vertical in-
a. Which variable, H or n, is the independent vari- tercept have?
able and which is the dependent variable? d. Suggest possible values for the domain and
b. Plot these points carefully on a sheet of graph range of this function.
paper and use the black thread method to locate 11. You have a fixed budget of $30 to spend on nuts and
and draw the line that seems to best fit the data Gummi Bear™ candy for a party. The nuts cost
points. $3 per pound, and the candy costs $2 per pound.
c. Estimate the equation of this line.
a. Write an equation expressing the relationship
d. What is the meaning of the slope of this line?
between the number of pounds of nuts and of
e. Use your answer to part (c) to estimate the num-
Gummi Bears that you can buy if you spend
ber of stories in a building 860 feet tall.
your budget completely. This equation is your
f. What is your best estimate of the height of a
budget constraint.
building that has 96 stories?
b. Graph the budget constraint, assuming that you
9. The table of values at the bottom of the page gives can buy any fractional amount of a pound. Label
data relating a car’s gas mileage to its weight. the intercepts.
a. Plot these points carefully on a sheet of graph c. What are the domain and range for this function?
paper and use the black thread method to locate d. Suppose that your roommate chips in an addi-
and draw the line that seems to best fit the data tional $30 for the party. Graph the new budget
points. constraint on the same set of axes used for the
b. Estimate the equation of this line. budget constraint graphed in part (b).

Weight (lb) 2100 2200 2400 2500 2800 3000 3200

Mileage (mpg) 37 34 29 27 26 25 23
Source: Student project.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 70

70 CHAPTER 2 Families of Functions

e. Keep the original budget at $30 and suppose 15. Suppose that a function f is increasing and concave up
that the Gummi Bears go on sale for half the and that f 160 2  250 and f 170 2  300. Which val-
price. Sketch the new budget constraint on the ues are possible and which are impossible? Explain.
same axes used in part (d). a. f 1652  270 b. f 165 2  275
f. Keep the original budget at $30 and suppose c. f 1652  280 d. f 11002  400
that the price of nuts suddenly doubles be- e. f 1100 2  450 f. f 11002  500
cause of a frost in the Southeast. Sketch the g. f 1402  100 h. f 140 2  150
new budget constraint on the same axes used i. f 1402  200
in part (d).
16. Suppose that a function f is decreasing and con-
12. For the implicit equation of a line 4p  3q  5, cave up, and that f 110 2  80 and f 1122  70.
find the following. Which values are possible and which are impossi-
a. An explicit function that gives p as a function ble? Explain.
of q. a. f 1112  78 b. f 111 2  75
b. The slope of the line in part (a). c. f 1112  72 d. f 115 2  50
c. An explicit function that gives q as a function e. f 1152  55 f. f 115 2  60
of p. g. f 15 2  100 h. f 15 2  105
d. The slope of the line in part (c). i. f 15 2  110
13. a. Find the slope and vertical intercept of each line 17. Draw the graph of a function f that is decreasing
given in normal form. and concave up. Mark three points on the curve: P
3y  2x  12 and 4x  5y  20. near the left, Q near the center, and R near the right.
These points determine three line segments: PQ,
b. Draw the graphs of the two lines on the same axes.
QR, and PR.
c. Find the point of intersection of the two lines
i. graphically; a. List the three line segments in the order of in-
creasing slopes.
ii. numerically by trial-and-error;
b. List the three segments in the order of increasing
iii. algebraically. steepness.
14. Repeat Problem 13 for the two lines 18. Repeat Problem 17 if the function is decreasing and
3y  2x  12 and 4x  5y  21. concave down.

Exercising Your Algebra Skills

2
Solve each formula for the indicated variable. 9. F  mvr , for v
1. A  bh, for h 2. C  2pr, for r Solve each equation in normal form for y in terms of x.
3. A  pr , for r
2
4. K  12 mv2, for m Identify the slope in each case.
5. K  12 mv2, for v 6. F 
GmM
d ,
2 for d 10. 4x  5y  20 11. 6x  5y  30
mv 2 12. 5x  4y  10 13. 2x  7y  9
7. T  2p 3 gl , for l 8. F  r , for r

2.4 Exponential Growth Functions

The population of Florida was 12.94 million in 1990 and has been growing as
shown in the following table of values. Let’s see if we can find a mathematical pat-
tern for the way in which this population is growing. If the population grows linear-
ly, the changes or increases in population from one year to the next, P, would all be
the same. Let’s check these differences.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 71

2.4 Exponential Growth Functions 71

Year Population P
1990 12.94
0.38  13.32  12.94
1991 13.32
0.38
1992 13.70
0.40
1993 14.10
0.41
1994 14.51
0.42
1995 14.93
0.43
1996 15.36
0.45
1997 15.81

Not only are the successive differences not constant, they are increasing. This
makes sense because as the population grows, there are more people to have babies.
Consequently, Florida’s population has been growing at a faster than linear rate.
We therefore need a concave up function to model this population over time.
Instead of taking differences, suppose that we take ratios of successive terms.
To do so, we divide the population in any year by the population in the preceding
year. This quotient gives
Population in 1991 13.32 million
  1.029,
Population in 1990 12.94 million
Population in 1992 13.70 million
  1.029,
Population in 1991 13.32 million
Population in 1993 14.10 million
  1.029,
Population in 1992 13.70 million
and so on. If you check the population figures for the subsequent years through
1997, you will find that each year the population grew by the same factor of
about 1.029.
Because the ratios of successive population values are constant, we have, for
any year,
Population next year
 1.029
Population this year
or
Population next year  1.029 . population this year.
If this trend continues, we can estimate Florida’s population in 1998 as
Population in 1998  1.029 . population in 1997  1.029 . 15.81  16.27.
The fact that Florida’s population next year is 1.029 times this year’s popula-
tion is equivalent to saying that
Population next year  1.029 . population this year
 11  0.029 2 . population this year
 population this year  0.029 . population this year.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 72

72 CHAPTER 2 Families of Functions

In other words, each year between 1990 and 1997, Florida’s population grew by
about 0.029  2.9% from one year to the next. The number 2.9% is called the
annual growth rate for the population.
Whenever the successive ratios are constant (here they all are 1.029), the func-
tion is an exponential function.
We now find an equation for this exponential function P1t2 , where t is the
number of years since 1990. The starting population, when t  0, is P10 2  12.94,
which we write as P0 . Then
when t  1, P11 2  1.029 . P0 , or 13.32;
when t  2, P12 2  1.029 . P11 2  1.029 . 11.029 . P0 2  11.0292 2 P0 , or 13.70;
when t  3, P132  1.029 . P12 2  1.029 . 11.0292 2 P0  11.029 2 3 P0 , or 14.10;
and so on. In general, after t years, the population of Florida is
P1t2  P0 . 11.0292 t  12.9411.0292 t.
This equation is called an exponential growth function with base 1.029. The
name exponential is used because the independent variable (in this case, t) occurs in
the exponent. The base (in this case, 1.029) is called the growth factor. It gives the
population each year as 1.029 times the population in the preceding year. The quan-
tity 0.029  2.9% is the associated annual growth rate. Note the relationship be-
tween the growth factor and the growth rate.

Growth factor  1  growth rate

In this formula, you must write the growth rate as a decimal, not as a percent. For
instance, if the growth rate for a process is 4%  0.04 each year, the associated
growth factor is 1  0.04  1.04.
Assuming that Florida’s population continues to grow with the same exponen-
tial pattern for the next 80 years, we can graph this population function as shown
in Figure 2.19. The function obviously is increasing. Moreover, the graph grows
faster and faster as time goes on, so the curve is concave up. This behavior is typi-
cal of an exponential growth function. Compare this function’s behavior with that
of an increasing linear function. Because a linear function grows at the same rate at
every point, its graph is a line. However, exponential growth functions such as this
one are curves that may seem to climb slowly at first but eventually climb extreme-
ly rapidly. This type of behavior explains why there is widespread concern about
P

140

120

100

80

60

40 P(t) = 12.94(1.029) t

20
t
FIGURE 2.19 0 20 40 60 80
Gord.3896.02.pgs 4/24/03 9:25 AM Page 73

2.4 Exponential Growth Functions 73

the exponential growth of the world’s population: Eventually there won’t be

enough land, food, and water to sustain everyone.
The graph shown in Figure 2.19 is only an approximation to the actual graph
of Florida’s population. We can’t have a fraction of a person, so the graph theoreti-
cally should be jagged with small steps up or down each time someone is born,
dies, or moves into or out of Florida. However, on a scale of millions of people,
such changes are insignificant, and our smooth curve actually is a good approxi-
mation to the population.
We summarize the formula for an exponential growth function and its param-
eters as follows.

Formula for an Exponential Growth Function

P is an exponential growth function of t with base c  1, if
P1t2  P0ct,

where P0 is the initial quantity (when t  0) and c is the growth factor by

which P changes when t increases by 1 unit.
Because c  1, we can write c  1  a, where a is the growth rate written as
a decimal.

For example, if a quantity (e.g., the balance in your bank account) is growing at
5% per year, the growth rate a  5%  0.05 and the associated growth factor is
c  1  a  1.05. The corresponding formula for the balance B in the account as a
function of time t is
B1t2  B0 . 11.052 t,
where B0 represents the initial or starting balance.
The growth factor c in any exponential growth function y  P0ct plays a role
similar to the slope in a linear function. The larger the growth factor c, the faster
the exponential function grows. Figure 2.20 shows a series of exponential growth
curves, with growth rates a of 3%, 4%, and 5% and corresponding growth factors c
of 1.03, 1.04, and 1.05. All start with the same initial value at time t  0, but the
larger the growth factor, the faster the curve grows. A curve corresponding to a
growth factor of 1.033, say, would lie between the curves for the growth factors of
1.03 and 1.04; the curve corresponding to a growth rate of 2.6% would lie below
the lowest of the three curves.
B(t)

5%

4%

3%

t
FIGURE 2.20 0
Gord.3896.02.pgs 4/24/03 9:25 AM Page 74

74 CHAPTER 2 Families of Functions

The coefficient P0 in an exponential function y  P0ct plays a role similar to the

vertical intercept for a line. If we set t  0, we get
y  P0c0  P0
because any number (other than 0) raised to the zero power is 1. Thus the initial
or starting value P0 of the exponential growth function represents the height at
which the exponential function crosses the vertical axis (when t  0). Figure 2.21
shows the graphs of three different exponential growth curves: y  1011.032 t,
y  3011.03 2 t, and y  6011.032 t. All have the same growth factor, 1.03, and so the
same growth rate, 0.03  3%, but all have different initial values. Note how all
three curves have similar shape, but each crosses the vertical axis (at t  0) at a dif-
ferent height.
y

60

30
10
t
FIGURE 2.21 0

Comparing Linear and Exponential Growth

The key fact about linear growth is that a linear function grows at a constant rate.
That is, every time the independent variable x, say, increases by 1, the linear function
grows by the same amount (equal to the slope), as illustrated in Figure 2.22(a). In
contrast, the key fact about exponential growth is that an exponential function grows
by a fixed percentage. Suppose that the growth factor is 1.20 so that the growth rate is
20%. Then every time the independent variable x increases by 1, the exponential
function grows by the same multiple, 1.2, as illustrated in Figure 2.22(b). The corre-
sponding values are then y0 , 1.2 y0 , 11.2 2 2 y0 , 11.22 3 y0 , and so on, and these values
eventually grow very rapidly as the exponent increases. As a result, eventually any ex-
ponential growth function will outstrip any linear function.
y y

1
m 1

1
m
1
1
m
1
1
m 1
1
1 1
x x
FIGURE 2.22 (a) (b)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 75

2.4 Exponential Growth Functions 75

Applications of Exponential Growth

In Examples 1 and 2 we apply the ideas and definitions on exponential growth.
E XAMPLE 1
In 1995, the population of Mexico was 93.7 million and growing at a rate of 2.2% a year.
a. Find a formula for the population of Mexico at any time t.
b. Predict the population of Mexico in 2003.
Solution

a. Because the annual growth rate for Mexico is 2.2%  0.022  a, the corresponding
growth factor is c  1  0.022  1.022. Let t be the number of years since 1995 and
M1t2 be the population of Mexico in millions. Then a formula for the Mexican pop-
ulation at any time t since 1995 is
M1t2  93.711.0222 t.
b. Assuming that this exponential growth pattern continues until 2003, we have t  8
years after 1995. We predict that the population of Mexico will be
M182  93.711.0222 8  111.52 million people,
as shown in Figure 2.23.
M(t)

111.52

93.7

t
FIGURE 2.23 0 8

◆
E XAMPLE 2
During one of New York City’s recent financial crises, someone discovered a million dol-
lar loan the city made to the U.S. Government in 1812. At first it appeared that the loan
had not been repaid. For a 6% annual compound interest rate, what would this amount
have become by the year 2000?
Solution The 6% growth rate corresponds to a growth factor of c  1.06 so that t years
after 1812, the amount would be
b1t2  b0 . 11.06 2 t  1,000,00011.06 2 t.
For 2000, t  2000  1812  188, and the resulting balance would be
b11882  1,000,00011.06 2 188
 $57,214,047,000.
As depicted in Figure 2.24, that would easily have solved the municipal finance problem
for many years to come. Unfortunately for New York City, the loan was later found to
have been repaid, with interest, in 1815.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 76

76 CHAPTER 2 Families of Functions

Interest on New York City's Loan?

50,000

Balance (millions of dollars)

40,000

30,000

20,000

10,000

0
1800 1825 1850 1875 1900 1925 1950 1975 2000
FIGURE 2.24 Year
◆
Think About This How much interest did New York City receive in 1815 from this loan? ❐

Doubling Time
One of the special characteristics of any exponential growth function is that it has
a unique doubling time—the time needed for the exponential function to double.
We illustrate this concept in Example 3.

E XAMPLE 3
Assuming that the exponential model P1t2  12.9411.0292 t for Florida’s population cre-
ated at the beginning of this section continues to hold far into the future, estimate the
population of Florida in (a) 2014, when t  24; (b) 2038, when t  48; and (c) 2062,
when t  72.
Solution We use exponential growth model
P1t2  12.9411.0292 t
to predict the following values.
a. P124 2  12.9411.0292 24  25.70  2 . 12.94
b. P148 2  12.9411.0292 48  51.04  4 . 12.94
c. P1722  12.9411.0292 72  101.35  8 . 12.94
◆
Let’s look at what these predicted population values indicate. After 24 years,
Florida’s population has doubled. After roughly another 24 years (i.e., t  48), it has
doubled again. After roughly another 24 years (i.e., t  72), the population has dou-
bled yet again. Therefore we say that the doubling time of Florida’s population is
about 24 years: If you take the population in any given year and compare it to the
population 24 years later, you will find that it has doubled.

Think About This To extrapolate far into the future, we must assume that the population continues to
grow exponentially at the same rate of 2.9% per year. The farther we project into the
future, the riskier our prediction becomes because other factors can affect the growth
rate. What are some? ❐
Gord.3896.02.pgs 4/24/03 9:25 AM Page 77

2.4 Exponential Growth Functions 77

Every population that grows exponentially has a fixed doubling time that de-
pends only on the growth rate or the growth factor, not on the size of the popula-
tion. The world’s population, with an annual growth rate of about 1.5%, has a
doubling time of about 38 years. (We show how to calculate doubling times later.)
The current population is about 6 billion, so there will be about 12 billion people
in 38 years and roughly 24 billion people in 76 years, all competing for an ever di-
minishing amount of resources. As another way of looking at it, if you live to be
76, the world’s population will quadruple during your lifetime.
y

2y

y doubles

FIGURE 2.25 Change in t is doubling time

The doubling time T for any exponential growth process is the same at any
point on the curve; that is, if you pick any point 1t, y2 on the exponential curve, the
value for y will always increase to 2y (it has doubled) after T time units. You can vi-
sualize what this means by looking at Figure 2.25.

Predicting with Exponential Growth Functions

The purpose of creating an expression for an exponential growth function is to an-
swer predictive questions about the quantity being modeled, as we illustrate in Ex-
amples 4 and 5.
E XAMPLE 4
Estimate when the population of Florida will reach 20 million.
Solution Our formula for the population of Florida is P1t2  12.9411.029 2 t, and we want
the value of t when the curve reaches a height of 20. We therefore must solve the equation
12.9411.0292 t  20
for t. We can solve this equation numerically by using trial-and-error by substituting dif-
ferent values of t until we get a value for 12.9411.0292 t that is very close to 20. (Your cal-
culator may have a table feature that allows you to generate a table of values and zoom in
with smaller and smaller steps to find the value of the independent variable that pro-
duces a given value—20 in this case—for the dependent variable.)
A simpler approach is to solve the equation graphically by drawing the graph of the
function on a function grapher and tracing along the curve to determine when the func-
tion reaches a height of 20, as shown in Figure 2.26. If necessary, this approach could
also involve zooming in to increase the level of accuracy. Alternatively, we could graph
the two functions, y  12.9411.0292 t and y  20, and find the point of intersection
using a function grapher. Whichever way we proceed, the solution is approximately
t  15.2 years from 1990 or early in 2005. (We develop an algebraic approach using log-
arithms for solving such an equation that yields an exact answer in Section 2.8.)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 78

78 CHAPTER 2 Families of Functions

P(t)

20

12.94

t
FIGURE 2.26 0 15.2

E XAMPLE 5
Estimate the doubling time for the population of Florida.
Solution We again use the exponential growth model P1t2  12.9411.029 2 t and we
now must find how long it takes for the population to double, that is, to reach
2  12.94  25.88. We therefore have to solve the equation
12.9411.029 2 t  2  12.94  25.88
for t. We solve this equation graphically by looking for the intersection of the two curves
y  12.9411.0292 t and y  25.88, as shown in Figure 2.27. This point is at t  24.2465,
so the doubling time for Florida’s population is about 24 14 years.
P(t)

25.88

12.94

t
FIGURE 2.27 0 24.26

◆
So far, we have thought of exponential functions as starting at time t  0. In
reality, the formula for any exponential function can be interpreted for negative
values of the independent variable, as we demonstrate in Example 6.

E XAMPLE 6
Use the model we constructed for the population of Florida to predict what the popula-
tion was in 1980.
Solution Our formula for the population of Florida is
P1t2  12.9411.0292 t,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 79

2.4 Exponential Growth Functions 79

where t is the number of years since 1990. The year 1980, 10 years before 1990, therefore
corresponds to t  10. The formula predicts that the population in 1980 was
P110 2  12.9411.0292 10  12.9410.751352  9.72 million people,
as illustrated in Figure 2.28. Note that this result is considerably below the 1990 popula-
tion value of 12.94 million people, as we would expect.

P(t)

12.94

9.72

t
FIGURE 2.28 –10 0

◆
Just as the domain of a linear function theoretically is the set of all real num-
bers from 
to 
, the domain of an exponential function is likewise theoreti-
cally from 
to 
. Of course, in any real-world setting, there may be practical
limitations to the domain. For instance, it wouldn’t make sense to use the function
to extrapolate the population of Florida 200 years into the past, as Florida became
a state only in 1845. Moreover, as we’ve stated before, extrapolating far into the fu-
ture or the past is risky because the trend in the data may not hold.
Also, Example 6 indicates that, when t 0, the values for the exponential
growth function continue to decrease from right to left. Figure 2.29 shows the typ-
ical graph of an exponential growth function y  kc x, with k  0. Note how it
grows in the expected way toward the right and decays to 0 toward the left. The rea-
son is that, as we move farther to the left of the vertical axis, the values of x become
ever more negative. Suppose that we write x  z. Recall one of the basic proper-
ties of exponents:
1
bz  .
bz
y

y = kc x

x
FIGURE 2.29 0
Gord.3896.02.pgs 4/24/03 9:25 AM Page 80

80 CHAPTER 2 Families of Functions

Consequently,
1
1.02910   0.75135.
1.02910
As the exponent becomes ever more negative,
1.02920  0.56454,
1.02930  0.42417,
1.029100  0.05734,
the values become ever smaller and eventually approach 0. We say that the curve
approaches the negative x-axis asymptotically because it never reaches 0 in any fi-
nite time interval. We call the horizontal axis a horizontal asymptote for the graph
of the exponential decay function. The range of any exponential growth function
y  kc x, with k  0, is therefore all positive values for y.

Finding an Exponential Function Through Two Points

We know that two points determine a line (because one and only one line can pass
through the two points). Similarly, two points also determine an exponential func-
tion in the sense that one and only one exponential function passes through the
two points, provided that the y-values for the points are either both positive or
both negative. Suppose that we have any two points 1x1 , y1 2 and 1x2 , y2 2 , where
x1 x2 and 0 y1 y2 , so the second point is to the right of and above the first
point and both points are above the x-axis, as shown in Figure 2.30. One and only
one exponential growth curve passes through the two points.
y

y2

y1
x
FIGURE 2.30 0 x1 x2

Think About This By drawing several sketches, convince yourself why it is not possible to draw an ex-
ponential growth curve through two points when one is above the x-axis and the
other is below the x-axis. Also, if the two points are both below the x-axis, what
should you expect about the sign of the coefficient k in y  kc x? ❐

We now determine a formula for the exponential function that passes through
two points. Doing so also gives us a way to find the growth rate for any exponential
process. For the equation of an exponential function y  kc x, values for the two
parameters k and c must be determined, which is why we use two points. We
demonstrate how to do so in Example 7.
E XAMPLE 7
The number of cell phones in use worldwide grew from 11 million in 1990 to 319 mil-
lion in 1998.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 81

2.4 Exponential Growth Functions 81

a. Assuming that the growth pattern was exponential, find the annual growth rate for
the number of cell phones in use and the equation of the exponential function that
models the number of cell phones in use.
b. Predict the number of cell phones in use in 2003.
Solution

a. Let t represent the number of years since 1990 and P the number of cell phones (in mil-
lions) in use. We then have the two points 10, 112 and 18, 319 2. The exponential growth
function has the form
P1t2  P0ct,
where the constants P0 and c must be determined. Substituting the coordinates of the
point 10, 11 2 into the function gives
P10 2  P0c0  P0  11,
because c0  1. Thus the exponential function becomes P1t2  11ct. Using the point
18, 3192 gives
P18 2  11c8  319.
Solving for c8 gives
319
c8   29.
11
Just as we solve x 2  10 for x by taking the square root of 10 or solve x 3  10 for x by
taking the cube root of 10, we solve c8  29 for c by taking the eighth root of 29. (We
discuss the details more formally in Section 2.7.) Thus
8
c 2 29  1.5234.
8
(Verify that 1 29  1.5234 by taking the eighth power of 1.5234.) For the growth
factor of 1.5234, the annual growth rate in the number of cell phones in use is
0.5234  52.34%. Moreover, the exponential function that models the growth in the
number of cell phones is
P1t2  1111.52342 t,
where t is the number of years since 1990.
b. Because 2003 is 13 years after 1990, we set t  13 as shown in Figure 2.31. We then
use this exponential model to predict that the number of cell phones in use in 2003 is
P113 2  1111.52342 13  2618.0 million,
or about 2.618 billion.
P(t)

2618.0

11
t
0 13
FIGURE 2.31
◆
Gord.3896.02.pgs 4/24/03 9:25 AM Page 82

82 CHAPTER 2 Families of Functions

By letting t represent the number of years since 1990, we simplified the work in
Example 7 to give the vertical intercept 10, 112 as one of the points. If we can’t do
so, things become more complicated, as shown in Example 8.
E XAMPLE 8
Find the equation of the exponential function that passes through the points 11, 6 2
and 12, 92.
Solution The desired exponential function has the form f 1x 2  kc x, where we must
find the correct values for the parameters k and c. Using the point 11, 62, we have
f 11 2  kc1  kc  6.
Using the point 12, 92 , we have
f 12 2  kc2  9.
From the first of these two equations, we solve for k and get k  6>c. We substitute this
term into the second equation to get

kc2  a b c2  6c  9,
6
c
and so
9
c  1.5.
6
Therefore
6 6
k   4,
c 1.5
and the desired exponential function is f 1x 2  411.5 2 x, as shown in Figure 2.32.

20

15
y = 4(1.5) x
10

x
FIGURE 2.32 0 1 2 3

Determining Whether a Set of Data Is Exponential

Recall the simple criterion that determines whether a set of data follow a linear
pattern: The successive differences in the dependent variable must be constant
when there is a constant difference between values of the independent variable.
Similarly, we can determine whether a table of data values 1t, y2 follows an expo-
nential pattern by looking at the successive ratios of the y values.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 83

2.4 Exponential Growth Functions 83

If the ratios of the successive values of the dependent variable are constant for
equally spaced t values, the y values follow an exponential pattern: y  k c t.

The common ratio is precisely the growth factor for the exponential growth
process if the t values increase by 1 unit. For instance, with Florida’s population
values from one year to the next, we found that the common ratio was 1.029,
which is the growth factor, and that the associated growth rate is 0.029, or 2.9%
per year.

E XAMPLE 9
One of the following functions is exponential and the other isn’t. Determine which is the
exponential function. The values are rounded to four decimal places.

x y x y
0 20.0 0 20.0
1 21.0 1 21.0

2 22.10 2 22.05

3 23.2775 3 23.1525

4 24.6425 4 24.3101

5 26.2650 5 25.5256

Solution We apply the criterion for an exponential pattern and examine the ratios of
successive terms for each function. For the first function the ratios are
21.0 22.10 23.2775
 1.05,  1.0524,  1.0533,
20.0 21.0 22.10
24.6425 26.2650
 1.0586, and  1.0658.
23.2775 24.6425
The successive ratios are not constant, so this function cannot be exponential.
For the second function the ratios are
21.0 22.05 23.1525
 1.05,  1.05,  1.05,
20.0 21.0 22.05
24.3101 26.5256
 1.04999, and  1.04999.
23.1525 24.3101
These ratios are essentially constant (the last two vary slightly because the entries listed
in the table were rounded), so we conclude that this function is indeed exponential.
◆
Rules for Exponents
Because exponential functions involve working with exponents, all the usual alge-
braic rules for manipulating exponents apply. As a reminder, we list some of the
fundamental definitions and algebraic rules for exponents.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 84

84 CHAPTER 2 Families of Functions

Definitions and Rules for Exponents

Property Example
1. a x . ay  a xy 10 . 102  110 . 10 . 10 2 . 110 . 10 2 
3

105  1032
ax 105 10 . 10 . 10 . 10 . 10
2.  axy, a  0   103  1052
ay 102 10 . 10
1ax 2 y  axy 1103 2 2  103 . 103  106  103 2
.
3.
4. a0  1 100  1
1 1
5. a1  , a  0 101 
a 10
1 1 1
6. an  n , a  0 103  3 
a 10 1000
n 3
7. a1>n  2a , a  0 101>2  210 , 101>3  2 10

Problems
1. The accompanying graph shows population growth c.
curves for four different nations. Which nation t 0 10 20 30
a. has the greatest growth rate? Q(t) 200 208 216.32 224.97
b. has the smallest growth rate?
3. Anne opens a bank account with $1200 at 4% annu-
c. has the largest initial population?
al interest. Bill opens an account with $1000 at 4.5%
d. has the smallest initial population?
annual interest. Christine opens an account with
e. Which nations have the same growth rate?
$1500 at 3.8% annual interest. Doug opens an ac-
P count with $1200 at 4.5% annual interest. Elka
A
B opens an account with $1300 at 4.25% annual inter-
C est. Sketch a graph showing the balances in the five
D accounts over time on the same set of axes. Be sure
to label which account belongs to which person.
4. Use the exponential growth function f 1t2 
12511.042 t to make a prediction for 2000 if (a) t is
the number of years since 1980, (b) t is the number
t of years since 1900, (c) t is the number of years
since the year 0.
2. Determine which of the functions are exponential. 5. In 1990, the United States imported $495 billion
For any exponential function, find the equation of the worth of goods. In 1998, the United States import-
function and use it to predict the next entry to extend ed $912 billion worth of goods. Assuming that the
the table of values. growth in imports has been following an exponen-
tial growth pattern, find an equation of the expo-
a.
x 0 1 2 3 nential function that models U.S. imports when
a. the independent variable t represents the num-
y 1000 1200 1440 1728 ber of years since 1990.
b. the independent variable t represents the num-
b.
t 0 1 2 3 ber years since 1900.
c. the independent variable t represents the num-
L(t) 300 308 320.2 335.5 ber of years since the year 0.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 85

2.4 Exponential Growth Functions 85

d. For the three functions you created in parts a. Find an expression for the population at any
(a)–(c), which parameters changed and which re- time t.
mained the same? Explain why the changes oc- b. What will be the population be in 2005?
curred. Explain why the parameters that stayed c. Estimate the doubling time for this population.
the same didn’t change. 10. The 1995 population of France was 58.1 million
e. Use each model from (a)–(c) to predict the and growing at an annual rate of 0.3%.
amount of imports in 2005.
a. Find an expression for the population at any
Source: 2000 Statistical Abstract of the United States. time t.
6. Match each formula with the corresponding table b. What will be the population in 2010?
of values. c. Estimate the doubling time for this population.
a. y  a11.1 2 s 11. In 1990, 1.36 billion metric tons of carbon dioxide
b. y  b11.05 2 s were emitted into the atmosphere in the United
c. y  c11.03 2 s States. In 1998, 1.595 billion metric tons were emitted.
a. Construct the exponential function giving the
i. amount of carbon dioxide emitted into the at-
s 2 3 4 5 6
mosphere as a function of the number of years
f (s) 1.06 1.09 1.13 1.16 1.19 since 1990.
b. Use the exponential function to estimate the
ii. amount emitted in 2004.
s 1 2 3 4 5 12. The population graph shown in the accompanying
g(s) 2.20 2.42 2.66 2.93 3.22 figure is growing exponentially.

P (thousands)
iii.
s 3 4 5 6 7 800

h(s) 3.47 3.65 3.83 4.02 4.22 700

600

7. In 1980, a total of $119 trillion was spent on food 500

and drinks in the United States. In 1994, the total 400
spent was $274 trillion. 300
a. Find the equation of the exponential function 200
that can be used to model the total spent on food 100
and drinks in the United States as a function of t
0 1 2 3 4 5 6 7
the number of years since 1980.
b. Use your model to predict the amount spent in
1990. a. Use the graph to estimate the doubling time of
c. What is your prediction for the total sales of the population.
food and drink in 2004? b. Verify graphically that the doubling time does
d. Estimate when the total sales will reach $500 tril- not depend on where you start on the graph.
lion if this exponential trend continues. 13. The world’s population passed 6 billion in late 1999
8. The 1990 population of Arizona was 3.7 million and is increasing at a rate of about 1.5% per year.
and growing at an annual rate of 1.7%. a. Find the world’s population 15 years later if this
a. Find an expression for the population at any trend continues.
time t. b. Estimate how long it will take the world’s popu-
b. What will be the population in 2005? lation to double.
c. Estimate the doubling time for this population. 14. Find a formula for the balance in a bank account in
9. The 1995 population of Venezuela was 21.8 million which $100 was deposited at 6% annual interest
and growing at an annual rate of 2.6%. compounded for 10 years.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 86

86 CHAPTER 2 Families of Functions

15. Find the balance after 1 year if $100 is deposited at tween 1995 and 2000. What would you predict as
an annual rate of 6% compounded quarterly in- the value for the Dow at the beginning of 2004?
stead of yearly. What is the balance after 10 years? 21. Repeat Problem 20, using the facts that the Dow
(Hint: What is the interest rate for each 3-month was 964 at the beginning of 1981 and was 11,358 at
period?) the beginning of 2000.
16. In 1998, the population of the United States was 22. a. Suppose that you’re an aggressive stockbroker
about 268.2 million with an annual growth rate of who is trying to convince a little old lady to in-
0.7%. At the same time, the population of Mexico vest her life savings with you. What argument
was about 100.1 million with an annual growth rate would you make based on your work on either
of 2.2%. If these growth rates continue, use either Problem 20 or 21 to convince her.
graphical or numerical methods to estimate when b. Now suppose that the little old lady is your
the population of Mexico will overtake that of the grandmother. What argument would you make
United States. based on your work on either Problem 20 or 21
17. According to an article in the New York Times on to convince her to be more conservative.
May 27, 1990, a wealthy Pennsylvania merchant 23. An exponential function f is such that f 10 2  512
named Jacob DeHaven loaned $450,000 to the and f 14 2  1250. Which of the values are possible
Continental Congress in 1776 to rescue the troops and which are impossible?
at Valley Forge. The descendants of Mr. DeHaven a. f 12 2  800 b. f 12 2  881 c. f 12 2  981
sued the U.S. government for what they believed
24. The net income of the Acme Company was $240 mil-
they were owed. The interest rate in effect in 1776
lion in 1990 and has been increasing at an annual
was 6% per year. How much did the family stand to
rate of 10% per year since. Over the same period,
collect in 1991, assuming that interest is com-
the net income of its chief competitor, the Finest
pounded annually?
Corporation, has been growing 8% annually from
18. The lily pads in a pond grow in such a way as to an income of $300 million in 1990. Which was the
double the area of the pond that they cover daily. richer company in 2000? Does Acme ever surpass
a. If the lily pads exactly cover the entire pond on Finest? If so, estimate when.
the 25th day, how much of the pond do they 25. (Extension of Problem 24) Suppose that Finest
cover on the 24th day? grew by a fixed amount of $25 million per year
b. Write an exponential function that models the since 1990 while Acme grew exponentially at an an-
fraction of the pond covered on any particular day. nual rate of 10%. By using trial and error, estimate
c. If the area of the pond is 40,000 sq ft, find the when Acme surpassed Finest.
area covered by the lily pads on the initial day. 26. When Steven was 5 years old, his grandmother de-
d. What area of the pond is covered by the lily pads cided to set up a trust account to pay for his college
at the end of 1 week? education. She wanted the account to grow to
19. Let f 1x 2 be an exponential function of x. If $80,000 by Steven’s 18th birthday. If she was able to
f 17 2  25.6 and f 182  28.8, find invest her money at 6% per year, how much did she
a. the growth factor; have to put into this trust account? (Note: This
b. the growth rate; amount is known as the present value of the invest-
c. the value of the function when x  10; ment. The $80,000 is known as the future value.)
d. a formula for f 1x2. 27. In Example 7 the number of cell phones in use in-
20. The Dow-Jones average of 30 industrial stocks is creased 29-fold, from 11 million in 1990 to 319 mil-
the most famous measure of performance of the lion in 1998. This is equivalent to a 3000% increase
New York Stock Exchange. At the beginning of 1995 over that 8-year period. Explain what’s wrong with
the Dow was 3834, and at the beginning of 2000 it the reasoning that says: If the number of cell phones
was 11,358. Assuming (incorrectly) that the Dow increased by 3000% over the 8 years, the annual
increased continuously over these 5 years and that growth rate is 81 of 3000% or 375%.
the pattern is exponential, find the exponential x5
28. Show that x 5>3  3
function that models the behavior of the Dow be- x
Gord.3896.02.pgs 4/24/03 9:25 AM Page 87

2.5 Exponential Decay Functions 87

a. numerically, by finding at least one value of x for b. graphically, by comparing the graphs of the two
which the two expressions are different; functions y  x 5>3 and y  x 5>x 3.

Exercising Your Algebra Skills

Simplify the following. 11. x 1>2x 3>4 12. y2>3y4>3
1. x 5 . x 3 x . x2 14. 1x 5 2 3
4
2. 13. z2>3z5>3
3. a8 . a4 4.
a15 15. 1x 3 2 5 16. 1a8 2 4
a6 Perform the following operations:
5. x 5 . x 3 6. a5 . a3 17. 1a3b5 2 4 18. 1a3  b2 2
r8
7. 4 8.
b15 19. 1a3  b2 2
r b6
w4 w7
9. 7 10.
w w4

2.5 Exponential Decay Functions

Prozac is one of the most widely used drugs to treat extreme depression. Once a med-
ication such as Prozac has been absorbed into the bloodstream, it eventually is elimi-
nated from the body by the kidneys, which purify the blood by filtering out foreign
chemicals. For now, let’s assume that a person takes a single dose of Prozac and that it
has been completely absorbed into the blood. It is reasonable to assume that, during
any fixed time period, a fixed percentage of any medication, including Prozac, is re-
moved from the bloodstream as the kidneys process the blood. In particular, the kid-
neys eliminate approximately one-fourth of the Prozac in the bloodstream during
any 24-hour period, so that 75% of the drug remains. (Note that this rate is specific to
Prozac and that other medications are washed out of the body at different rates.)
Suppose that the original dosage of Prozac is 80 mg (milligrams). We want
to develop a formula for the amount D1t2 present at any time t. Clearly, it must
be a decreasing function because the level of the drug in the bloodstream is de-
caying over time.
We start with D10 2  80 mg. After the first 24 hours, one quarter of 80 mg, or
20 mg, of the Prozac is eliminated, leaving three quarters of the 80 mg, or 60 mg, of
the Prozac in the bloodstream. After one 24-hour period, when t  1, the amount
of Prozac in the system is
D 11 2  0.75180 2  60 mg.
After a second 24-hour period, the kidneys remove 25% of the remaining 60 mg of
Prozac, so 15 mg are eliminated, leaving 75% of the remaining 60 mg of Prozac.
Thus, when t  2,
D 12 2  0.75160 2  0.7510.752 180 2  10.75 2 2 180 2 mg.
After the third 24-hour period, 25% of the remaining Prozac is eliminated, leaving
D 13 2  0.75D 12 2  0.7510.752 2 180 2  10.752 3 180 2.
Similarly,
D 142  0.75D 13 2  10.752 4 180 2
Gord.3896.02.pgs 4/24/03 9:25 AM Page 88

88 CHAPTER 2 Families of Functions

and
D 15 2  0.75D 14 2  10.752 5 1802 .
In general, after t days the amount of Prozac in the bloodstream is given by
D 1t2  8010.752 t.
This function has the same form, y  kct, as the exponential growth functions
presented in Section 2.4 except that the base c is 0.75, which is less than 1. It is an ex-
ample of an exponential decay function, and its graph is shown in Figure 2.33. Note
that the behavior is that of a decreasing, concave up function. Each step down is small-
er than the previous one. This result makes sense because, as the amount of Prozac re-
maining in the bloodstream gets smaller, there is less of the drug left to eliminate, and
the amount of decrease in drug strength diminishes every successive day.
D
D(t)
80
70
60

Drug level
50
D(t) = 80(0.75) t
40
30
20
10
t t
FIGURE 2.33 0 FIGURE 2.34 0 1 2 3 4 5

You can see this numerically by calculating the values of the function previ-
ously given:
D 10 2  80, D 11 2  60, D 12 2  45, D 13 2  33.75, D 14 2  25.3125, . . . ,
which is a decreasing, concave up pattern. If you continue these calculations, you will
find that the values eventually approach 0 asymptotically; that is, the drug level never
reaches 0 in any finite time interval, as illustrated in Figure 2.34. Thus, the horizontal
axis is a horizontal asymptote for the graph of the exponential decay function.
In general, the graph of any exponential decay function, y  kct, with
0 c 1, is a decreasing, concave up curve that approaches 0 as t gets larger and
larger. In comparison, the graph of any exponential growth function, y  kc t, with
c  1, is an increasing, concave up curve. Because the base c for an exponential
decay function is between 0 and 1, we call it the decay factor.
Often, we are told that a process is decaying at a given rate—say, 12% per year.
The 12%  0.12 is known as the decay rate and the associated decay factor c is
Decay factor  1  decay rate,
where the decay rate must be written as a decimal. Thus
c  1  0.12  0.88,
because 88% (or 0.88) of the original amount is left. By comparison, for exponen-
tial growth, recall that
Growth factor  1  growth rate.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 89

2.5 Exponential Decay Functions 89

Note that whether we have an exponential decay function such as y  8010.752 t

for the level of Prozac or an exponential growth function such as y  12.9411.0292 t
for the population of Florida, it is still an exponential function and the same techniques
that we introduced in Section 2.4 apply. The only difference is that, for an exponential
growth function, c  1, whereas for an exponential decay function, 0 c 1.
E XAMPLE 1
Find the amount of Prozac in the bloodstream after 1 week.
Solution We use the formula for the exponential decay function,
D 1t2  8010.752 t,
we previously constructed. After 1 week, t  7 days, so the level of Prozac will be
D 17 2  8010.752 7  10.679,
or about 10 23 mg.
◆
E XAMPLE 2
Estimate how long it takes until the level of Prozac in the bloodstream drops to 2 mg.
Solution Using the formula for the level of Prozac, we have to find t so that
D 1t2  8010.752 t  2.
Using either numerical or graphical methods, we find that t  12.8 days, as shown in
Figure 2.35.
D(t)

80

2 t
FIGURE 2.35 0 12.8

◆
We summarize the formula for an exponential decay function and its parame-
ters as follows.

Formula for an Exponential Decay Function

P is an exponential decay function of t with base c, 0 c 1, if
P1t2  P0ct,
where P0 is the initial quantity (when t  0) and c is the decay factor by
which P changes when t increases by 1 unit. Because 0 c 1, we write
c  1  a, where a is the decay rate, written as a decimal.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 90

90 CHAPTER 2 Families of Functions

The larger the decay rate a, and hence the smaller the decay factor c, the faster
the exponential decay function approaches 0, as illustrated in Figure 2.36.

c = 0.9

c = 0.8
c = 0.7
t
FIGURE 2.36 0

For example, if a quantity is decreasing at the rate of 12% per hour (e.g., the ef-
fectiveness of a medication in the body), the decay rate is a  0.12 and the decay fac-
tor is c  1  a  0.88. This reflects the fact that, if 12% of the quantity is removed
each hour, then 88% of the quantity remains at the end of the hour. The correspon-
ding formula for the exponential decay function that models the quantity Q is
Q1t2  Q0 . 10.88 2 t,
where Q0 is the initial amount of the quantity at time t  0.

Half-life
Just as the doubling time for an exponential growth process is the time needed for
the quantity to double, the half-life for an exponential decay process is the time T
needed for the quantity to be reduced by half. You can visualize what this means by
looking at Figure 2.37.

y
y halves
1
2
y

FIGURE 2.37 Change in t is half-life

Note that the half-life T for any specific process is the same at any quantity
level; no matter which point 1t, y2 you select, the quantity will decrease to 12 y after
T time units.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 91

2.5 Exponential Decay Functions 91

E XAMPLE 3
Estimate the half-life of Prozac in the bloodstream following an 80 mg dose.
Solution The exponential decay function that models the amount of Prozac in the blood-
stream is
D 1t2  8010.752 t.
We want to find the time t needed for this level to drop to 1 12 2 80  40 mg, so we must
solve the equation
8010.75 2 t  40.
Using either numerical or graphical methods, as shown in Figure 2.38, we get
t  2.4 days. Therefore, no matter what level of Prozac is in the blood at any specific
time, the level will be down by half about 2.4 days, or 58 hours, later.

D(t)

80

( 12 )80

t
FIGURE 2.38 0 2.4

Radioactive Decay
One of the characteristics of any radioactive substance, such as radium or urani-
um, is that it transforms, or decays, to some other element, often lead, as time pro-
gresses. This decay is accompanied by the release of energy, called radioactivity,
which can be detected and measured. More specifically, the rate at which an ele-
ment decays is distinctive for that element. That is, during any fixed length of time,
the same percentage of the mass of a radioactive element will decay. For instance,
over the course of any 100-year period, approximately 4.3% of any radium present
will decay to lead, leaving 95.7% of the radium at the end of 100 years, as illustrat-
ed in Figure 2.39. Thus, if someone had put aside R0  100 grams of radium in the

down 4.3%

down 4.3%
Quantity

down 4.3%

FIGURE 2.39 Time

Gord.3896.02.pgs 4/24/03 9:25 AM Page 92

92 CHAPTER 2 Families of Functions

year 1900, we would expect to find only R112  95.7 grams by the year 2000. By
the end of a second century, the amount of radium left would be
R12 2  0.957 R11 2  10.9572 2 R0
and by the end of a third century it would be
R13 2  0.957 R12 2  10.9572 3 R0 .
In general, the amount of radium present after t centuries is modeled by the
exponential decay function
R1t2  10.9572 t R0
for any t.
Alternatively, because 4.3%  0.043 is the decay rate for this exponential
decay process, the decay factor is 1  0.043  0.957. Thus, if the initial amount
of radium is R0 , we can use the general formula for an exponential decay func-
tion to get R1t2  10.9572 t R0 .
Figure 2.40 shows a graph of the amount of radium as a decaying exponen-
tial function of time. The amount of radium begins decreasing relatively rapidly,
then decreases more slowly, and eventually approaches the time axis as a hori-
zontal asymptote.

R0
Radium

FIGURE 2.40 0 Time

E XAMPLE 4
Estimate the half-life of radium.
Solution We want to determine the value of t for which

R1t2  10.9572 t R0  R0 .
1
2
We first divide both sides of this equation by R0 to obtain

10.957 2 t  .
1
2
If we now use either numerical or graphical methods, as shown in Figure 2.41, we find
that t  15.77 centuries. That is, the half-life for radium is approximately 1577 years.
(The actual value for its half-life is closer to 1590 years; our calculations were based on
the fact that approximately 4.3% of the radium decays to lead each century, and this
rounding produced an error.)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 93

2.5 Exponential Decay Functions 93

R(t)

R0

( 12 )R0

t
FIGURE 2.41 0 15.77

◆
Think About This What is the actual percentage of radium that decays into lead each year, based on
its half-life of 1590 years? ❐

In all of the examples so far, time is the independent variable. Example 5 illus-
trates a situation in which the independent variable in an exponential function
may represent some other quantity.

E XAMPLE 5
The strength of any signal in a fiber-optic cable, such as the type used for telephone and
other communication lines, diminishes 15% every 10 miles.
a. Find an expression for the strength of a signal remaining after a given number of 10-
mile lengths.
b. How much of the signal is left after 100 miles?
c. How far does a signal go until its strength is down to 1% of the original level?
Solution
a. If the signal diminishes by 15% every 10 miles of cable, after each 10-mile stretch,
only 85% of the original signal strength remains. Let S0 be the initial strength of some
signal and let S1n2 be the strength of the signal remaining after n 10-mile lengths.
Therefore, after the first 10-mile length of cable 1n  1 2, 85% of S0 is left, so
S11 2  0.85S0 .
Similarly, after the second 10-mile length 1n  22, 85% of S112 ,the signal strength
remaining after the first 10-mile length, is left. That is,
S12 2  0.85S11 2.
Continuing this pattern, we get
S10 2  S0 ,
S11 2  10.85 2S0 ,
S12 2  10.85 2S11 2  10.852 10.852 S0  10.85 2 2S0 ,
S13 2  10.85 2S12 2  10.852 10.852 2S0  10.852 3S0 ,
and so on. After n 10-mile lengths of a cable,
S1n 2  S0 . 10.85 2 n,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 94

94 CHAPTER 2 Families of Functions

which is an exponential decay function with decay factor c  0.85.

b. After 100 miles, or n  10 ten-mile lengths, the fraction of the original signal
strength remaining is
S110 2  S0 . 10.85 2 10  0.1969 S0,
so just under 20% of the original signal strength is left.
c. To find out how far the cable can go until only 1% of the signal strength is left, we
must find the value of n for which the strength remaining is 0.01 S0 , or
S1n2  S0 . 10.852 n  0.01S0 .
If we divide both sides of this equation by the initial signal strength S0 , we get
10.85 2 n  0.01.
Solving this equation either numerically or graphically, as shown in Figure 2.42, we
find that n  28. Therefore the signal deteriorates by 99% after about 28 ten-mile
lengths, or about 280 miles.

S(t)

S0

0.01S0
t
FIGURE 2.42 0 28

◆
Think About This In practice, this model suggests that fiber-optic signals need to be boosted if they
are to go any great distance. For instance, if a booster station can clearly detect a
signal at 1% of its original strength, such stations would have to be located every
280 miles. Suppose that the equipment used can clearly detect a signal at 0.1% of
its original level. How far apart would the booster stations have to be? ❐

Determining Whether a Set of Data Is Exponential Growth

or Exponential Decay
In Section 2.4, we presented a simple criterion for recognizing that a set of data fol-
lows an exponential growth pattern: The successive ratios of the values of the de-
pendent variable y are constant for equally spaced t values. The same criterion
applies if the values of y are decreasing in an exponential decay pattern. In this
case, the common ratio is precisely the decay factor for the process if the values of
t increase by 1 unit. In general, the ratio criterion works whether the data values are
increasing or decreasing. A common ratio greater than 1 gives the growth factor for
an exponential growth process; a common ratio less than 1 gives the decay factor
for an exponential decay process.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 95

2.5 Exponential Decay Functions 95

Finally, we consider some parallels between the family of linear functions and
the family of exponential functions. The general formula for a linear function is
y  mx  b, and the general formula for an exponential function is y  kc x, so
both are two-parameter families. For linear functions the more important param-
eter usually is the slope, and its sign determines whether the function increases or
decreases. For exponential functions the more important parameter is the growth
or decay factor c, and whether its value is greater than 1 or less than 1 determines
whether the exponential function increases or decreases.
The following problems include both exponential growth and exponential
decay situations because you need to learn to distinguish between them.
Problems
1. Determine which of the six functions could be expo- c.
nential functions of the form f 1x 2  kc x and which t 0 10 20 30
cannot be exponential. Explain your reasoning. Q(t) 400 288 207.36 149.30

(a) y (b) y 3. Which of the following pairs of points can deter-

mine an exponential function of the form y  Ac x
and which cannot. For those that can, sketch the
graph of the exponential function and indicate the
sign of A and whether the growth or decay factor c
x x
is greater than or less than 1.

(c) y (d) y
(a) y (b) y

x x x x

(e) y (f) y

(c) y (d) y

x x
x x
2. Determine which of the functions are exponential.
For any exponential function, find the equation of
the function and use it to predict the next entry to
extend the table of values. 4. Decide which situations represent exponential
a. growth, exponential decay, linear increase or de-
x 0 1 2 3 crease, or none of these patterns.
y 2000 1800 1620 1458 a. The value for a rare bottle of wine goes up $50
each year.
b. The value for a piece of sculpture increases 15%
b.
t 0 1 2 3 each year.
c. A 3-year labor contract calls for yearly increases
L1t 2 300 240 190 150 of $800.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 96

96 CHAPTER 2 Families of Functions

d. A 3-year labor contract calls for an increase of (c) y

5% the first year, 4% the second year, and 3% the

third year. (2, 12)
e. The value of a car drops by 40% each year.
f. The average cost of a home computer for the first- (1, 6)
time buyer has been dropping by $300 each year.
x
g. The number of new cases of a disease reported
over the last decade has been dropping by 12%
each year. 7. In 1980, about 27,700 cases of tuberculosis were
reported in the United States. In 1997, there were
5. The accompanying figure shows the graph of the
19,900 such cases. Source: U.S. Centers for Disease Con-
price of each of seven collectible toys as a function
trol and Prevention.
of time. Match each scenario with one of the graphs
and write a brief scenario for each of the remaining a. Write an exponential decay function that models
graphs. the number of reported cases of TB as a function
of time.
Price b. Predict the number of cases in 2004.
c. Estimate how long it will take for the number of
(ii) reported cases to drop to 10,000.
(i) 8. In 1940, there were 6,102,000 farms in the United
States. By 1997, the number of farms had dropped
(iii)
to 1,912,000.
(v) Source: 2000 Statistical Abstract of the United States.
a. Assuming that the pattern of decay is exponen-
tial, find the equation of a model that can be
(iv) used to predict the number of farms.
(vi)
(vii) b. Use your model to predict the number of farms
t
in 1980. How close is your prediction to the cor-
rect value of 2,440,000 farms?
a. The price of the toy increased by 10% each year. c. Predict the number of farms in 2005.
b. The price of the toy increased by 6% each year. d. If the trend continues, estimate when there will
c. The price of the toy dropped by $5 each year. be 1 million farms.
d. The price of the toy remained steady. e. Write a paragraph describing the long-term im-
6. Find possible equations for the exponential func- plications if this trend continues.
tions graphed in (a)–(c). 9. When a person smokes a cigarette, about 0.4 mg of
(a) y nicotine is absorbed into the blood. About 35% of
the nicotine is washed out of the blood every hour.
(2, 36) a. Find the equation of a function that models
the level of nicotine in the blood after a single
cigarette.
b. Use your model to estimate how long it takes for
(0, 4)
x the amount of nicotine in the blood to drop to
0.005 mg.
(b) y 10. The amount of the drug ampicillin (a form of peni-
cillin) in the bloodstream decreases by about 42%
(–1, 8) every hour.
a. If the dosage of ampicillin is 250 mg, write a
(1, 2) function that can be used to model the level of
x
ampicillin in the blood as a function of time, if
one dose is taken.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 97

2.5 Exponential Decay Functions 97

b. How much ampicillin is left in the blood after b. What percentage of the carbon-14 remains after
5 hours? a thousand years.
c. Estimate how long it will take for the level of 15. The filter in a swimming pool removes 30% of all
ampicillin to drop to 1 mg. impurities in the water every hour it operates.
11. A hospital patient is administered 3 mg of mor- a. Find an expression for the level of impurities left
phine to control his pain. About 31% of the mor- in the pool after n hours, if no further impurities
phine in the blood is washed out every hour. are added.
a. Construct a function that models the level of b. How much is left after 5 hours?
morphine in the blood after one dose. 16. Use the information in Example 5 to estimate the
b. How much morphine remains in the blood after half-life of a signal in a fiber-optic cable. What does
4 hours? it mean?
c. Estimate how long it will take for the amount of
17. One of the major problems associated with any organ
morphine left to drop to 0.2 mg.
transplant is the long-term risk of rejection, despite
12. The level of pollution in the Great Lakes is a major patients’ taking anti-rejection drugs for the rest of
concern to environmentalists. their lives. The percentage of individuals who have
a. In Lake Erie, about 38% of the pollutants are not rejected a transplanted organ can be modeled by
washed out each year if no pollutants are added. an exponential decay function as a function of time in
Write a function that models the level of pollu- years. According to one study, the half-life of kidney
tants in the lake as a function of time. transplants done in 1988 was 9.1 years; according to
b. How long will it take for 90% of the pollutants to another study, the half-life of kidney transplants done
be washed out of Lake Erie if no further pollu- in 1996 was projected to be 13.3 years. Is this later re-
tants are added? sult good or bad news? Explain your reasoning.
c. In Lake Superior, about 0.053% of the pollutants 18. According to a medical study, the half-life of kidney
are washed out each year if no further pollutants transplants was 13.3 years.
are added. Write a function to model the level of a. Write a formula for an exponential function that
pollutants in Lake Superior as a function of time. can be used to model the percentage of kidney
d. How long will it take for 90% of the pollutants to transplant recipients who haven’t rejected the
be washed out of Lake Superior if no further kidney as a function of time.
pollutants are added? b. What percentage of kidney transplant recipients
13. One of the major concerns about above-ground do you predict will still have their new kidneys
nuclear testing is that it produces strontium-90, a functioning after 10 years?
radioactive element whose half-life is 29 years and c. How long will it take until the percentage of kid-
which has worked its way into the food chain. That ney transplant recipients having their new kid-
is, strontium-90 from fallout is deposited on grass, neys will be down to 20%?
eaten by cows, carried into their milk, and eventual- 19. Treatments for different kinds of cancer are usually
ly finds its way onto the kitchen table. Suppose that, reported in terms of the percentage of patients who
as a result of a single nuclear explosion, the amount survive for 5 years after receiving the treatment, be
of strontium-90 in a particular valley exceeds it surgery, chemotherapy, or radiation therapy. The
health limits by a factor of 10. Estimate how long it percentage who survive can be modeled by a expo-
will take for the strontium-90 to decay to the safety nential decay function. The 5-year survival rate for
level. early stage malignant melanoma, a particularly se-
14. Carbon-14, a radioactive form of carbon, is used in vere type of skin cancer, is 80%.
the carbon-dating process to measure the age of ob- a. What percentage of patients having this treat-
jects. About 0.012% of the carbon-14 decays into ment will survive 10 years?
carbon-12 every century. b. Use the information given to write an exponen-
a. Write a function for the amount of carbon-14 tial decay function that models the percentage of
remaining in an object that originally contained patients treated for melanoma who survive any
C0 grams of carbon-14. given length of time t in years.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 98

98 CHAPTER 2 Families of Functions

c. What is the half-life for survival among patients c. Suppose that you are told that the amount of the
having this treatment? substance present after 30 days is 400 grams. Use
20. The 5-year survival rate for stage I lung cancer (the this information and the amount left after
mildest and earliest form) treated by surgery is 60% 10 days to estimate the amount present after
to 70%. 20 days. Is the actual value higher or lower than
your estimate? How do you know?
a. Use the middle value of 65% to write an expo-
d. How might you use the results from (b) and (c)
nential decay function that models the percent-
to come up with a better estimate of the amount
age of patients treated for stage I lung cancer
of radioactive material present after 20 days?
who survive any given length of time t in years.
b. What is the half-life for survival among patients 24. A certain radioactive isotope has a half-life of 20 days.
having this treatment? Suppose that 800 mg are present initially and con-
c. Repeat parts (a) and (b), using the lowest sur- sider a 60-day time period. Let r1 represent the aver-
vival rate of 60%. age daily rate of decrease of the isotope over the full
d. Repeat parts (a) and (b), using the highest sur- 60-day period, let r2 be the average daily rate of de-
vival rate of 70%. crease over the first 30-day period, and let r3 be the
21. In 1990, 442.2 million prerecorded cassette tapes and average daily rate of decrease over the last 30 days.
865.7 million CDs were sold in the United States. In List these three rates in increasing order without
1998, 158.5 million cassettes tapes and 1,124.3 mil- calculating their values.
lion CDs were sold. Assume for now that the patterns 25. The function shown in the accompanying figure is a
of sales for both items are exponential functions. modified exponential function of the form y  A 
a. Find the equation for the number of cassette B . c x, with c 1. Find appropriate values for the
tapes sold as an exponential function of time. three constants A, B, and c.
b. Find the equation for the number of CDs sold as y
an exponential function of time.
c. What is the practical significance of the growth
or decay factors and growth or decay rates in 10
parts (a) and (b)?
(1, 5)
d. If the trends in sales of both items were indeed
exponential functions, estimate when the num- x
(0, 0)
ber of CDs sold overtook the number of cas-
sette tapes sold.
26. You have been asked to design a slide at a water
22. An exponential function f is such that f 11 2  96 amusement park that extends vertically from point
and f 152  6. Which of the values are possible and A to point B. A person sliding down it will speed up
which are impossible. due to the force of gravity. For the three possible
a. f 132  24 b. f 132  51 c. f 13 2  65 shapes of the slide shown, along which will a person
23. Suppose that a scientist has some initial amount R0 make the trip from A to B most rapidly? Give rea-
of a radioactive substance whose half-life is meas- sons for your answer. (The specific curve along
ured on a scale of days. which an object will slide without friction from A
to B in the shortest possible time is known as the
a. Sketch the graph of the amount of this substance
brachistochrone and was first solved by Jacques
present as a function of time.
Bernoulli in about 1700.)
Use the concavity of your graph from part
(a) to answer the following questions.
A A A
b. Suppose that you measure the amount of the
substance after 10 days and find that 800 grams
are left and after 11 days that 750 grams are left.
Use this information to estimate the number of
grams remaining after 20 days. Is the actual
value higher or lower than your estimate? How B B B
do you know? (i) (ii) (iii)
Gord.3896.02.pgs 4/24/03 9:25 AM Page 99

2.6 Logarithmic Functions 99

Exercising Your Algebra Skills

Simplify the expressions by using properties of exponents. 9. 12x 2 5 10. 10.7x 2 10
1 1 Write each expression as the product of two terms, one
1. 2m . 2n 2. u . v
2 2 an exponential term and the other a constant.
3. 53 . 5x 4. 42 . 43x 11. 3x2 12. 5x2
5. 35 . 32a 6. 24 . 23w 1 4x3
13. 103x1 14. a b
103x 43x 2
7. 8. 3x
102x 4

2.6 Logarithmic Functions

In Section 2.4, we constructed an exponential function to approximate the popula-
tion (in millions) of Florida as
P  f 1t2  12.9411.0292 t,
where t is the number of years since 1990. Using this model, we can predict Flori-
da’s population at any given time, assuming that the growth rate remains 2.9%
each year.
In Example 4 of Section 2.4, we estimated (using both numerical and graphical
methods) that Florida’s population will reach 20 million in early 2005, when
t  15.2. This problem involved finding the value of t for which
f 1t2  12.9411.0292 t  20.
Because this exponential function is always increasing, we know that there
must be only one value of t when P  20. We can always find an approximate value
for t numerically or graphically, as we did in Sections 2.4 and 2.5. We now develop
an algebraic approach for solving such equations exactly rather than approximate-
ly. We want a process that extracts the variable t from the exponent in P  kct. This
process involves a new function called the logarithm. As with exponential func-
tions, logarithms have a base b. Although logarithms can have any base b (denoted
logb), we work primarily with logarithms to base 10.

Definition of Logarithms to Base 10

log10 x  y means 10 y  x.
The logarithm to the base 10 of x is that power of 10 needed to produce x.

For instance,
log10100  log10102  2
because 2 is that power of 10 needed to produce 100, or 102  100. Also,
log101000  log10103  3
because 3 is that power of 10 needed to produce 1000, or 103  1000. Similarly,
log10 10.1 2  log10101  1
Gord.3896.02.pgs 4/24/03 9:25 AM Page 100

100 CHAPTER 2 Families of Functions

because 0.1  1>10  1>101  101 and 1 is the power to which 10 must be
raised to produce 0.1, or 101  0.1.
The logarithm to the base 10 of x, log10 x, is usually written simply as log x. Be-
cause the logarithm is a function, it would be preferable to write log1x 2 rather than
just log x. But log1x 2 is not standard usage, so we avoid it. However, we do use
parentheses for expressions such as log15x 2.
The definition of the logarithm also gives two useful formulas.

Fundamental Logarithmic-Exponential Identities

log110x 2  x, for all real x

10log x  x, for all x  0

Because these formulas hold for all appropriate values of x, they are called
identities. Think about the two results to be sure that you understand them thor-
oughly. For the first identity, log110x 2 is that power of 10 needed to produce 10x.
Clearly, that power must be x itself. For instance, log1101.234 2  1.234. For the sec-
ond identity, the exponent in 10log x is log x. In other words, log x is the power of 10
that gives the number x. For instance, 10log 50.7  50.7. The second property allows
us to undo an equation involving logarithms. We discuss why the second identity is
restricted to positive values of x later in this section.

Using the Logarithm

The principal reason for introducing logarithms here is to solve equations of the form
cx  A
for the variable x in the exponent when the quantities c and A are known. For in-
stance, we might want to solve 3x  8 for the variable x in the exponent. To do so,
we apply the following fundamental property of logarithms.

log1c x 2  x . log c, c0

A proof of this formula can be found in any algebra textbook.

E XAMPLE 1
Solve for x in the equation 3 x  8.
Solution To extract x from the exponent, we take the logarithm of both sides of the
equation:
log13x 2  log 8.
We use the above fundamental property of logarithms to get
log13x 2  x . log 3  log 8,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 101

2.6 Logarithmic Functions 101

where log 3 and log 8 are both just numbers. Finally, we divide both sides by log 3 to ob-
tain
log 8
x  1.893,
log 3
using a calculator. The graph of the function is depicted in Figure 2.43. To verify this re-
sult, we substitute x  1.893 into the original equation and get
31.893  8.0018.
We would have gotten 8 exactly if we hadn’t rounded log 8>log 3  1.893.

FIGURE 2.43 0 x = 1.893

◆
We now show how to obtain an exact solution to the question on the growth of
Florida’s population posed at the beginning of this section.

E XAMPLE 2
Determine when the population of Florida will reach 20 million.
Solution We begin with the equation
P  f 1t2  12.9411.0292 t  20.
Dividing both sides of the equation by 12.94, we get

11.0292 t 
20
.
12.94
We now take logarithms of both sides and use the fundamental property of logarithms
to get

log11.029t 2  t log11.0292  log a b.

20
12.94
Dividing both sides by log11.029 2 gives
log120>12.94 2
t
log11.029 2
,

which is the exact solution. When we perform the actual calculations, we get t  15.23.
Thus Florida’s population will reach 20 million about 15 14 years after 1990, or early in
2005, as depicted in Figure 2.44.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 102

102 CHAPTER 2 Families of Functions

P(t)

20

12.94

t
FIGURE 2.44 0 15.2

◆
Think About This In the solution to Example 2, we intentionally left the quantity in the form 20>12.94
to avoid possible rounding errors when dividing it out. What happens to the
final answer if you perform the division operation early in the solution and
round differently? See what happens, for instance, if you use 20>12.94  1.5 or
20>12.94  1.55 or 20>12.94  1.546. ❐

In Example 3 of Section 2.5, we estimated numerically and graphically that

the half-life of Prozac is approximately 2.4 days based on the exponential decay
function
D 1t2  8010.752 t
that models the amount of Prozac in the bloodstream following an 80 mg dose. We
now show how to obtain the exact answer using logarithms.

E XAMPLE 3
Determine the half-life of Prozac in the bloodstream.
Solution To find the half-life exactly, we must find the time t needed until the original
80 mg drug level falls to 40 mg. Therefore we must solve the equation
8010.75 2 t  40.
If we divide both sides by 80, we get

10.75 2 t 
40
 0.5.
80
We now take logarithms of both sides and use the fundamental property of logarithms
to find
log10.75 2 t  t log10.752  log10.5 2.
When we divide both sides by log10.75 2, we get
log10.5 2
t  2.4094.
log10.75 2
Thus, no matter what level of Prozac is in the blood at any given time, the level will drop
by half about 2.4 days later, as illustrated in Figure 2.45.
Gord.3896.02.pgs 11/21/03 10:25 AM Page 103

2.6 Logarithmic Functions 103

D(t)

80

( 12 )80

t
FIGURE 2.45 0 2.4

◆
Properties of Logarithms
To use logarithms, you need to know their basic properties.

Properties of Logarithms
1. log1Ax 2  x . log A
2. log1A . B2  log A  log B
3. log1A>B2  log A  log B

Proofs of all three of these properties can be found in any algebra textbook.
The first property is the tool we used to extract a variable from the exponent.
The second property lets us simplify the logarithm of a product by writing it as the
sum of the individual logarithms; for instance,
log15 . 12 2  log 5  log 12.
Check this result on your calculator. Also,
log1100x 2  log 100  log x  2  log x.
The third property lets us simplify the logarithm of a quotient by writing it as
the difference of the individual logarithms; for instance,

log a b  log 9  log 4.

9
4
Check this result on your calculator. Also,

log a b  log 1000  log x  3  log x.

1000
x

Think About This Is log19>4 2 the same as 1log 9 2 > 1log 4 2? Why or why not? Try it on your calculator
to see. ❐

Think About This Is log110002>log1x 2  3>log x the same as log 1000  log x? Why or why not?
Graph both y  log110002 >log1x 2 and y  3>log x to see whether it is true. ❐

Note that all three properties of logarithms apply to logarithms with any base
b, not just the base 10. Furthermore, these properties give us some alternative tools
for solving some of the problems that we have already encountered. In Example 4,
Gord.3896.02.pgs 4/24/03 9:25 AM Page 104

104 CHAPTER 2 Families of Functions

we show how to use the second and third properties of logarithms to determine
once more when the population of Florida will reach 20 million.

E XAMPLE 4
Find when Florida’s population will reach 20 million by using properties of logarithms.
Solution We again have to solve the equation
12.9411.0292 t  20.
In Example 2, we began by dividing both sides by 12.94. Instead, suppose that we start by
taking logarithms of both sides of the equation:
log312.9411.0292 t 4  log 20;
log 12.94  log11.0292 t  log 20; log1AB2  log A  log B
log 12.94  t . log 1.029  log 20. log1Ax 2  x log A
To solve for t, we subtract log 12.94 from both sides of the equation:

t . log11.029 2  log 20  log 12.94  log a b. log A  log B  log a b

20 A
12.94 B
Dividing by log11.0292, we get
log120>12.942
t  15.23,
log11.029 2
which is the same result as in Example 2.
◆

Behavior of the Logarithmic Function

For any value of x  0, there is a single corresponding value of log x, so log x is a func-
tion of x. We call this function the logarithmic function or simply the log function.
Let’s now consider the behavior of the log function. Recall that the logarithm
of a number x represents that power of 10 needed to produce x. Because no power
of 10 ever produces 0 (10 raised to what power is 0?), log 0 is undefined. Similarly,
because every power of 10 is positive, log x is not defined for negative values of x.
Thus the domain of the logarithmic function is x  0.
However, it is possible to have a negative power of 10, so log x can be negative.
For instance, 100.25  0.56234 means that log 0.56234  0.25. Thus the range
of the logarithmic function includes both positive and negative values. Figure 2.46
demonstrates that the logarithm of a number between 0 and 1 is negative and that

y = log x
log x not defined log x positive
x
(1, 0)

log x negative
FIGURE 2.46
Gord.3896.02.pgs 4/24/03 9:25 AM Page 105

2.6 Logarithmic Functions 105

the logarithm of any number larger than 1 is positive. Finally, because log 1  0
(since 100  1), the range of the log function consists of all real numbers.
We use these ideas to examine the behavior of the log function f 1x2  log x.
First, let’s look at some values for this function. We know that
log 1  0,
log 10  1,
log 100  2,
log 1000  3,
o
log 1,000,000  6,

and so on. Clearly, log x is an increasing function, at least for x  1. Because the
successive values grow more and more slowly, it is concave down. In fact, the most
significant feature of the growth pattern for the logarithmic function is that it
grows extremely slowly. Note what happens with the above values for the log
function—to gain just one unit vertically requires going 10 times as far horizon-
tally. Thus you need an extremely large value of x to make log x large. For instance,
what value of x is needed to make log x  100? By the definition, x must be 10100
because log110100 2  100. Consequently, it takes an incredibly long time for the
log curve to reach a height of 100. The log function goes to infinity as x increases,
although it does so exceedingly slowly.
The log function is not defined for x  0 or for negative values. But what hap-
pens for small positive values of x? Consider the values
log11 2
 0,
 log1101 2
log10.1 2  1,
log10.012 log1102 2  2,
log10.0012 log1103 2  3,
o
log10.0000012  log1106 2  6,
and so on. As x gets closer and closer to 0, log x becomes more and more negative.
Thus the line x  0 (the y-axis) is a vertical asymptote of the graph of y  log x be-
cause the curve gets closer and closer to this line as x gets closer and closer to 0, but
the curve never reaches it. This vertical asymptote reinforces the fact that the loga-
rithmic function is not defined for x  0, and so the graph of y  log x has no
y-intercept. It does, however, have an x-intercept at x  1 because log11 2  0, as
illustrated in Figure 2.46.

Comparing Exponential and Logarithmic Functions

From the definition of the logarithm, it is clear that there is a close interrelation-
ship between y  log x and the exponential function y  10x. To investigate this
relationship, we start with the graphs of the two functions shown in Figure 2.47.
Both are clearly increasing functions. The exponential function is concave up,
whereas the log function is concave down. However, the main difference in their
growth patterns is that the exponential function y  10x grows extremely rapidly
but the logarithm function y  log x grows extremely slowly.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 106

106 CHAPTER 2 Families of Functions

y
y = 10 x

(b, a) y=x
y

(1, 10) (1, 10)

y = 10 x

(10, 1) (a, b)
(0, 1) y = log x
x
(0, 1) y = log x (10, 1)
(1, 0)
x
(1, 0)
FIGURE 2.47 FIGURE 2.48

Figure 2.48 shows something striking about the graphs of the two functions
y  10x and y  log x: They are reflections of each other about the diagonal line
y  x and thus are symmetric about this line (see Appendix D). Let’s see why.
We know that
log 10  1,
so the point 110, 12 is on the graph of y  log x. By the definition of the logarithm,
log 10  1 means 101  10.
But 101  10 tells us that the point 11, 102 satisfies the equation 10x  y, so the
point 11, 10 2 is on the exponential graph y  10x. The points 110, 1 2 and 11, 102
are reflections of each other about the line y  x. In general, if the point 1a, b2 is
on the graph of y  log x, then
log a  b.
This expression is equivalent to saying that
10b  a,
which means that the point 1b, a2 is on the graph of the exponential function
y  10x. Hence the log graph and the exponential graph are reflections of each
other about the line y  x. We investigate this phenomenon in more depth in
Section 2.9.

Applications of Logarithmic Functions

Logarithms have many applications. For instance, chemists use a quantity known as
the pH to measure how acidic a water solution is. The pH is based on the concentra-
tion of hydrogen ions (measured in moles per liter) in the solution. The hydrogen-
ion concentration of pure water is 107 moles per liter. Thus the pH of pure water is
pH  log1concentration2  log1107 2  17 2log 10  7 . 1  7,
which is used as the reference point for a neutral solution. Water solutions whose
pH values are less than 7 are said to be acidic, whereas water solutions with pH val-
ues greater than 7 are said to be basic or alkaline. The lower the pH, the more acidic
the solution; the higher the pH, the more basic the solution. Orange juice, which is
Gord.3896.02.pgs 4/24/03 9:25 AM Page 107

2.6 Logarithmic Functions 107

somewhat acidic, has a hydrogen-ion concentration of 2  104 moles per liter

and so its pH is
pH  log12  104 2  3log 2  log1104 2 4
 log 2  log1104 2
 log 2  14 2log 10
 0.301  4 . 1  3.7.
Hydrochloric acid, with a hydrogen-ion concentration of 101 moles per liter, has
a pH of 1, which indicates that it is extremely acidic. In comparison, household
ammonia, with a pH of 11.5, is extremely basic.

Think About This Human blood has a hydrogen-ion concentration of 4  108. What is its pH? Is
blood slightly or extremely basic? ❐

Note that each one-point decrease in pH represents a tenfold increase in the

hydrogen-ion concentration.
The crust of the Earth is composed of about 20 rigid plates that “float” on liq-
uid magma (the molten material beneath the Earth’s crust). The study of this phe-
nomenon is called plate tectonics. A geologic fault, such as the famous San Andreas
Fault in California, is the space between two plates. As plates move, they bump into
one another and sometimes one plate passes slightly under another, causing the
upper plate to shift and heave. The result is an earthquake on the Earth’s surface.
There are about a million earthquakes, mostly very minor, each year. American
seismologist Charles Richter developed a way of measuring the intensity of an
earthquake. The Richter scale is based on the idea that there is a minimum notice-
able, or threshold, level of earthquake intensity, denoted by I0 . The energy involved
in a threshold level earthquake is approximately equal to the energy released by
10,000 atomic bombs. Any stronger quake has an intensity denoted by I. The
Richter scale relates the magnitude R of an earthquake to its intensity, or

R  log a b .
I
I0
That is, the magnitude given by the Richter scale measurement is the logarithm of
the ratio of the actual intensity to the threshold level.
The largest recorded earthquake, which occurred in Japan in 1933, had magni-
tude R  8.9 on the Richter scale. Let’s see just how powerful this quake was. We
have

R  log a b  8.9,
I
I0
so when we take powers of 10 of both sides of the equation,
I
10log 1I>I02   108.9  794,328,235.
I0
Therefore this quake had an intensity almost 800 million times greater than the
threshold level!
How are different measurements on the Richter scale related? For instance, if the
measurement for one earthquake is double that of another, how much greater is it?
What does a one point change in magnitude represent?
Gord.3896.02.pgs 4/24/03 9:25 AM Page 108

108 CHAPTER 2 Families of Functions

E XAMPLE 5
How does a magnitude 5 earthquake on the Richter scale compare to a magnitude 6 quake?
Solution If R  5, we have

log a b  5.
I
I0
We undo the logarithm by taking powers of 10 of both sides of the equation and use the
fundamental identity
10log x  x
to get
I
10log 1I>I02   105  100,000.
I0
For I>I0  100,000, we get I  100,000I0, so the intensity of a magnitude 5 quake is
100,000 times the threshold level. This quake’s energy is equivalent to roughly
100,000  10,000  109, or 1 billion, atomic bombs exploding simultaneously.
Similarly, consider an earthquake measuring R  6 on the Richter scale. We now get
I>I0  106  1,000,000.
The intensity of this quake is 1 million times the threshold level. Thus an increase of
1 Richter unit corresponds to a tenfold increase in the intensity of the earthquake.
◆

Suppose that one earthquake has a reading twice that of another on the Richter
scale. How much stronger is it? Is it four times as strong? Is the relative intensity the
same? Does it depend on the value for R? Let’s compare R  4 to R  2 to see what
happens. With R  4, we have

log a b  4,
I
I0
so when we take powers of 10 of both sides of the equation,
I
10log 1I>I02   104
I0
and therefore
I  104 . I0 .
Hence a magnitude 4 quake is 10,000 times the intensity of the threshold level. For
R  2, we have

log a b  2
I
I0
so that
I
10log 1I>I02   102
I0
Gord.3896.02.pgs 11/21/03 10:25 AM Page 109

2.6 Logarithmic Functions 109

and therefore
I  102 . I0 .
Hence a magnitude 2 quake is 100 times the intensity of the threshold level. Con-
sequently, a magnitude 4 quake is actually 104>102  100 times stronger than a
magnitude 2 quake.

Changing Bases
Throughout this book, we work with logarithms to the base 10 to undo expo-
nential functions of the form y  k . 10x. However, as we stated earlier, it is pos-
sible to have bases other than 10—say, c  2 or c  1.029—as the base for an
exponential function y  kc x. Each possible base gives rise to a corresponding
logarithmic function. For instance, we could work with logarithms to the base 2,
written log2x.

Definition of Logarithms to Base c

logc x  y means cy  x, x  0.

The logarithm to the base c of x is that power of c needed to produce x.

In practice, there is one particular base other than 10 that is widely used: the
number e  2.71828. . . . The logarithm corresponding to base e is called the
natural logarithm. It is especially important in calculus and many of the sciences.
Even though we could write the natural logarithm as loge x, it is customarily writ-
ten ln x.
Although log1010  1, we have ln 10  2.3026 because e2.3026  2.718282.3026 
10.0001. Similarly, whereas log10 100  2, we have ln 100  4.6052 because
e4.6052  2.718284.6052  100.003.
We previously said that all the properties of logarithms apply no matter what
base is used. Thus the following are properties of the natural logarithm.

Properties and Identities for the Natural Logarithm

1. ln1Ap 2  p ln A
2. ln1A . B2  ln A  ln B
3. ln1A>B2  ln A  ln B
4. ln e x  loge e x  x
5. eln x  x, if x  0

If different bases are used, we must be able to convert either an exponential

function or a logarithm in one base into an exponential function or a logarithm in
a different base. That is, for any number x, how do we convert from cx to 10x or
convert log x to ln x and vice versa? Let’s first look at converting bases of exponen-
tial functions.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 110

110 CHAPTER 2 Families of Functions

E XAMPLE 6
We found that the population of Florida can be modeled by the exponential function
P1t2  12.9411.029 2 t. Convert this function to an equivalent expression that involves
(a) base 10 and (b) base e.
Solution

a. Suppose that we try to find the appropriate power q so that

11.0292 t  10q.
Using properties of logarithms, we take logs of both sides and get
t log11.0292  q log 10  q . 1  q.
The formula for the population of Florida becomes
P1t2  12.9411.0292 t  12.94110q 2  12.94110t log11.0292 2  12.941100.0124t 2,
since log11.0292  0.0124. Alternatively, we might think of this result as coming from
10t log 11.0292  110log 11.0292 2 t  11.0292 t.
b. To convert the formula for the Florida population to an equivalent formula using
base e, we use the property eln x  x:
11.029 2 t  1eln 1.029 2 t  1e0.0286 2 t  e0.0286t.
Therefore
P1t2  12.9411.0292 t  12.941e0.0286t 2.
◆
We have three formulas for the population of Florida:
P1t2  12.9411.0292 t;
P1t2  12.941100.0124t 2 ;
P1t2  12.941e0.0286t 2 .
These three formulas are mathematically equivalent—only the bases are different.
Graph these three functions using your function grapher to convince yourself that
they are identical, except perhaps for slight differences due to rounding.
In general, to convert an exponential function y  kc x from base c to base 10,
where y  k . 10mx, we write c  10m so that m  log c. To convert an exponential
function y  kc x from base c to base e, where y  kemx, we write c  em, so that
m  ln c.
Now let’s consider how to convert a logarithm in one base to a logarithm in
another base. We begin by looking at some typical values of log x and ln x,
rounded to four decimal places, as shown in the following table. To see if there is
any clear relationship between the two sets of logarithmic values, we plot the val-
ues of ln x versus log x, as shown in Figure 2.49.
The graph shows that there is a linear pattern. The line passes through the ori-
gin, so its vertical intercept is 0 and we can write
ln x  m log x
for some constant of proportionality m. You can also see this from the ratio of ln x and
log x for any value of x—in every case, the ratio is approximately 2.3026. Check this
Gord.3896.02.pgs 4/24/03 9:25 AM Page 111

2.6 Logarithmic Functions 111

y
x log x ln x
2.5
1 0 0
2
2 0.3010 0.6932
1.5

ln x
3 0.4771 1.0986
1
4 0.6021 1.3863
0.5
5 0.6990 1.6094
x
6 0.7782 1.7918 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
log x
7 0.8451 1.9459
FIGURE 2.49
8 0.9031 2.0794

9 0.9542 2.1972
10 1 2.3026

on your calculator by using several different values of x. Note that Figure 2.49 shows
the values of ln x plotted against those of log x. If we plotted either ln x against x or log
x against x, we would get the usual graph of a logarithmic function—one that is in-
creasing and concave down. Only the graph of ln x versus log x results in a line.
The value of the constant of proportionality m  2.3026 is also the slope of
the line through the points shown in Figure 2.49. Thus we can write
ln x  2.3026 log x
for any x. Because 2.3026 appears in the last row of the preceding table as the value
of ln 10, we can rewrite this relationship as
ln x  1ln 102 log x,
or equivalently,
ln x
log x  .
ln 10
Rewriting this equation to highlight the base of the logarithm, we get
log e x
log10 x  .
log e10
In fact, if we perform the identical analysis with any other base (say, c instead of e), we
obtain the comparable result for changing between base 10 and base c, for any c; or
logc x
log10 x  .
logc10

Problems
d. y  10log 1x 2
2 2
1. The graphs of the following functions may surprise c. y  log 10x
you. Use your function grapher to graph each func- e. y  log 3 x
f. y  log110>6x 2
tion and then explain what you see and why, using 2. Use your function grapher to draw simultaneously
the properties of logarithms. the graphs of y  log12x 2, y  log13x 2, and
a. y  log 102x b. y  log12x 2  log1x 2 y  log15x 2 . For each function, use the properties
Gord.3896.02.pgs 4/24/03 9:25 AM Page 112

112 CHAPTER 2 Families of Functions

of logarithms to explain why you get the graphs dioactive form of carbon, decays into carbon-12 with a
you do. half-life of 5730 years.
3. The population of Argentina was 34.6 million in 10. The famous Cro-Magnon cave paintings are found
1995 and was growing exponentially at an annual in the Lascaux Cave in France. If the level of car-
rate of 1.3%. bon-14 radioactivity in charcoal in the cave is ap-
a. Find an expression for Argentina’s population proximately 14% of that of living wood, estimate
at any time t, where t is the number of years the date when the paintings were made.
since 1995. 11. The level of carbon-14 in a charred roof beam found
b. What population would you predict for 2005 if in a 1950 excavation of an ancient Babylonian city is
the present trend continues? about 61% of the level in living wood. Estimate when
c. Use logarithms to find the doubling time exactly. the fire occurred.
4. The population of Kenya is growing exponentially. Its 12. The well-preserved body of a Stone Age man was
population was 23.3 million people in 1988 and found in melting snow in the northern Italian Alps
28.3 million in 1995. in 1991. Examination of a tissue sample from the
a. Find an expression for the population at any time body indicated that 47% of the carbon-14 present
t, where t is the number of years since 1988. in the body at the time of death had decayed. When
b. What would be the population in 2005? did the man die?
c. Use logarithms to find the doubling time. 13. Several groups of scientists were allowed to test the
5. Because of ardent fishermen during the summer Shroud of Turin, the supposed burial cloth of
months, the population of fish in a lake is reduced Jesus, in 1991. They found that the cloth contained
by 10% each week. Find the half-life of this dwin- 91% of the amount of carbon 14 contained in
dling fish population, using logarithms. newly made cloth of the same material. Based on
this information, how old is the Shroud of Turin?
6. The Best Company earned $50 million in 2000,
and its income is growing at a rate of 2% per year. 14. A radioactive substance decays exponentially so that
The Acme Corporation earned $30 million that after 10 years, 40% of the initial amount R0 remains.
year, and its income is growing at a rate of 6.5% a. Find an expression for the quantity remaining
per year. When will Acme overtake Best in annual after t years.
income? b. How much will be present after 25 years?
7. a. Find the doubling time for annual growth rates c. What is the half-life of the substance?
of 3%, 4%, 5%, 6%, and 7%. d. How long will it be before only 2% of the origi-
b. Consider the doubling time d as a function of nal amount is left?
growth rate r. Plot your results from part (a) and 15. In an effort to reduce the breeding rate of a strain of
decide what type of function seems to fit the be- pesticide-resistant mosquitoes in the southeastern
havior pattern you observe. United States, a group of scientists released large
8. Bankers use a technique called the Rule of 70 to esti- numbers of sterilized male mosquitoes to mate with
mate the doubling time for money invested at differ- the fertile females who would consequently produce
ent interest rates, dividing 70 by 100 times the interest no offspring. Suppose that effort reduced the mosqui-
rate. Thus for an interest rate of 10%  0.10, to population by 2% per month.
bankers estimate the doubling time to be a. What percentage of the original population P0
70 70 would remain after 1 year?
.   7 years. b. How long would it take to lower the population
100 0.10 10
by half?
Use your results from Problem 7 to test the accura- c. How long would it take for the population to fall
cy of this method. to 10% of the original level?
9. Determine when the cost of first-class postage for a 16. In computer science, the efficiency of algorithms
letter will reach $1, given that first class postage rose (methods for accomplishing a certain task) are
to 29¢ in 1990 and to 37¢ in 2002. often analyzed by how long it takes to perform the
Problems 10–13 are based on the carbon dating operation with n objects. Typically, as n increases,
process to measure the age of objects. Carbon-14, a ra- the time involved for the operation increases signif-
Gord.3896.02.pgs 4/24/03 9:25 AM Page 113

2.7 Power Functions 113

icantly. Two algorithms used to put a set of names b. A loud noise of about 150 decibels will cause
in alphabetical order are compared. For one algo- deafness. How much more intense is this level
rithm, the time needed to order n names, as a func- than the threshold level?
tion of n, is B1n2  12 n2. The time for the second c. An aircraft taking off has a loudness level of
algorithm, as a function of n, is S1n2  n log n. about 120 decibels. How much more intense is
Which method is faster? Explain. this level than the threshold level?
17. How much stronger is a magnitude 6 earthquake d. How loud (that is, how many decibels) is a
than a magnitude 3 earthquake? sound whose intensity is 1 million times the
18. How much stronger is threshold level?
e. The noise level from a rock band is about 100 bil-
a. a magnitude 7 quake than a magnitude 5 quake?
lion times higher than the threshold level. What
b. a magnitude 7 quake than a magnitude 4 quake?
is the decibel measure of this noise level?
19. Let I0 be the minimum (or threshold) level of
20. Convert the formula D1t2  8010.752 t for the level of
sound that can be heard by human beings. If the in-
Prozac in the bloodstream following an initial dose of
tensity of a particular sound is I, the magnitude of
80 mg to equivalent formulas with base 10 and base e.
the sound, measures in decibels d, is given by
21. The population of the world can be modeled by the
d  10 log a b .
I
function P1t2  611.0152 t, where t is the number of
I0 years since 1999. Convert this formula to equivalent
a. Normal conversation measures about 60 decibels. formulas with base 10 and base e.
How much more intense is this level than the
threshold level?

Exercising Your Algebra Skills

Simplify each expression by using the properties of 13. 411.05 2 x  511.042 x 14. 310.7 2 x  610.52 x
logarithms. 15. log x  2 16. log x  0.5
1. log x  log x 2  log x 3 2. log x  log 1x Without using a calculator, evaluate each term.
3. log x  log y  log x  log y 2
2 3
17. log 1,000,000 18. log 0.001
2
4. log1x>y2  log1y>x2 5. log 10x 19. log11>10,0002 20. log 210
6. 10log1x 2
2

Determine which of the following are true for all values

Solve each expression for x. of x and which are not by using algebra.
7. 7x  11 8. 1.05x  2 21. log1x 2  1 2  log1x  1 2  log1x  1 2
9. 0.4x  0.6 10. 311.04 2 x  5 22. log1x 2  1 2  log1x 2 2  log112
11. 1210.86 2 x  3 12. 910.17 2 x  0.25 23. log11>x 2  log x  log11 2

2.7 Power Functions

The area A of a circle with radius r is given by
A  f 1r2  pr 2.
The surface area S of a sphere with radius r is given by
S  g 1r2  4pr 2.
(Picture a tennis ball whose surface is made up of four roughly circular regions, as
shown in Figure 2.50). The volume V of the sphere is given by
4
V  h1r2  pr 3.
3
Gord.3896.02.pgs 4/24/03 9:25 AM Page 114

114 CHAPTER 2 Families of Functions

FIGURE 2.50 Tennis ball cover split apart

Similarly, the inverse square law of gravitation describes how the force of grav-
ity of one object on any other object in the universe varies with distance. In partic-
ular, the gravitational force F on a unit mass at a distance d from the center of the
Earth is given by
k
F or F  kd 2,
d2
where k is a positive constant.
All four of these functions are examples of power functions, so called because
the independent variable is raised to a constant power. In each case, the dependent
variable is a constant multiple of some power of the independent variable. In gen-
eral, a power function is any function of the form
y  f 1x 2  k x p,
where k and p are any constants, positive or negative. (Compare this expression for
a power function with an exponential function of the form y  kc x, where the in-
dependent variable x is the exponent and the base c is a constant, as shown in Fig-
ure 2.51.) Note that the family of power functions y  kx p is a two-parameter
family with parameters p and k.

Exponential function Power function

Variable

y = kc x y = kx p

FIGURE 2.51 Constant

The definition of a power function includes the special case where p  1,

which gives the linear function y  kx 1  kx that passes through the origin. The
definition also includes functions such as
y  x 2, y  x 3, and y  x 4,
as well as the case where p is a positive or negative fraction or a decimal, such as
y  x 1>2, y  x 1>3, y  x 3>2, and y  x 2.83.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 115

2.7 Power Functions 115

In algebra, fractional exponents are usually introduced purely as a means for sim-
plifying operations with terms involving radicals. Recall that
3 8 5
x 1>2  2x , x 1>3  2 x, x 5>8  2x ,
and, in general,
x m>n  2x m  A 2x B .
n n m

Power functions of the form y  f 1x 2  k x m>n arise naturally in many appli-

cations. For instance, biologists have found a relationship between the weight W of
large flying birds and their wingspan S. This relationship can be modeled by the
power function
W  f 1s2  0.15S 9>4.
This function gives the weight that can be supported by a given wingspan.
For example, the wingspan S of a condor is about 10 feet. According to this
model, its weight is approximately
W  0.15110 2 9>4  27 lb.
To perform this calculation we must either first raise S  10 to the ninth power
and then take the fourth root of the result, so that
W  0.15110 2 9>4  0.153 110 2 9 4 1>4  0.153 1,000,000,0004 1>4  27
or first take the fourth root of S and then raise the result to the ninth power, so that
W  0.15110 2 9>4  0.153 110 2 1>4 4 9  0.153 1.7782794 9  27.
Symbolically, we write

S  12
4 9 4
S 9>4  2 S2 9.
When you use a calculator to evaluate such an expression, be careful to use paren-
theses around the fractional exponent, as in
0.15*10^(9/4);
without the parentheses, the rules for the order of operations will give you a very
different answer.
The graph of the power function W  0.15S 9>4 is shown in Figure 2.52. Note
that the pattern is that of an increasing, concave up function; thus, as the wingspan
S of a bird increases, its weight W increases even more rapidly. Consequently, heav-
ier birds require relatively smaller wingspans in order to fly, which is likely contrary
to intuition.

W = 0.15S 9/4
Weight

FIGURE 2.52 Span

Gord.3896.02.pgs 4/24/03 9:25 AM Page 116

116 CHAPTER 2 Families of Functions

Think About This The largest known bird is the Steller’s eagle, with a wingspan of 8 feet. Estimate the
weight of an adult Steller’s eagle. ❐

Aeronautical engineers use the same principle—based on a similar relation-

ship between wingspan and the weight of a plane—when designing new aircraft.

Behavior of Power Functions for x  0

Recall that for exponential functions the value of the base c in y  kc x leads to dif-
ferent behavior patterns—exponential growth when c  1 and exponential decay
when 0 c 1. Similarly, the behavior of power functions depends on the size of
the constant power p in y  k x p. To simplify things initially, we let k  1 so that
we can consider the more basic power function y  x p.
The three different behavior patterns for power functions are illustrated in Fig-
ure 2.53: the graphs of y  x 2, y  x 1>2, and y  x 2 (along with the graph of y  x
for reference) for x  0. Note how the graph of y  x 2 (with p  2) is increasing and
concave up; that of y  x 1>2 (with p  12) is increasing and concave down, and that of
y  x 2 (with p  2) is decreasing and concave up. The specific values p  1
(when we get a line through the origin) and p  0 (when we get a horizontal line at
height y  1 2 are critical in separating one kind of behavior from another.

y = x –2
y = x2
y=x

y = x 1/2

(1, 1)

x
FIGURE 2.53 0

Let’s investigate these cases in more detail by looking at a variety of different

power functions of each type. Figure 2.54(a) shows the graphs of three related

y y = x 2.5 y y

y = x3 y = x2

y = x 3/4

y = x 1/2

y = x 1/4 (1, 1)
(1, 1)
(1, 1) y = x –2
y = x –1
y = x –3
x x x
0 0 0
(a) p > 1 (b) 0 < p < 1 (c) p < 0

FIGURE 2.54
Gord.3896.02.pgs 4/24/03 9:25 AM Page 117

2.7 Power Functions 117

power functions, y  x 2, y  x 2.5, and y  x 3, all of which behave in the same

manner as y  x 2. Similarly, Figure 2.54(b) shows the graphs of three other
power functions, y  x 1>4, y  x 1>2, and y  x 3>4, all of which behave in a pat-
tern similar to that for y  x 1>2. Finally, Figure 2.54(c) shows the graphs of
y  x 1, y  x 2, and y  x 3, all of which behave in a third manner, similar to
y  x 2.
These graphs suggest the following facts about power functions for x  0:

1. If p  1, the power function y  x p is increasing and concave up.

If 0 p 1, the power function y  x p is increasing and concave
down.
If p 0, the power function y  x p is decreasing and concave up.
2. Every power function y  x p passes through the point 11, 12.
3. If p  0, the power function y  x p passes through the origin.
If p 0, the power function y  x p rises toward the positive y-axis
as x gets closer to 0 (the y-axis is a vertical asymptote) and decays to-
ward the positive x-axis as x approaches
.

Statement 2 is true because 1 raised to any power is 1 (that is, 1p  1 for any p).
The first part of Statement 3 is obvious because 0 raised to any positive power will
be 0 (that is, 0 p  0 for p  0). As for the second part of Statement 3, if the power
p is negative, we can write it as p  q so that
1
x p  x q  ,
xq
using one of the basic rules for exponents from algebra. Obviously, we can’t have
x  0 because the quotient 1>x q is not defined at 0. However, the closer x is to 0,
the closer x q is to 0 also, and therefore the larger 1>x q is. That is, the graph of any
power function of the form y  x p  x q must always rise and approach the
positive y-axis as x approaches 0. The y-axis is a vertical asymptote for these
curves because they approach it more and more closely but never reach it. Also, if
p 0, as x increases, x p  x q  1>x q becomes smaller and eventually ap-
proaches 0. So the x-axis is a horizontal asymptote for any power function with
p 0.

Behavior of Power Functions for x  1

It is also evident from Figure 2.54 that the behavior patterns for these power func-
tions are different when x is between x  0 and x  1 compared to when x  1.
Again, the point 11, 12 serves as a demarcation between the different behaviors. To
see the differences more clearly, in Figure 2.55 we zoom in on all the graphs shown
in Figure 2.54 to examine what happens when x  1 for different values of p. (In-
cidentally, in all these cases, you should think of x as the variable and p as a param-
eter that takes on a fixed value to produce a particular curve whose values depend
on x.)
Figure 2.55 suggests the following additional fact about power functions.
Gord.3896.02.pgs 4/24/03 9:25 AM Page 118

118 CHAPTER 2 Families of Functions

y y y
y = x 2.5
y = x 3/4
y = x3
y= x2
(1, 1)
y = x 1/2

y = x 1/4 y = x –1

(1, 1)
(1, 1) y = x –2
y = x –3
x x x
0 0 0
(a) p > 1 (b) 0 < p < 1 (c) p < 0

FIGURE 2.55

4. If p  0, the larger p is, the larger x p is when x  1.

If p 0, the more negative p is, the more rapidly x p dies out as x
increases.

Thus, for instance, when x  1,

x 5  x 4  x 3  x 2,
and also
x 5 x 4 x 3 x 2.
These relationships are shown numerically in Table 2.1 for p  0. For each
value of x  1, not only do the higher powers of x get larger, but they get larger
much faster.

TABLE 2.1

x2 x5 x  10 x  20
y  x2 4 25 100 400
y  x3 8 125 1000 8000
y  x4 16 625 10,000 160,000
y  x5 32 3125 100,000 3,200,000

We can prove this fact algebraically. For instance, if x  1 and we multiply

both sides of this inequality by the positive term x 3, we get
x 3 . x  x 3 . 1 so that x 4  x 3.
This comparison gets more pronounced as x gets larger and larger. Compare the
values in the table for x  10 and x  20. As x gets ever larger (i.e., as x approach-
es infinity, denoted by x S
), any positive power completely overwhelms, or
dominates, any smaller power.
Now let’s look at the case when p 0. For each value of x  1, Table 2.2
shows that, not only do the negative powers of x get smaller as x increases, but the
more negative the power, the faster the function y  x p dies out.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 119

2.7 Power Functions 119

TABLE 2.2

x2 x5 x  10 x  20
y  x2 0.25 0.04 0.01 0.0025
y  x3 0.125 0.008 0.001 0.000125
y  x4 0.0625 0.0016 0.0001 0.00000625
y  x5 0.03125 0.00032 0.00001 0.0000003125

Behavior of Power Functions for 0  x  1

The preceding conclusions are based on what happens to power functions when
x  1. Now let’s see what happens to these same power functions when 0 x 1,
as shown in Figure 2.56. Note that the behavior patterns are reversed from those when
x  1. In particular, the graphs suggest the following fact about power functions.

5. If p  1 or p 0, the larger p is, the smaller x p is when 0 x 1.

If 0 p 1, the larger p is, the larger x p is when 0 x 1.

y y y

(1, 1) y = x –2
(1, 1) y = x –3

y = x 3/4

y = x –1

y = x2 y= x 1/2
y = x3

y = x 1/4
(1, 1)
y = x 2.5
x x x
0 0 0
(a) p > 1 (b) 0 < p < 1 (c) p < 0

FIGURE 2.56

Thus, for instance, when x is between 0 and 1,

x 5 x 4 x 3 x 2,
and
x 2 x 3 x 4 x 5.
Thus, as x approaches 0, the more positive the power p, the faster a power function
dies out near the origin. Similarly, the more negative the power p, the faster a
power function grows toward infinity as x approaches 0.

Think About This What do the graphs of y  x 0.99  x 199>1002 and y  x 1.01 look like? How do they
behave compared to the line y  x? ❐

Think About This What do the graphs of y  x 0.01, y  x 0, and y  x 0.01 look like? How do they
compare to one another? ❐
Gord.3896.02.pgs 4/24/03 9:26 AM Page 120

120 CHAPTER 2 Families of Functions

Applications of Power Functions

In Sections 2.2 and 2.3, we dealt with many situations that led to equations involv-
ing linear functions. In Sections 2.4–2.6, we similarly explored some situations
leading to equations involving exponential functions of the form
y  kct.
In many of these situations, we had a value for y and had to solve for the un-
known t in the exponent by using logarithms. Similarly, situations often arise that
lead to equations involving power functions of the form
y  k x p,
where a value for y is given and we have to solve for the unknown x, which in this
case is raised to a constant power.
As we discussed in several simple examples in previous sections, we solve the
power function equation x 2  25 by taking the square root of both sides of the equa-
tion to get x  5. (To envision the two solutions, imagine the graph of the function
y  x 2: It reaches a height of 25 in two places, one for x 0 and another for x  0.)
Similarly, we solve the equation x 3  27 by taking the cube root of both sides of
the equation to get
3
x 2 27  3.
Similarly, we solve the equation x 7  50 by taking the 7th root of both sides of the
equation (equivalently by raising both sides to the 17 power):
7
x 2 50  501>7  1.7487.

E XAMPLE 1
A bald eagle weighs about 16 pounds. Use the relationship W  0.15S 9>4 to estimate its
wingspan.
Solution We have to solve for the eagle’s wingspan S corresponding to W  16 pounds
in the equation
0.15S 9>4  16.
We first divide both sides by 0.15 to get
S 9>4  106.667.
To solve for S, we have to undo the 94 power, which we do by raising both sides of this
equation to the 49 power:
1S 9>4 2 4>9  S 1  S  106.6674>9  7.968,
or about 8 feet.
◆

In Example 2 we show how these ideas about power functions arise in the con-
text of one of the most useful applications of radioactive decay. Scientists routine-
ly use a process known as carbon-dating to establish the age of fossils. It is based on
the fact that carbon-14 decays to carbon-12 with a half-life of 5730 years.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 121

2.7 Power Functions 121

E XAMPLE 2
Crater Lake in Oregon was formed as the result of a volcanic eruption. A charcoal sam-
ple from a tree that burned during the eruption contains about 46% of the carbon-14
found in live trees.
a. What is the decay factor for carbon-14?
b. What was the approximate date for the formation of Crater Lake?

Solution Because the half-life of carbon-14 is 5730 years and slightly more than 50%
of the radioactive carbon has disintegrated, we expect that the time involved is some-
what more than 5730 years; we might estimate, say, about 6000 years. Now let’s find out
more precisely.
a. The exponential decay function that models the radioactive decay process is

R1t2  R0ct
for some initial quantity R0 of the radioactive carbon and some decay factor c 1.
Because the half-life of carbon-14 is 5730 years, we substitute t  5730 into the ex-
pression for the function to get
1
R157302  R0c5730  R0
2
so that
1
c5730  .
2
We solve this equation for c by extracting the 5730th root of 12 , to get the decay factor:
1 1>5730
c a b  0.99988.
2

b. We know that

R1t2  R0 . 10.999882 t.
We now have to find how long it takes for the carbon-14 to decay to the point where
only 46% of R0 is present. Thus we want to find t when
R1t2  R0 . 10.999882 t  0.46R0 .
Dividing through by R0 gives
10.999882 t  0.46.
To solve this exponential equation, we take logs of both sides and get
log10.999882 t  t log10.999882  log10.462.
Therefore
t  log10.462 >log10.999882  6470.685,
or about 6471 years ago, as shown in Figure 2.57. We thus conclude that Crater Lake
was formed in roughly 4471 B.C.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 122

122 CHAPTER 2 Families of Functions

R(t)

R0

0.46(80)

t
FIGURE 2.57 0 6471

◆
Note that we solved two very different mathematical problems in Example 2. In
part (a), we had the power function equation c5730  12 and solved for c by taking the
5730th root. In part (b), we had the exponential function equation 10.999882 t  0.46
and solved for t by using logarithms. In general, we take roots to undo powers and so
solve equations of the form x n  A involving power functions. Similarly, we use loga-
rithms to extract variables from an exponent and so solve equations of the form
bx  A involving exponential functions.
Unfortunately, neither operation will do anything useful to solve an equation
such as
3x  x 4,
which involves both an exponential function and a power function. This equation
can be solved numerically or graphically to find an approximate solution that is ac-
curate to any desired degree of accuracy, by using a graphing calculator or a com-
puter graphics program. (We ask you to solve it as a problem at the end of the
section.) But, the equation cannot be solved algebraically to find an exact solution.

Fitting Power Functions to Two Points

We have seen that two points determine a line or an exponential function because
only one line or one exponential function passes through those points. Similarly,
two points also determine a power function, which we illustrate in Examples 3–5.
E XAMPLE 3
Find the power function that passes through the points 11, 5 2 and 14, 602.
Solution A power function is of the form y  kx p, for some constants k and p. We sub-
stitute the coordinates of the first point 11, 52 into this expression to obtain
k . 11 2 p  k  5.
Therefore the expression for the power function reduces to y  5x p. We use the second
point, 14, 602 , to determine the power p. When we substitute the coordinates x  4
and y  60, we get
514 2 p  60.
We divide both sides of this equation by 5 to obtain
4p  12.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 123

2.7 Power Functions 123

To extract the unknown p from the exponent, we take logs of both sides to get
log14p 2  p log 4  log 12,
so that
log 12
p  1.792.
log 4
Consequently, the power function that passes through the two points 11, 52 and 14, 602 is
y  5x 1.792.
As depicted in Figure 2.58, this function is increasing and concave up.

80

60
y = 5x 1.792
40

20

x
FIGURE 2.58 0 1 2 3 4 5

◆
Example 4 illustrates a somewhat more complicated situation.

E XAMPLE 4
Find the power function that passes through the points 12, 5 2 and 14, 602.
Solution When we substitute the coordinates of the first point 12, 5 2 into the equation
of a power function y  k x p, we get
k . 2p  5,
which involves both unknowns k and p. Similarly, when we substitute the coordinates of
the second point 14, 60 2 , we get
k . 4p  60.
To eliminate one of the unknowns, we divide the second equation by the first:
k . 4p 4p 60
 p  12.
k . 2p 2 5
Because
4p 4 p
 a b  2p,
2p 2
the preceding equation becomes
2p  12.
We solve for p by using logarithms:
log 2p  p log 2  log 12
Gord.3896.02.pgs 4/24/03 9:26 AM Page 124

124 CHAPTER 2 Families of Functions

so that
log 12
p  3.584963.
log 2
The desired power function is therefore y  k . x 3.584963. To determine the value of k, we
use the first point 12, 5 2 and obtain
5  k . 23.584963.
Therefore
5
k 3.584963  0.41667,
2
and the power function, which is shown in Figure 2.59, is
y  0.41667x 3.584963.

80

60

40 y = 0.41667x 3.584963

20

x
FIGURE 2.59 0 1 2 3 4 5

E XAMPLE 5
Biologists have long observed that the larger the area of a region, the more species in-
habit it. The relationship is best modeled by a power function. The island of Puerto Rico
contains 40 species of reptiles and amphibians on its 3459 square miles. The nearby is-
land of Hispaniola (comprising Haiti and the Dominican Republic) contains 84 species
on 29,418 square miles.
a. Determine a power function that relates the number of species of reptiles and am-
phibians on a Caribbean island to its area.
b. Use the relationship from part (a) to predict the number of species of reptiles and
amphibians on Cuba, which measures 44,218 square miles.
Solution

a. We want a power function of the form S  kAp, where S is the number of species, A
is the area in square miles, and k and p are constants that must be determined. Using
the information on Puerto Rico, where S  40 and A  3459, we have
k . 134592 p  40, (1)
which involves both p and k. The data on Hispaniola, A  29,418 and S  84, give
k . 129,4182 p  84, (2)
Gord.3896.02.pgs 4/24/03 9:26 AM Page 125

2.7 Power Functions 125

so we now have two nonlinear equations in the two unknowns k and p. We can elim-
inate the unknown k by dividing Equation (2) by Equation (1). We then get
k . 1294182 p 84
  2.1.
k . 134592 p
40
We cancel the common factor k to get
29418 p
 2.1.
3459 p
Using one of the properties of exponents, we find that this equation reduces to
29418 p
a b  18.5052 p  2.1.
3459
To solve for p, we take logs of both sides of this exponential equation to get
log18.5052 p  p log18.5052  log12.1 2,
from which we find that
log12.1 2
p  0.3466.
log18.5052
Substituting this value into Equation (1) gives
k . 13459 2 0.3466  40,
so that
40
k  2.3739.
134592 0.3466
Thus the power function that models the number of species of reptiles and amphib-
ians on a Caribbean island having area A is
S  2.3739A0.3466.
Note from the graph of this function shown in Figure 2.60 that it is an increasing, concave
down function, which is what we would expect from a power function with p  0.3466.

100

S = 2.3739A0.3466
Number of species

80

60

40

20

0 5000 10000 15000 20000 25000 30000

FIGURE 2.60 Area of island (sq mi)

b. For the area of Cuba, 44,218 square miles, we use this formula to estimate that there are

S  2.37391442182 0.3466  96.745,

or about 97 reptile and amphibian species on Cuba.

◆
Gord.3896.02.pgs 4/24/03 9:26 AM Page 126

126 CHAPTER 2 Families of Functions

Power Functions with Integer Powers

So far, we have restricted our attention to power functions with x  0 because most
power functions of the form y  x m>n with rational exponents are not defined when
x 0. However, if the power is an integer, either positive or negative, the power
function y  x n (n an integer) is well defined for all values of x, both positive and
negative. So, when the power is an integer, we extend the domain to include all real x.
Let’s look at the case where the power n is a positive integer so that these power
functions include y  x, y  x 2, y  x 3, y  x 4, y  x 5, . . . , for all real x. Note
from Figure 2.61 that the graphs of these power functions fall into two groups: func-
tions with odd powers and functions with even powers.

y y

y = x5 y = x3 y = x6 y = x4

y = x2

y=x

FIGURE 2.61 (a) Odd powers (b) Even powers

Even Positive Integer Powers

We look at the even power functions: y  x 2, y  x 4, y  x 6, . . . , as shown in Fig-
ure 2.61(b). Their common characteristics are the following.
◆ All even power functions decrease at first (until x  0) and then increase as
x increases from left to right, so all are U-shaped with a turning point at the
origin.
◆ The higher the power n, the flatter the curve is as it passes through the origin.
◆ All the even powers are concave up everywhere, so they do not have a
point of inflection.
◆ Not only does each curve pass through the origin and the point 11, 1 2 , but
each one also passes through the point 11, 1 2 [because 11 2 n  1 if n is
an even integer].
◆ All even power functions are symmetric about the y-axis (the left and right
halves of the curves are mirror images.) (See Appendix D.)
Examine some of the even power functions on your own, using your function
grapher, to convince yourself that these properties are valid.

Odd Positive Integer Powers

Next, let’s examine the power functions with odd powers: y  x, y  x 3,
y  x 5, . . . , as shown in Figure 2.61(a). They have the following characteristics in
common.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 127

2.7 Power Functions 127

◆ They all are increasing everywhere as x increases from left to right.

◆ If n  1, all the odd power functions are concave down when x 0 and
concave up when x  0. This change in concavity at x  0 means that every
odd power function except the line y  x 1  x has a point of inflection at
the origin, where it is growing most slowly.
◆ The higher the power n, the flatter the curve is as it passes through the origin.
◆ Not only does each curve pass through the origin and the point 11, 1 2, but each
one also passes through the point 11, 1 2 [because 11 2 n  1 if n is an
odd integer].
◆ All odd power functions are symmetric about the origin; that is, the portion
of each curve in the third quadrant is the mirror image of the correspon-
ding portion in the first quadrant.
Examine some of the odd power functions on your own, using your function
grapher, to convince yourself that these properties are valid.

Problems
1. Which of the pairs of points shown can determine a 2. Match each formula for a power function with one of
power function of the form y  kx p and which the graphs. Explain the reasons for your decisions.
cannot. For those that do, sketch the graph of the y y
power function, indicate the sign of the coefficient (2)
(1) (4)
k, and tell whether the power p is less than 0, be- (3)
(5)
tween 0 and 1, or greater than 1.
(6)

(a) y (b) y
x x

(i) (ii)

y (8)

x x (9)
(7)

(c) y (d) y
x

(iii)

a. y  x 0.8 b. y  x 0.2 c. y  x 1.4

x x d. y  x 1.8
e. y  x 1.6
f. y  x 1.4
g. y  x 0.8
h. y  x 0.6
i. y  x 0.4
3. Identify which of the functions in parts (a)–(n) are
(e) y (f) y exponential functions, which are power functions,
and which are neither. Give the reasons for your
decisions.
a. f 1x 2  40x 1.05 b. f 1x 2  4011.052 x
c. f 1x 2  d. f 1x 2   2.4
1 3
11.4 2
x x x
x
e. f 1t2  5t 3.7 f. f 1q 2  1.09q  4.37
Gord.3896.02.pgs 4/24/03 9:26 AM Page 128

128 CHAPTER 2 Families of Functions

g. f 1t2  1210.35 2 t h. f 1t2  5 2t c. According to the ideal gas law, when a gas is kept
i. f 1s2  2s  3 j. f 1r2  43pr 3
2 at a constant temperature, the pressure P is in-
k. f 1z2  z z l. f 1x2  x x
. 3>5 versely proportional to the volume V.
m. f 1w2  w 3 2. w
n. f 1u2  711.62 2 u d. The force F of gravity between two objects is in-
versely proportional to the square of the distance
4. Match each formula with its corresponding table of d between them.
values. e. The square of the diameter d of the long bone in
a. y  4x 1.2 b. y  5x 0.8 c. y  411.2 2 x. the leg of many animals is proportional to the
i. cube of the length L of the bone.
x 2 3 4 5 6 f. The cube of the surface area S of many verte-
f (x) 8.71 12.04 15.16 18.12 20.96 brate mammals is proportional to the square of
their body mass m.
ii. g. The fourth power of the rate R at which air flows
x 1 2 3 4 5 into and out of the lungs of many vertebrate
g(x) 4.80 5.76 6.91 8.29 9.95 mammals is proportional to the cube of their
body mass m.
iii. h. The fourth power of the speed s at which many
x 2 4 6 8 10 mammals can trot is proportional to their body
h(x) 9.19 21.11 34.34 48.50 63.40 mass m.
i. The cube of the speed s at which most birds fly is
5. Data from three different functions are shown in proportional to the square of their body mass m.
the tables of values. One function is exponential, j. The square of the swimming speed s of most
one has the form y  ax 2, and one has the form species of fish is proportional to the length L of
y  bx 3. Which function is which? their bodies.
i. k. The fifth power of the radius r of the shock wave
x 3 3.5 4 4.5 5 after the explosion of a nuclear bomb is propor-
f (x) 28.8 39.2 51.2 64.8 80.0 tional to the square of the time t since the bomb
exploded.
ii. 7. Use the formula relating the weight of a large flying
x 3 3.5 4 4.5 5 bird to its wingspan to explain why a 15 pound
g(x) 4.39 5.01 5.71 6.51 7.42 turkey with a wingspan of about 2.5 feet can’t soar
like an eagle.
iii. 8. A full grown African vulture has a 9-foot wingspan.
x 3 3.5 4 4.5 5
Based on the model relating weight to wingspan, how
h(x) 10.80 17.15 25.60 36.45 50.00 much does a vulture weigh?
9. The largest known flying creature, with a wingspan of
6. For each relationship, (i) identify which of the quan- 40 feet, was the pterosaur that lived 65 million years
tities should be considered the independent variable ago. Assuming that the formula for birds also applies
and which the dependent variable; (ii) write an to flying dinosaurs, estimate the weight of an adult
equation expressing the dependent variable in terms pterosaur. What can you conclude from your answer?
of the independent variable to create a power func-
10. Find a formula expressing the volume V of a sphere
tion that represents the relationship; and (iii) sketch
as a function of its surface area S.
a rough graph of the function based on the value of
the power p. 11. Find the power function that passes through the
following pairs of points.
a. 11, 3 2 and 14, 6 2 b. 11, 32 and 14, 8 2
a. If a car is traveling at a constant rate, the distance
c. 11, 3 2 and 14, 102 d. 15, 202 and 16, 30 2
d that it travels is proportional to the time t that
e. 11, 10 2 and 14, 52 f. 12, 202 and 15, 8 2
it travels.
b. The distance d that an object falls under the in-
fluence of gravity is proportional to the square 12. Police sometimes use the formula s  230kd to
of the time t that it is falling. estimate the speed s in miles per hour that a car was
Gord.3896.02.pgs 4/24/03 9:26 AM Page 129

2.7 Power Functions 129

going if it left a set of skid marks d feet long. The 15. a. Use your function grapher to plot on the same
coefficient k depends on the road conditions (dry screen the graphs of the power functions y  x1>2,
or wet) and the type of pavement. For instance, x 1>3, and x 1>4 for the interval 0  x  0.2. Deter-
k  0.8 for dry concrete, k  0.4 for wet concrete, mine an appropriate range for y so that all powers
k  1.0 for dry tar, and k  0.5 for wet tar. will be distinguishable in the viewing rectangle.
a. A car left a set of skid marks 120 feet long on dry b. Plot the same graphs for 0  x  2 and deter-
concrete. How fast was it going? mine an appropriate range for y.
b. Suppose that the concrete pavement in part (a) c. Plot the same graphs for 0  x  20 and deter-
was wet. How fast was the car going? mine an appropriate range for y.
c. If the car in part (a) left skid marks 240 feet long, d. What happens if you use the interval
how fast was it going? 2  x  2?
d. Suppose that a car is going 50 mph on a dry tar 16. What happens to
surface when the driver slams on the brakes. a. x 3 as x S
? as x S 
?
How far will it skid? b. x 3 as x S
? as x S 
?
e. Suppose that the tar pavement in part (d) was c. x 1>3 as x S
? as x S 
?
wet. How far will the car skid? d. x 1>3 as x S
? as x S 
?
13. Scientists are actively investigating the potential of e. x 3 as x S
? as x S 
?
using windmills to generate electricity. They have f. x 3 as x S 0?
found that, for moderate wind speeds, the power P in 17. In 1990, 442.2 million prerecorded audio cassette
watts generated by a windmill is related to the wind tapes were sold, and 865.7 million CDs were sold in
speed v in miles per hour according to the equation the United States. In 1998, 158.5 million cassette
P  0.015v3. tapes were sold, and 1,124.3 million CDs were sold.
a. How much power is generated by a steady wind Assume for now that the patterns of sales for both
at 10 mph? items are power functions.
b. How much power is generated by a steady wind a. Find the equation for the number of cassette
at 20 mph? tapes sold as a power function of time.
c. Based on your results in parts (a) and (b), by b. Find the equation for the number of CDs sold as
what factor does doubling the wind speed in- a power function of time.
crease the power generated? c. If the trends in sales of both items were indeed
d. Compare the power generated by a steady wind power functions, find when the number of CDs
at 5 mph to that of a steady wind at 10 mph. sold overtook the number of cassette tapes sold.
Does doubling of the wind speed increase the 18. In the accompanying figure let R be the radius of
power generated by the same factor found in the Earth (about 3960 miles). Find an expression
part (c)? for the distance D to the horizon from a point at a
e. Suppose that a certain community has power height of H miles above the Earth’s surface. (Hint:
needs for an additional 250 kilowatts of electric-
ity and can anticipate winds on the average of
12 mph. How many windmills would be needed H D
to meet the added electric demand?
f. What wind speed would be needed to light a
100-watt light bulb?
R
14. a. Use your function grapher to plot on the same R
screen the graphs of the power functions y  x 2,
x 5, and x 8 for the interval 0.2  x  0.2. Deter-
mine an appropriate range for y so that all powers
will be distinguishable in the viewing rectangle.
b. Plot the same graphs for 2  x  2 and deter-
mine an appropriate range for y.
c. Plot the same graphs for 20  x  20 and de-
termine an appropriate range for y.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 130

130 CHAPTER 2 Families of Functions

Recall that, for a circle, any tangent line is perpendi- the value of D, so you might be tempted to ap-
cular to a radius.) proximate the distance D using the simpler for-
19. The observation deck of the Empire State Building in mula D  27920H  89 1H . Determine whether
New York is 1250 feet high. If you’re standing there, using this approximation is reasonable by completing
complete the phrase: “On a clear day, you can see. . . .” the following table comparing the estimated value for
(Hint: Use the formula you created in Problem 18.) this distance with the actual value.
20. Ultra high frequency (UHF) TV transmissions trav-
el along a line of sight from a transmitter as far as H D  89 2H D  2H2  7920H
the horizon. In the Chicago area, the UHF stations
0.1 mile
broadcast from a transmitter atop the 1454-foot
1  0.275 mile  14 mile2 high Sears tower. What is 1 mile
the greatest distance that someone could receive a
10 miles
UHF signal from the tower?
21. Suppose that a mast 250 feet (about 201 mile) high is 100 miles
being planned for the Sears tower to extend the
broadcast range of UHF stations. How much far- 26. a. Find, correct to three decimal places, all values of
ther would the signal extend? How much larger a x for which x 4  3x by graphing the two func-
receiving area would be covered? tions y  x 4 and y  3x. (Hint: Use different
windows to convince yourself that you have lo-
22. NASA’s space shuttles orbit the Earth at altitudes of
cated all points of intersection of the two curves.)
about 200 miles. Find the maximum line-of-sight
b. Repeat part (a) by creating the function y 
transmission distance from the shuttle to the sur-
x 4  3x and looking for all the points where y  0.
face of the Earth. Approximately how large a receiv-
ing area on the Earth is in range of this shuttle? 27. Consider the function f 1x 2  x 2 and let P be the
point on the curve where x  0, R be the point
23. Communications satellites orbit the Earth in geosyn-
where x  2, and Q be the midpoint where x  1.
chronous orbits (carefully chosen heights and veloc-
Find the slopes of the three line segments PQ, QR,
ities so that they appear to be permanently above a
and PR. How does the slope of PR compare to the
fixed point on the surface of the Earth as the Earth
slopes of the other two segments?
rotates). Suppose that such a satellite is in orbit at a
height of 23,000 miles above a point on the equator. 28. Repeat Problem 27, using the function g 1x 2  x 3.
The radius of the Earth is about R  3960 miles, so Does the relationship among the three slopes you
the distance around the equator is approximately found in Problem 27 also hold for g?
2pR  24,880 miles. Consequently, a point on the 29. Consider the function f 1x 2  x 2 and let P be the
equator is rotating at a velocity of about 1037 mph. point where x  a, Q be the point where
Find the orbital velocity of such a communication x  a  h, and R be the point where x  a  2h,
satellite in a geosynchronous orbit. for any quantity h  0. Find the slopes of the three
24. Explain why it isn’t possible to have a communica- line segments PQ, QR, and PR. Show that the slope
tions satellite whose signals cover a full half the of PR is the average of the other two slopes.
Earth’s surface. 30. Consider the sequence of values 105, 104, 103, 102,
25. Using R  3960 miles for the radius of the Earth, the and 101 and use it to provide a reason for defining
formula you found in Problem 18 for the line-of- 100  1. What about 101?
sight distance to the horizon from a height of H miles 31. By trial and error, determine the largest power of 10
is D 1H2  1H 2  2RH  1H 2  7920H . When that your calculator can handle. What is the small-
H is small, the term H 2 seemingly has little effect on est positive number?

Exercising Your Algebra Skills

Use the properties of exponents to evaluate each term 3. 84>3 4. 84>3
(do not use a calculator). Simplify.
1. 91>2 2. 91>2 5. x 4 . x 3 6. x 6 . x 8
Gord.3896.02.pgs 4/24/03 9:26 AM Page 131

2.8 Comparing Rates of Growth and Decay 131

r8 z12 x 3>4 a2>3

7. 8. 11. 12.
r4 z9 x 5>4 a5>3
9. x 3>4 . x 5>4 10. a2>3 . a5>3 a5>3
13. 2>3
a

2.8 Comparing Rates of Growth and Decay

Most of the families of functions that we’ve discussed in this chapter—linear, ex-
ponential, and logarithmic—are either strictly increasing or strictly decreasing. If
we restrict our attention to nonnegative values of x, power functions also are either
strictly increasing or strictly decreasing. Let’s summarize what we know so far.

Function Equation Behavior Graph

Linear y  mx  b Strictly increasing when m  0. y
The more positive the slope, the faster the rate of in-
crease. y = mx + b
m>0
Strictly decreasing when m 0.
The more negative the slope, the greater the rate of x
decrease.
y

y = mx + b
m<0

Exponential y  cx Strictly increasing when the growth factor c  1. y

(for c  0) The larger c is, the faster the function grows.
y = cx
Strictly decreasing when the decay factor 0 c 1. c>1
The smaller c is, the faster the function decays toward 1
zero. x
y
Exponential graphs are always concave up.
y = cx
c<1
1
x

y
Power y  xp Strictly increasing when p  0.
The larger p is, the faster the function grows beyond
x  1.
y = xp
If p  1, the graph is concave up—it grows more
0<p<1
and more rapidly.
y = xp
If 0 p 1, the graph is concave down—it grows p>1
more and more slowly. x
Gord.3896.02.pgs 4/24/03 9:26 AM Page 132

132 CHAPTER 2 Families of Functions

Strictly decreasing for x  0 when p 0. y

The more negative p is, the faster the function decays

toward zero. y = xp
p<0
If p 0, the graph is concave up, for x  0.

Logarithmic y  log x Strictly increasing. y

x0 Logarithmic graphs are always concave down.
y = log x

x
1

This summary of information compares the growth or decay rate of one func-
tion in a family to that of other functions in the same family. In this section, we
look at two other ways to compare rates of growth. At a local level, we look at how
fast a single function is growing or decaying at different points. At a global level,
we look at how quickly functions in one family grow or decay compared to how
quickly functions in a different family grow or decay. In particular, we want to an-
swer two questions: (1) which family of functions grows fastest? and (2) which
family of functions decays to zero fastest?

Exponential Versus Linear Functions: Which Grow Faster?

We saw in Section 2.4 that any exponential growth function y  kcx with c  1
will eventually grow faster than any linear function with positive slope. The rea-
son is that the multiplicative effect of the growth factor c in an exponential func-
tion is greater than the additive effect of the slope in a linear function. Similarly,
any exponential decay function will eventually decrease more slowly than any
linear function with a negative slope.

Exponential Versus Power Functions: Which Grows Faster?

Power functions of the form y  x p, p  1 and exponential growth functions of
the form y  c x, c  1 both grow rapidly as x increases. But, do they grow at
roughly the same rate, or does one grow much faster than the other?
Consider y  2x and y  x 4 for x  0. We know that every power function of
the form y  x p with p  0 passes through the origin and that every exponential
curve of the form y  cx crosses the vertical axis at y  1. So let’s begin by com-
paring these two functions for small values of x. The local, or close-up, view, in Fig-
ure 2.62 shows that, between x  0 and x  1, the graph of y  2x is above the
graph of y  x 4 but that the power function seems to be growing more rapidly. If
we extend the interval somewhat, we find that by x  2 the power function has
surpassed the exponential function. (Where does that happen?) On a somewhat
larger scale, Figure 2.63 reveals that the power function continues to pull away
from the exponential function.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 133

2.8 Comparing Rates of Growth and Decay 133

y y

30
2 y = x4
y= 2x

1
y = x4
y = 2x

x
0 1 x
0 3
FIGURE 2.62 FIGURE 2.63

However, in Figure 2.64, which shows the interval from x  0 to x  20, the ex-
ponential curve has again overtaken the power curve. (Where does that happen?)
The still larger view from x  0 to x  25 in Figure 2.65 shows that, for large x-val-
ues, y  x 4 is insignificant compared to y  2x. In fact, y  2x is growing so much
faster than y  x 4 that its graph appears almost vertical in comparison to the rela-
tively slow growth of y  x 4. Verify this comparison numerically by trying several
different values of x—say, x  1, x  2, x  10, and x  50 (but don’t go too far
because you may exceed your calculator’s capacity).

y
y
250,000

y = 2x 2,000,000
y = 2x

y = x4

y = x4

x
x 0 25
FIGURE 2.64 20 FIGURE 2.65

Think About This Use your function grapher to find, correct to two decimal places, all the points
where y  2x and y  x 4 intersect. ❐

This behavior pattern is typical of any power function y  x p, with p  1

compared to any exponential function y  c x, for c  1. Although the exponen-
tial function starts more slowly than the power function for small values of x, the
exponential function eventually dominates y  x p for any value of p  1, and so
it always wins the race toward infinity.

Think About This Plot y  3x and y  x 5 for 0  x  20 with 0  y  300,000 to see where the ex-
ponential function overtakes the power function. ❐

You have already seen that a positive constant multiple doesn’t change the overall
shape or behavior of a function. For instance, the power function y  5000x 4 has the
same shape as the power function y  x 4, but it grows more rapidly because the first
Gord.3896.02.pgs 4/24/03 9:26 AM Page 134

134 CHAPTER 2 Families of Functions

is 5000 times larger for any x value. We’ve already shown that the exponential function
y  2x eventually overtakes the power function y  x 4. It also eventually overtakes the
power function y  5000x 4; it just takes longer. The only question is: Where does that
happen? In the long run, the exponential function invariably wins the race to infinity.
E XAMPLE 1
Estimate the point x where f 1x 2  1.05x finally overtakes g 1x 2  x 10.
Solution We know that the power function g 1x 2  x 10 grows very rapidly and that the
exponential function f 1x 2  1.05x has a fairly small growth factor of 1.05. Let’s look at
their respective function values numerically for different values of x.

x f (x)  (1.05)x g (x)  x 10

10 1.62889 1010

100 131.501 10010  1102 2 10  1020

1000 1.5463  1021 100010  1103 2 10  1030

10,000 7.816  10211 10,00010  1104 2 10  1040

From this comparison, it is evident that the exponential function has overtaken the
power function sometime after x  1000 but long before x  10,000, where f 1x 2 
11.05 2 x has far exceeded the value of g 1x 2  x 10. Suppose we narrow our search by try-
ing a few additional values of x.

x f (x)  (1.05)x g (x)  x 10

2000 2.3911  1042 1.024  1033

1500 6.0806  1031 5.7665  1031

1499 5.7911  1031 5.7282  1031

1498 5.5153  1031 5.6901  1031

We conclude that the exponential function f 1x 2  11.052 x finally overtakes the

power function g 1x 2  x 10 just before x  1499, as illustrated in Figure 2.66. If we zoom
in, either numerically on the table or geometrically on the graph, we find that the point of
intersection of the two functions occurs near x  1498.718 and y  5.71169  1031.
y

6 × 10 31

y = x10
y = 1.05 x

x
FIGURE 2.66 0 500 1000 1500

◆
Gord.3896.02.pgs 4/24/03 9:26 AM Page 135

2.8 Comparing Rates of Growth and Decay 135

Now let’s compare a decaying exponential function and a power function

with a negative exponent. Consider y  1 12 2 x  10.5 2 x and y  x 2  1>x 2. Both
graphs eventually approach the x-axis as a horizontal asymptote, but which one
approaches the x-axis faster? We could compare them graphically or numerical-
ly by trying several large values of x, such as x  100 or x  1000. Alternatively,
we can reason that, because 2x is eventually larger than x 2, we know that
10.5 2 x  1 12 2 x  1>2x is eventually smaller than x 2  1>x 2. So the graph of
y  1 12 2 x eventually falls below the graph of y  x 2, as shown in Figure 2.67.
y

y = x –2

y = 2 –x

x
FIGURE 2.67 0

This behavior pattern is typical. All decaying exponential functions invariably

approach 0 faster than any power function with a negative exponent as x S
. A
power function could begin dropping at a faster rate when compared to an expo-
nential function; for instance, compare y  x 100 with y  0.9x. However, as
x S
, the exponential function eventually decays faster than the power function
to win the race toward 0. The only question is: When does the decaying exponen-
tial function overtake the decaying power function on the way to zero? As we have
demonstrated, this point of intersection can be approximated with any desired de-
gree of accuracy by using either numerical or graphical methods.

Logarithmic Functions Versus

Power Functions: Which Grows Faster?
We know that a logarithmic function and a power function with power 0 p 1
both increase and are concave down. A table of values for f 1x 2  log x reveals that
the logarithm grows very slowly as x increases beyond x  1. Figure 2.68 shows the
y

y = x 1/3

y = log x

x
(0, 0) (1, 0)
FIGURE 2.68
Gord.3896.02.pgs 4/24/03 9:26 AM Page 136

136 CHAPTER 2 Families of Functions

graphs of y  log x and y  x 1>3. In fact, log x grows more slowly than any positive
power of x. In the long run, any power function with 0 p 1 beats the logarith-
mic function in the race toward infinity.

Think About This Compare the behavior of different power functions (for example, y  x 1>2 or
y  x1>10 2 to y  log x on your function grapher to convince yourself of how
slowly the log function grows. ❐

Furthermore, both logarithmic functions and power functions with power

0 p 1 grow more slowly than linear functions with slope m  0. The reason
is that the linear function y  x is a power function with p  1, which grows
faster than any power function with a smaller power p.
In summary, we have the following facts:

Comparisons When x is Large

Concave Up Growth Functions
Power functions with power p  1 grow faster than linear functions with
slope m  0.
Exponential functions with growth factor c  1 grow faster than power
functions with power p  1.

Concave Down Growth Functions

Power functions with power 0 p 1 grow more slowly than linear
functions with slope m  0.
Logarithmic functions grow more slowly than power functions with
power 0 p 1.

Decay Functions
Exponential functions with decay factor 0 c 1 decay more rapidly
than power functions with power p 0.

Average Rate of Growth

We have often described an increasing, concave up function as “increasing faster
and faster” or “increasing at an increasing rate,” although we never precisely de-
fined what this means. We now formalize this concept by building on what we
know about lines. One of the main characteristics of a linear function is that it
grows (or decays) at a constant rate. That is, for each fixed increase (say, x) in the
independent variable x, the line rises (or falls) the same amount y no matter what
point on the line we use, as shown in Figure 2.69. The constant ratio y>x is the
slope of the line.
Now let’s try this with an exponential growth function y  f 1x 2  kcx. At dif-
ferent points on the curve, move the same horizontal distance x to the right and
determine the corresponding vertical change y, as shown in Figure 2.70. At point
P, there is a relatively small change in y; at point Q, the corresponding change in y
is somewhat larger; and by the time we get to point S, the corresponding change in
y is considerably larger.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 137

2.8 Comparing Rates of Growth and Decay 137

y y

∆y
∆x S
∆y ∆x
∆x
∆y R
∆x Q ∆x
∆y P ∆x
∆x ∆x

x x
FIGURE 2.69 FIGURE 2.70 0

For instance, suppose that y  f 1x 2  1.2x and that we take steps of size
x  0.5. We start at x  0, where y  1.20  1. When we move x  0.5 to the
right, we get to x  0.5 at a height of 1.20.5  1.095. The change in height is
y  1.095  1  0.095, as shown in Figure 2.71.
y

y = 1.2x

1.5774
∆y = 1.5774 – 1.44 = 0.1374
1.44
∆x = 0.5

1.095
1 ∆y = 1.095 – 1 = 0.095
∆x = 0.5

x
FIGURE 2.71 0 x = 0.5 x=2 x = 2.5

We now repeat this process, starting at x  2 where y  f 12 2  1.22  1.44.

We again move x  0.5 to the right, getting to the point x  2.5 and
y  f 12.5 2  1.22.5  1.5774. The corresponding change in height is now
y  1.5774  1.44  0.1374, as shown in Figure 2.71. Thus the same size step to
the right has resulted in a considerably larger increase in the change in height of the
exponential growth function.

Think About This Repeat the preceding argument by starting from the point x  3 and moving to
the right by x  0.5. How does the change in y compare to the values of y that
we found? ❐

To measure how rapidly this, or any other function y  f 1x 2 is increasing (or

decreasing), we consider the ratio y>x, called the average rate of change of the
function over an interval, which we discussed briefly in Section 1.1. We have
y f 1x2 2  f 1x1 2
Average rate of change of f from x1 to x2   .
x x2  x1
In the special case where the function is linear, this ratio is simply the slope of the line.
E XAMPLE 1
Find the average rate of change of the linear function f 1x2  3x  5 between x  1
and x  9.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 138

138 CHAPTER 2 Families of Functions

Solution The average rate of change is

f 19 2  f 11 2 13 . 9  5 2  13 . 1  5 2 32  8 24
    3,
x 91 8 8
which is the slope of the line.
◆
E XAMPLE 2
Find the average rate of change of the exponential function f 1x 2  1.2x (a) between
x  0 and x  0.5 and (b) between x  2 and x  2.5.
Solution

a. The average rate of change of f 1x 2  1.2x between x  0 and x  0.5 is

f 10.5 2  f 10 2 1.095  1
  0.19.
x 0.5
b. The average rate of change of f 1x 2  1.2x between x  2 and x  2.5 is
f 12.5 2  f 12 2 1.5774  1.44 0.13744
   0.27488.
x 0.5 0.5
◆
Note how the average rate of change between x  2 and 2.5 is considerably
larger than that between x  0 and 0.5. The function is growing ever faster as we
move to the right.
Instead of thinking of the average rate of change of a function y  f 1x 2 from
x1 to x2 , we can also think of it as the average rate of change from any point x  x0
to x  x0  x. We then have
f 1x0  x 2  f 1x0 2
Average rate of change  ,
x
as illustrated in Figure 2.72. Geometrically, the average rate of change is the slope
of the line segment through the two points 1x0, f 1x0 2 2 and 1x0  x, f 1x0  x 2 2.
The more positive this slope is, the faster the curve is increasing. Thus, for an in-
creasing, concave up function, the average rate of change increases as we move
from left to right.
y

y = f (x)

f (x 0 + ∆x) – f (x 0) f (x 0 + ∆x)
m=
∆x

∆y = f (x 0 + ∆x) – f (x 0)

f (x 0)
∆x

x
FIGURE 2.72 x0 x 0 + ∆x

If f 1x 2 is a decreasing function, the average rate of change will be negative, just

as the slope is negative for a decreasing linear function.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 139

2.8 Comparing Rates of Growth and Decay 139

Think About This What is the average rate of change of f 1x2  x 1  1>x from x  1 to x  1.02? ❐

E XAMPLE 3
Find the average rate of change of f 1x 2  log x between x  2 and x  2.4.
Solution In going from x  2 to x  2.4, we have a step of x  0.4, as depicted in Fig-
ure 2.73. Therefore the average rate of change between x  2 and x  2.4 is
f 12.4 2  f 12 2 log 2.4  log 2
  0.19795.
x 0.4
y
0.380 – 0.301
m=
0.4 log 2.4 = 0.380

∆y = 0.380 – 0.301 = 0.079

log 2 = 0.301
∆x = 0.4
y = log x

x
FIGURE 2.73 0 2 2.4

◆
Problems
1. Use your function grapher to graph y  x 3 and a. f 1x 2  4x  9 between x  2 and x  5
y  2x. Determine appropriate x- and y-scales to b. f 1x 2  4x  9 between x  2 and x  3
obtain the diagrams shown below. c. f 1x 2  3x  4 between x  2 and x  5
2. a. For what values of x is 2x  x 2? d. f 1x 2  3x  4 between x  2 and x  3
b. For what values of x is 3x  x 3? e. 4x  3y  12 between x  2 and x  5
c. For what values of x is 4x  x 4? 6. Prove that the average rate of change for any linear
3. Use your function grapher to estimate when y  function f 1x 2  mx  b on any interval from x1 to
10.62 x overtakes y  x 6 as they both decay to zero. x2 is equal to the slope m.
4. Estimate where f 1x2  x 0.15 finally overtakes g 1x 2  7. Find the average rate of change of f 1x2  x 2 (a) be-
log x on their “turtle versus snail” race toward infinity. tween x  0 and x  1, (b) between x  0 and
5. For each linear function, find the average rate of x  2, and (c) between x  1 and x  2. Put them
change on the indicated intervals. in ascending numerical order.

(a) y (b) y (c) y

y = x3
y = 2x
y = 2x

y = x3

y = 2x
y = x3
x x
Gord.3896.02.pgs 4/24/03 9:26 AM Page 140

140 CHAPTER 2 Families of Functions

8. Consider the function f 1x2  x 2. the three line segments PQ, QR, and PR. List
a. Find the average rate of change of f between these line segments in the order of increasing
x  0 and 0.01. slope (smallest to largest).
b. Find the average rate of change of f between
x  1 and 1.01. y
c. Find the average rate of change of f between
x  2 and 2.01. R

d. Based on your results from parts (a)–(c), can you

predict the average rate of change of f between Q
x  3 and 3.01? between x  4 and 4.01?
P
e. What happens to your answers in parts (a)–(c) if x
x  0.001 instead of 0.01?
9. a. In parts (b), (c), and (d), you are asked to find
the average rate of change of f 1x 2  1x be- b. Repeat part (a) if the function is increasing and
tween x  0 and 1, between 1 and 2, and be- concave down instead.
tween 0 and 2. Before calculating these values, 12. Consider the function f shown and the point A at
predict the numerical order, from smallest to x  a on the curve. Determine the point:
largest, of these three quantities.
b. Find the average rate of change of f between y
x  0 and 1.
c. Find the average rate of change of f between
x  1 and 2.
d. Find the average rate of change of f between
x  0 and 2.
10. The functions y  x 2 and y  2x intersect at x  2 A
and at x  4; the functions y  x 3 and y  3x in- x
tersect at x  3 and at x  2.478. In general, for x=a

x  0 the graphs of f 1x2  x p and g 1x 2  px inter-

sect at two points except for one specific value of p a. B at x  b giving the interval from a to b over
(with p  1) for which the curves intersect at only which the change in f is least.
one point. Use your function grapher and trial and b. C at x  c giving the interval from a to c over
error to locate the one special value of p (accurate which the change in f is greatest.
to two decimal places) for which the curves y  x p c. D at x  d giving the interval from a to d over
and y  px intersect at only one point. which the average rate of change in f is least.
11. a. Use the three points P, Q, and R shown on the d. E at x  e giving the interval from a to d over
accompanying graph of y  f 1x2 to determine which the average rate of change in f is greatest.

2.9 Inverse Functions

Countries using the metric system report temperatures in degrees Celsius. Thus an
American visiting Canada who wants to know the temperature in degrees Fahren-
heit must be able to convert the Celsius readings to Fahrenheit readings by using
the formula
9
F  C  32.
5
Gord.3896.02.pgs 4/24/03 9:26 AM Page 141

2.9 Inverse Functions 141

In this relationship the Fahrenheit measurement is a function of the Celsius

measurement. Canadian visitors to the United States face the reverse problem: They
must convert Fahrenheit readings to Celsius readings. They can do so by solving the
preceding formula algebraically for C as a function of F, getting the related function

1F  32 2 .
5
C
9
These two functions have the effect of undoing each other. For that reason, they are
called inverse functions.
In general, we write the inverse of a function f as f 1 and read it as “f inverse.”
The inverse of a function f is a function that reverses or undoes f. The two func-
tions relating F and C, which give temperature conversions between the two sys-
tems of measurements, represent a pair of inverse functions.
Suppose that we have a function y  f 1x2 that represents some quantity or
process of interest to us. Typically, we can ask two types of predictive questions.
The first question is: Determine the value of y corresponding to a particular
value of x. All we need do is substitute the value of x in the expression for the
function. The second question is: Determine when the quantity achieves a par-
ticular level—that is, find the value of the independent variable x that produces a
given value for the dependent variable y. Here we must undo the given function,
which is what the inverse function is all about. To be able to answer this question
requires the existence of an inverse function and the ability either to find its
equation algebraically or to estimate its values graphically or numerically.
When a function is given in a table, finding the inverse function is trivial, as we
demonstrate in Example 1.
E XAMPLE 1
Table 2.3 gives the average distance D from the sun (in millions of miles) for each of the
planets as a function of its average speed S (in miles per hour). Find the inverse function.
Solution For this function, we think of the average speed S of each planet as the inde-
pendent variable and the average distance D from the sun as the dependent variable, so

TABLE 2.3

Planet Speed Distance D  f (S)

Mercury 4,110 36.0
Venus 7,671 67.2
Earth 10,605 92.9
Mars 16,153 141.5
Jupiter 55,171 483.3
Saturn 101,164 886.2
Uranus 203,459 1782.3
Neptune 318,790 2792.6
Pluto 418,744 3668.2
Gord.3896.02.pgs 4/24/03 9:26 AM Page 142

142 CHAPTER 2 Families of Functions

D  f 1S2. The corresponding inverse function f 1 simply reverses the role of the two vari-
ables. Thus the average distance D from the sun of each planet is now the independent vari-
able and the average speed S of each planet is the dependent variable. We write S  f 1 1D 2.
Table 2.4 simply interchanges the columns for S and D from Table 2.3, as shown.

TABLE 2.4

Planet Distance Speed S  f 1(D)

Mercury 36.0 4,110
Venus 67.2 7,671
Earth 92.9 10,605
Mars 141.5 16,153
Jupiter 483.3 55,171
Saturn 886.2 101,164
Uranus 1782.3 203,459
Neptune 2792.6 318,790
Pluto 3668.2 418,744

◆
In general, if 1a, b2 is a point on the graph of a function y  f 1x 2, then 1b, a2
must be a point on the inverse function x  f 1 1y2. Thus, finding the inverse for
any function given in a table is trivial, assuming that the inverse function exists.
For relatively simple functions given by formulas, determining the inverse func-
tion f 1 for a given function f is straightforward. We just solve the original expres-
sion algebraically for the independent variable in terms of the dependent variable,
as illustrated in Example 2.

E XAMPLE 2
Find the inverse function to the Celsius to Fahrenheit conversion function F  95 C  32.
Solution To find the inverse function, we first subtract 32 from both sides of the equa-
tion and get
9
C  F  32.
5
Multiplying both sides of the equation by 59 gives

1F  32 2 .
5
C
9
So C  59 1F  32 2 is the inverse function.
◆
Many functions, however, do not have an inverse, as we show later. Even when
a function f does have an inverse, it is not always possible to find a formula for the
inverse f 1 algebraically. Fortunately, as Examples 3 and 4 indicate, most of the
Gord.3896.02.pgs 4/24/03 9:26 AM Page 143

2.9 Inverse Functions 143

common functions that we have discussed do have inverses and we can find ex-
pressions for them algebraically.

E XAMPLE 3
a. Find the inverse function for the exponential function P1t2  12.9411.029 2 t that
models the population of Florida, where t is the number of years since 1990 and P is
the population in millions.
b. What does this inverse function tell us?
c. What are reasonable values for the domain and range of this inverse function?
Solution

a. Since P  12.9411.0292 t, we have to solve for t as a function of P. First, we divide

both sides by 12.94:

 11.029 2 t.
P
12.94
To solve for t, we take logs of both sides of this equation:

log a b  log11.0292 t  t log11.0292.

P
12.94
Therefore

1P2.
log1P>12.942 1
t f
log11.0292
We can stop here or we can use properties of logarithms to simplify this expression.
Using the property that the logarithm of a quotient is the difference of the logs, we get
log1P2  log112.94 2
tf 1
1P2 
log11.0292
or, using the approximate values of log 12.94 and log 1.029, we have
log1P2  1.1119
t  80.545 log1P2  89.669.
0.0124
This logarithmic function is the inverse to the function modeling Florida’s population.
b. The inverse function gives the number of years since 1990 (the value of t) that it takes
for the population of Florida to reach any given level P.
c. For the inverse function, the independent variable is the value of the population P
and the dependent variable is the number of years t since 1990. Therefore the domain
of the inverse function consists of all reasonable values for P—say, from 5 million to
a maximum of 50 million people. The range consists of all corresponding values of t.
To find these values, we use the equation for the inverse function that we obtained in
part a. If we substitute P  5 into the preceding equation, we get
tf 1
15 2  80.545 log15 2  89.669  33.4.
According to this model, the population of Florida was 5 million about 33 years be-
fore 1990, or in 1957. Similarly, substituting P  50 yields
tf 1
1502  80.545 log150 2  89.669  47.2,
Gord.3896.02.pgs 4/24/03 9:26 AM Page 144

144 CHAPTER 2 Families of Functions

so the population of Florida will reach 50 million about 47 years after 1990, or in
early 2037. Consequently, a reasonable range for the inverse function is t from 33.4
to 47.2 years, which corresponds to about 1957 to about 2037.
◆
In general, the inverse of any exponential function will be a logarithmic func-
tion (and vice versa) because the logarithm undoes the exponential function.

E XAMPLE 4
We have shown that the power function W  f 1S2  0.15S 9>4 can be used to model the
weight W of large birds as a function of their wingspan S.
a. Find the inverse function for f.
b. What does the inverse function tell us?
c. What wingspan would allow a 15 pound turkey to fly?
d. What is a realistic domain and range for the inverse function?
Solution

a. We have
W  f 1S2  0.15S 9>4.
To solve for S, we first divide both sides of the equation by 0.15:
S 9>4  W>0.15  6.667 W.
To find S, we must undo the 49 power, so we raise both sides of this equation to the 49
power:
1S 9>4 2 4>9  S  16.667W 2 4>9,
using properties of exponents. Therefore the inverse function is
S  16.667 2 4>9 W 4>9  2.324 W 4>9.

b. The inverse function gives the wingspan S in feet needed to support in flight a bird
that weighs W pounds.
c. If a turkey weighs 15 pounds, this formula predicts that, in order for the turkey to fly,
it would need a wingspan of
S  2.324115 2 4>9  7.7436 feet.
Since this is about three times the actual wingspan of a turkey, it isn’t able to fly.
d. For the original function f, the independent variable is the wingspan S and the de-
pendent variable is the weight W of a bird. For the inverse function f 1, the inde-
pendent variable is the weight W and the dependent variable is the wingspan S. If we
consider reasonably large birds that weigh between 2 pounds and 20 pounds, say, the
domain for the inverse function f 1 will be between W  2 and W  20. To find the
corresponding range, we use the equation for S we obtained in part (a):
f 12 2  2.32412 2 4>9  3.16;
1

f 1 120 2  2.324120 2 4>9  8.80.

Thus the range of the inverse function is from about 3 feet to almost 9 feet.
◆
Gord.3896.02.pgs 4/24/03 9:26 AM Page 145

2.9 Inverse Functions 145

Note that the inverse function to the power function W  f 1S2  0.15S 9>4
turned out to be another power function, S  f 1 1W 2  2.324W 4>9. In general,
the inverse of any power function, if it exists, is a power function.
Further, note how the powers of the two functions W  f 1S2  0.15S 9>4 and
S  f 1 1W 2  2.324W 4>9 compare algebraically: Each power is the reciprocal of
the other. This result is analogous to what happens with y  x 3: We solve for x by ex-
3
tracting the cube root to get x  y1>3  1y . Thus the inverse function for y 
f 1x 2  x is x  y  f 1 y2 . Similarly, if we need to find the value of x for which
3 1>3 1

175
x 175  200, then x  2200  2001>175  1.03074.
If we need to determine the value of the base b for which
1 1>2004
b 2004 1 2004 1
 , then b  B 2  a b  0.99965.
2 2

To verify these results, just calculate 11.030742 175 and 10.999652 2004. Although these
numbers may seem bizarre to you, we perform such operations routinely in later
chapters because they allow us to answer some interesting questions.
Examples 3 and 4 illustrate two important, though different, situations. To
extract an unknown variable that appears as the base in a power function
y  x p, take the corresponding pth root of both sides of the equation. To extract
an unknown variable that appears in the exponent of an exponential function
y  c x, take the logarithm of both sides of the equation. We summarize this in-
formation as follows. Be sure that you understand the difference between these
two situations.

To solve for the variable x from a power function y  x p, we extract the

pth root. That is,
if y  f 1x 2  x p, then x  f 1
1 y2  y1>p .
If p is a fraction whose denominator is even, we must have x  0 as the
domain for f.
The inverse of a power function is another power function.
To solve for the variable x from an exponential function y  c x, we take
logarithms. That is, for c  10,
if y  f 1x 2  10x, then x  f 1
1 y2  log y.

The inverse of an exponential function is a logarithmic function and vice

versa.

Extracting the appropriate root from the preceding functions was quite sim-
ple. Unfortunately, complications can arise, depending on the behavior of the
function, as we illustrate in Example 5.

E XAMPLE 5
The function y  f 1t2  16t 2  48t  6 models the height y of a ball thrown
straight up as a function of time t. Find how long it takes the ball to reach a height of
35 feet.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 146

146 CHAPTER 2 Families of Functions

Solution For the desired height of y  35 feet, we want to find the corresponding
value of t, so we have to solve the equation
16t 2  48t  6  35.
Figure 2.74 shows the graph of the function f 1t2. Note that the horizontal line corre-
sponding to the height y  35 crosses the curve twice, once when t  0.84 seconds (on
the way up) and again when t  2.16 seconds (on the way down).
y

y = —16t 2 + 48t + 6

y = 35

t
FIGURE 2.74 0 0.84 2.16

◆
Example 5 illustrates the fact that not every function has an inverse. In this case,
two different values of t correspond to a height of 35 feet. Consequently, the func-
tion y  f 1t2  16t 2  48t  6 does not have an inverse because t is not a func-
tion of y—at least one value of y 1 y  35 2 leads to two different values of t. That is,
we can’t undo the effects of the original function f uniquely. We discuss this situa-
tion in more detail later.
We have said that a function f and its inverse f 1 undo each other. To show
what this means, suppose that we start with a number x in the domain of f. The
function f carries x into the corresponding value of y  f 1x 2 in the range of f, as il-
lustrated in Figure 2.75. Similarly, if we start with any value of y in the range of f ,
then f 1 maps y into the value of x associated with it so that f 1 1 y2  x. That is,
xf 1
1 y2 and y  f 1x 2 .
f

x y = f (x)
y
x = f –1(y)
FIGURE 2.75 f –1

Again, consider Figure 2.75. For the original function f, the circle on the left rep-
resents the domain of f (the allowable values of x) and the circle on the right rep-
resents the range of f (the corresponding values of y). For the inverse function
f 1, the circle on the right represents the domain of f 1 (the allowable values of
y) and the circle on the left represents the range of f 1 (the corresponding values
of x).
In particular, suppose that x0 is any specific value of x and that y0  f 1x0 2 , so
that f transforms x0 into y0 . If we follow this by applying f 1 to y 0 , we get
f 1 1y0 2  x0 , returning to the original x0 value. That is, f 1 undoes the effect of f
on any value x0 . (We consider the idea of applying one function after another in
detail in Section 4.6.)
Gord.3896.02.pgs 4/24/03 9:26 AM Page 147

2.9 Inverse Functions 147

Similarly, if y0 is any specific value of y and x0  f 1

1 y0 2, then f 1x0 2  y0 .
That is, f undoes the effect of f 1 on any value y0 .
We can represent these ideas pictorially as follows.

1
If f and f are inverse functions,
y  f 1x 2 1 y2;
f " f 1 " 1
for any x: x xf

1x 2 y  f 1 x 2.
f 1 " 1 f "
for any y: y xf

For instance, the exponential function and the logarithmic function are inverses of
each other, which simply restates the relationships
log110x 2  x and 10log y  y.

Think About This Consider the model for the population of Florida P  f 1t2  12.9411.0292 t and its
inverse
log1P2  log112.94 2
tf 1
1P2 
log11.0292

(from Example 3.) Select any year—say, 1996 when t  6 —and verify that if
P  f 1t2, then t  f 1 1P2. ❐

Determining the Existence of an Inverse Function

Based on the results of Example 5 on the height of a ball thrown vertically upward,
we know that not every function f has an inverse f 1. In particular, no power func-
tion with an even power can have an inverse. To see why, consider the function
f 1x 2  x 2, whose graph is the parabola shown in Figure 2.76. The inverse func-
tion, if it exists, should give the value of x that produces any given height y. Sup-
pose that we start at a height of y  25. Clearly, there are two different points on
the parabola at a height of 25, one corresponding to x  5 and the other corre-
sponding to x  5. We cannot reverse the process uniquely to get only one value
of x for a given y, so y is not a function of x and f 1x 2  x 2 does not have an inverse.
y

25

y = x2

x
–5 0 5
FIGURE 2.76

However, we can restrict the domain of this function to produce a partial in-
verse. Suppose that we limit our attention to nonnegative values of x and consider
the function g 1x 2  x 2, for x  0. In this case, if we take any positive value for y
Gord.3896.02.pgs 4/24/03 9:26 AM Page 148

148 CHAPTER 2 Families of Functions

(say, y  25,) we can undo the function g by taking the square root and accepting
only the nonnegative value, x  5. Thus the function g 1x 2  x 2 with domain re-
stricted to x  0 has an inverse, g 1 1 y2  1y . Alternatively, if we restrict our do-
main to values of x  0, we could also uniquely undo the results of squaring and
get the inverse h1 1 y2  1y .
Is there a simple criterion to determine whether a function f has an inverse?
Definitely! Again, we know that the function f 1x 2  x 2 has an inverse if we re-
strict its domain to either x  0 or x  0. It does not have an inverse if we allow
the domain to include both positive and negative values for x. When we restrict
the domain to x  0, we consider only the right-hand side of the parabola where
the function is strictly increasing. When we restrict the domain to x  0, we con-
sider only the left-hand side of the parabola where the function is strictly decreas-
ing. In both instances, the restricted function has an inverse. On the one hand,
when we allow both positive and negative values for x, the function first decreases
and then increases and thus does not have an inverse. On the other hand, the
function h1x 2  x 3 has an inverse x  h1 1 y2  1 3
y without any restrictions on
x. We also know that this function is strictly increasing for all values of x, as shown
in Figure 2.77.

y
y=x

y = x3

3
y = √x

FIGURE 2.77

These observations suggest a simple criterion for functions to have an inverse:

The function must be either strictly increasing or strictly decreasing. We call such a
function monotonic. Compare the two functions shown in Figure 2.78. The one on
the left is strictly increasing and, for any desired height y, we can undo the effect of
the function and come back to a unique x that produced that particular y. In con-
trast, the one on the right increases and decreases over different intervals. There are
some heights, such as y3 and y4 , that occur several times, so different x-values cor-
respond to the same y-value. Thus it is not possible to find the unique x that pro-
duces a particular y-value.

Increasing and decreasing

Strictly increasing on different intervals
y y

y2 y4

x1 y3
x2 x x

FIGURE 2.78 y1
Gord.3896.02.pgs 4/24/03 9:26 AM Page 149

2.9 Inverse Functions 149

In summary, all linear, exponential, and logarithmic functions have inverses

because they are either strictly increasing or strictly decreasing. Any power func-
tion that is restricted to x  0 also has an inverse.
There is an alternative criterion analogous to the Vertical Line Test that we dis-
cussed in Section 1.4 for determining whether a curve represents a function. Recall
that a curve represents a function if every vertical line crosses the curve at most
once. In other words, for any value of x, one and only one value of y corresponds to
that x. In an analogous way, we can use a Horizontal Line Test to determine
whether a function has an inverse: If every horizontal line crosses the curve at most
once, a function f has an inverse f 1. In other words, for any height y, one and only
one value of x corresponds to that height.

Behavior of the Inverse Function

At times, expressing both f and f 1 as functions of the same variable is desirable so
that their graphs can be drawn on the same set of axes and their behaviors com-
pared easily. For instance, we previously showed that
y  f 1x 2  x 3 and x  f 1
1 y2  2
3
y
are inverse functions. Instead, let’s write these two related functions so that both
are functions of the same independent variable, x:
y  f 1x 2  x 3 and 1
1x 2  2
3
yf x.
Figure 2.77 displayed the graphs of both f and f 1. Notice that the graphs of the
function f 1x 2  x 3 and its inverse f 1 1x 2  1x  x 1>3 are mirror images of each
3

other about the line y  x. (Note that, if we didn’t interchange x and y for the in-
3
verse function, the two formulas y  x 3 and x  1y  y1>3 would represent iden-
tical curves, so that we would see only one curve.)
Similarly, the exponential function y  10x and the logarithmic function
y  log x are inverse functions. If y  f 1x 2  10x, we can solve for x by taking the
logarithm of both sides to get
log y  log 10x  x  f 1
1y2 .
We now interchange the variables so that x is also the independent variable for the
inverse function and write y  f 1 1x 2  log x. Again, note that the graphs of
these two functions are mirror images of each other about the line y  x, as shown
in Figure 2.79.
y
y = 10 x

(b, a) y=x

(1, 10)

(10, 1) (a, b)
(0, 1) y = log x
x
(1, 0)

FIGURE 2.79
Gord.3896.02.pgs 4/24/03 9:26 AM Page 150

150 CHAPTER 2 Families of Functions

In general, the graphs of a function f and its inverse f 1 are always mirror im-
ages of each other about the line y  x. For the inverse to exist, the function f must
be either strictly increasing or strictly decreasing. Consequently, the graph of the in-
verse function f 1 also is monotonic, and both functions increase or both decrease.
However, there are no clear patterns for the concavity of f and f 1. Both f
and f 1 can be concave up, both can be concave down, each can have opposite
concavity, or both can have no concavity (if their graphs are lines).
Note that f 1 1x2 is not the same as 1> f 1x 2. For instance,
if f 1x 2  x 3, then f 1
1x 2  x 1>3,
whereas 1> f 1x 2  1>x 3  x 3, which is not the same as x 1>3. (Check their graphs
to convince yourself.) Similarly,
if g 1x 2  10x, then g 1 1x 2  log x,
but 1>g 1x 2  1>10x  10x, which is not the same as log x. (Check their graphs to
convince yourself.) Figure 2.80 shows the graphs of g 1x 2  10x and 1>g 1x 2  10x.
The graphs are not mirror images of each other about the line y  x. Using the
symmetry condition, describe where the graph of the inverse function of g would be
in Figure 2.80.
y

y = 10 −x y = 10 x

y=x

x
FIGURE 2.80

Finding the Inverse Function

Suppose that we know that a function f has an inverse because it is either strict-
ly increasing or strictly decreasing. Can we always find the inverse function f 1?
If the formula for the function is quite simple, we might be able to undo the
equation algebraically to obtain a formula for f 1. For instance, we demonstrat-
ed earlier that we can undo the conversion formulas between the Fahrenheit and
Celsius temperature scales, getting

f 1x 2  x  32 and 1x 2  1x  32 2;
9 1 5
f
5 9
the algebra is simple because the relationship is linear. Similarly, we can undo the
equation of an exponential function to get the logarithmic function and vice versa
so that
g 1x 2  10x and g 1 1x 2  log x, for x  0.
Also, we can undo the relationship between the squaring and square root functions
algebraically, obtaining
h1x 2  x 2 and h1 1x 2  2x , for x  0.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 151

2.9 Inverse Functions 151

In Example 6, we illustrate these ideas with a somewhat more complicated function.

E XAMPLE 6
For the function y  g 1x 2  1  1>x with domain x  0, find g 1. Analyze the behav-
ior of g and g 1.
Solution As shown in Figure 2.81, the graph of the function g is strictly decreasing, so
its inverse exists. And because the domain of g is x  0, we know that 1>x  0. Thus
1  1>x must be larger than 1, and therefore the range of g must be y  1.
y
x=1

g –1(x) y=x

g(x)
1 y=1

x
FIGURE 2.81 1

Next, let’s find a formula for the inverse function. If y  1  1>x,

1
 y  1.
x
Taking the reciprocal of both sides, we get

 g 1 1y2,
1
x y  1.
y1
Interchanging the roles of x and y to use the same independent variable yields

 g 1 1x 2,
1
y x  1.
x1
The graphs of the function g and its inverse g 1 are also shown in Figure 2.81. As expect-
ed, they are mirror images of the other about the line y  x.
Further, the original function g is a decreasing function; it decays from a vertical as-
ymptote at x  0 toward a horizontal asymptote of y  1. The graph of g 1 also decreases
from a vertical asymptote at x  1 toward a horizontal asymptote of y  0. Note that the
vertical and horizontal asymptotes are interchanged and that both curves are concave up.
◆
Unfortunately, solving for an inverse function algebraically, as we did in Exam-
ple 6, usually is not possible. Hence, we usually have to resort to numerical or
graphical methods to estimate values for the inverse function; that is, given a par-
ticular value for y, we can determine the corresponding value of x by examining ei-
ther the graph of the original function or successive numerical estimates.
E XAMPLE 7
For the function f 1x 2  2x  3x, (a) explain why f 1
exists and (b) estimate the value
of f 1 110 2 .
Gord.3896.02.pgs 4/24/03 9:26 AM Page 152

152 CHAPTER 2 Families of Functions

Solution

a. As shown in the graph of f in Figure 2.82, the function appears to be strictly increas-
ing, which means that the inverse function exists.
y

800

600

400

200

x
FIGURE 2.82 –2 2 4 6

b. Unfortunately, for the expression y  2x  3x, it is not possible to solve for x in terms
of y, so we are unable to find a formula for f 1, even though we do know that f 1 ex-
ists. Graphically, the curve clearly passes the level y  10 at some point x, so we must
estimate, either numerically or graphically, the value of x for which f 1x 2  10. We
know that f 112  2  3  5 and that f 12 2  22  32  4  9  13, so the desired
value of x must be between 1 and 2. We can zoom in either by checking further nu-
merical values or by examining the graph of the function between 1 and 2, as shown in
Figure 2.83. Either way, we find that x  1.73, so f 11.732  21.73  31.73  10.007.
y

10

x
FIGURE 2.83 0 1.73

Problems
1. Which of the following functions have inverses? b. Your distance from New York on an airplane flight
Explain why or why not. For any function having from New York to San Francisco as a function of
an inverse, describe what the inverse function the time t since takeoff.
tells you. c. The height of the student who is numbered n on
a. The height of water after t minutes in a child’s your instructor’s class roster.
pool that you are filling at a steady rate, using a d. The amount that the nth customer in line at
garden hose. Burger Heaven pays for lunch.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 153

2.9 Inverse Functions 153

e. The length of the fingernail on your right index 5. The table of values gives the time T needed for a
finger if t is the number of hours since you last Trans Am to accelerate from zero to the indicated
clipped your nails. final speed v.
f. The amount spent by a family to heat their home
if T is the temperature at which they set the Final speed, v (mph) 30 40 50 60 70
thermostat.
g. The depth of the snow on a person’s front lawn Time, T (sec) 3.00 4.29 5.52 7.38 9.81
in Buffalo as a function of the time t elapsed
from October 1 to the following March 1. a. Explain why this set of data represents a func-
h. The total amount of snow that falls on the per- tion and why it has an inverse.
son’s lawn in part (g) as a function of the time t b. Explain what the inverse function tells you. What
elapsed from October 1 to the following March 1. is f 1 15.522 ? Estimate the value of f 1 172.
2. For the function f shown in the accompanying fig- 6. We know that 1 inch is equivalent to about 2.54 cen-
ure, estimate the value for x that corresponds to timeters.
y a. Write a formula for the function f that gives an
object’s length C in centimeters as a function of
5
its length I in inches.
4
b. Find a formula for the inverse function f 1 and
explain what f 1 tells you, in practical terms.
3 7. Find the inverse function of p1t2  11.042 t.
2 8. Find the inverse function of f 1t2  50110 2 0.1t.
9. Suppose that the temperature of an object is being
1
measured to the nearest degree on both the Fahren-
x heit and Celsius scales. In general, which reading
–2 –1 1 2 3 4 5 would you expect to be more accurate? Why?
–1 10. For each function f shown, sketch the graph of the
inverse function f 1 on the same set of axes.
a. y  0; b. y  2;
(a) y (b) y
c. y  5; d. y  1.
Then plot the resulting points and use them to
sketch the graph of the inverse function f 1.
3. Consider the function f with values given in the fol-
lowing table.
x x
0 0
x 0 1 2 3 4 5
(c) y (d) y
f (x) 2.94 2.48 2.05 1.84 1.44 1.12

a. What is the domain of f? What is the range?

b. Create a table of values for f 1. What are its do-
main and range?
x x
4. Use your function grapher to decide which functions 0 0
have inverses. For those functions that do, estimate
the value for f 1 110 2. 11. Suppose that a function f is increasing and concave
a. f 1x 2  x 3  9x 2  5x  5 up. By thinking of its inverse as the reflection about
b. f 1x2  x 3  2x 2  5x  5 the line y  x, explain why f 1 is also increasing. Is
c. f 1x2  2x  x 2 d. f 1x 2  2x  x 2 it concave up or concave down? What happens if f is
e. f 1x2  2x  x 3 f. f 1x 2  2x  x 3 increasing and concave down?
Gord.3896.02.pgs 4/24/03 9:26 AM Page 154

154 CHAPTER 2 Families of Functions

12. Repeat Problem 11 if the function f is decreasing 15. The level of Prozac in the blood can be modeled
and concave up; the function f is decreasing and by the function P1t2  8010.752 t.
concave down. a. Find a formula for the inverse function.
13. On the same set of axes, sketch the graph of f 1 b. Use the inverse function to determine how long it
that corresponds to the function f shown in the ac- will take until the level of Prozac drops to 25 mg.
companying figure. 16. The temperature of a chicken cooking in an oven
y can be modeled by the function T1t2  350 
5 31010.992 t.
a. Find a formula for the inverse function.
4
b. Use the inverse function to determine how long
3
it will take until the temperature of the chicken
reaches 175°.
2 f
17. In Problem 20 of Section 1.3 we introduced a func-
1
tion f that represents a simple replacement code in
which each letter of the alphabet is replaced by a
x different letter according to f 1A2  M, f 1B2  D,
f 1C2  K, f 1D2  V, f 1E 2  X, f 1F2  B,
–2 –1 1 2 3 4 5

f 1G2  P, f 1H2  T, f 1I2  J, f 1J2  S,

–1

f 1K2  Z, f 1L2  Q, f 1M 2  H, f 1N2  O,

f 1O2  A, f 1P2  L, f 1Q2  W, f 1R2  C,
–2

14. Use the quadratic formula f 1S2  F, f 1T2  Y, f 1U 2  R, f 1V2  G,

f 1W 2  I, f 1X2  U, f 1Y2  N, and f 1Z2  E.
b 2b2  4ac
t a. Explain why this function has an inverse f 1.
2a
b. Use the inverse to decode the message
to solve the quadratic equation 16t 2  48t 
6  35 from Example 5 modeling the height of a JF YTJF HMYT?
ball. What is the significance of the two roots of
this equation?

Exercising Your Algebra Skills

Solve for the unknown in each equation. For Problems 13–16, solve for the indicated variable in
1. c25  14 2. 0.07t  3 each formula to find the inverse function.
3. 0.84k  0.20 4. m1995  4 13. F  ma, for a 14. E  mc2, for m
5. 17b8  32 6. 25c9  8 1
15. P  kVT, for V 16. K  mv2, for v
7. 411.02 2 x  7 8. 210.75 2 t  1 2
For Problems 9–12, solve for the independent variable
in terms of the dependent variable for each function.
9. y  f 1x2  12x 7>2 10. y  f 1t2  1211.062 t
11. Q  g 1w2  27w3>4 12. L  g1t2  12510.92 2 t

Chapter Summary

In this chapter, we covered the following ideas and approaches relating to families of
functions.
◆ Important behavior characteristics of four important families of functions—
linear functions, exponential functions, logarithmic functions, and power
functions.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 155

Review Problems 155

◆ How to find the slope and the equation of a line.

◆ What the slope of a line means.
◆ A criterion for knowing when a set of data follows a linear pattern.
◆ How to estimate the equation of a line that captures the linear pattern in a set
of data.
◆ How to set up and solve problems involving linear processes.
◆ The behavior of exponential growth and exponential decay functions.
◆ What the growth and decay rates and the growth and decay factors mean.
◆ A criterion for knowing when a set of data follows an exponential pattern.
◆ How to find the doubling time for an exponential growth process and the
half-life for an exponential decay process.
◆ How exponential behavior compares to linear behavior.
◆ How to find the exponential function that passes through two points.
◆ How to set up and solve problems involving exponential processes.
◆ The behavior of logarithmic functions.
◆ How to set up and solve problems involving logarithmic functions.
◆ How to use logarithms with bases other than 10.
◆ The behavior of power functions when the power is greater than 1, between
0 and 1, and negative.
◆ How to find the power function that passes through two points.
◆ How to set up and solve problems involving power functions.
◆ How power function behavior compares to exponential behavior.
◆ How logarithmic function behavior compares to power function behavior.
◆ How to determine whether a function has an inverse.
◆ What the inverse function tells you.
◆ How to set up and solve problems involving inverse functions.

Review Problems
1. For each of the curves shown, suggest any types of than 1. If you suggest a power function, indicate
functions that might have the indicated behavior whether the power p is positive or negative and
pattern. If you suggest an exponential function, in- whether p is greater than 1 or less than 1.
dicate whether the base c is greater than 1 or less

(a) y (b) y (c) y

x x x
Gord.3896.02.pgs 4/24/03 9:26 AM Page 156

156 CHAPTER 2 Families of Functions

2. Identify each function as linear, exponential, log- i. y  511.082 x j. y  x 2.5

arithmic, or power. In each case, explain your k. y  x 2.5 l. y  x 0.25
reasoning.
a. y b. y
y
A
B

x x
C
x
D
c. y d. y

x x
y
F G

e. y f. y

x x

H
I x
g. y  1.05 x
h. y  x 1.05 0

i. y  10.72 x j. y  x 0.7
k. y  1> 1x l. 5x  3y  15 y
m. n.
x y x y J

0 3 0 5

1 5.1 1 7
K
2 7.2 2 9.8

3 9.3 3 13.72

L
3. Match each formula for a function with one of the x
0
graphs (A)–(L). Because more than one function
from the same family appears, match each member
of that family to the most appropriate graph.
4. The accompanying figure shows the graphs of the
a. y  3x  3 b. y  2x  3 values of shares of 7 stocks as functions of time.
c. y  3  2x d. y  2x  3 Match each scenario with one of the graphs and
e. y  x  3 f. y  510.922 x write a brief scenario for each of the remaining
g. y  510.972 x h. y  511.032 x graphs.
Gord.3896.02.pgs 4/24/03 9:26 AM Page 157

Review Problems 157

Price 7. a. The profits of Alamo Paper Company are grow-

ing by $100,000 each year. In 1990, its profits
(1) were $1.5 million. Determine the profit func-
tion P1t2, where t represents the number of
(2)
years since 1990. Draw a graph of y  P1t2 and
(3) determine the year in which profits first exceed
$2 million.
(5)
b. Ord Paper Company had profits of $950,000 in
1990, and its profits are growing at the rate of
10% each year. Determine when its profits first
(4)
(6) exceed $2 million.
(7)
t
c. Use your function grapher to determine when the
profits of Ord first exceed the profits of Alamo.
a. The value of the stock increased 12% per year. 8. When the bald eagle was formally put on the en-
b. The value of the stock increased 8% per year. dangered species list in 1967, there were about 800
c. The value of the stock dropped by $4 each year. of the eagles in the United States. As a result of
d. The value of the stock increased by $6 each year. eagle protection and restoration efforts, the bald
e. The value of the stock remained steady. eagle was removed from the list in 1994 when its
5. The points 11, 12, 12, 5 2, 13, 20 2, 14, 50 2 , and
population was just over 8000.
15, 1002 are plotted in the accompanying figure and a. Determine the doubling time of the eagle popula-
a line is drawn through them. tion assuming the growth pattern is exponential.
b. Estimate the number of bald eagles in the United
100 States in 2005, assuming that the growth trend
continues.
50
c. When can you expect the eagle population to
reach 20,000?
20 9. In preparing a holiday cranberry mold, a cook
added boiling water at 212°F to the fruit and gela-
5 tin mixture, which was then poured into the mold
and put into a 40° refrigerator. After 30 minutes,
1 the temperature of the mixture was 148°. The tem-
perature F1t2 at time t (in minutes) is given by
F1t2  17211  a2 t  40, where a is a constant.
1 2 3 4 5
What is the temperature of the mixture 3 hours after
a. What is wrong with the graph as it is drawn? the mold was put in the refrigerator?
b. Draw a correct graph of these points and sketch 10. From 1980 to 1998, the number of workers, in mil-
a smooth curve that passes through them. lions, covered by Social Security can be approxi-
c. Describe the behavior pattern in the function mated by a linear function of time t. Use the data in
based on these points. Identify a possible type of the table below and the black thread method to find
function that might be an appropriate model for the equation of a line that fits the data.
these values. Explain your reasoning. Use this linear function to estimate the number
6. A function F1t2 is exponential with known values of workers covered by Social Security in 2005.
F102  5 and F132  8.5. Determine the function
and give the growth factor.

Year 1980 1985 1990 1992 1993 1994 1995 1996 1997 1998

Workers (millions) 140.4 150.9 164.0 167.5 169.1 170.7 172.9 174.8 177.0 179.1
Source: 2000 Statistical Abstract of the United States
Gord.3896.02.pgs 4/24/03 9:26 AM Page 158

158 CHAPTER 2 Families of Functions

11. The following table shows the number of deaths in e. Use both models to predict the amount of taxes
the United States resulting from accidental poison- collected in 2010. Which model seems more ac-
ing by drugs and medications in various years. Use curate? Explain.
the black thread method to find the equation of a 14. The median family income I in the United States
linear function that fits the data. What is the slope was about $17,700 in 1980 and rose to about
of the line? What does this slope represent? Use the $37,000 in 1997. Let t be the number of years since
equation to predict the number of deaths from ac- 1980.
cidental poisoning in 2003.

t Linear Model Exponential Model

Year 1980 1990 1994 1995 1996
0 $17,700 $17,700
Deaths 2492 4506 7828 8000 8431
5 ? ?
Source: 2000 Statistical Abstract of the United States
10 ? ?
12. Find possible equations for the function represent-
15 ? ?
ed by each graph.
17 $37,000 $37,000
a. y
25 ? ?
(–3, 21)
30 ? ?
(0, 8)
a. If you assume that the increase in family income
x
has been linear, find an equation for the line rep-
b. y resenting income I in terms of t. Use this equa-
tion to complete the second column of the table.
(–3, 21)
b. If you assume that the increase in family income
has been exponential, find an equation of the
(0, 8) form I  I0c t to represent family income levels
since 1980 and complete the third column of the
x
table.
y c. On the same set of axes, sketch the graphs of the
c.
(4, 18)
functions you obtained in parts (a) and (b).
d. Use the equations from parts (a) and (b) to pre-
dict the median family income in 2003 for both
(2, 7) types of growth.
e. Suppose that both predictions from part (d)
x seem unreasonable. Can you suggest any other
types of functions that might be a better fit?
13. In 1990 (when t  0), the IRS collected $1055 bil- 15. (Continuation of Problem 14) Suppose that you
lion in taxes. In 1995, the IRS collected $1573 billion. learn that the median family income in 1990 was
a. Construct the linear function giving the amount about $28,900.
of taxes collected by the IRS as a function of time t. a. Which of the two models in Problem 14 now
b. Use the linear function to estimate the amount seems more accurate?
of taxes collected in 2003. b. If you plot the three data points corresponding to
c. Construct the exponential function giving the 1980, 1990, and 1997, how would you describe
amount of taxes collected as a function of time t. the shape of the graph of median family income
d. Use the exponential function to estimate the as a function of time? What is the significance of
amount of taxes collected in 2003. this shape?
Gord.3896.02.pgs 4/24/03 9:26 AM Page 159

Review Problems 159

16. For each function, draw the inverse function, if one 20. a. Since 1960, the price of an ice cream cone in one
exists, on the same axes. If the function has no in- southern city has been growing approximately ex-
verse, explain why. ponentially according to the function f 1t2  Act.
y If the price of a one-scoop cone was 20¢ in 1960
a.
and $1.80 in 2000, (i) determine the function f
and (ii) predict what the price of such a cone will
be in 2005.
b. The average price of a ticket to a first-run movie
was $2.00 in 1960. This price has been growing
x
exponentially and in 2000 was $9.00. Which of
the prices, for ice cream or for movies, is grow-
y
b. ing faster?
c. When can you expect the ice cream and movie
prices to be the same if they each continue to
grow in the same way?
d. How much would a ticket to the movies cost at
x the time you found in part (c)?
e. Would you use your model to predict the answer
y you got in parts (c) and (d)? Explain.
c.
21. The aim of a college administration is to reduce
the number of students who need remedial work
in English by 10% each year. At the time the policy
was put into place, 1600 students were enrolled in
remedial English classes. If this program is suc-
x
cessful, how many students will be enrolled in re-
medial English in 3 years? How long will it take for
17. For each function, give its domain and find the in- the number of students enrolled in such classes to
verse function. be reduced to one section of 15 students?
a. f 1t2  0.5log12t  42 b. g 1x 2  x 3  6 22. The level of a drug in the bloodstream decreases at
18. The function F1x2 is either linear or exponential. a rate of 30% of the drug per hour. Assume that
From the values in the table, decide which is the the initial dose is 150 mg. How long does it take to
correct type and find a formula for F. bring the drug level down to under 20 mg? How
long does it take to bring the drug level down to
5% of the original level?
x 1 2 3 4 5

F(x) 6 9 13.5 20.25 30.375

19. Match each of the functions f, g, and h in the table

to the behavior described as
a. increasing, straight line.
b. increasing, concave down.
c. increasing, concave up.

x 1 2 3 4 5
f (x) 2.70 3.64 4.92 6.34 8.96

g(x) 1.4 4.6 7.8 11.0 14.2

h(x) 5.10 5.19 5.27 5.34 5.40

Gord.3896.02.pgs 4/24/03 9:26 AM Page 160
gord.3896.03.pgs 4/24/03 9:53 AM Page 161

3
Fitting Functions to Data

3.1 Introduction to Data Analysis

When the space shuttle Challenger exploded some 90 seconds after launch on Jan-
uary 28, 1986, it left the entire nation in a state of shock. The shuttle accident was
especially traumatic to the many school children who were watching the launch
live on television because Christa McAuliffe, a teacher, was a member of the crew.
The Challenger explosion also left behind many unanswered questions. Among
the most important were:
◆ What went wrong?
◆ Could the problem have been anticipated?
◆ If it could have been anticipated, why wasn’t it?

The first question is technical, the second is mathematical, and the third is politi-
cal. In this section, we address the first two questions.
The Challenger disaster involved two factors: a component known as an O-ring
and the air temperature at launch. The O-ring is a very thin ring (37.5 feet in diame-
ter but only 0.28 inches thick) that seals the connections, or joints, between different
sections of the shuttle engines. The locations of the six O-rings are indicated by the ar-
rows shown in Figure 3.1. On the morning of the Challenger launch, the air tempera-
ture was 31°F, which was considerably colder than the temperature at any previous
launch. In fact, the coldest temperature recorded at any previous launch was 53°F.
The Rogers Commission, which studied the Challenger disaster, focused on the
O-rings as a possible cause of the explosion because there had been problems with
O-rings on previous flights. In fact, the night before the launch, some of the project
engineers, as part of the standard prelaunch routine, had thought about the six
O-rings and questioned whether the Challenger should lift off because of the pre-
dicted overnight temperatures. The reasoning that went into the flight decision is
worth considering because it demonstrates the important role that mathematical
analysis can play in making informed decisions.
Twenty-four shuttle flights preceded the Challenger flight. On seven of them,
relatively minor problems had occurred with the O-rings. In reviewing these previ-
FIGURE 3.1 ous incidents, the engineers examined the data shown in Figure 3.2. Note that the

161
gord.3896.03.pgs 4/24/03 9:53 AM Page 162

162 CHAPTER 3 Fitting Functions to Data

horizontal axis indicates the air temperature at launch, and the vertical axis gives
the number of O-rings affected. Each black dot represents a particular shuttle
launch. Thus the shuttle launched at 53°F experienced problems with three differ-
ent O-rings. Even though the graph shows one dot for 70°, two shuttles actually
were launched at this temperature and both had problems with one O-ring. Never-
theless, examination of this graph (as the shuttle engineers probably did) reveals
no consistent pattern indicating that the lower the air temperature at launch, the
more likely there will be O-ring problems. In fact, note that the shuttle launch at
75°F had problems with two different O-rings. (Interestingly, it was the previous
launch of the Challenger.) Consequently, the data didn’t give the engineers any
solid reason for canceling the Challenger launch the following morning.

4 4
Number of O-rings affected

Number of O-rings affected

3 3

2 2

1 1

0 0
50 55 60 65 70 75 80 85 50 55 60 65 70 75 80 85
Calculated joint temperature (°F) Calculated joint temperature (°F)

FIGURE 3.2 FIGURE 3.3

Unfortunately, the engineers didn’t realize that the data in Figure 3.2 is just
part of the story. It reflects only those launches during which there were O-ring
problems. What is missing are those launches that were trouble-free. From the full
set of data for all 24 previous shuttle flights shown in Figure 3.3, a striking pattern
emerges: Almost all the problem launches occurred at low temperatures, and all
the problem-free launches occurred at temperatures above 65°F. These results sug-
gest that problems with O-rings are unlikely on warm days, but there may likely be
a problem on a cool day. And, the predicted temperature of 31°F when the
Challenger was due to lift off was far colder than the temperature for any previous
launch. Had the engineers looked at all the data, there is no way that they could
have allowed the Challenger to lift off.
Moreover, we can go beyond just an eyeball examination of the data in look-
ing for trends. Recall that in Chapter 2 we introduced the “black thread method”
for estimating the line that is the best fit to a set of data points. The graph of the
data points is known as a scatterplot. In Figure 3.3 the scatterplot represents the
number of O-ring problems as a function of air temperature. A line is not a good
fit to this set of data because the pattern is clearly concave up. Instead, a curve
such as the one shown superimposed over the data points in Figure 3.4 suggests
a decaying exponential function or a power function with p  0. Although ex-
trapolating beyond the region of the data points (in this case from 53°F to 80°F)
usually is risky, there is little doubt that launching the Challenger at 31°F was
even riskier!
Most scientific, engineering, and technical work involves collecting and work-
ing with data and, more important, basing decisions on what can be inferred from
gord.3896.03.pgs 4/24/03 9:53 AM Page 163

3.1 Introduction to Data Analysis 163

Number of O-rings affected

3

0
30 40 50 60 70 80 90
FIGURE 3.4 Temperature (°F)

the data. To make intelligent decisions, you must understand all the information
that a set of data imparts. Thus you need techniques for displaying the data in a
form that makes it easy to extract the relevant information. Unfortunately, these
techniques haven’t been a major focus in past mathematical and technical training.
The Challenger situation illustrates the changing role of mathematics and how
people use it. Applying mathematics is not a matter of “Here’s an expression. Fac-
tor it.” or “Here’s an equation. Solve it.” Rather, in the real world you will often face
a situation about which a decision must be made. You need to be able to view that
situation mathematically (i. e., create an appropriate mathematical model), identi-
fy the appropriate question to ask, obtain the solution (often with some electronic
tool), interpret the solution in terms of the original problem, and communicate
that solution effectively to others. The emphasis is much more on reasoning and
judgment, not just on mechanical operations.
In Chapter 2 we discussed families of functions that have various behavior pat-
terns. In this chapter, we consider the problem of finding the function that best fits
a set of data. If the data fall into a roughly linear pattern, either increasing or de-
creasing, we want to find the line that is the best possible fit to the data. Similarly, if
the data fall into certain nonlinear patterns, we want to find the function that best
fits the data. Figure 3.5 shows three different data sets that clearly are not linear
patterns. Figure 3.5(a) shows an increasing, concave up pattern that could be mod-
eled by either an exponential growth function with base c  1 or a power function
with power p  1. Figure 3.5(b) shows a decreasing, concave up pattern that could
be modeled by either an exponential decay function with base c  1 or a power
function with power p  0. Figure 3.5(c) shows an increasing, concave down pat-
tern that could be modeled by either a power function with power 0  p  1 or a
logarithmic function. Once we have written a formula for the function that fits a
set of data, we can then use it to answer questions of a predictive nature.
The techniques that we discuss here are the methods by which we construct func-
tions. With this approach, preventable disasters such as the Challenger explosion aren’t
likely to recur.
(a) y (b) y (c) y

FIGURE 3.5 x x x
gord.3896.03.pgs 4/24/03 9:53 AM Page 164

164 CHAPTER 3 Fitting Functions to Data

3.2 Linear Regression Analysis

We now consider fitting a line to a set of data from a somewhat more sophisticated
viewpoint than the black thread method. We now want to determine the single best
line to fit a set of points.
Suppose that we have a set of n measurements on two presumably related
quantities x and y: 1x1 , y1 2, 1x2 , y2 2 , . . . , 1xn , yn 2 , where x1 and y1 are the coordi-
nates of the first point, x2 and y2 are the coordinates of the second point, and so on.
For instance, the coordinates might represent people’s heights and weights; stu-
dents’ high school averages and college GPAs; or the gross receipts (box office,
cable and broadcast TV, videotapes, etc.) of the movie industry, in billions of dol-
lars, in different years, as given in the following table.

Year 1990 1993 1994 1995 1996 1997

Gross receipts 39.98 49.80 53.50 57.18 60.28 63.01
Source: 2000 Statistical Abstract of the United States.

We begin by drawing a scatterplot of the data, as shown in Figure 3.6(a). The

data appear to fall in a roughly linear pattern with a positive slope. We can draw
many different lines that all seem to capture the overall pattern of the points in the
scatterplot, as shown in Figure 3.6(b). However, none of them can possibly pass
through all the points. In fact, a good fit line may not necessarily pass through any of
the points. Our objective is to determine the best linear fit to this set of data: the one
line that comes closest to all the data points.

(a) (b)
Movie Industry Receipts (billions)

Movie Industry Receipts (billions)

70 70

60 60

50 50

40 40

30 30

0 90 92 94 96 98 0 90 92 94 96 98
Years Years

FIGURE 3.6

We now write the equation of a line in the form y  ax  b instead of the

more usual form y  mx  b to give a better match to the display on most calcu-
lators. As always, the coefficient of the independent variable represents the slope,
and the constant gives the vertical intercept regardless of the letters used.
The key to finding this line is the phrase “comes closest to all the points.” The
most common interpretation of this phrase is that the sum of the squares of all the
vertical distances from the points to the line should be a minimum, as shown in
gord.3896.03.pgs 4/24/03 9:53 AM Page 165

3.2 Linear Regression Analysis 165

Figure 3.7. (If we used only the actual vertical distances, rather than their squares,
some would be positive, others would be negative, and they would tend to cancel
each other when added.) We discuss these vertical distances, called residuals, in
more detail in Section 3.6.
y
y = a + bx

(x1, y1)

y1 − (a + bx1) y2 − (a + bx2)
(x2, y2) a + bx2
a + bx1

FIGURE 3.7 FIGURE 3.8

Suppose that the equation of the best-fit line is y  ax  b, where a and b are,
for now, unknown constants. For the first data point 1x1 , y1 2 , the point on the line
with the same x-coordinate x  x1 has height y  ax1  b. Hence the vertical dis-
tance from the data point to the line is y1  1ax1  b 2, as illustrated in Figure 3.8.
For the second data point 1x2 , y2 2 , the vertical distance to the line is similarly
y2  1ax2  b2 , and so on for the rest of the data points. The corresponding
squares of these vertical distances, for each of the n points, are
3y1  1ax1  b2 4 2, 3 y2  1ax2  b2 4 2, ..., 3yn  1axn  b2 4 2.
To measure how close the line comes to all n data points, statisticians use the sum
of the squares of these differences:
3 y1  1ax1  b 2 4 2  3y2  1ax2  b 2 4 2  . . .  3 yn  1axn  b2 4 2.
Exactly one line corresponds to a minimum value for the sum of these squares.
It is known as the least squares line, or more commonly, the regression line. The
formulas1 for the coefficients a and b in the equation of the regression line are built
into most sophisticated calculators, usually under the STAT (statistics) menu and
may be marked as LinReg or LinR. Routines for these calculations also are widely
available in many computer packages, including spreadsheets such as Excel™.
To find the equation of this line with your calculator, you typically have to ac-
cess the STAT menu, select the data option, clear any old data, enter the new data,
and then go back to the main STAT menu and select calculate linear regression.
Most calculators will give you values for the coefficients a and b in the regression
equation y  ax  b, and the value of a third quantity that we discuss later in this
section. See the detailed instructions in your calculator’s manual.

1
The regression coefficients are
n g 1xy2  1 gx 2 1 gy2 1 gx 2 2 1 gy2  1 gxy2 1 gx2
a and b 
n1 gx 2 2  1 gx2 2 n1 gx 2 2  1 gx2 2
,

where n  number of data pairs 1x, y2 , gx 2  the sum of the squares of the x’s, 1 gx 2 2  the
square of the sum of the x’s, and g 1xy2  the sum of the products of x and y in each pair.
gord.3896.03.pgs 4/24/03 9:53 AM Page 166

166 CHAPTER 3 Fitting Functions to Data

We illustrate how to apply these ideas in Example 1.

E XAMPLE 1
(a) Find the equation of the regression line that best fits the data on gross receipts of the
movie industry, in billions of dollars, as a function of time. (b) What does the slope of
this line mean in this context? (c) Use this function to estimate the gross receipts in 2003.
Solution

a. Let the independent variable x represent the number of years since 1990 and let the
dependent variable y be the gross receipts, in billions of dollars, for each year. We
have the following table of values.

Year 0 3 4 5 6 7

Gross receipts 39.98 49.80 53.50 57.18 60.28 63.01

We enter these values for x and y into the statistics routine of a graphing calculator or
computer and run the linear regression routine. The calculator or computer re-
sponds with the equation of the regression line
y  3.3435x  40.027.
The graph of this line superimposed over the scatterplot of the data is shown in Fig-
ure 3.9. Note that the line seems to be an excellent fit to all the data points.
b. The slope, 3.3435, of this regression line tells us that the gross receipts of movies are
increasing, on average, by about $3 13 billion each year.
c. Having this linear model for the gross receipts as a function of time, we use it to pre-
dict the receipts in 2003, when x  13 years since 1990. We then get
y  3.3435113 2  40.027  $83.49 billion.
Movie industry receipts (billions)

70

60

50

40

30

0 3 6 9
FIGURE 3.9 Years since 1990

◆
In general, when working with years or other “large” numbers, scaling them
down as we did in Example 1 usually is helpful. For instance, we let x be the num-
ber of years since 1990.
We next look at a somewhat more complicated example based on the data pre-
sented in Table 3.1 giving world record times (in minutes: seconds) for the mile
run and the year the record was set.
gord.3896.03.pgs 4/24/03 9:53 AM Page 167

3.1 Most Recent First Level Head 167

E XAMPLE 2
(a) Find the equation of the regression line that best fits the data in Table 3.1 on the
world records for the mile run as a function of time. (b) What does the slope of this line
mean in this context? (c) Use this function to estimate the world record in 2005.

TABLE 3.1 World Records in the Mile Run

Year Record time Winner
1911 4:15.4 John Paul Jones, United States
1913 4:14.6 John Paul Jones, United States
1915 4:12.6 Norman Taber, United States
1923 4:10.4 Paavo Nurmi, Finland
1931 4:09.2 Jules Ladoumegue, France
1933 4:07.6 Jack Lovelock, New Zealand
1934 4:06.8 Glen Cunningham, United States
1937 4:06.4 Sidney Wooderson, Great Britain
1942 4:06.2 Gunder Haegg, Sweden
1942 4:06.2 Arne Andersson, Sweden
1942 4:04.6 Gunder Haegg, Sweden
1943 4:02.6 Arne Andersson, Sweden
1944 4:01.6 Arne Andersson, Sweden
1945 4:01.4 Gunder Haegg, Sweden
1954 3:59.4 Roger Bannister, Great Britain
1954 3:58.0 John Landy, Australia
1957 3:57.2 Derek Ibbotson, Great Britain
1958 3:54.5 Herb Elliott, Australia
1962 3:54.4 Peter Snell, New Zealand
1964 3:54.1 Peter Snell, New Zealand
1965 3:53.6 Michel Jazy, France
1966 3:51.3 Jim Ryun, United States
1967 3:51.1 Jim Ryun, United States
1975 3:51.0 Filbert Bayi, Tanzania
1975 3:49.4 John Walker, New Zealand
1979 3:49.0 Sebastian Coe, Great Britain
1980 3:48.9 Steve Ovett, Great Britain
1981 3:48.8 Sebastian Coe, Great Britain
1981 3:48.7 Steve Ovett, Great Britain
1981 3:47.6 Sebastian Coe, Great Britain
1985 3:46.5 Steve Cram, Great Britain
1993 3:44.4 Noureddine Morceli, Algeria

Solution

a. Let the independent variable x represent the number of years since 1900 and let the de-
pendent variable y be the time, in minutes, of each record-breaking mile run. The times
given in Table 3.1 are in the form minutes : seconds, so it is necessary to recalculate
each value in terms of minutes. For instance, the first entry of 4:15.4, or 4 minutes and
15.4 seconds, is equivalent to 4  15.4>60  4.2567 minutes. Making this conversion
for all values in the table, we obtain
gord.3896.03.pgs 4/24/03 9:53 AM Page 168

168 CHAPTER 3 Fitting Functions to Data

x 11 13 15 ... 93

y 4.2567 4.2433 4.2100 ... 3.7400

Using either a calculator or computer program, we get the equation of the regression
line as
y  0.00663x  4.335.
Figure 3.10 shows the graph of this line superimposed over the scatterplot of the
data; the line seems to fit all the data points well.

4.3

4.2

4.1

4.0

3.9

3.8

3.7

FIGURE 3.10 0 1910 1930 1950 1970 1990 2010

b. The slope, 0.00663, of this regression line tells us that the world record is dropping,
on average, by 0.00663 minutes, or 0.498 seconds, each year.
c. Assuming that this linear model for the world-record times in the mile run contin-
ues to hold, we can use it to predict the world record in 2005, when x  105 years
since 1900. We then get
y  0.006631105 2  4.335  3.639 minutes
or about 3 minutes and 38.3 seconds.
◆
Think About This Does the world record time of 3 minutes and 38.3 seconds in 2005 calculated in Ex-
ample 2 seem reasonable? What is the predicted value for the world record in the mile
run in 2500 based on the linear model? Is that value reasonable? Similarly, what hap-
pens if you predict back to 1492? Do you get a reasonable value? Why or why not? ❐

Recall that, if you use the equation of the regression line to predict values of y
corresponding to values of x within the interval of data values (called interpolating),
the results are usually quite reasonable. Recall also that, if you try to use the regres-
sion equation well beyond the interval of data values (called extrapolating), the re-
sults become extremely questionable. Thus you should not extrapolate too far into
either the future or the past. The domain of the mathematical model should not be
extended too far beyond the given data.

The Correlation Coefficient

There is another major concern regarding the use of the regression equation. If you
take any set of measurements relating two variables, you can always construct the re-
gression equation based on the data. However, the results are completely meaningless if
gord.3896.03.pgs 4/24/03 9:53 AM Page 169

3.2 Linear Regression Analysis 169

the two variables are not linearly related or even are totally unrelated. For example, you
could collect data on people’s telephone numbers and their social security numbers,
construct a scatterplot, and calculate the corresponding regression equation. But the
two variables are unrelated, so the results of predicting people’s social security numbers
from their phone numbers would be of no value.
Therefore we need a way to determine whether, in fact, a linear relationship ex-
ists between two quantities. The most common way of detecting such a relation-
ship is by using a quantity known as the linear correlation coefficient, or simply the
correlation coefficient. This quantity is denoted by r and is always a number be-
tween 1 and 1.
◆ Values of r close to 1 indicate a high degree of positive correlation be-
tween x and y. That is, they are likely related via a linear relationship, and
the regression line will have positive slope. For example, we would expect a
high positive correlation between a company’s profits and its sales: As sales
go up, profits usually go up also.
◆ Values of r close to 1 indicate a high degree of negative correlation between
x and y. They are likely related by a linear relationship, and the regression line
will have negative slope. For instance, there is a high negative correlation be-
tween a car’s gas mileage and its weight—as weight goes up, gas mileage goes
down, and vice versa. Similarly, there is a high negative correlation between
the literacy rate and the infant mortality rate in any nation.
◆ Values of r close to 0 indicate little or no correlation, and we would con-
clude that there is no linear relationship between the variables. For in-
stance, there is no correlation between students’ social security numbers
and their telephone numbers.
You can visualize the different cases by looking at the scatterplots shown in
Figure 3.11. In Figure 3.11(a), the data points lie more or less along a rising line
and the value of the correlation coefficient will be positive and relatively close to 1.
In Figure 3.11(b), the points are scattered about a downward-sloping line, which
means that the correlation coefficient is negative and relatively close to 1. In Fig-
ure 3.11(c), there is no clear pattern for the points, and so the correlation coeffi-
cient is relatively close to 0, indicating that there is no linear relationship between
the variables.

positive r close to +1 negative r close to −1 r close to 0

(a) (b) (c)

FIGURE 3.11
gord.3896.03.pgs 4/24/03 9:53 AM Page 170

170 CHAPTER 3 Fitting Functions to Data

As with the calculation of the regression coefficients, the correlation coeffi-

cient2 usually is obtained either with a computer program or a calculator, not by
hand. The important thing is knowing how to interpret the results, not how to cal-
culate the regression equation and the correlation coefficient.

E XAMPLE 3
Find the correlation coefficient for the data on the mile run records and interpret it.
Solution When we enter the recalculated data values from Example 2 in a calculator or
program, we get a correlation coefficient of r  0.9899. The fact that r is negative sim-
ply reflects the downward trend in the data and the negative slope in the equation of the
regression line y  0.00663x  4.335. The value of r  0.9899 is extremely close to
1, which suggests that there is negative correlation between the world record in the
mile run and the year.
◆
Incidentally, if the data points all lie exactly on a line, the regression line found
by the calculator or program actually passes through the points and the correlation
coefficient is either 1 or 1, depending on whether the slope is positive or negative.

Determining When the Level of Correlation Is Significant

To use the correlation coefficient properly, you must be able to distinguish between
significant correlation and no correlation. Typically a set of data represents just a
sample from a much larger population of possible data values. For instance, if we
were studying the relationship between gas mileage and vehicle weight, we certain-
ly couldn’t collect data on every car on the road—only on a random sample of all
cars. Presumably the sample would be representative of all cars, but there is no
guarantee of that. It is conceivable that we chose an extremely unrepresentative
sample. What’s worse, there usually is no way of telling simply by looking at the
values that a sample is unrepresentative.
No statistical conclusion is ever 100% certain because that would require en-
compassing every conceivable sample from a population and that is not possible.
Statisticians have developed a set of techniques that allow us to draw conclusions
that are correct with 95% certainty or, equivalently, that are correct 95% of the
time, to account for the possibility that a sample may contain several very unrepre-
sentative observations. (Statisticians also work with 90% and 99% levels of certain-
ty, but we consider only 95% certainty here.)
Suppose that you have a random sample of n data points and use either a com-
puter program or a graphing calculator to graph the scatterplot and calculate both
the regression line and the correlation coefficient r. If the points on the scatterplot
clearly do not appear to fall in a linear pattern, you should expect that there is little
or no linear correlation between the two variables. Even if the pattern seems to be
roughly linear, you must still use the information provided by the correlation coef-
ficient to decide whether it indicates a significant level of correlation between the
two variables.

2
The correlation coefficient is:
n1 gxy2  1 gx2 1 gy2
r
2n1 gx 2  1 gx 2 2 2n1 gy2 2  1 gy2 2
2
gord.3896.03.pgs 4/24/03 9:53 AM Page 171

3.2 Linear Regression Analysis 171

The size of the sample also comes into play here. If we take a relatively small
sample—say, n  5 values—and this sample contains a nonrepresentative data
point, that point will have a significant impact on the results. If the sample con-
tains n  50 random values, the effect of a single nonrepresentative data point will
likely be diluted. If the sample consists of n  500 random values, the effect of any
single nonrepresentative data point is almost certain to be negligible.
Table 3.2. contains a partial set of so-called critical values for the correlation co-
efficient r. These critical values separate what we interpret as correlation from
what we interpret as no correlation. These critical values change, depending on
the size of the sample, n—the larger the sample size, the smaller the critical value.
We write rn to indicate the critical value of r based on n data points.

TABLE 3.2 Critical Values of r

n rn n rn n rn
3 0.997 13 0.553 27 0.381
4 0.950 14 0.532 32 0.349
5 0.878 15 0.514 37 0.325
6 0.811 16 0.497 42 0.304
7 0.754 17 0.482 47 0.288
8 0.707 18 0.468 52 0.273
9 0.666 19 0.456 62 0.250
10 0.632 20 0.444 72 0.232
11 0.602 21 0.433 82 0.217
12 0.576 22 0.423 92 0.205

To use the table, compare the value of r from the sample data to the correspon-
ding critical value rn shown in the table. For instance, a sample of size n  10 has a
critical value of r10  0.632. If the correlation coefficient for the data is greater than
0.632—say, r  0.758—we can conclude with 95% certainty that there is positive
correlation between the two variables. If the correlation coefficient for the data is
less than 0.632—say, r  0.685—we can conclude with 95% certainty that
there is negative correlation between the two quantities being studied. But, if the
value for r is between 0.632 and 0.632—say, r  0.446 or r  0.583—we
cannot conclude that there is any linear correlation between the two variables. In
other words there does not appear to be a linear relationship between x and y. Fig-
ure 3.12 illustrates the process. Of course, there still may be a nonlinear functional
relationship between x and y; we examine such cases later in this chapter.

Negative Positive
Correlation No Correlation Correlation

FIGURE 3.12 –1 –0.632 0.632 +1

Note that the critical value for a sample of size n  20 is r20  0.444, whereas
the critical value for a sample of size n  10 is r10  0.632. As we pointed out, a
larger sample is more likely to be representative of the population, so the evidence
for correlation does not have to be quite as great. With a small sample, the value for
r must be very close to 1 or 1 for us to conclude with any certainty that there is
correlation. For instance, the critical value for a sample of size n  3 is r3  0.997.
gord.3896.03.pgs 4/24/03 9:53 AM Page 172

172 CHAPTER 3 Fitting Functions to Data

A more extensive table of critical values for r is reproduced on the inside back
cover for easy reference.

E XAMPLE 4
Decide whether there is a significant level of correlation between the world record times
for the mile run and the year in which the record was set in Examples 2 and 3.
Solution In Example 3 we found that the value for the correlation coefficient was
r  0.9899, based on the n  32 entries in Table 3.1. The corresponding critical
value is then r32  0.349. Because the value for the data is more negative than this
critical value (i.e., closer to 1), we conclude that a significant level of correlation ex-
ists between the two variables.
◆
In our examples, we consistently use X as the independent variable and Y as the
dependent variable for the output from either the computer or calculator. We then
have to interpret what the X- and Y-variables represent in the context of each indi-
vidual situation.

E XAMPLE 5
A study is conducted to determine the relationship between a person’s height in inches
and shoe size, based on the following set of data pairs:
166, 92, 163, 72, 167, 8 12 2, 171, 102 , 162, 62 , 165, 8 12 2, 172, 122 , 168, 10 12 2, 160, 5 12 2 , 166, 8 2 .

a. Determine the value of the correlation coefficient and decide whether it is significant.
b. Find the equation of the regression line relating height to shoe size, based on this sample.
c. Use the equation to predict the most likely shoe size for a person who is 70 inches tall
and for someone who is 61 inches tall.
d. Use the equation to predict the most likely height of a person whose shoe size is 9.
Solution

a. Using a calculator or software package, we find that the correlation coefficient for
this set of data is r  0.951, which suggests a high degree of positive correlation be-
tween a person’s height and shoe size. Because there are n  10 data points, the
critical value for r is r10  0.632 from Table 3.2. Because 0.951  0.632, we can
conclude with 95% certainty that a significant level of positive correlation exists
between a person’s height and shoe size.
b. Figure 3.13 shows the scatterplot of the data and the graph of the associated regression
line. The equation of the regression line, as given by either a calculator or computer, is
Y  0.51X  25.016,
which is equivalent to
S  0.51H  25.016
for a person’s shoe size S as a function of the person’s height H in inches.
c. Using this regression equation for a person who is 70 inches tall, we estimate that the
shoe size corresponding to H  70 is
S  0.51170 2  25.016  10.684  10 12 .
gord.3896.03.pgs 4/24/03 9:53 AM Page 173

3.2 Linear Regression Analysis 173

12
11
10

Shoe size
9
8
7
6
5
H
0 60 62 64 66 68 70 72
FIGURE 3.13 Height (inches)

Similarly, for H  61, we estimate that the corresponding shoe size is

S  0.51161 2  25.016  6.094  6.
d. For a shoe size of S  9, we need to solve for H in
S  0.51H  25.016  9.
We add 25.016 to both sides of the equation to get
0.51H  9  25.016  34.016,
so that
H  34.016>0.51  66.7.
We conclude that the person is about 66 34 inches tall.
◆

In the context of Example 5, it seems sensible to think of shoe size as a function

of a person’s height. Thus we chose height H as the independent variable. However, it
is also appropriate to interchange the roles of S and H and think of height H as a
function of shoe size S.
Suppose that you have determined, with 95% certainty, that there is a high de-
gree of linear correlation between two variables. What does that tell you? In Exam-
ple 5, a high correlation coefficient means that there is a linear relationship between
H and S. That is, the two variables can be related with a linear equation—namely,
the regression equation. However, the fact that two quantities are correlated and
that a linear relationship exists does not mean that there is a cause-and-effect rela-
tionship between them. For example, several studies have found a high positive cor-
relation between teacher salaries in a school district and the amount of alcohol
consumed by the students. That is, if teacher salaries are low, the level of student
drinking also is low; if teacher salaries are high, considerable drinking is going on. Is
it reasonable to conclude that one causes the other? Could the level of student
drinking be reduced by lowering teacher salaries in a district?
A positive correlation between two quantities says nothing more than that a lin-
ear relationship exists between them. Other factors could contribute to both. For in-
stance, high teacher salaries typically reflect a school district in a relatively affluent
community where the students are likely to have relatively large amounts of their
own money to spend on alcohol.
gord.3896.03.pgs 4/24/03 9:53 AM Page 174

174 CHAPTER 3 Fitting Functions to Data

E XAMPLE 6
One of the most famous moments in the history of science was Galileo’s reported experi-
ment of dropping various objects from the top of the 180 foot high Leaning Tower of Pisa
and discovering that they fell at the same rate, regardless of their weight. The following
table gives the speed v, in feet per second, of an object dropped from the top of the tower
measured at half second intervals until it hits the ground. The negative values for the speed
simply reflect the convention that velocity is considered positive when an object is moving
upward and negative when an object is moving downward. Note how the object starts to
fall slowly and then accelerates. (Incidentally, these values are considerably more accurate
than anything Galileo could have measured at the end of the fourteenth century.)

Time, sec 0 0.5 1.0 1.5 2.0 2.5 3.0

Speed, ft/sec 0 16.0 32.1 47.9 64.1 80.2 96.1

a. Does the correlation coefficient indicate a significant level of correlation between the
two variables?
b. Find the line that best fits the experimental data, giving the speed v as a function of time t.
c. What is the significance of the slope in the equation of the line that fits this data?
Solution

a. We start with the scatterplot of the data as shown in Figure 3.14 and observe that
the pattern looks extremely linear. The value for the correlation coefficient is
r  0.999997. The correlation coefficient is negative because the trend in the data
is downward, so the slope of the regression line should be negative. The fact that
the value for r is so close to 1 means that there is a significant level of correlation
between the two variables and that the regression line almost perfectly fits the data
points. The speed of the falling object at any time is a linear function of t.

Time
0 1 2 3 4

–20

–40
Speed

–60

–80

–100

FIGURE 3.14 –120

b. Using a calculator or computer package, we obtain the equation of the line that
best fits this data as
Y  32.05X0.01786,
or, in terms of the variables in this situation,
v  32.05t  0.01786.
Note that the constant term 0.01786 is very close to zero, and we know that the velocity
of any falling object at time zero is zero. The slight discrepancy occurs because of possible
gord.3896.03.pgs 4/24/03 9:53 AM Page 175

3.2 Linear Regression Analysis 175

inaccuracies in measurements or rounding errors. We conclude that a better model for

the speed of a falling object might be
v  32.05t.
c. The slope of the regression line is 32.05. Because v is measured in feet per second
and t is measured in seconds, the units for the slope are feet per second per second or
feet per second squared, written ft>sec2. The slope is negative because the object is
falling downward ever faster. The specific value for the slope, approximately
32 ft>sec2, is the acceleration due to gravity.
◆
E XAMPLE 7
Psychologists claim that the human mind has a preference for rectangles having a certain
shape with the longer side approximately 1.6 times the shorter side. (The exact value used is
known as the golden ratio, which is equal to 11  15 2 >2  1.618.) If this claim is correct,
we might expect that artists would have naturally used this ratio in their paintings because
it is visually pleasing. Table 3.3 shows a selection of art masterpieces and their actual di-
mensions, in centimeters. Draw the scatterplot and find the regression line that best fits the
data. What evidence does this result provide that artists instinctively use the golden ratio?

TABLE 3.3

Artist Title Dimensions (cm)

da Vinci Mona Lisa 77  53.5
Rubens Landscape with the Chateau of Steen 134.5  236.7
Goya The Third of May, 1808 270  410
Rembrandt Self Portrait 133.6  103.8
Vermeer Young Woman with a Water Jug 45.7  40.6
T. Rousseau A Meadow Bordered by Trees 41.6  61.9
Millet The Gleaners 83.8  111.8
Constable The Haywain 130.1  185.4
Manet The Fifer 160  97.5
Monet Regatta at Argenteuil 47.5  74.3
Renoir Le Moulin de la Galette 130.7  175.3
Degas Prima Ballerina 58.3  42
Cassatt The Bath 100.3  66
Cezanne Still Life with Apples 43.5  54
Seurat The Bathers 202  300.3
Van Gogh The Starry Night 73.7  92.1
Picasso The Old Guitarist 121.3  82.7
H. Rousseau The Dream 200  300
Matisse The Joy of Life 171.3  238
Chagall I and the Village 191.5  151
Dali The Persistence of Memory 24  33

Solution To simplify things, we ignore the fact that some paintings are in portrait style
(taller than they are wide) while others are in landscape style (wider than they are tall).
We use the shorter dimension W (for width) as the independent variable and the longer
dimension H (for height) as the dependent variable. The scatterplot for the data is
shown in Figure 3.15. The corresponding correlation coefficient is r  0.9894, which in-
dicates a high degree of linear correlation between the two variables. The associated re-
gression line, which is superimposed over the data in the scatterplot, is
gord.3896.03.pgs 4/24/03 9:53 AM Page 176

176 CHAPTER 3 Fitting Functions to Data

450
400

Longer Dimension (cm)

350
300
250
200
150
100
50

0 50 100 150 200 250 300

FIGURE 3.15 Shorter Dimension (cm)

H  1.5104W  7.0017.
The slope of the regression line is 1.5104, which is fairly close to the golden ratio 1.618.
Also, the line passes relatively close to the origin, so the slope is approximately equal to
the ratio of height to width. Note that the points are all clustered fairly tightly about the
regression line, so we can conclude that most of these artists used a proportion close to
the golden ratio for these canvases.
◆
So far we have investigated only the possibility that a linear relationship exists
between two variables. We then used the correlation coefficient to detect and meas-
ure only the strength of the linear relationship. In the following sections, we con-
sider ways of detecting, measuring, and calculating a nonlinear (exponential,
power, or logarithmic) relationship between two quantities.

Problems
1. Match each value for the correlation coefficient r
with its scatterplot (i)–(v).
a. r  0.962 b. r  0.781 100 50
c. r  0.434 d. r  0.837 80 40
e. r  0.998 60 30
40 20
20 10

0 5 10 15 20 25 0 5 10 15 20 25
(i) (ii)

50 100 100
40 80 80
30 60 60
20 40 40
10 20 20

0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
(iii) (iv) (v)
gord.3896.03.pgs 4/24/03 9:53 AM Page 177

3.2 Linear Regression Analysis 177

2. Suppose that you have data on each of the following 5. According to a leading road and track magazine,
pairs of variables. For which pairs would you expect the following information gives the time in seconds
the correlation coefficient to be close to 1, for for a Mercedes to accelerate from zero to the indi-
which would you expect r to be close to 1, and for cated speed in miles per hour:
which would you expect r to be close to 0?
a. A young child’s height and weight. Speed (mph) 30 40 50 60 70 80
b. The age of a car and its book value.
Time (sec) 3.6 5.0 7.0 9.1 11.9 15.2
c. A child’s shoe size and the number of words in
the child’s vocabulary. Does the corresponding correlation coefficient in-
d. The number of hours per week that a student dicate a significant level of linear correlation be-
studies and the amount of money spent on food tween the two variables? If so, determine the
each week. equation of the regression line that best fits the
e. The number of hours per week that a student data. Estimate how long it would take a Mercedes to
studies and the resulting GPA. accelerate to 45 mph and to 90 mph. Which is more
f. The number of cartons of cigarettes sold in the likely to be accurate? Why?
United States and the tax revenue on cigarettes.
6. The following table gives the percentage of the U. S.
g. The number of hours a person sleeps nightly, on
gross domestic product (GDP) spent on health care
average, and the number of push-ups the person
over the years.
can do.
3. What is wrong with each statement?
Year 1960 1965 1970 1975
a. The correlation coefficient for the number of
push-ups an athlete can do and the time it takes Percentage 4.4 5.7 7.7 8.3
the athlete to run a mile is 1.25. Year
b. For a set of 1x, y2 data, the regression equation is
1980 1985 1990 1995

Y6.5922.158X and the correlation coef- Percentage 10 10.2 12.2 13.7

ficient is r  0.87. Source: 2000 Statistical Abstract of the United States.
c. If the correlation coefficient for a set of 1x, y2
data is negative, it follows that the smaller x is, Is there a significant level of linear correlation be-
the smaller y will be. tween these two variables? If so, what is the regres-
sion line? Estimate the percentage of GDP spent on
4. The following table gives some measurements on health care in 2000. If current trends continue,
the rate of chirping (per second) of the striped when will 20% of GDP be spent on health care?
ground cricket as a function of the temperature.
7. Repeat Problem 6, using 60, 65, 70, . . . , 90 for the
years instead of 1960, 1965, . . . How do the results
T (°F) 89 72 93 84 81 75 70
for the equation of the line of best fit and for the
correlation coefficient change? What would happen
Chirps 20 16 20 18 17 16 15 if you used 0, 5, 10, . . . , 30 for the years instead?
T (°F) 82 69 83 80 83 81 84 76 8. The electrical resistance R of a piece of metal de-
pends on the temperature T of the metal. An exper-
Chirps 17 15 16 15 17 16 17 14 iment was conducted by measuring the resistance
Source: Adapted from George W. Pierce, The Songs of Insects. Boston: in ohms in a piece of wire at different temperatures
Harvard University Press, 1948. in degrees Celsius. The results were as follows:

Determine the equation of the line that best fits this T 33.2 40.6 45.3 51.8 58.4
set of data. How does it compare to the equation we
estimated by eye in Example 4 of Section 2.3? Does R 4.71 4.80 4.93 5.02 5.17
the value of the correlation coefficient indicate a T 63.8 71.0 76.9 80.6 90.1
high degree of correlation between chirp rate and
air temperature? R 5.34 5.39 5.52 5.55 5.75
gord.3896.03.pgs 4/24/03 9:53 AM Page 178

178 CHAPTER 3 Fitting Functions to Data

Determine the correlation coefficient and the table below shows the total payroll, in millions of
equation of the regression line that best fits the dollars, for each of the major league teams during
data. Does the value for the correlation coefficient the 2000 baseball season and the number of games
suggest a linear relationship between the two each team won out of the maximum of 162.
quantities? a. Find the equation of the line that best fits the
9. Consider the data in Example 5 on shoe size versus number of wins during the 2000 baseball season
height. as a function of a team’s payroll.
a. Interchange the roles of H and S to consider S as b. Is there a significant level of correlation between
the independent variable and then find the cor- the two variables?
responding linear regression equation and corre- c. Based on your linear model in part (a), how many
lation coefficient. wins would $100 million have bought during this
b. How is this correlation coefficient related to the season?
one we obtained in Example 5? d. Do the regression equation and the correlation co-
c. In Example 5, we found S  0.51H  25.016. efficient support or contradict the claim that the
Solve this equation algebraically for H as a func- teams with larger bankrolls are buying more wins?
tion of S. e. Write a short essay with your views on this
d. Explain why the results in (a) and (c) differ. issue. Be sure to include appropriate mathemat-
(Hint: The linear regression equation is based on ical arguments, based on your findings in parts
minimizing the sum of squares of the vertical (a) and (b).
distances from the points to the line.) 11. The great chemist Mendeleev once conducted an
10. Many baseball fans are concerned that the team experiment relating the solubility of sodium nitrate
owners with the most money are able to “buy” the in water to the temperature of the water, in degrees
best players and so dominate their leagues. The Celsius. He obtained the data on the next page.

Team Payroll Wins Team Payroll Wins

New York Yankees $113.37 87 Toronto Blue Jays 59.22 83
Atlanta Braves 95.01 95 Tampa Bay Devil Rays 55.16 69
Los Angeles Dodgers 94.22 86 San Francisco Giants 54.24 97
Boston Red Sox 93.87 85 Houston Astros 52.00 72
New York Mets 89.75 94 Chicago Cubs 51.08 65
Arizona Diamondbacks 80.76 85 Chicago White Sox 36.94 95
Cleveland Indians 78.72 90 Philadelphia Phillies 36.68 65
St. Louis Cardinals 72.38 95 Cincinnati Reds 35.13 85
Seattle Mariners 62.55 91 Milwaukee Brewers 33.77 73
Texas Rangers 61.36 71 Oakland A’s 32.69 92
Detroit Tigers 60.60 79 Pittsburgh Pirates 31.94 69
Baltimore Orioles 59.22 74 Montreal Expos 27.97 67
Anaheim Angels 58.74 82 Florida Marlins 25.86 79
Colorado Rockies 56.05 82 Kansas City Royals 24.47 77
San Diego Padres 54.68 76 Minnesota Twins 15.82 69
gord.3896.03.pgs 11/21/03 10:30 AM Page 179

3.2 Linear Regression Analysis 179

e. What might be appropriate values for the do-

Temperature 0 4 10 15 main and range of this linear function?
Solubility 66.7 71.0 76.3 80.6 12. The table at the bottom of the following page
gives Olympic gold medal times, in seconds, for
Temperature 21 29 36 51 68
the 100-meter freestyle for men and women. For
Solubility 85.7 92.9 99.4 113.6 125.1 each data set, find the best-fit line. What is the
practical significance of each line’s slope? Be-
a. Based on the scatterplot of the data and the cause the slopes are different, determine the
value of the correlation coefficient, is it rea- point where the two lines intersect and tell what
sonable to use a linear function to model this this point means. Is it reasonable?
relationship? 13. In a standard experiment conducted in introducto-
b. Find the equation of the linear function that ry physics class, an object was dropped and its
best fits these data. speed v measured every tenth of a second. The fol-
c. Use your model to estimate the solubility at a lowing table gives the speed, in centimeters per sec-
water temperature of 40°C. ond, at different times. Find the equation of the
d. Explain why using the linear model to predict line that best fits the data. What are the units for
a value for the solubility S when T  5 the slope?
would not be appropriate.

t 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

v 0 98 196 295.6 391.8 490.2 587.8 606.1 783.8 881.7 979.9

14. Use the fact that 1 inch equals 2.54 cm to verify that c. Find the equation of the line that best fits the ex-
the value you obtained for the slope in Problem 13 perimental data.
(the acceleration due to gravity in the metric sys- d. What are the units for the slope of the line? What
tem) is equivalent to the value we obtained in Ex- is the significance of the slope?
ample 6 (U. S. customary units of measurement). e. Adjust the equation of the line so that it passes
15. A spring is mounted from the ceiling and hangs through the origin. (Note: The resulting relation-
straight down. When a mass is attached to the end of ship is known as Hooke’s law after the British sci-
the spring, as shown in the accompanying figure, the entist Benjamin Hooke and the slope, which
spring lengthens. The following data values were ob- depends on the particular spring used, is called
tained in an experiment where different masses m, the spring constant.)
in grams, were attached and the associated lengths 16. The following table gives estimates for the average
L, in centimeters, of the spring were recorded. temperature, in degrees Celsius, at the Earth’s sur-
face, worldwide, in different years.
m 0 100 200 300
L 0 3.9 7.9 12.0
Year 1880 1900 1920
m 400 500 600 700
Temperature 13.8 13.95 13.9
L 16.0 20.1 24.1 28.2
Year 1940 1960 1980 1999

Temperature 14.15 14.0 14.2 14.4

a. Decide which variable is independent and which

is dependent. a. Find the equation of the line that best fits these
b. Does the correlation coefficient indicate a sig- data.
nificant level of linear correlation between the b. Assuming that the trend continues, predict the
two variables? average surface temperature in 2020.
gord.3896.03.pgs 4/24/03 9:53 AM Page 180

180 CHAPTER 3 Fitting Functions to Data

17. The following table shows the trend in worldwide a. Find the equation of the linear function that best
grain production (wheat, rice, and corn, primarily), fits these data.
in millions of tons. b. What does the model predict for the amount of
grain produced in 2010?
Year 1965 1970 1975 1980 c. When does the model predict that the total
amount of grain produced will reach 2000 mil-
Amount 905 1079 1237 1430 lion tons?
Year 1985 1990 1995 1999 d. Write a paragraph describing the implications
of the fact that grain production is growing
Amount 1647 1769 1713 1855 roughly linearly while the population is grow-
Source: Lester R. Brown et al., Vital Signs 2000: The Environmen- ing roughly exponentially.
tal Trends That Are Shaping Our Future.

Olympic Gold Medal Times in Swimming

MEN’S TIMES WOMEN’S TIMES
Year Swimmer Time Year Swimmer Time
1908 Charles Daniels (US) 65.60 1912 Fanny Durack (AUST) 82.20

1912 Duke Kahanamoku (US) 63.40 1920 Ethelda Bleibtrey (US) 73.60

1920 Duke Kahanamoku (US) 61.40 1924 Ethel Lackie (US) 72.40

1924 Johnny Weissmuller (US) 59.00 1928 Albina Osipowich (US) 71.00

1928 Johnny Weissmuller (US) 58.60 1932 Helene Madison (US) 66.80

1932 Yasuji Miyazaki (JAP) 58.20 1936 Hendrika Mastenbroek (NETH) 65.90

1936 Ferenc Csik (HUN) 57.60 1948 Greta Andersen (DEN) 66.30

1948 Walter Ris (US) 57.30 1952 Katalin Szoke (HUN) 66.80

1952 Clarke Scholes (US) 57.40 1956 Dawn Fraser (AUST) 62.00

1956 Jon Hendricks (AUS) 55.40 1960 Dawn Fraser (AUST) 61.20

1960 John Devitt (AUS) 55.20 1964 Dawn Fraser (AUST) 59.50

1964 Don Schollander (US) 53.40 1968 Margo Jan Henne (US) 60.00

1968 Michael Wenden (AUS) 52.20 1972 Sandra Neilson (US) 58.59

1972 Mark Spitz (US) 51.22 1976 Kornelia Ender (E GER) 55.65

1976 Jim Montgomery (US) 49.99 1980 Barbara Krause (E GER) 54.79

1980 Jorg Woithe (E GER) 50.40 1984 Nancy Hogshead (US) 55.92

1984 Rowdy Gaines (US) 49.80 1988 Kristin Otto (E GER) 54.93

1988 Matt Biondi (US) 48.63 1992 Zhuang Yong (CHI) 54.65

1992 Aleksandr Popov (USSR) 49.02 1996 Le Jingyi (CHI) 54.50

1996 Aleksandr Popov (USSR) 49.02 2000 Inge de Bruijn (NETH) 53.83

2000 Pieter van den Hoogenband (NETH) 48.30

gord.3896.03.pgs 4/24/03 9:53 AM Page 181

3.3 Fitting Nonlinear Functions to Data 181

18. Use of the automobile has sparked immense c. According to this model, predict the cost of a
changes in human culture, both for good and for plan that provides 2000 free minutes.
bad. The following table shows the growth in the 20. Consider the four points 10, 02, 10, 12 , 11, 0 2 and
number of cars in use throughout the world, in mil- 11, 1 2. Two reasonable guesses for the best-fit line are
lions, since 1950. (a) the diagonal line passing through 10, 0 2 and
11, 1 2 and (b) the horizontal line y  12 . Using the
criterion that the best-fit line is the one with the
Year 1950 1960 1970 1975 1980 minimal value for the sum of the squares of the ver-
Cars 8 13 23 25 29 tical distances from each of the points to the line, de-
cide which of these two lines is a better fit. Find the
Year 1985 1990 1993 1995 1997 1999 actual best-fit line by using your graphing calculator.
How high is the value for the correlation coefficient?
Cars 32 36 34 36 38 39
21. The accompanying graph shows the scatterplot for
the points 11, 48 2, 12, 682 , 13, 932 and 14, 1142 along
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
That Are Shaping Our Future.
with the line y  20x  30.

a. Find the equation of the linear function that best

fits these data.
b. What is the significance of the slope of this line? 115
c. What does the model predict for the number of 105
cars in use worldwide in 2025? 95
d. When does the model predict that the number 85
of cars in use will reach 50 million? 75

19. An ad for wireless phone service listed the following 65

prices for monthly service: 55

45
$19.99 for 325 free minutes
$29.99 for 750 free minutes 0 1 2 3 4
$39.99 for 1100 free minutes
$79.99 for 3000 free minutes a. Find the sum of the squares associated with this
a. Find the equation of the linear function that best line.
fits these data. b. What changes would you make (increase or de-
b. What is the practical significance of the slope crease) to the slope and/or the vertical intercept to
and the vertical intercept? get a better fit?

3.3 Fitting Nonlinear Functions to Data

Although linear regression and correlation are extremely powerful methods, not all
relationships between two quantities are linear. If a scatterplot suggests a known
nonlinear pattern, you need the ability to fit the appropriate function, such as an
exponential function, a power function, or a logarithmic function, to the data.
Just as graphing calculators and software packages can calculate the equation
of the line that best fits a set of data, they can also find the equation of: (1) an ex-
ponential function of the form y  acx (or the form y  abx, which is the format
used by most calculators); (2) a power function of the form y  kx p (or the form
y  ax b , which is what most calculators show); or (3)a logarithmic function of the
form y  a  b ln x that fits the data. We show how to find such functions in the
examples in this section.
gord.3896.03.pgs 4/24/03 9:53 AM Page 182

182 CHAPTER 3 Fitting Functions to Data

E XAMPLE 1 The Growth of the U.S. Population from 1780 to 1900

The population of the United States, in millions, from 1780 to 1900, is shown
in Table 3.4. The corresponding scatterplot of population versus time is shown in
Figure 3.16.
P

80
70
60

Population 50
40
30
20
10
t
0 2 4 6 8 10 12
FIGURE 3.16 Decade

a. Find an appropriate function that best fits these population data and discuss its char-
acteristics.
b. Find the correlation coefficient and discuss its significance.
Solution

a. To simplify the calculations, we let the independent variable t represent the number
of decades since the year 1780. That is, 1780 corresponds to t  0, 1790 corresponds
to t  1, and so on until 1900, which corresponds to t  12.
The growth pattern for the U.S. population shown in Figure 3.16 clearly is not
linear; it is concave up. Based on our discussions of population growth in Chapter 2, we
expect the pattern is likely exponential. To check, we calculate the ratio of successive
terms, as shown in Table 3.4. The ratios are roughly constant, which suggests that the
pattern is roughly exponential. The discrepancies may be due to other factors (polit-
ical or economic, say), which could give the population a spurt in one decade while
slowing its growth during another time period, such as during the Civil War.

TABLE 3.4 U.S. Population (1780–1900)

Decade Year Population Ratio
0 1780 2.8 1.39  3.92.8
1 1790 3.9 1.36
2 1800 5.3 1.36
3 1810 7.2 1.33
4 1820 9.6 1.34
5 1830 12.9 1.33
6 1840 17.1 1.36
7 1850 23.2 1.35
8 1860 31.4 1.27
9 1870 39.8 1.26
10 1880 50.2 1.25
11 1890 62.9
1.21
12 1900 76.0
gord.3896.03.pgs 4/24/03 9:54 AM Page 183

3.3 Fitting Nonlinear Functions to Data 183

To determine an exponential function that fits these population values, we enter the
data in a calculator, say, and select exponential regression. The calculator responds
with the values of the parameters a  3.069 and b  1.321 for an exponential func-
tion of the form y  abx, so the function is
Y  3.06911.321 2 X
or equivalently, in terms of our variables,
P1t2  3.06911.3212 t,
where t is the number of decades since 1780.
The growth factor, 1.321, is very close to most of the ratios of successive popula-
tion values we calculated in Table 3.4. This value indicates that the U.S. population
was growing at a rate of about 32.1% per decade from 1780 to 1900, which is about
3% per year.
b. Figure 3.17 shows the original population data with this exponential function super-
imposed. The curve fits the data well. Moreover, the corresponding correlation coef-
ficient r  0.998 is very close to 1. For the n  13 data points, the table on the inside
back cover gives the critical value of r13  0.553, so the level of correlation is defi-
nitely significant. Therefore the exponential function is a good fit to the population
data from 1780 through 1900.
P

100

80 P = 3.069(1.321) t
Population

60

40

20

t
0 2 4 6 8 10 12
FIGURE 3.17 Decade

◆
Think About This For each of the decades t  0, 1, . . . , 12 since 1780, use the exponential function
P1t2  3.06911.3212 t to create a table of predictions for the U.S. population based
on this model. Use these values to then calculate the error between the actual value
of the population and this predicted value for each decade. ❐

We could also analyze this set of population data by letting t represent the
year itself, 1t  1780, 1790, . . . 2 or letting t represent the number of years since
1780 1t  0, 10, 20, . . . 2 . Each produces equivalent exponential functions with
different bases, but each function is an appropriate model for the situation. As a re-
sult, you must be careful to keep track of what the variables used represent.
You may wonder why we have considered only the U.S. population up to 1900
and not beyond. The reason is that the population does not follow an exponential
pattern quite as closely thereafter. Various factors, such as limitations on immigra-
tion, changes in lifestyle reflected in smaller families, and the end of westward ex-
pansion came into play during the twentieth century to slow the rate of population
gord.3896.03.pgs 4/24/03 9:54 AM Page 184

184 CHAPTER 3 Fitting Functions to Data

growth. We introduce a more sophisticated mathematical model for population

growth that takes such factors into account in Chapter 5 and consider the question
of fitting a better mathematical function to the data on the U.S. population up to
the present day.
The methods we used in Example 1 also apply to situations that can be mod-
eled by an exponential decay function, as illustrated in Example 2.

E XAMPLE 2 Level of a Drug in the Body

L-Dopa is a drug used to control the symptoms of Parkinson’s disease. The following
data give the amount L of L-Dopa in the bloodstream, in nanograms per milliliter, t
minutes after the drug was absorbed into the blood.

t 0 20 40 60 80 100 120 140 160 180 240 300

L 2950 2600 1550 1100 900 725 600 510 440 300 250 225
Source: Douglas Brown & Tom Timchek; unpublished manuscript.

Find an exponential function that models the level of L-Dopa in the blood as a function
of time and check its correlation coefficient for significance.
Solution The scatterplot shown in Figure 3.18 suggests that a decaying exponential
function is a reasonable model. An alternative might be to use a decaying power func-
tion with a negative power p. But a power function has a vertical asymptote at t  0 and
we are told that a fixed amount of the drug is in the blood at time t  0. Therefore a
power function would not be an appropriate choice.

3000
Level of L-Dopa

2000

1000

t
0 60 120 180 240 300
FIGURE 3.18 Time (minutes) from absorption

Using a calculator, we find that an exponential decay function that fits these data is
L1t2  2147.810.99092 t,
which indicates that the level of L-dopa in the blood decreases by almost 1% every
minute. This function is superimposed over the data in Figure 3.19 demonstrating a rea-
sonably good fit. The corresponding correlation coefficient is r  0.9545; it is negative
gord.3896.03.pgs 4/24/03 9:54 AM Page 185

3.3 Fitting Nonlinear Functions to Data 185

because the trend in the data is decreasing. Also, the critical value of r with n  12 data
points is r12  0.576, indicating that the fit is good.

3000

Level of L-Dopa
2000

1000

t
0 60 120 180 240 300
FIGURE 3.19 Time (minutes) from absorption

◆
We next consider fitting a power function to a set of data.

E XAMPLE 3 Number of Species on an Island

Biologists have long observed that the larger the area of a region, the more species that
inhabit it. The following table gives some data on the area (in square miles) of various
Caribbean islands in the Greater and Lesser Antilles and estimates on the number of rep-
tile and amphibian species living on each. (Note that Trinidad has been omitted because
the island is exceptionally rich in species and tends to distort the data. Trinidad is only 7
miles off the coast of Venezuela, and many species have been able to emigrate easily from
the mainland to the island.)
a. Determine a function that models the relationship between the number of species N
living on one of these islands and the area A of the island and find the correlation co-
efficient.
b. The area of Barbados is 166 square miles. Estimate the number of species of reptiles
and amphibians living there.

N
Island Area N
100
90 Redonda 1 3
80
Number of Species

70 Saba 4 5
60
50
Montserrat 40 9
40
Puerto Rico 3459 40
30
20 Jamaica 4411 39
10

0 15,000 30,000 45,000

A Hispaniola 29,418 84
FIGURE 3.20 Area (square miles) Cuba 44,218 76
gord.3896.03.pgs 4/24/03 9:54 AM Page 186

186 CHAPTER 3 Fitting Functions to Data

Solution

a. It is reasonable to think that the number of species depends on the area of the island,
so area A is the independent variable and the number of species N is the dependent
variable. Figure 3.20 shows the scatterplot of the data. Don’t be misled by the fact that
the last point is somewhat lower than the preceding one; it is not reasonable to expect
that a larger area will necessarily be home to fewer species, so the trend should not
turn back down. Note that the points corresponding to the first few entries in the table
are very close together because of the large horizontal scale needed. The points appear
to lie on top of one another, which can very easily distort our perception of the actual
pattern in the data.
The overall data pattern suggests either a power function with a positive power
p  1 or a logarithmic function, both of which are increasing and concave down.
However, a theoretical island of zero area would be home to no species, so the func-
tion should pass through the origin. Consequently, the model of choice is a power
function.
Using a calculator to construct a power function to fit these data, we obtain
N  3.055 A 0.310.
In Figure 3.21 this function is superimposed on the original data and captures the
trend in the data reasonably well. The corresponding correlation coefficient
r  0.998 indicates a very high level of correlation between the variables.

100
90
80
Number of Species

70
60
50
40
30
20
10
A
0 15,000 30,000 45,000
FIGURE 3.21 Area (square miles)

b. The area of Barbados is A  166 square miles, so the model predicts that

N  3.055 1166 2 0.310  14.902,

or about 15 species of reptiles and amphibians live there.
◆

Biologists have found that comparable results apply for virtually any other en-
vironment, large or small, and for any other species. Typically the power p is rela-
tively close to 0.3. Because 100.3  1.995, biologists use the rule of thumb that a
tenfold increase in the size of an environment leads to roughly double the number
of species inhabiting it.
gord.3896.03.pgs 4/24/03 9:54 AM Page 187

3.3 Fitting Nonlinear Functions to Data 187

E XAMPLE 4 The Spread of AIDS

The following table shows the accumulated total number of reported cases of AIDS in
the United States since 1983. Find an exponential function that can be used to model the
spread of AIDS and check its correlation coefficient for significance.

Year 1983 1984 1985 1986 1987 1988 1989 1990

Number of AIDS Cases 4,589 10,750 22,399 41,256 69,592 104,644 146,574 193,878
Year 1991 1992 1993 1994 1995 1996 1997 1998

Number of AIDS Cases 251,638 326,648 399,613 457,280 528,144 594,641 653,084 701,353
Source: U.S. Centers for Disease Control and Prevention.

Solution The scatterplot of the data points is shown in Figure 3.22. The pattern is in-
creasing and concave up for the most part. However, note that the rate of growth seems
to be diminishing in the last couple of years shown. It is reasonable to model the growth
in the total number of reported cases of AIDS with an exponential function where the
independent variable is the time t since 1980 and the dependent variable is the number
of cases of AIDS A.

A A
Number of AIDS Cases (thousands)
Number of AIDS Cases (thousands)

750 750

600 600

450 450

300 300

150 150

t t
0 1983 1986 1989 1992 1995 1998 0 1983 1986 1989 1992 1995 1998

FIGURE 3.22 FIGURE 3.23

The exponential function to fit this data, obtained by using a calculator, is

A1t2  5413.511.36262 t ,
where t is measured in years since 1980. The growth factor of 1.3626 indicates that the
overall growth rate is over 36% per year. The corresponding correlation coefficient is
r  0.9483. Based on n  16 data points, the critical value for r is r16  0.497, so there
is a significant level of correlation, which suggests that this function fits the data well.
However, when we superimpose this exponential function over the data, as shown in
Figure 3.23, the curve doesn’t match the pattern of the data. In particular, the exponen-
tial function grows more slowly than the number of cases of AIDS from 1987 to 1994
and thereafter grows much more rapidly.
◆
Because the spread of AIDS has been slower than exponential growth, it
makes sense to try a different model—say, a power function—which we illustrate
in Example 5.
gord.3896.03.pgs 4/24/03 9:54 AM Page 188

188 CHAPTER 3 Fitting Functions to Data

E XAMPLE 5 The Spread of AIDS (Continued)

Find a power function that can be used to model the spread of AIDS and compare the
results to those obtained with the exponential function in Example 4.
Solution We use the data in Example 4, with t representing the number of years since
1980 and A the total number of cases of AIDS in the United States. Using a calculator, we
obtain the power function that fits these data:
A1t2  233.287t 2.865.
The corresponding correlation coefficient is r  0.9961, which is considerably higher than
the value of r  0.9483 for the exponential function. This result suggests that the power
function fits the data better than the exponential function. Superimposing this power func-
tion over the data, as shown in Figure 3.24, confirms that the power function fits the data bet-
ter than the exponential function. However, like the exponential function, the power function
also is growing more rapidly than the number of cases of AIDS over the last few years shown.
A
Number of AIDS Cases (thousands)

750

600

450

300

150

FIGURE 3.24 t
0 1983 1986 1989 1992 1995 1998

◆
Together, Examples 4 and 5 illustrate an important point: When faced with a set of
data, you should not be content to consider only one type of function as a potential
model if there are several different families of functions that have the same behavior
pattern. Rather, it would make sense for you to examine what happens when you fit var-
ious types of functions to the data and then decide which one is the best fit. We return
to this example later in the book to show the effects of using other functions as models.
In Section 1.1, we discussed the growth in life expectancy over the years since
the beginning of the twentieth century. The graph showing this trend is presented
in Figure 3.25. Let’s investigate this situation.

70
Life expectancy (years)

60

50

40

30
t
0 1900 1920 1940 1960 1980 2000
FIGURE 3.25 Year of birth
gord.3896.03.pgs 4/24/03 9:54 AM Page 189

3.3 Fitting Nonlinear Functions to Data 189

E XAMPLE 6 Increase in Life Expectancy

The following table gives the life expectancy for children born in the United States in the
years shown since the beginning of the twentieth century.
a. Find an appropriate function to fit these data and check its correlation coefficient for
significance.
b. Predict the life expectancy of a child born in 2008.

Year, t 1900 1915 1930 1945 1960 1975 1990 2000

Life Expectancy, L 47.3 54.5 59.7 65.9 69.7 72.6 75.4 76.4
Source: 2000 Statistical Abstract of the United States.

Solution

a. The increasing, concave down pattern shown in Figure 3.26 suggests that either a power
function with 0  p  1 or a logarithmic function would be an appropriate choice for
a model. However, the function modeling life expectancy clearly cannot pass through
the origin (life expectancy was not 0 in the year 0), so a power function is not an appro-
priate choice. Note that the data values are growing more and more slowly over time, so
we might expect that a logarithmic function would be a reasonable model for the data.
Using a calculator, we find that the logarithmic function that fits these data is
L  561.93 ln t  4192.2,
L L

70 70

60 60
Longevity L

Longevity L

50 50

40 40

30 30
t t
0 1900 1920 1940 1960 1980 2000 0 1900 1920 1940 1960 1980 2000
Year of birth Year of birth

FIGURE 3.26 FIGURE 3.27

where t is the actual year. The result is given in terms of the base e  2.71828, which we
discussed in Section 2.6. Superimposed over the data in Figure 3.27, this function is a
reasonably good fit. The corresponding correlation coefficient is r  0.9820. The critical
value for n  8 data points is r8  0.707, so we have a high level of positive correlation.
b. When t  2008, the logarithmic function predicts a life expectancy for someone
born in 2008 of
L12008 2  561.93 ln 120082  4192.2  81.2 years.
◆
E XAMPLE 7
An object is dropped from the top of the 1450-foot-high Sears Tower in Chicago. The
following set of measurements show how far the object has fallen after each second.
gord.3896.03.pgs 4/24/03 9:54 AM Page 190

190 CHAPTER 3 Fitting Functions to Data

Time (sec) 1 2 3 4 5 6 7 8

Distance Fallen (ft) 16 64 144 256 401 574 786 1022

Find a function that gives the distance the object has fallen after any time t.
Solution From the scatterplot shown in Figure 3.28, it is clear that our best choices are
either an exponential function or a power function because the pattern is increasing and
concave up. However, at time t  0 the object has fallen a distance of 0 feet, so the curve
should pass through the origin. Thus an exponential function is not appropriate. Using
a calculator, we find that the power function fit is
D  16.004t 1.99977
with a correlation coefficient r  0.999999, which suggests a virtually perfect fit.
D

1200

1000

800

600

400

200
t
FIGURE 3.28 0 2 4 6 8

◆
In fact, the actual formula for the distance fallen, based on an application of
Newton’s laws of motion, is precisely
D  16t 2.
The coefficient 16 is measured in feet per second per second. Because this coefficient
actually is one half the quantity known as the acceleration due to gravity (denoted by
g  32 ft>sec2), we can write the formula as
1
D  gt 2.
2
Usually, knowing how high the falling object in Example 7 is above the ground
is far more important than how far it has fallen. If the object is dropped from a
height of 1450 feet and the distance it falls in t seconds is 16t 2, then subtracting the
distance fallen from the initial height of 1450 feet gives
H  1450  16t 2
as the object’s height above the ground at any time t. We now can easily answer the
question of how long it takes for the object to hit the ground. We set the height
H  0, and solve the resulting equation for t:
16t 2  1450
1450
t2   90.625
16
t  290.625  9.5 sec.
gord.3896.03.pgs 11/21/03 10:30 AM Page 191

3.3 Fitting Nonlinear Functions to Data 191

Thus it takes about 9.5 seconds until impact.

In general, if an object is dropped from any initial height y0 , its height y above
ground level at any time t is given by
y  y0  16t 2.
In the preceding examples, we illustrated the power of regression methods to
create a function to model a set of data. However, a number of other important is-
sues remain. First, we didn’t discuss how the different functions are calculated,
based on the data. Second, a number of problems can arise if you enter certain
kinds of numbers into either a calculator or computer program and then ask the
machine to calculate the best-fit function. In order to know what kinds of numbers
to avoid, you have to understand how the different functions are calculated. Third,
if certain kinds of numbers have to be avoided, you need a way to circumvent the
problems that can arise and create functions to fit the actual data values. Fourth,
most calculators use one method to find the equations of the functions that fit a set
of data, whereas some mathematical software packages, such as Derive™ and
Maple™, use a very different method. As a result, different functions from the same
family of functions will be obtained, depending on the technology employed.
Again, which function is the best to use requires an understanding of the methods
employed. Fifth, in each of the preceding examples, we were able to decide on a sin-
gle function to use. However, recall how when we weren’t satisfied with the results
of fitting an exponential function to the data on the spread of AIDS, we then tried
a power function. When you try different functions, how do you decide which
function is the best choice if all are appropriate mathematical models?
We discuss all these issues in the following sections. In this section, we focused
on creating a single function to fit a set of data and using that function to answer
predictive questions in that context.

Problems
1. For each scatterplot, select the type of function—
exponential, power, or logarithmic—that is the
most reasonable candidate to fit the data or decide
that none are appropriate.

(vii) (viii)

2. For each table of values, select the type of func-

tion—linear, exponential, power, or logarithmic—
that is most likely to fit the data.
(i) (ii) (iii)
a. b.
x y x y
1 14.8 1 14.5

2 7.4 2 11.3

3 3.6 3 8.2

(iv) (v) (vi) 4 1.9 4 4.9

5 0.9 5 1.6
gord.3896.03.pgs 4/24/03 9:54 AM Page 192

192 CHAPTER 3 Fitting Functions to Data

c. d. 200 minutes and (b) the length of time required

x y x y for the level of L-dopa in the blood to drop to
1 2 1 2.8 100 nanograms per milliliter?
8. Use the model created in Example 6 to estimate
2 9.8 2 4.8
(a) the life expectancy of a child born in 2004 and
3 25.1 3 8.1 (b) the year in which the life expectancy of a new-
born will be 85 years?
4 48.2 4 14.3
9. The table shows the number of species of nonflying
5 161.5 5 24.5 mammals living on various islands in the English
e. f. Channel, as well as the areas of the islands, in square
x y x y kilometers.
1 0 1 5
Island Herm Sark Alderney Guernsey Jersey
2 5.5 2 6.0
Area 1.3 5.2 7.9 63.5 116.3
3 8.7 3 6.8
Species 2 2 3 5 9
4 11.2 4 7.4
Source: Kevin Mitchell et al., Mathematical Models of Biological Systems. 1998.
5 12.8 5 7.7
a. Find the power function that models the num-
3. In Example 1, we created an exponential model for ber of species on these islands as a function of
the growth of the U.S. population from 1780 to 1900. area.
a. Use the model to predict the population in 1820 b. How large an island would be needed to support
and 1850. 15 species of nonflying mammals?
b. Calculate the error of using this model to predict 10. The table shows estimates for the number of species
the population in 1820 and 1850 by comparing of plants in different coastal regions of California
the values of the function to the actual popula- and the area, in square miles, of each region.
tion values in the table.
c. Use the model to predict the U.S. population in Region Area Species
1920 and 1950.
d. How large are the errors in these estimates? (The Tiburon Peninsula 5.9 370
actual U.S. population was 105.7 million in 1920 San Francisco 45 640
and 150.7 million in 1950.)
e. Based on these values, has the U.S. population Santa Barbara area 110 680
been growing faster or slower than the exponen- Santa Monica Mountains 320 640
tial model?
Marin County 529 1060
4. The Caribbean island of Guadeloupe covers 687 square
miles. Estimate the number of species of reptiles and Santa Cruz Mountains 1386 1200
amphibians that live on the island according to the
model in Example 3. Monterey County 3324 1400

5. How large a Caribbean island would be necessary, San Diego County 4260 1450
according to the model created in Example 3, to Source: Kevin Mitchell et al., Mathematical Models of
support 25 species of reptiles and amphibians? Biological Systems. 1998.
6. Use the model created in Example 4 to estimate a. Find the power function that models the num-
(a) the total number of cases of AIDS in the United ber of plant species in coastal regions of Califor-
States through 2001 and (b) the year in which the nia as a function of the area of the region.
total number of cases reaches 1 million. b. The area of Baja California is 24,210 square
7. Use the model created in Example 2 to estimate miles. What is your prediction for the number of
(a) the amount of L-dopa in the blood after t  species of plants in Baja?
gord.3896.03.pgs 4/24/03 9:54 AM Page 193

3.3 Fitting Nonlinear Functions to Data 193

c. The actual number of plant species in Baja Cali- a. Let t be the number of years since 1890. Deter-
fornia is about 1450. Because Baja California is mine the best linear, exponential, and power
considerably south of the other regions listed in functions to model the number of high school
the table, what does that suggest to you about the graduates as a function of t.
importance of latitude (north–south distance) to b. Use each function to predict the number of high
the validity of the model? school graduates in 2010. Which prediction
11. The accompanying table shows the number of seems the most reasonable? the least reasonable?
households, in millions, with cable television in c. Use each function to predict the year in which
various years. there will be 5 million high school graduates.
d. Which function seems to give the most reason-
able prediction? the least reasonable?
Year 1977 1980 1983 1986 1988 13. The table gives the number of violent crimes per
Number 12.1 17.7 34.1 42.2 48.6 100,000 people in the United States since 1960.

Year 1990 1992 1994 1996 1998 2000

Year 1960 1965 1970 1975
Number 54.9 57.2 60.5 64.7 67.0 68.5
Number of Crimes 175 200 360 490
Source: World Almanac and Book of Facts.
Year 1980 1985 1990 1995
a. Find a linear function, an exponential function,
Number of Crimes 580 550 750 685
and a power function that fit these data using t
as the number of years since 1975. Source: FBI, Crime in the United States.

b. Using each function, predict the total number of

a. Find the best linear and exponential fits to this
households with cable TV in the year 2005.
set of data, where t represents the number of
c. Which prediction do you think is most reason-
years since 1950.
able? Explain.
b. Use both models to predict the number of vio-
12. The table gives the total number, in thousands, of high lent crimes, per 100,000 people, that occurred in
school graduates in the indicated years since 1900. 2000.
c. With both models, predict when the number of
violent crimes will reach 1000 per 100,000 people.
Year High School Grads d. What is the doubling time for the exponential
1900 95 model?
14. According to Motor Trend magazine, the following
1910 156
data are the times (in seconds) it takes a Trans Am
1920 311 to accelerate from zero to the indicated speed.
1930 667
Speed (mph) 30 40 50 60 70
1940 1221
Time (sec) 3.00 4.29 5.52 7.38 9.81
1950 1200

1960 1858 a. Determine the best linear, exponential, and

1970 2889 power functions to model the acceleration time
as a function of the final speed.
1980 3043 b. Estimate how long it will take a Trans Am to ac-
celerate to 45 mph; to 80 mph; to 90 mph.
1990 2586
c. Which estimated time is most likely to be accu-
2000 2839 rate? (You might want to compare these data to
Source: Digest of Education Statistics 2000, the corresponding information about the Mer-
U.S. Department of Education. cedes in Problem 5 of Section 3.2.)
gord.3896.03.pgs 4/24/03 9:54 AM Page 194

194 CHAPTER 3 Fitting Functions to Data

15. According to the U.S. Department of Education, c. What is the significance of the positive slope for
the following data are the numbers, in thousands, the linear fit?
of college degrees awarded during the indicated d. What is the significance of the growth factor in
year from 1900 to 1995. the exponential fit being greater than 1?
18. The Dow-Jones average of 30 major industrial
stocks is probably the most widely watched meas-
Year 1900 1910 1920 ure of the stock market. The following table shows
College Graduates 30 54 73 the Dow-Jones average at the beginning of each
year since 1980.
Year 1930 1940 1950 1960

College Graduates 123 223 432 530 Year 1980 1981 1982 1983 1984 1985 1986

Year 1970 1980 1990 1995 Dow 839 964 875 1047 1259 1212 1547

College Graduates 878 935 1017 1165 Year 1987 1988 1989 1990 1991 1992 1993
Source: U.S. National Center for Educational Statistics. Dow 1896 1939 2169 2753 2634 3169 3301

Year 1994 1995 1996 1997 1998 1999 2000

a. Determine the best linear, exponential, power,
and logarithmic functions to model the number Dow 3758 3834 5177 6447 7965 9184 11,358
of college graduates as a function of t, the num- Source: Wall Street Journal.
ber of years since 1890.
b. Use each of the four functions to predict the a. Find the best linear, exponential, and power
number of college graduates in 2005. functions to fit these values for the Dow, using t
c. Which of the four predictions seems the most to represent the number of years since 1979.
reasonable? the least reasonable? b. Which of the three functions seems to best fit
d. Use each of the four models to predict the year in these data? Explain.
which there will be 2 million college graduates. c. What is your prediction, based on this function,
e. What is the doubling time for the exponential for the current value of the Dow? Check a newspa-
model? per or listen to the business news on the radio or
16. a. Compare the results in Problem 15 to the results television to find out how close your prediction is
in Problem 12 concerning the number of high to the actual value.
school graduates as a function of time. In partic- d. The Dow closed above 4000 for the first time on
ular, for the two exponential growth models, February 23, 1995. Which of the three models
which model is growing faster? comes closest to predicting that date?
b. If you project forward indefinitely with both e. The Dow closed above 5000 for the first time on
models, when will the number of college diplo- November 21, 1995. Which of the three models
mas awarded surpass the number of high school comes closest to predicting that date?
diplomas awarded? Discuss the reasonableness of f. Based on your best-fitting function, when do
this scenario. you predict the Dow will first reach 14,000?
17. a. Use the set of data from Problem 12 on high 19. The following table shows the cumulative number of
school diplomas awarded and the data from
HIV infections, worldwide, in millions, since 1980.
Problem 15 on college degrees awarded to create
a table consisting of the number of college de- a. Assuming that the number of people infected with
grees and the number of high school diplomas HIV is growing exponentially, find the exponen-
awarded each year from 1900 to 1990. tial function that fits these data, where the variable
b. Determine the linear, exponential, power, and t represents the number of years since 1979.
logarithmic functions that fit these data to con- b. What is the growth factor in your function and
struct four models relating the number of college what does it mean?
degrees awarded as a function of the number of c. Predict the number of people who will have
high school diplomas awarded the same year. been infected with HIV by 2005.
gord.3896.03.pgs 4/24/03 9:54 AM Page 195

3.3 Fitting Nonlinear Functions to Data 195

function that fits the data, where the variable t

Year 1980 1981 1982 1983 1984 represents the number of years since 1979.
Number 0.1 0.3 0.7 1.2 1.7 b. What is the growth factor in your function and
what does it mean?
Year 1985 1986 1987 1988 1989 c. Predict the total number of people who will have
Number 2.4 3.4 4.5 5.9 7.8 died from AIDS by 2005.
d. Find the power function that fits the data, where
Year 1990 1991 1992 1993 1994 the variable t represents the number of years
Number 10.0 12.8 16.1 19.7 23.8 since 1979.
e. Predict the number of people who will have died
Year 1995 1996 1997 1998 1999 from AIDS by 2005.
f. Does the exponential function or the power
Number 28.3 33.5 38.9 44.1 49.9
function seem to fit the data better? Explain.
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
That Are Shaping Our Future. 21. A growing percentage of the world’s population is now
living in urban areas, as shown in the following table.
d. How does the growth in the number of people
infected with HIV compare to the growth of this Year 1950 1955 1960 1965 1970
exponential function? That is, does the function
grow faster or slower than the actual data? Percentage 29.7 31.6 33.6 35.5 36.7
e. Find the power function that fits these data, Year 1975 1980 1985 1990 1995 1999
where the variable t represents the number of
years since 1979. Percentage 37.8 39.4 41.2 43.2 45.3 47.0
f. Predict the number of people who will have Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
been infected with HIV by 2005. That Are Shaping Our Future.
g. How does the growth in the number of people a. Find the equation of the linear function that best
infected with HIV compare to the growth of this fits the data, where t is the number of years since
power function? That is, does the power func- 1945.
tion grow faster or slower than the actual data? b. Predict the percentage of the world’s population
h. Does the exponential function or the power that will live in urban areas in 2020, based on the
function seem to fit the data better? Explain your linear function.
reasoning. c. Find the equation of the exponential function
20. The following table shows the cumulative number that fits the data.
of deaths from AIDS worldwide, in millions, since d. Predict the percentage of the world’s population
1984. that will live in urban areas in 2020, based on the
exponential function.
Year 1984 1985 1986 1987
e. Find the equation of the power function that fits
the data.
Number 0.1 0.2 0.3 0.5 f. Predict the percentage of the world’s population
that will live in urban areas in 2020, based on the
Year 1988 1989 1990 1991 1992 1993
power function.
Number 0.8 1.2 1.7 2.4 3.3 4.4 g. Which of the three functions appears to fit the
data best? Which of the three functions appears
Year 1994 1995 1996 1997 1998 1999
to give the most reasonable prediction for the
Number 5.7 7.3 9.2 11.3 13.7 16.3 percentage of the world’s population living in
Source: Lester R. Brown et al.,Vital Signs 2000: The Environmental Trends
urban areas in 2020? Which gives what appears to
That Are Shaping Our Future. be the least reasonable prediction? Explain.
22. The following table shows worldwide consumption
a. Assuming that the number of deaths from AIDS of natural gas, measured in the equivalent of mil-
is growing exponentially, find the exponential lions of tons of oil, over time.
gord.3896.03.pgs 4/24/03 9:54 AM Page 196

196 CHAPTER 3 Fitting Functions to Data

f. Predict the worldwide consumption of natural

Year 1950 1960 1970 1975 gas in 2020, based on the power function.
Natural Gas Used 187 444 1022 1199 g. Which of the three functions appears to fit the
data best? Which of the three functions appears
Year 1980 1985 1990 1995 1999 to give the most reasonable prediction for the
Natural Gas Used 1406 1640 1942 2116 2301 worldwide consumption of natural gas in 2020?
Which gives what appears to be the least reason-
Source: Lester R. Brown et al.,Vital Signs 2000: The Environmental Trends
That Are Shaping Our Future.
able prediction? Explain.
23. As the pressure of a liquid goes up in a confined
a. Find the equation of the linear function that space (say, in a pressure cooker), the boiling point
best fits the data, where t is the number of years also goes up. The following table gives the tempera-
since 1940. ture T of the boiling point of water, in degrees Cel-
b. Predict the worldwide consumption of natural sius, at various vapor pressures, P, in kilo-pascals.
gas in 2020, based on the linear function. a. Find the logarithmic function that fits the data.
c. Find the equation of the exponential function How good is the fit?
that fits the data. b. Use your function from part (a) to find the boiling
d. Predict the worldwide consumption of natural point of water when the vapor pressure is 6.2 kilo-
gas in 2020, based on the exponential function. pascals.
e. Find the equation of the power function that fits c. What vapor pressure is needed if the boiling
the data. point of the water is 120°C?

Pressure 0.61 1.22 2.34 4.25 7.38 12.34 19.93 31.18 47.37 70.12 101.32

Temp. 0° 10° 20° 30° 40° 50° 60° 70° 80° 90° 100°
Source: CRC Handbook of Chemistry and Physics, 1996.

3.4 How to Fit Exponential and

Logarithmic Functions to Data
In Section 3.2, we developed the ideas of correlation and linear regression that
allow us to determine whether a linear relationship exists between two variables,
and if so, to find the equation of the regression line that best fits the data. In Sec-
tion 3.3, we considered the natural extension of these ideas to fit nonlinear func-
tions such as exponential, power, and logarithmic functions to a set of data.
However, the approach we used there was just to quote the results obtained from a
calculator or a spreedsheet. Let’s now look at how the calculator finds these equa-
tions. In the process, you will come to understand some of the cautions we men-
tioned at the end of Section 3.3.
One of the key approaches used in practice when a set of data does not fall in a
linear pattern is the following three-step process.
1. Transform the data in some way so that the resulting transformed values
fall in a roughly linear pattern. We call this linearizing the data.
2. Once the data have been linearized, find the regression line and correlation
coefficient for the transformed data.
3. Undo the transformation by using the appropriate inverse function to produce
the equation of the nonlinear function that fits the original set of data.
Here, we consider modeling a data pattern with an exponential or a logarithmic
function; in Section 3.5, we consider modeling a data pattern with a power function.
gord.3896.03.pgs 4/24/03 9:54 AM Page 197

3.4 How to Fit Exponential and Logarithmic Functions to Data 197

Fitting Exponential Functions to Data

Let’s begin by considering the population of the United States, in millions, from 1780
to 1900, as shown in Table 3.4 (Example 1 of Section 3.3) with the corresponding scat-
terplot in Figure 3.29. Again, the independent variable t is the number of decades
since 1780. As we showed there, the successive ratios are fairly constant, which sug-
gests that the pattern is roughly exponential. The calculator gives the exponential
function as P1t2  3.069 11.3212 t with a correlation coefficient of r  0.998. Now
let’s see how this function is actually obtained, using transformations.

80
70
60

Population
50
40
30
20
10
t
0 2 4 6 8 10 12
FIGURE 3.29 Decade

E XAMPLE 1
Construct the exponential function that fits the data on the U.S. population from
1780–1900, using the transformation approach.
Solution To determine the exponential function that fits the population values shown
in Table 3.4, we first transform the population data by taking the logarithm of each pop-
ulation value, as shown in Table 3.5. (We discuss the reason for doing so later.)
The resulting scatterplot of log(population) versus time in decades, shown in Figure 3.30,
now indicates a roughly linear pattern. Checking the differences in successive values under
log(population) in Table 3.5 reveals that they are roughly equal. We therefore find that the
best linear fit to this set of transformed data, using the ideas from Section 3.2, is

TABLE 3.5 Logarithms of Population Values

Y (= log P)
Decade Population log(population)
2
0 2.8 0.447
1.8
1 3.9 0.591
1.6
2 5.3 0.724
log (Population)

1.4
3 7.2 0.857
1.2
4 9.6 0.982
1
5 12.9 1.111
0.8
6 17.1 1.233
0.6
7 23.2 1.365
0.4
8 31.4 1.497
X (= t)
0 2 4 6 8 10 12 9 39.8 1.600
Decade 10 50.2 1.701
11 62.9 1.799
FIGURE 3.30 12 76.0 1.881
gord.3896.03.pgs 4/24/03 9:54 AM Page 198

198 CHAPTER 3 Fitting Functions to Data

Y  0.121X  0.487,
where Y  log1population 2 and X is the number of decades since 1780.
The corresponding correlation coefficient is r  0.998. Comparing this value to
the critical value of r13  0.553 from the table on the inside back cover shows that, with
95% certainty, there is positive correlation between the decade and the logarithm of the
population from 1780 through 1900. Note that the value for the correlation coefficient,
r  0.998, for the linear fit to the transformed data values is exactly the same value that
a calculator gives as part of its exponential regression calculations.
However, in the regression equation Y  0.121X  0.487, the independent variable
X represents t and the dependent variable Y represents log P. That is, the regression equa-
tion actually represents
log P  0.121t  0.487.
We now undo the original transformation (the logarithm) of the data by applying the
inverse function, which is an exponential function with base 10, and by using the appro-
priate properties of exponents. We obtain
P  10log P  100.121t0.487 10log u  u
 100.121t . 100.487 10uv  10u . 10v
10u p  110u 2 p
.
 3.06911.3212 t,
where t is the number of decades since 1780. This new equation has the form P0ct for an
exponential function. This expression is identical to the exponential function that the
calculator gives automatically.
◆
Why the Transformation Approach Works
Let’s explore why we took the logarithm of the population values. Suppose that the
scatterplot for a set of data appears to follow an exponential pattern so that we
hope to fit an exponential function of the form y  f 1x 2  kcx to the data for
some constants c and k. If we take logarithms of both sides of this equation and use
properties of logs, we get
log y  log1kcx 2
 log k  log1c x 2 log AB  log A  log B
 log k  x log c log Ap  p log A
 1log c2 x  log k.
Because log c and log k are constants, this expression has the form
Y  aX  b,
where
Y  log y, a  log c, and b  log k.
Thus, if y is an exponential function of x, then log y is a linear function of x, and
this transformation (taking the logarithm of the y values) linearizes the data, as il-
lustrated in Figure 3.31. We then find the coefficients a and b using the linear re-
gression technique. Finally, we undo the transformation by taking powers of 10 of
both sides of the equation and apply the appropriate properties of exponents and
logarithms to get the desired exponential function, as illustrated in Example 1.
Example 2 demonstrates that this approach—based on transforming the data
values from 1x, y2 to 1x, log y2 —actually works.
gord.3896.03.pgs 4/24/03 9:54 AM Page 199

3.4 How to Fit Exponential and Logarithmic Functions to Data 199

y Y (= log y)

y = kc x
Y = (log c)X + log k

(X, log y)
(x, y)
x X (= x)

FIGURE 3.31 (a) (b)

E XAMPLE 2
Show that the linearization technique works for points on the exponential function
y  5 . 2 x.
Solution We select the points on the curve y  5 . 2x corresponding to x  0 (so that
y  5 . 20  5 2 , x  1, x  2, x  3, and x  4, as shown in Figure 3.32(a). These
points give us the first two columns in the following table. When we transform this set of
data by taking the logarithms of the y-values, we get the third column.
The scatterplot of the transformed data points 1x, log y2 shown in Figure 3.32(b) sug-
gests that the points appear to lie exactly on a straight line. If we take any two of the

y log y

2
80 1.8
1.6
60 1.4
1.2
40 1
0.8
0.6
20
0.4
0.2
x x
FIGURE 3.32 0 1 2 3 4 0 1 2 3 4

transformed points, we find that the slope of the line joining them is 0.301. Further-
more, the y-intercept is b  0.699. Thus the equation of the line through the trans-
x y log y formed data points is Y  0.301X  0.699, which is equivalent to
0 5 0.699 log y  0.301x  0.699.

1 10 1 We now undo the logarithm by taking powers of 10 on both sides of this equation:
10log y  y  100.301x0.699 10log u  u
 110 2 110 2
2 20 1.301 0.301x 0.699
10uv  10u . 10v
 1100.301 2 x 15.00032 10u p  110u 2 p
3 40 1.602 .

4 80 1.903  5.000311.99992 x,
which is effectively the exponential function y  5 . 2x that we started with when round-
ing errors are eliminated.
◆
We summarize this approach as follows.
gord.3896.03.pgs 4/24/03 9:54 AM Page 200

200 CHAPTER 3 Fitting Functions to Data

If the 1x, y2 data appear to follow an exponential pattern y  kc x,

1. plot log y versus x to linearize the data;
2. find the best linear fit,
log y  ax  b,
to the transformed 1x, log y2 data; and
3. undo the logarithmic transformation by taking powers of 10.

Virtually every graphing calculator has a built-in routine as part of its statis-
tics capabilities to find the exponential function that fits a set of data based on this
procedure.
Caution: In the process of transforming to log y versus x, all the y-values must
be positive because the logarithm is not defined for negative or zero values.
Many scientific software packages, as well as most spreadsheets, contain rou-
tines that will fit an exponential (or other) function to a set of data. However, some
mathematical software packages, such as Derive™, Maple™ and Mathematica™,
calculate a best fitting exponential function by applying the least squares criterion
directly. They find the values for k and c in y  kc x that minimize the sum of the
squares of the vertical deviations between the data points and the curve, as illus-
trated in Figure 3.33. The results obtained by using this method can be somewhat
different from the results obtained with the transformation approach to linearize
the data values. We use the transformation approach throughout the book and all
answers shown are based on it. Also, whenever we speak of a best fitting (nonlin-
ear) function, it is in the context of the transformation approach.

FIGURE 3.33

The transformation method also applies to situations that can be modeled by

an exponential decay function, as demonstrated in Example 3.

E XAMPLE 3
The following table and the graph shown in Figure 3.34 give the amount L of L-Dopa in
the bloodstream, in nanograms per milliliter, t minutes after the drug was absorbed into
the blood. Find an exponential function that models the level of L-Dopa in the blood as
a function of time, using the transformation approach.
gord.3896.03.pgs 4/24/03 9:54 AM Page 201

3.4 How to Fit Exponential and Logarithmic Functions to Data 201

t 0 20 40 60 80 100 120 140 160 180 240 300

L 2950 2600 1550 1100 900 725 600 510 440 300 250 225
Source: Douglas Brown and Tom Timchek; unpublished manuscript.

3000

2500

Level of L-Dopa
2000

1500

1000

500

t
0 60 120 180 240 300
FIGURE 3.34 Time (minutes) from absorption

Solution In Example 2 of Section 3.3, we found that the calculator gave

L1t2  2147.810.99092 t
as the exponential decay function that best fits these data.
To find this exponential function by using the transformation approach, we first
look at log L versus t.

t 0 20 40 60 80 100 120 140 160 180 240 300

log L 3.470 3.415 3.190 3.041 2.954 2.860 2.778 2.708 2.643 2.477 2.398 2.352

The corresponding scatterplot, shown in Figure 3.35, is reasonably linear. Applying a

calculator’s linear regression routine to this transformed data, we find that the resulting
regression equation that is the best linear fit to this transformed data is
Y  0.00396X  3.3320,
Y = log L

3.5
3.3
3.1
Log L

2.9
2.7
2.5
2.3
t
0 60 120 180 240 300
FIGURE 3.35 Time (minutes) from absorption

where Y  log L. The corresponding correlation coefficient is r  0.9545. Note that

both the slope and the correlation coefficient are negative because the trend in the
gord.3896.03.pgs 4/24/03 9:54 AM Page 202

202 CHAPTER 3 Fitting Functions to Data

data is decreasing. Also, the critical value is r12  0.576, indicating a significant level of
correlation.
Because we transformed the data by plotting log L versus t, the linear function we
got actually represents
log L  0.00396t  3.3320.
To undo the logarithm transformation, we take powers of 10 on both sides:
L  10log L  100.00396t3.3320 10log u  u
 100.00396t . 103.3320 10uv  10u10v
 1100.00396 2 t . 2147.8 10u p  110u 2 p
.

 2147.8 . 10.99092 t
so that the exponential decay model is
L1t2  2147.810.99092 t,
which is identical to the expression that the calculator gave.
◆

Fitting Logarithmic Functions to Data

Suppose now that a set of data falls in the increasing, concave down pattern shown
in Figure 3.36. Such a shape might suggest a logarithmic function of the form
y  f 1x 2  a log x  b.

FIGURE 3.36

(This scatterplot may also suggest a power function with 0  p  1; we consider

such cases in Section 3.5.) This expression for the logarithmic function suggests
that we should transform the original data set by comparing y to log x rather than
comparing y to x.
However, most calculators and computer packages that perform this calcula-
tion use natural logarithms with base e rather than logarithms with base 10. Thus
we actually fit logarithmic functions of the form
y  f 1x 2  a ln x  b
to the data. So, to linearize data that follows a logarithmic pattern, we plot y versus
ln x.
In Example 6 of Section 3.3, we created the logarithmic function
L  561.93 ln t  4192.2 to model the growth in life expectancy over the years
since the beginning of the twentieth century. The graph of this trend is shown in
Figure 3.37.
gord.3896.03.pgs 4/24/03 9:54 AM Page 203

3.4 How to Fit Exponential and Logarithmic Functions to Data 203

90

Life expectancy (years)

75

60

45

30

15

0 t
1900 1915 1930 1945 1960 1975 1990 2000
FIGURE 3.37 Year

E XAMPLE 4
The following table shows the trend in life expectancy for children born in the years
shown since the beginning of the twentieth century. Find the best logarithmic function
to fit these data.

Year, t 1900 1915 1930 1945 1960 1975 1990 2000

Life Expectancy, L 47.3 54.5 59.7 65.9 69.7 72.6 75.4 76.4
Source: 2000 Statistical Abstract of the United States.

Solution To fit a logarithmic function to the data, we need to compare life expectancy
L to the natural logarithm ln t of the year, as follows.

ln t 7.5496 7.5575 7.5653 7.5730 7.5807 7.5883 7.5959 7.6009

L 47.3 54.5 59.7 65.9 69.7 72.6 75.4 76.4

The graph of the transformed data 1ln t, L2 shown in Figure 3.38 appears to be reason-
ably linear and the best linear fit to this transformed data is
Y  561.93 X  4192.2,
Y=L L

80 80

70 70

60 60

50 50

40 40
X = ln t
0 7.54 7.55 7.56 7.57 7.58 7.59 7.6 7.61 0 1900 1920 1940 1960 1980 2000

FIGURE 3.38 FIGURE 3.39

gord.3896.03.pgs 4/24/03 9:54 AM Page 204

204 CHAPTER 3 Fitting Functions to Data

which really represents

L  561.93 ln t  4192.2.
Because this expression already gives L as a function of t, we need not detransform the
equation in order to solve for L. Figure 3.39 shows this function superimposed over the
data, and it is a reasonably good fit, though it doesn’t bend as much as the data set does.
The corresponding correlation coefficient is r  0.9820, which indicates a high level of
positive correlation compared to the critical value r8  0.707. Incidentally, note that this
function is identical to the one found by calculator in Example 6 of Section 3.3.
◆
We summarize this approach as follows.

If the data appear to follow a logarithmic pattern,

1. plot y versus ln x to linearize the data; and
2. find the best linear fit,
y  a ln x  b,
to the transformed data.

Incidentally, a plot of either log y versus x (for an exponential fit) or y versus

log x (for a logarithmic fit) is known as a semi-log plot. You may have used semi-log
paper in some of your laboratory courses to plot experimental data that follow ei-
ther an exponential or a logarithmic pattern.

Problems

In each problem use the transformation approach to find the

35
exponential or logarithmic function that best fits the data.
Deaths per 100,000 women

30
1. (a) Repeat the analysis of the growth of the U.S. Lung cancer
25 Stomach cancer
population presented in Example 1, but concen-
trate on the period from 1780 to 1890 instead of 20
1780 to 1900. (b) Does this exponential function 15
give a better fit? How do you know? 10
2. The accompanying graph shows the number of 5
deaths per 100,000 women in the United States from
both stomach cancer and lung cancer since 1930. 0 1930 1940 1950 1960 1970 1980 1990
Both sets of data appear to be exponential functions. Year

a. Find the best exponential function that fits each

d. Can you give any reasons why the trend for
set of data.
stomach cancer (like that for most other can-
b. What are your estimates for the number of
cers) is decreasing exponentially, while the trend
deaths per 100,000 women from each type of
for lung cancer is rising exponentially?
cancer in 2005?
c. About how many women died in 1990 from 3. The table shows the growth of the federal debt in
stomach and lung cancer according to these billions of dollars from 1940 to 2000.
models if there were approximately 100 million a. Determine the exponential function that best
adult women in the United States at that time? fits these data.
gord.3896.03.pgs 4/24/03 9:54 AM Page 205

3.4 How to Fit Exponential and Logarithmic Functions to Data 205

b. Use this model to predict the federal debt in e. Consult the current edition of the Statistical Ab-
2005. stract of the United States to find the correct
value for the median family income in 2000.
Year 1940 1950 1960 1970 1980 1990 2000 How close is your prediction in part (a) to the
actual value?
Debt 651 257 291 381 909 3207 5686
7. Suppose that the pattern you found in Problem 6
Source: 2000 Statistical Abstract of the United States. for the growth in median family income continues
4. The table shows the growth of the U.S. population without change. In addition, suppose that infla-
in millions from 1940 to 1990. tion “remains under control” for the foreseeable
future and is limited to about 3% per year. Write a
a. Determine the exponential function that best short interpretation of what these two trends, if
fits these data. they continue without change, mean in terms of
b. Use this model to predict the U.S. population in the standard of living in 20 years.
2000.
c. According to the 2000 census, the actual U.S. pop- 8. The table shows the Dow-Jones average for 30 indus-
ulation in 2000 was 281.4 million. Did the model trial stocks at the beginning of each year since 1990.
give a reasonably accurate prediction? Explain.
Year 1990 1991 1992 1993 1994
Year 1940 1950 1960 Dow 2753 2634 3169 3301 3758
Population 131.7 150.7 179.3 Year 1995 1996 1997 1998 1999 2000
Year 1970 1980 1990 Dow 3834 5177 6447 7965 9184 11,358
Population 203.3 226.5 248.7 Source: Wall Street Journal.

5. Use the data in Problems 3 and 4 including the fact a. Find the exponential function that best fits these
that the U.S. population was 281.4 million in 2000 data.
to construct a set of values representing the average b. What is your prediction, based on this function,
amount of the national debt per person in the Unit- for the current value of the Dow? Check a news-
ed States every 10 years from 1940 to 2000. Deter- paper or listen to the business news on the radio
mine an appropriate function that best fits these or television to find out how close the prediction
data. If these trends continue, estimate your share is to the actual value.
of the debt in 2005. c. The Dow closed above 4000 for the first time on
6. The table at the bottom of the page shows the me- February 23, 1995. How close did the model
dian family income, in current dollars, in the Unit- come to predicting that date?
ed States from 1970 to 1992. d. The Dow closed above 5000 for the first time on
November 21, 1995. How close did the model
a. Determine the best exponential function to fit
come to predicting that date?
these data on median family income by year.
e. Based on your best fitting function, when do you
b. Use your result in part (a) to predict the median
predict the Dow will first reach 14,000?
family income in 2000 and in 2005.
f. How do the results of this problem compare to
c. Using your model, determine when median
those of Problem 18 in Section 3.3, which was
family income will reach $50,000.
based on the value of the Dow-Jones average
d. What is the doubling time for the exponential
from 1980 to 2000?
model?

Year 1970 1975 1980 1985 1990 1993 1995 1997 1999

Median family income 9,867 13,719 21,023 27,735 35,353 36,959 40,611 44,568 48,950
Sources: 2000 Statistical Abstract of the United States and U.S. Bureau of the Census.
gord.3896.03.pgs 4/24/03 9:54 AM Page 206

206 CHAPTER 3 Fitting Functions to Data

9. In Problem 18 of Section 3.3, you were asked to find c. According to your model, what do you predict
the exponential function that best fits the Dow- for the total wind energy generating capacity in
Jones average from 1980–2000. 2010?
a. Suppose that your grandmother wants to invest 12. The table shows the worldwide production of photo-
heavily in the stock market. Write a paragraph voltaic cells used for collecting solar energy over
based on the results of Problem 8 above or Prob- time. The units are the energy equivalent of the cells
lem 18 of Section 3.3 and your interpretations that in megawatts.
might convince her to be more conservative.
b. Suppose that you are the aggressive stockbroker
who is trying to convince her to invest heavily. Year 1975 1980 1985 1990
Write another paragraph based on the results Solar energy 1.8 6.5 22.8 46.5
from either Problem 8 above or Problem 18 of
Section 3.3 that might convince her to let you in- Year 1993 1995 1997 1999
vest her life savings for her.
Solar energy 60.1 78.6 125.8 201.3
10. The table shows worldwide car production, in
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental
millions. Trends That Are Shaping Our Future.

Year 1950 1960 1970 1980 1990 1999 a. Find the exponential function that best fits these
data.
Cars 9 14 22 30 38 39 b. What is the doubling time for this exponential
Source: Lester R. Brown et al., Vital Signs 2000: The Environmen- function? Explain what it means.
tal Trends That Are Shaping Our Future. c. Write a paragraph comparing the growth rate for
the total energy equivalent of the photovoltaic
a. Find the exponential function that best fits this cells to the growth rate for the total wind energy
data. generating capacity in Problem 11. What is the
b. According to your model, what do you predict likely long-term significance of this difference?
for the number of cars produced in 2010? d. According to your model, what do you predict
11. The table shows worldwide wind energy generating for the total energy equivalent of the photovolta-
capacity, in megawatts, over time. ic cells produced in 2010?
13. There is considerable discussion worldwide about
the growing levels of carbon dioxide 1CO2 2 in the
Year 1980 1985 1988 1990
atmosphere because of its effects on global warm-
Wind Energy 10 1020 1580 1930 ing. The table below shows the atmospheric con-
centrations of CO2 , in parts per million, over time.
Year 1992 1995 1997 1999
a. Find the equation of the exponential function
Wind Energy 2510 4820 7640 13840 that best fits these data.
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends b. What is the doubling time for this exponential
That Are Shaping Our Future. function? Explain what it means.
c. What does the model predict for the atmospher-
a. Find the exponential function that best fits the ic concentration of CO2 in 2010?
data. d. When does the model predict that the carbon
b. What is the doubling time for this exponential dioxide concentration will reach 400 parts per
function? Explain what it means. million?

Year 1960 1970 1975 1980 1985 1990 1993 1995 1997 1999

CO2 Concentration 316.7 325.5 332.0 338.5 345.7 354.0 357.0 358.8 363.9 368.4
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That Are Shaping Our Future.
gord.3896.03.pgs 4/24/03 9:54 AM Page 207

3.4 How to Fit Exponential and Logarithmic Functions to Data 207

14. Tourism is a booming global pastime and has be- d. When does the model predict that the number
come an important economic base for many na- of phones in use will reach 1 billion?
tions. The following table shows the growth in the e. The exponential function P1t2  3.611.0132 t can
number of international tourist arrivals, in millions, be used to model the growth of the world’s pop-
since 1950. ulation. Clearly, the growth factor for the popula-
tion is considerably smaller than the growth
factor for the number of phones. If both trends
Year 1950 1960 1970 1975 1980 continue, when will there be one phone for every
Tourists 25 69 166 223 286 person living? (Hint: Think about units.)
16. The table shows the growth in the number of cell
Year 1985 1990 1993 1995 1997 1999
phones, in millions, in use throughout the world
Tourists 328 459 519 569 620 657 since 1985.
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
That Are Shaping Our Future.
Year 1985 1988 1990 1991 1992
a. Find the equation of the exponential function Cell Phones 1 4 11 16 23
that best fits these data.
b. What is the doubling time for this exponential Year 1993 1994 1995 1996 1997 1998
function? Explain what it means. Cell Phones 34 55 91 142 215 319
c. What does the model predict for the number of
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
international tourists in 2010? That Are Shaping Our Future.
d. When does the model predict that the number
of international tourists will reach 1 billion? a. Find the equation of the exponential function
15. The table shows the growth in the number of tele- that best fits these data.
phones, in millions, in use throughout the world b. What is the doubling time for this exponential
since 1960. function? Explain what it means.
c. What does the model predict for the number of
cell phones in use in 2020?
Year 1960 1970 1975 1980 1985 d. When does the model predict that the number
Telephones 89 156 229 311 407 of cell phones in use will reach 1 billion?
17. The table below shows the growth in the number of
Year 1990 1993 1995 1997 1998
computers connected to the Internet, in thousands,
Telephones 520 606 691 788 844 throughout the world since 1985.
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends a. Find the equation of the exponential function
That Are Shaping Our Future. that best fits these data.
b. What is the doubling time for this exponential
a. Find the equation of the exponential function function? Explain what it means.
that best fits these data. c. What does the model predict for the number of
b. What is the doubling time for this exponential computers that can access the Internet in 2020?
function? Explain what it means. d. When does the model predict that the number
c. What does the model predict for the number of of computers that can access the Internet will
phones in use in 2020? reach 250 million?

Year 1985 1988 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

Internet
Connections 2.3 80 376 727 1313 2217 5846 14,352 21,819 29,670 43,230 72,398
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That Are Shaping Our Future.
gord.3896.03.pgs 4/24/03 9:54 AM Page 208

208 CHAPTER 3 Fitting Functions to Data

Exercising Your Algebra Skills

Several sets of data have been linearized to find the best 3. Y  1.0729  0.0223X
exponential fit. The calculator gives the following equa- 4. Y  0.3010  0.0706X
tions for the lines that best fit the transformed data. 5. Y  0.3522  1.0843X
Undo the transformations to get the corresponding
6. Y  1.3015  0.7840X
exponential functions.
7. Y  0.8525  1.2733X
1. Y  0.7782  0.0219X
8. Y  1.581  0.903X
2. Y  1.3010  0.0128X

3.5 How to Fit Power Functions to Data

Suppose that we have a growth pattern in a set of data that seems to be less extreme
than exponential growth or exponential decay or one that is concave down. We
might then suspect that a power function of the form y  f 1x 2  kx p, where k and
p are two constants to be determined, is the most appropriate model. As with the
exponential function in Section 3.4, we need a way to linearize the data with an ap-
propriate transformation so that we can apply linear regression techniques. We
then undo the transformation to find the power function that fits the original data.
To begin, we take logarithms of both sides of the equation y  k . x p and use
properties of logs:
log y  log1kx p 2
 log k  log1x p 2 log1u . v2  log u  log v
 log k  p log x log1u p 2  p log u
Because both log k and p are constants, we can interpret the equation
log y  log k  p log x
as saying that log y is a linear function of log x. That is,
log y  a log x  b
is of the form
Y  aX  b,
where
a  p and b  log k.
Therefore, if y is a power function of x, we can linearize the original data set by tak-
ing the logarithms of both the x- and y-values. Thus, if y versus x follows a power
function pattern, log y versus log x will follow a linear pattern. We then apply the
linear regression technique to determine a and b and finally undo the transforma-
tion to get the desired values for p  a and k  10b.

E XAMPLE 1
Show that the linearization technique works for points on the power function curve
y  5x 2.
gord.3896.03.pgs 4/24/03 9:54 AM Page 209

3.5 How to Fit Power Functions to Data 209

Solution We select the points on the curve y  5x 2 corresponding to x  1 (so

y  5 . 1  5 2 , x  2, x  3, and x  4. These points give the first two columns in the
2

following table. Figure 3.40 shows the curve through these points. When we transform this
set of data by taking the logarithms of both the x- and y-values, we get the other two
columns in the table.

y
x y log x log y
1 5 0 0.699
80 (4, 80)
2 20 0.301 1.301

3 45 0.477 1.653
y= 5x 2 4 80 0.602 1.903

10
(1, 5)
x
FIGURE 3.40 0 1 2 3 4 5

Figure 3.41 shows the transformed data points 1log x, log y2, and they appear to lie
on a line. Taking any two of the transformed points gives the slope of the line as 2. Fur-
thermore, the y-intercept is b  0.699, so
log y  2 log x  0.699
 log x 2  0.699. p log x  log x p

Y (= log y)

2
1.8
1.6
1.4
log y

1.2
1
0.8
0.6
X (= log x)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
FIGURE 3.41 log x

We undo the logs by taking powers of 10 on both sides of the equation:

10log y  y  101log x 0.6992
2
10log u  u
 100.699 . 10log x 10uv  10u . 10v
2

 5 . x 2. 10log u  u
So the original data points do indeed lie on the curve y  5x 2.
◆
We now apply these ideas to a variety of real world examples.
gord.3896.03.pgs 4/24/03 9:54 AM Page 210

210 CHAPTER 3 Fitting Functions to Data

E XAMPLE 2
In Example 3 of Section 3.3, we used data on the number of species of reptiles and am-
phibians living on different Caribbean islands in the Greater and Lesser Antilles to find a
relationship between the number of species N and the area A of the habitat. We found
the power function N  3.055A0.310. Find this function, using the transformation ap-
proach for the data given in the following table.
N
Island Area N
100
Redonda 1 3
Number of Species

80
Saba 4 5
60
Montserrat 40 9
40
Puerto Rico 3459 40
20
Jamaica 4411 39
A
0 15,000 30,000 45,000 Hispaniola 29,418 84
FIGURE 3.42 Area (square miles)
Cuba 44,218 76

Solution Figure 3.42 shows the scatterplot for these data. The increasing and concave
down pattern suggests a power function with 0  p  1 as the model of choice, as discussed
in Example 3 of Section 3.3. This choice is reinforced because such a function passes through
the origin and, if there were an island with A  0 area, it would be home to N  0 species.
To fit a power function N  kAp to the original data using the transformation ap-
proach, we extend the preceding table to include two additional columns, one for log N
and the other for log A. We calculate these quantities and then plot log N versus log A, as
shown in Figure 3.43.
Y = log N
Island Area N log A log N
2
Redonda 1 3 0 0.48
1.6
Saba 4 5 0.602 0.70
1.2
Montserrat 40 9 1.602 0.95
0.8 Puerto Rico 3459 40 3.539 1.60
0.4 Jamaica 4411 39 3.645 1.59
x = log A Hispaniola 29,418 84 4.469 1.92
FIGURE 3.43 0 1 2 3 4 5
Cuba 44,218 76 4.646 1.88

These points fall in a roughly linear pattern. Using the linear regression routine on a
calculator, we find that the equation of the regression line for the transformed data
points (log A, log N) is
Y  0.310X  0.485.
In the present context, this equation is equivalent to
log N  0.310 log A  0.485
 log1A0.310 2  0.485. p log u  log u p
gord.3896.03.pgs 4/24/03 9:54 AM Page 211

3.5 How to Fit Power Functions to Data 211

To solve for N, we use properties of exponents and logarithms:

N  10log N  103log 1A 2 0.4854
0.310
10log u  u
 10log1A 2 . 100.485 10uv  10u . 10v
0.310

 A0.310 13.0552. 10log u  u

Thus we have the power function N  3.055A0.310, which is the same function we found
in Section 3.3 using a calculator regression routine for the power function. When we su-
perimpose this power function over the original data values as shown in Figure 3.44, it
fits the data well. In fact, the correlation coefficient for the data is r  0.9982, which also
indicates that the fit is extremely good.
N

100
Number of Species
80

60

40

20

A
0 15,000 30,000 45,000
FIGURE 3.44 Area (square miles)

◆
We summarize this approach as follows.

If the data appear to follow a power function pattern y  kx p,

1. plot log y versus log x to linearize the data;
2. find the best linear fit,
log y  a log x  b,
to the transformed data; and
3. undo the logarithmic transformation by taking powers of 10.

E XAMPLE 3
The following table gives the takeoff weights, in thousands of pounds, of various jet liners
(weight of the plane plus fuel plus passengers) and their wingspans, in feet.
a. Decide which variable is the independent variable and which is the dependent vari-
able.
b. Find a power function that fits these data, using the transformation approach.

Solution

a. An aeronautical engineer designing a new plane would likely start with the desired
load, obtain the total weight, and then calculate the wingspan needed to support
that load. Thus the weight W is the independent variable and the wingspan S is the
dependent variable.
gord.3896.03.pgs 4/24/03 9:54 AM Page 212

212 CHAPTER 3 Fitting Functions to Data

Airplane Weight Wingspan

Boeing 707 330 145.7
Boeing 727 209.5 108

Boeing 737 117 93

Boeing 747 805 195.7

Boeing 757 300 156.1

DC8 350 148.5

DC9 121 93.5

DC10 572 165.4

Source: Michael Taylor (ed.), Jane’s Encyclope-
dia of Aviation. Crescent Books, 1993.

b. Figure 3.45 shows the scatterplot of the data with S as a function of W. The pattern is
increasing and concave down, so a power function with a power p between 0 and 1 is
reasonable. Also, it makes sense to have a function that passes through the origin—a
plane that weighs 0 pounds will have a wingspan of 0 feet. To find the power function,
we first transform the data by plotting log S versus log W, as shown in Figure 3.46. The
pattern of the transformed data is quite linear, and the line that best fits it is
Y  0.3942X  1.157

S Y = log S

200 2.4

2.3
Wingspan

150
2.2

100 2.1

50 2

1.9
W
0 200 400 600 800 1000
X = log W
Takeoff weight 0 2 2.2 2.4 2.6 2.8 3

FIGURE 3.45 FIGURE 3.46

The correlation coefficient is r  0.9688, which indicates a high level of correlation

between the variables. In this context, this linear function is equivalent to
log S  0.3942 log W  1.157
 log1W 0.3942 2  1.157. p log u  log u p
We find S by taking powers of 10 on both sides of this equation:
10log S  S  103log 1W 2 1.157 4
0.3942
10log u  u
 10log 1W 2 . 101.157
0.3942
10uv  10u # 10v
 W 0.3942 114.352 , 10log u  u
gord.3896.03.pgs 4/24/03 9:54 AM Page 213

3.5 How to Fit Power Functions to Data 213

which is the desired power function. Figure 3.47 shows this function superimposed
over the original data points, and it fits the measurements well.

200

150

100

50

W
FIGURE 3.47 0 200 400 600 800 1000

Summary of Curve Fitting Procedures

If the data appear to follow an exponential pattern, y  kc x,
1. plot log y versus x to linearize the data;
2. find the best linear fit, log y  ax  b, to the transformed data; and
3. undo the transformation, using the inverse function.
If the data appear to follow a power function pattern, y  kx p,
1. plot log y versus log x to linearize the data;
2. find the best linear fit, log y  a log x  b, to the transformed data;
3. undo the transformation, using the inverse function.
If the data appear to follow a logarithmic pattern, y  k ln x  d,
1. plot y versus ln x to linearize the data; and
2. find the best linear fit, y  a ln x  b, to the transformed data.

Cautions When Transforming Data with Logarithms

You should be aware of one major problem when using any of the curve fitting
procedures involving logarithmic transformations of the data. Recall that the loga-
rithmic function y  log x is defined only for values of x greater than 0. Thus, if
any of your data values are 0 or negative, you cannot take their logarithms. Often
you can circumvent this difficulty by redefining the independent variable. For ex-
ample, suppose that the data represent values of a quantity versus time starting in
1950. You could count the years since 1950, but then 1950 corresponds to t  0,
which causes a problem when you take logs. Alternatively, you could count the
years since 1900 because 1950 then corresponds to t  50, which circumvents the
problem of log 0 being undefined. For that matter, you simply could count the
years from 1949 so that 1950 corresponds to t  1. You could even use the full year
1950 itself, although having t  1950 creates potential rounding errors owing to
gord.3896.03.pgs 4/24/03 9:54 AM Page 214

214 CHAPTER 3 Fitting Functions to Data

the size of the numbers, as we previously discussed. Whatever you do, be sure to
keep track of what your independent variable represents.

Cautions when Fitting Power Functions to Data

Some serious problems can arise when you are fitting power functions to data.
Consider the data values shown.

x 1 2 3 4 5

y 400 410 430 460 500

The scatterplot shown in Figure 3.48 indicates clearly that the pattern in the data is
increasing and concave up. Using a calculator, we find that the power function that
best fits these data is y  387.43 x 0.129, with correlation coefficient r  0.907. The
correlation coefficient indicates a fairly good fit (the critical value for r with n  5
data points is r5  0.878). However, superimposing the graph of this power function
on the scatterplot, as shown in Figure 3.49, indicates that this function is concave
down. This result is also evident because the power p  0.129  1. Clearly, this
power function completely misses the trend in the data. Let’s see why.

y y

550 550

500 500

450 450

400 400

350 350
x x
0 1 2 3 4 5 0 1 2 3 4 5

FIGURE 3.48 FIGURE 3.49

We know that every power function with a positive power p passes through the
origin. Although five points are listed in the preceding table, the act of fitting a
power function to the data automatically introduces the origin as an additional
point, as shown in the following table.

x 0 1 2 3 4 5

y 0 400 410 430 460 500

Worse, considering the large jump from the origin to the point (1, 400), we see that
any power function that attempts to follow the trend in all six points must be con-
cave down. As a result, a power function is not an appropriate model to use for this
set of data.
gord.3896.03.pgs 4/24/03 9:54 AM Page 215

3.5 How to Fit Power Functions to Data 215

In general, whenever you attempt to fit a power function to a set of data, you
must look at the numbers and the scatterplot to decide whether a power function
is an appropriate model. You must also keep in mind that the point (0, 0) is auto-
matically added to the data if the pattern is increasing.
You can circumvent this problem if you adjust the original data values before
fitting a power function and then undo the adjustment after the fact. To avoid the
large jump from the origin to the given data values, first adjust or shift the y-values
down (we discuss such shifts formally in Section 4.4) by reducing each y-value in
the original table by 399. We then get the values in the following table.

x 1 2 3 4 5

y 400 410 430 460 500

Y  y  399 1 11 31 61 101

Using a calculator, we find that the power function that best fits these adjusted
values is Y  y  399  1.1877x 2.86. The corresponding correlation coefficient
r  0.995 is considerably larger than the value r  0.907 obtained for the unad-
justed values. More important, the power is now p  2.86, which is greater than 1
and this function is concave up and so better captures the trend in the data. Final-
ly, having shifted all the y-values down by 399, we now adjust the function up by
adding the same amount, 399 to the expression and so obtain
y  1.1877 x 2.86  399.
The result is not a pure power function because of the additional term 399. Figure 3.50
shows this function superimposed over the original data points; it is a much better fit
to the original data, especially the first few data points.

550

500

450

400

350
x
FIGURE 3.50 0 1 2 3 4 5

In general, we recommend the following approach when fitting a power func-

tion to a set of data. Unless there is a natural origin for the data (as there was in Ex-
ample 2 on the number of species of reptiles and amphibians on an island—a
theoretical island with zero area would have zero species), you should think about
adjusting your data values by shifting the y-values down (or up) by some appropri-
ate amount. Unfortunately, there is no simple rule to know by how much to adjust
the y-values—it depends on the data you have. In practice, it may be a good idea to
try different adjustments in the y-values to see which accomplishes the best fit.
gord.3896.03.pgs 4/24/03 9:54 AM Page 216

216 CHAPTER 3 Fitting Functions to Data

After you have found the best power function for the adjusted data values, undo
the shift by adding (or subtracting) the same amount back in the expression.
There is another complication to keep in mind when you fit a power function
to data. In Examples 2 and 3 of Section 2.3, we considered data on U.S. imports in
1990 ($495 billion) and in 1998 ($912 billion) and constructed the line through
these two points, using first t  0 in 1990, then t  0 in 1900, and finally t  0 in
year 0. In each case, the slope of the resulting line was the same, although the ver-
tical intercept changed to reflect the starting point. Further, no matter which
equation we use, we get the same prediction for a particular year. This principle
holds in general for the line through any points.
Similarly, in Problem 5 of Section 2.4, you were asked to examine the same
issue with exponential functions. You should have found that in all three cases the
growth factor remained the same, although the vertical intercept for the three ex-
ponential functions did change to reflect the starting point. Also, you will get the
same predictions for a given year by using any of these equations.
Unfortunately, this principle does not carry over to power functions. If you
change the meaning of the independent variable (say, t) the function changes total-
ly. Recall that every power function with power p  0 passes through the origin.
Thus, when you find the equation of an increasing power function passing through
two points or the power function that fits three or more points, that function auto-
matically passes through an additional point at the origin.
Imagine what happens when you have data points between the years 1991 and
2000, say. Your choice of what t represents means that you are repositioning the
“origin” and the resulting power function is forced through a different extra point
(the new “origin”). For instance, if you let t represent the number of years since
1990, the origin is very close to the data points. If you let t represent the number of
years since 1900, the origin is quite far from the data points. And if you let t repre-
sent the year itself, the origin is extremely far from the data points. In each case,
you will get a power function with very different values for the parameters p and k
in y  kx p. Moreover, any predictions based on one expression will be different
from those based on a different expression. We illustrate this by fitting a power
function to the two data points on U.S. imports.

E XAMPLE 4
In 1990, the United States imported $495 billion worth of goods from abroad. In 1998, the
United States imported $912 billion worth. Assuming that the growth in imports follows
a power function pattern, find an equation of the power function that models U.S. im-
ports by using the independent variable t to represent (a) the number of years since 1989,
(b) the number of years since 1900, and (c) the number of years since year 0. (d) Compare
the results obtained from the three approaches. (e) Use each model to predict the amount
of imports in 2005.
Solution

a. With t  0 in 1989, we have the points (1, 495) and (9, 912). Using a calculator, we
find that the corresponding power function is G1 1t2  495t 0.278 with correlation
coefficient r  1.
b. With t  0 in 1900, the points are (90, 495) and (98, 912). Using a calculator, we
find that the corresponding power function is G2 1t2  4.69 . 1012t 7.176 with correla-
tion coefficient r  1.
gord.3896.03.pgs 4/24/03 9:54 AM Page 217

3.5 How to Fit Power Functions to Data 217

1200
1200
1000
Imports ($ billions)

Imports ($ billions)
800
800
600

400 400

200

0 2 4 6 8 10 0 20 40 60 80 100
Years since 1989 Years since 1900
(a) (b)

1200

1000
Imports ($ billions)

800

600

400

200

0 1988 1990 1992 1994 1996 1998 2000

Year
FIGURE 3.51 (c)

c. With t  0 in the year 0, the points are (1990, 495) and (1998, 912). Some calculators
give the corresponding power function as G3 1t2  1.736 . 10500t 152.312 with correla-
tion coefficient r  1. Other calculators give an overflow error because the numbers
are too large.
d. Note that all three choices led to a correlation coefficient of r  1, which suggests
that each of the three functions perfectly fit the data. However the three coefficients
and the three powers bear no relationship to one another. In fact, the power func-
tion in part (a) has p  0.278, which is between 0 and 1, so that power function is
increasing and concave down. The power functions in parts (b) and (c), in com-
parison, both have powers greater than 1, so they are increasing and concave up.
Figures 3.51(a)–(c) illustrate these results.
e. With t  0 in 1989, the year 2005 corresponds to t  16 and the function from part
(a) yields
G1 116 2  4951160.278 2  1069.918.
With t  0 in 1900, 2005 corresponds to t  105 and the function from part (b)
yields
G2 1105 2  4.6911012 2 . 1057.176  1497.019.
With t  0 in the year 0, we have t  2005 and the function in part (c) gives
G3 12005 2  1.736110500 2 12005152.312 2  1553.083.
gord.3896.03.pgs 4/24/03 9:54 AM Page 218

218 CHAPTER 3 Fitting Functions to Data

The results clearly are very different and, given the actual data values, the prediction
based on the formula from part (a) with t representing the number of years since
1989 appears to be the most accurate.
◆
Example 4 emphasizes that, whenever you work with a power function, you
must think very carefully about the variable you are using and consequently your
choice of origin. If you have more than two data points, it might also make sense
for you to try some different interpretations for the independent variable—for in-
stance, years since 1989, years since 1988, and years since 1987 to determine which
function best fits the data.

Problems
1. Use the model created in Example 3 relating the b. Using this function, estimate the distance from
wingspan of a jet aircraft to its total load to answer the left end of the dial to radio station 880; to
the following questions. radio station 1270.
a. If a new jet is being designed to have a total take- c. What station would be 6 cm from the left end of
off load of 500 thousand pounds, what wing- the dial?
span will be necessary to support the load? d. Determine whether the same function fits the
b. A super-jumbo jet is being designed to carry a comparable set of readings on your radio.
total weight of 1.2 million pounds. What wing- e. Determine the function that fits the readings on
span will be needed to support it? the FM band on your radio.
c. The wingspan of a jet is 175 feet. What is the 4. Weight-lifting competitions are divided into body
maximum takeoff weight that can be supported? weight classes. Championships are often based on
2. In Example 3, we constructed a model where the the total weight each lifter can lift in the press, the
wingspan S of a jet is a function of its total weight W. snatch, and the clean-and-jerk. The table shows the
Interchange the roles of S and W so that the weight is total weight W, in pounds, lifted in several body-
a function of the wingspan. Use the data in Example 3 weight classes as a function of the body weight B, in
to find the corresponding power function. How do pounds, of the lifter.
the parameters in this new power function compare
to those in the power function in Example 3? Body Weight 123 130 148 166 182 197
In Problems 3–13, use the transformation approach to Weight Lifted 750 776 851 912 966 1023
find a power function that fits the data.
Source: Thomas A. McMahon and John Tyler Bonner, On Size and
3. Marc notices that the radio frequency numbers on Life. Scientific American Library: 1983.
the AM dial of his stereo don’t seem to lie in a linear
pattern. He measures the distances, in centimeters, a. Explain why a power function would be an ap-
from the extreme left end of the dial to each of the propriate fit for these data.
numbers printed and gets the readings shown in b. Find the power function that best fits these data.
the accompanying figure. c. What total weight could be lifted by a 225 pound
super-heavyweight lifter according to this model?
D d. What total weight could be lifted by the prover-
bial 98-pound weakling?
5. The running speed of animals appears to be related
E to their overall body lengths. The table gives the
lengths L of various organisms, in centimeters, and
their top running speed S in centimeters per second.
a. Determine the power function that relates the a. Explain why a power function would be the
distance D to the station numbers n shown. function of choice to model this data.
gord.3896.03.pgs 4/24/03 9:54 AM Page 219

3.5 How to Fit Power Functions to Data 219

centimeters, and their top flying speed, S, in cen-

Animal Length (cm) Speed (cm/sec) timeters per second.
Clover mite 0.08 0.85 a. Find the power function that best fits these data.
b. Estimate the top flying speed for a bird that is 80 cm
Ant 0.42 6.5
long according to this model.
Deer mouse 9.0 250 c. Estimate the length of a bird that could fly at a
speed of 2000 cm/sec.
Zebra-tail lizard 15.0 720
d. Estimate the flying speed of a pterodactyl that
Eastern chipmunk 16.0 480 was approximately 240 cm long.
7. The swimming speed of animals also appears to be
Iguana 24.0 730
related to their overall body length. The table gives
Gray squirrel 25.0 760 the lengths L of various organisms, in centimeters,
and their top swimming speed S in centimeters per
Red fox 60.0 2000
second.
Cheetah 120.0 2900
Source: Thomas A. McMahon and John Tyler Bonner, On Size and
Life. Scientific American Library: 1983. Animal Length (cm) Speed (cm/sec)

b. Find the power function that best fits these data Paramecium 0.02 0.1
according to this model. Water mite 0.13 0.4
c. Estimate the length of an animal that could run
at a speed of 1500 cm/sec. Goldfish 0.7 75
d. If this model applies to humans, estimate the European dace 10.0 130
best running speed for a man who is 6 feet tall.
(Hint: 1 inch  2.54 cm.) Based on your an- Herring 30.0 440
swer, does this model apply to humans? Explain. Penguin 75.0 380
6. The flying speed of animals also appears to be relat-
ed to their overall body length. The accompanying Tuna 98.0 2080
table gives the lengths L of various organisms, in Dolphin 220.0 1030

Blue whale 2600 1030

Animal Length (cm) Speed (cm/sec) Source: Thomas A. McMahon and John Tyler Bonner, On Size
and Life, Scientific American Library: 1983.
Fruit fly 0.2 190
Horse fly 1.3 660
a. Find the power function that best fits these data.
Hummingbird 8.1 1120 b. Estimate the best swimming speed for a fish that
Dragonfly 8.5 1000 is 50 cm long according to this model.
c. Estimate the length of a fish that could swim at a
Bat 11.0 690 speed of 1000 cm/sec.
Common swift 17.0 2550 8. The table gives the weights, in pounds, of a variety of
full-grown flying birds and their wingspans in feet.
Flying fish 34.0 1560
a. Decide which variable is the independent vari-
Pintail duck 56.0 2280 able and which is the dependent variable.
Swan 120.0 1880
b. Find a power function that fits these data.
c. If a falcon has a wingspan of 3.3 feet, predict its
Pelican 160.0 2280 weight based on your model.
Source: Thomas A. McMahon and John Tyler Bonner, On Size d. If a turkey weighs 15 pounds, predict the
and Life. Scientific American Library: 1983. wingspan it would need in order to fly.
gord.3896.03.pgs 11/21/03 10:30 AM Page 220

220 CHAPTER 3 Fitting Functions to Data

d. A habitat for 200 species of birds is to be estab-

Bird Weight Wingspan lished in this region of the world. How large must
Vulture 18.7 9.3 the habitat be to support that many species?
10. Consider a bicyclist riding around a circular track
Eagle 12 7.5
with a constant radius. To keep making the turn,
Horned Owl 5 5.0 the bicyclist and the bike must lean sideways toward
the center of the circle. The angle A of the lean from
Golden Eagle 13 7.3
the vertical is related to the velocity of the bike. The
Red tailed hawk 4 4.0 table shows the velocity v in meters per second and
the angle A to the vertical in degrees.
Harris hawk 2.6 3.2

Turkey vulture 6.5 6.0

Angle 3 5.7 13.7 22.9 28.6 40.1
Whooping Crane 16.1 7.5
Velocity 1.3 2.0 3.1 4.0 4.7 5.7
Bald Eagle 16 7.5
Source: Thomas A. McMahon and John Tyler Bonner, On Size
Condor 22 9.9 and Life, Scientific American Library: 1983.

Blue-footed booby 4 3.0

Crow 1 2.9 A
Source: U.S. Fish and Wildlife Services.

9. The table gives the number of species of birds living

on various islands in the East Indies versus the area,
in square miles, of each island.

Island Area Number of Species a. What is the independent variable and what is the
dependent variable? Explain.
New Guinea 312,000 540
b. What is the range for the function relating these
Borneo 290,000 420 two variables?
c. Find the power function that best fits these data.
Philippines 144,000 368
d. What velocity is a bicyclist going if the angle of
Celebes 70,000 220 lean is 50° from the vertical?
e. If a bicyclist is going 3.5 meters per second, what
Java 48,000 337
angle from the vertical is needed to stay on the
Ceylon 25,000 232 circle?
11. Consider a bicyclist riding around a circular track
Flores 8870 143
with a constant velocity. To keep making the turn,
Timor 18,000 137 the bicyclist and the bike must lean sideways toward
Source: Kevin Mitchell et al., Mathematical Models of Bio- the center of the circle. The angle of lean to the ver-
logical Systems. 1998. tical is related to the radius of the circle. The table
shows the radius r in meters, and the angle A to the
a. Decide which variable is the independent variable vertical in degrees.
and which is the dependent variable. a. What is the independent variable and what is the
b. Find a power function that fits these data. dependent variable? Explain.
c. The island of Sumba, which has an area of 4600 square b. What is the range for the function relating these
miles, is in this region. Estimate how many species of two variables?
birds live on Sumba. c. Find the power function that best fits these data.
gord.3896.03.pgs 4/24/03 9:54 AM Page 221

3.6 How Good Is the Fit 221

13. Dennis is a gunnery officer for M1A1 Abrams tanks in

Angle 38.4 34.4 29.8 24.6 20.6 the National Guard. He has determined the distance,
Radius 0.8 0.9 1.05 1.3 1.74 in meters, to a potential target and the time, in sec-
onds, needed to locate, identify, and fire at that target.
Source: Thomas A. McMahon and John Tyler Bonner, On
Size and Life, Scientific American Library: 1983.

d. What radius circle is needed for a bicyclist lean- Range 250 500 1000 1500 2000 2500 3000
ing 15° from the vertical?
e. If a bicyclist is going around a circle that has a Time 3 4 5 6.5 9 11 16
radius of 1.5 meters, what angle from the vertical Source: Student project.
is needed for the byciclist to stay on the circle?
12. The following table shows the total worldwide car- a. Determine the power function that fits these data.
bon emissions, in millions of tons of carbon, from b. An enemy tank is at a distance of 1750 meters.
fossil fuel burning over the years. How long will it take for an M1A1 tank crew to
a. Find the equation of the power function that locate, identify, and then fire at the target?
best fits these data. c. If it takes 10 seconds for an M1A1 tank crew to
b. What does the model predict for the amount of fire at a target, what is the range to the target?
carbon emitted by burning fossil fuels in 2010? d. An allied tank is 3000 meters away. How long does
c. When does the model predict that the total the crew of an M1A1 tank have to realize that the
amount of carbon emitted from burning fossil target is actually a friend and avoid firing at it?
fuels will reach 7000 million tons?

Year 1950 1960 1970 1975 1980

Carbon Emissions 1612 2535 3998 4518 5156

Year 1985 1990 1993 1995 1997 1999

Carbon Emissions 5271 5946 5896 6212 6349 6307

Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That Are
Shaping Our Future.

Exercising Your Algebra Skills

Several sets of data have been linearized to find the best 3. Y  1.0792  1.5X
power fit. The following equations are for the lines that 4. Y  0.2218  0.4X
best fit the transformed data. What are the correspon- 5. Y  0.3522  1.0843X
ding power functions?
6. Y  1.3015  0.7840X
1. Y  0.9031  1.5X
7. Y  0.8525  1.2733X
2. Y  0.6990  0.7X
8. Y  0.817  2.015X

3.6 How Good Is the Fit?

Throughout this chapter, we have looked for the function within each appropri-
ate family of functions that best fits a set of data values in the least squares sense.
If the data points fall in a linear pattern, we use the regression line. If the data
points fall in some other recognizable pattern and if appropriate software is
available, we can apply the least squares criterion directly to construct the best fit
gord.3896.03.pgs 4/24/03 9:54 AM Page 222

222 CHAPTER 3 Fitting Functions to Data

function in any particular family. Alternatively, we can use the built-in routines
of graphing calculators or spreadsheets to find the function that best fits the
data, based on the transformation approaches described in Sections 3.3 and 3.4.
In doing any of these things, we face two problems:
1. being able to assess how well a particular curve fits the set of data; and
2. deciding which function is a better fit when there are several reasonable
candidates.
Unfortunately, there is no single, clear-cut way of resolving these problems. In this
section, we look at several different ways to make such decisions.

Interpreting the Correlation Coefficient

Comparing the quality of the best fit in one family of functions to the best fit in
another involves some deep and subtle issues and we can only touch the surface
here. First, recall that the correlation coefficient measures only how well the best
linear function fits any set of data, be it the original or the transformed data. If r
is not close enough to either 1 or 1, there is no point in using that function as
a model to describe the pattern in the data.
Suppose that we have sufficiently high values for r for several different families of
functions—say, linear, exponential, and power. It is tempting to think that the func-
tion having the largest value for r is the best fit among them, but unfortunately that
is not necessarily the case, especially if the values for r are relatively close—say, 0.93,
0.95, and 0.90. Thus you should not make a decision about which function is the best
fit by using r as the only criterion.

Interpreting the Residuals

The residuals associated with a particular fit provide considerably more information
about how good the fit is and how to compare different fits for the same set of data.
Recall that, for a linear fit to the original data, the residuals are the differences between
the actual y-values and the predicted values based on the regression line, as shown in
Figure 3.52(a). When we transform the data and obtain the regression line for the
transformed data, then the residuals are the differences between the transformed data
values (Y) and the predicted values based on the regression line for the transformed
data 1 y  mX  b2 . The differences between the original y-values and the predict-
ed values based on the detransformed function—say, an exponential function—are
not residuals; they are called errors, as shown in Figure 3.52(b).
For a linear fit, if the actual height is y and the predicted value given by the regres-
sion line is yp  ax  b, the corresponding residual is y  yp  y  1ax  b 2. If a
data point lies above the regression line, the residual is positive; if a data point lies
below the regression line, the residual is negative, as shown in Figure 3.53. If the fit is
good, roughly half the residuals should be positive and roughly half should be negative
because the regression line should pass, more or less, midway between the data points.
We plot the residuals corresponding to the data in the scatterplot shown in Figure
3.53 on a different set of axes as shown in Figure 3.54. This plot is called a residual plot,
Points on the horizontal axis of the residual plot represent data points with zero resid-
ual—that is, data points that fall exactly on the regression line. Whenever a data point
lies above the regression line, the residual is positive and the associated point in the
residual plot is above the horizontal axis by that amount. Similarly, whenever a data
gord.3896.03.pgs 4/24/03 9:54 AM Page 223

3.6 How Good Is the Fit? 223

Data Set (Linear Pattern)

y y

(x, y)
regression
residual line

x x

(a)

Data Set (Nonlinear Pattern) Transformed Data Data Set

y Y y
(X, Y) (x, y)

residual
error

curve of
best fit
regression y = f(x)
line
Y = mX + b
x X x

(b)

FIGURE 3.52

point lies below the regression line, the residual is negative and the associated point in
the residual plot is below the horizontal axis by that amount. (Note that the horizontal
axis shown in the residual plot has no direct relationship to the regression line.)

Residual

+ regression
(x, y)
line
+ − x
(x, yp)
−
y − yp +
y − yp −

FIGURE 3.53 FIGURE 3.54

E XAMPLE 1
In Example 1 of Section 3.2, we looked at the gross receipts of the movie industry, in bil-
lions of dollars, in different years since 1990 based on the following table.
gord.3896.03.pgs 4/24/03 9:54 AM Page 224

224 CHAPTER 3 Fitting Functions to Data

Year 0 3 4 5 6 7

Total Receipts 39.98 49.80 53.50 57.18 60.28 63.01

Source: 2000 Statistical Abstract of the United States.

We found the regression line to be y  3.3435t  40.027, where t is the number of years
since 1990. Use this equation to calculate the residuals and create the residual plot.
Solution We show the scatterplot of the data with the regression line y  3.3435t
 40.027 superimposed in Figure 3.55. Now consider the first data point, where t  0
and y  39.98 billion dollars. The predicted height of the corresponding point on the
regression line is
yp  3.343510 2  40.027  40.027  40.03.

y
Movie industry receipts ($ billions)

70

60

50

40

30
t
0 3 6 9
FIGURE 3.55 Years since 1990

This predicted value (the height to the line when t  0) is larger than the actual value
39.98, so the data point lies below the regression line. The residual for this point is then
y  yp  39.98  40.03  0.05,
so the corresponding point in the residual plot that we draw later will lie below the hor-
izontal axis there. We perform comparable calculations for each of the remaining data
points and list the results in the following table.

Year 0 3 4 5 6 7

Total Receipts y 39.98 49.80 53.50 57.18 60.28 63.01

Predicted yp 40.03 50.06 53.40 56.74 60.09 63.43

Residual 0.05 0.26 0.10 0.44 0.19 0.42

We now plot the residuals in the corresponding residual plot shown in Figure 3.56.
As we indicated the first residual point lies slightly below the horizontal axis because the
actual data point lies slightly below the regression line. The same is true of the second
point. The third, fourth, and fifth points all lie above the horizontal axis because each
corresponds to a point that lies above the regression line in Figure 3.55. Finally, the last
point lies below the horizontal axis because the actual data point lies below the height
predicted by the regression line.
gord.3896.03.pgs 4/24/03 9:54 AM Page 225

3.6 How Good Is the Fit? 225

0.50

Residuaals
0 1 2 3 4 5 6 7

–0.50

FIGURE 3.56 Years since 1990

Interpreting a Scatterplot
Figure 3.57 shows both a scatterplot and the associated residual plot for a set of
data in which a linear function fits the data well. Note that the regression line es-
sentially passes through the middle of the cluster of data points so that roughly half
the points lie above the regression line and roughly half lie below it. Also, note that
in the residual plot roughly half the residuals are positive (corresponding to data
points above the regression line) and roughly half are negative (corresponding to
data points below the regression line). Moreover, the residuals seem to be scattered
randomly both above and below the horizontal axis; there does not appear to be
any pattern to their locations.

y Residual

x
FIGURE 3.57

In contrast, consider the scatterplot and the associated residual plot shown in Fig-
ure 3.58. The data fall in a linear pattern, but the regression line does not appear to be
a particularly good fit; it is distorted by the presence of two points that are far from the
line. How does this poor fit show up in the residual plot? Again, note that roughly half
the residuals are above the horizontal axis and roughly half are below it. However, this
time the fact that most of the residuals on the left are negative (because most of the
data points on the left fall below the regression line) is significant. Similarly, most of
the residuals on the right are positive. Thus rather than being scattered randomly
above and below the horizontal axis, the residuals have a pattern. This pattern in the
residuals indicates that the fit is not good.
Consider now the scatterplot of a different set of data and its associated resid-
ual plot shown in Figure 3.59. The data values can likely be modeled by either an
exponential growth function or a power function with p  1. For comparison, the
regression line is superimposed on the scatterplot, even though the line does not
gord.3896.03.pgs 4/24/03 9:54 AM Page 226

226 CHAPTER 3 Fitting Functions to Data

y Residual

x
FIGURE 3.58

capture the trend in the data. Note that the data points on the left lie above the line,
the points in the middle lie below the line, and the points on the right lie above the
line. This behavior is reinforced by the residual plot in which the points fall into a
U-shaped pattern with the middle residual points falling below the horizontal axis.
As was the case in Figure 3.58, the existence of this pattern indicates that the linear
fit is not a good one.
y Residual

x
FIGURE 3.59

Figure 3.60 shows the best exponential fit to the data in Figure 3.59 and the
residual plot associated with the transformed data. The exponential curve is a good
fit to the data. Also, the residuals are small with no obvious pattern; they are scat-
tered about the horizontal axis in an apparently random pattern.

y Residual

x
FIGURE 3.60

In general, just by looking at the scatterplot, you may not be able to see that a
particular curve fits a set of data well. We have demonstrated previously that the
scales of the graph can distort an image when either or both variables extend over
a very large interval of values, preventing us from recognizing either a good or a
bad fit. We can overcome this difficulty by examining and interpreting the residual
plot associated with the regression line for the transformed data.
gord.3896.03.pgs 4/24/03 9:54 AM Page 227

3.6 How Good Is the Fit? 227

◆ If a particular fit is good, the residuals should display no pattern.

◆ If the residuals do display a pattern (e.g., rising and falling, U-shaped, or pe-
riodic), the fit is likely not a good one and you should look further for a bet-
ter fit among other families of functions.
◆ If the residuals are extremely large numerically, you should be suspicious of
the fit, even if the fit on the scatterplot looks good.
Furthermore, if only one or two residuals are extremely large compared to all
the others, that might indicate an error in keying in data values, errors in the actu-
al measurement, or highly unrepresentative values, called outliers. Check your val-
ues to be sure that there is no input error. If there is none, you might want to
experiment by removing the outliers from the data set and recomputing the re-
gression equation to determine whether there is a dramatic improvement in the fit.
However, any final report you write on your results should mention the apparent-
ly unrepresentative outliers, even if you don’t use them in the calculations.
Once you have convinced yourself that you have found an appropriate fit for
the data, the linear correlation coefficient r provides a measure of the degree of
agreement. However, keep in mind that you can calculate a correlation coefficient
for any set of data, either a set of original values or a transformed set of values.
Several different fits for the same set of data may have a significant level of correla-
tion. (Think about the graphs of functions: Over a relatively small interval, a line,
an exponential function, and a power function may appear almost identical, but
when you extend them sufficiently, they clearly move away from one another.)
Thus a good strategy is first to find the best fit and only then consider the as-
sociated correlation coefficient to verify that it is significant. You can, and probably
should, try different fits for a set of data, one after the other, until you come up
with the best curve among the families of linear, exponential, power, and logarith-
mic functions. Of course, a set of data may not fall in any of these patterns, but a
great many reasonable data sets do. In later chapters, we extend these ideas to cases
in which the possible patterns include quadratic and higher degree polynomials,
periodic functions, or other common patterns. Two or more different patterns
may apply to different portions of a set of data. For instance, a pattern could start
with linear growth and later appear to have exponential growth.

Interpreting the Sum of the Squares

Probably the best way to compare two or more functions to decide which is the
better fit to a set of data is based on the least squares criterion. The objective in
fitting any function, whether linear or nonlinear, to a set of data is to minimize
the sum of the squares of the differences between the actual heights y of the data
points and the predicted heights yp  f 1x 2 based on the function. That is, for
each of the data points 1x1 , y1 2, 1x2 , y2 2 , . . ., 1xn , yn 2, look at the vertical distances,
yi  f 1xi 2, for i  1, 2, 3, . . ., n, from the point to the curve and then calculate
the sum of the squares of these vertical differences:
n

a 3 yi  f 1xi 2 4
2
i1

(where g i1 indicates the sum of all the terms for each value of i from 1 through
n

n as discussed in Appendix A3.) This quantity is often calculated automatically in

many mathematical software packages. It is also fairly easy to calculate using a
gord.3896.03.pgs 4/24/03 9:54 AM Page 228

228 CHAPTER 3 Fitting Functions to Data

graphing calculator in the statistics or table mode. If functions from several differ-
ent families fit a set of data well, you could then compare the sum of the squares for
each function and select the function having the smallest sum of squares value as
the best fit among the choices.
However, the sum of the squares should be only one of several criteria used.
The fact that the sum of the squares is smaller for one function than for another
function doesn’t necessarily mean that it is the best fit to the data. Other factors
about the situation may not be reflected in the data. For instance, you may have
measurements on the growth of bacteria in a test tube that might suggest exponen-
tial growth, but you know that such growth cannot continue indefinitely, so an ex-
ponential function will model the population only for a short while. You should
examine the scatterplot for the presence of outliers that will distort any predicted
model. You should look at the residual plots to see if there is any pattern to the
residuals that will provide additional insight into which function is the best fit.
Another quantity is often used in the sciences to measure how well a function fits
a set of data. Instead of looking at just the sum of the squares of the vertical distances
between the data points and the approximating function, finding the average of this
sum over all the data points and then taking the square root of the result may be con-
venient. Doing so gives the root-mean-square (RMS) value associated with the func-
tion. It is calculated from the formula
g i1 3yi  f 1xi 2 4 2
n

.
B n
In some sense, this measures the “average” distance of the data points from the
curve. The smaller the average distance is, the better the fit.
Statisticians use a slightly different measure of the average distance of the
points from the curve, called the standard error of the estimate. It is given by
g i1 3yi  f 1xi 2 4 2
n

.
B n2
This discussion about which fit is the best—and the lack of consensus about
how to determine it—reinforces the point we made at the beginning of this section
that the issues are subtle and clearly depend on the set of data you are studying.
Each set of data has to be approached and examined on its own merits.

Kepler’s Third Law of Planetary Motion: An Application

In 1619, Johannes Kepler published his third law concerning the motion of the plan-
ets around the sun. This law expresses the relationship between the average distance
D of a planet from the sun and the period t (the length of a year) for that planet to
complete a full orbit about the sun. Kepler’s work was based on the best experimen-
tal data available at the time. In fact, astronomers then weren’t aware of the existence
of the three outer planets: Uranus, Neptune, and Pluto. We use the following current
data to determine the relationship between D and t, where D is the average distance
of each planet from the sun, in millions of miles, and t is the length of its year, in days.
To give you some perspective, the Earth takes 365 days (its year) to complete a full
orbit about the sun and the average distance to the sun is 92.9 million miles.
First, we must decide which is the independent variable and which is the de-
pendent variable. Although time is usually the independent variable, in this case it
makes more sense to think of the length of the year t for a planet as a function of
gord.3896.03.pgs 4/24/03 9:54 AM Page 229

3.6 How Good Is the Fit? 229

Planet Period t Distance D

Mercury 88 36.0

Venus 225 67.2

Earth 365 92.9

Mars 687 141.5

Jupiter 4,329 483.3

Saturn 10,753 886.2

Uranus 30,660 1782.3

Neptune 60,150 2792.6

Pluto 90,670 3668.2

the planet’s distance D from the sun—the farther a planet is from the sun, the
longer it takes to complete a full orbit.
We next look at the scatterplot of the data, as shown in Figure 3.61 where the av-
erage distance D from the sun is plotted along the horizontal axis and the period t is
plotted along the vertical axis. By eye, it appears that the points fall in a roughly lin-
ear pattern, so we first investigate the results of fitting a linear function to this data.
t

100,000
Length of year t (days)

80,000

60,000

40,000

20,000

D
0 1000 2000 3000 4000
FIGURE 3.61 Average distance D (million miles)

E XAMPLE 2
Find the linear function that best fits the data on planetary motion and discuss how
good the fit is.
Solution The linear regression equation for this data set is
Y  24.038X  4584.52,
or equivalently in terms of our variables here,
t  24.038D  4584.52.
The correlation coefficient is r  0.9887, which indicates a high level of correlation be-
cause the critical value for n  9 data points is r9  0.666.
Figure 3.62 shows this line superimposed over the data points. Although at first glance
it appears that the data points fall in a linear pattern, actually they do not. Consider the scale
for the scatterplot. The horizontal scale extends from 36 million miles out to 3668 million
gord.3896.03.pgs 4/24/03 9:54 AM Page 230

230 CHAPTER 3 Fitting Functions to Data

(or 3.668 billion) miles. Furthermore, all the data points fall rather far from the regression
line. These discrepancies may seem small to the eye, but they actually are enormous consid-
ering the size of the quantities involved. Hence the residuals—the differences between the
actual values and the predictions based on the linear function—may be quite large.
t

100,000

Length of year (days)

80,000

60,000

40,000

20,000

D
0 1000 2000 3000 4000
FIGURE 3.62 Average distance (million miles)

Using the linear regression equation, we predict values for the length of the year for
each planet based on its distance from the sun and so find the corresponding residuals.
They are shown in the following table.

Distance Actual Predicted period Residual 

Planet D period t t  24.038D  4584.5 actual  predicted
Mercury 36.0 88 3719.1 3807.1

Venus 67.2 225 2969.1 3194.1

Earth 92.9 365 2351.4 2716.4

Mars 141.5 687 1183.1 1870.1

Jupiter 483.3 4,329 7033.1 2704.1

Saturn 886.2 10,753 16,718.0 5965.0

Uranus 1782.3 30,660 38,258.4 7598.4

Neptune 2792.6 60,150 62,544.0 2394.0

Pluto 3668.2 90,670 83,591.7 7078.3

Note that the predictions for the length of a year for the four innermost planets are neg-
ative, which means that the year has a negative number of days! In fact, the most accu-
rate prediction based on the linear regression model (i.e., the smallest residual) is off by
1870 days! As we said previously, the apparently small discrepancies between the data
points on the scatterplot and the regression line actually are immense when we take the
scale into account. Consequently, we conclude that the linear fit is definitely not appro-
priate for this set of data, despite the high level of correlation.
◆
When we examine the scatterplot of the planetary data with the regression
line superimposed, as shown in Figure 3.62, there clearly is a concave up pattern
to the data. This pattern suggests that we should use either an exponential growth
function or a power function with p  1. In Example 3, we show what happens
with the exponential function.
gord.3896.03.pgs 4/24/03 9:54 AM Page 231

3.6 How Good Is the Fit? 231

E XAMPLE 3
Find the exponential function that best fits the data on planetary motion and discuss
how good the fit is.
Solution The exponential regression equation for this data set is
Y  484.17611.001722 x
or equivalently in terms of our variables here,
t  484.17611.001722 D.
The correlation coefficient is r  0.8937, which indicates a relatively high level of corre-
lation. Comparing it to the critical value r9  0.666 based on n  9 data points indi-
cates a significant level of correlation, although the correlation is considerably lower
than that for the linear fit. We show the exponential function superimposed over the
scatterplot in Figure 3.63, and observe that it is not a particularly good fit.

100,000
Length of year (days)

80,000

60,000

40,000

20,000

D
0 1000 2000 3000 4000
FIGURE 3.63 Average distance (million miles)

To check the accuracy of this function, we look at the errors—the differences between
the actual values and the predicted values—we get when using this exponential formula. Re-
call that, when we are working with a linear function, the errors are the same as the residuals
but the term residual applies only to a linear fit. We show these errors in the following table.

Distance Actual Predicted Period Error 

Planet D Period t t  484.176 11.001722 D Actual  Predicted
Mercury 36.0 88 515.1 427.1

Venus 67.2 225 543.4 318.4

Earth 92.9 365 568.0 203.0

Mars 141.5 687 617.5 69.5

Jupiter 483.3 4,329 1,111.0 3218.0

Saturn 886.2 10,753 2,220.3 8532.7

Uranus 1782.3 30,660 10,356.5 20,303.5

Neptune 2792.6 60,150 58,782.2 1367.8

Pluto 3668.2 90,670 264,694.8 174,024.8

gord.3896.03.pgs 4/24/03 9:54 AM Page 232

232 CHAPTER 3 Fitting Functions to Data

The best prediction, for the length of the year on Mars, is off by almost 70 days. The
worst prediction, for Pluto, is off by more than 174,000 days and this error is almost twice
the actual length of the year there. Consequently, we conclude that the exponential fit is
very poor for this set of data, despite the fact that the correlation coefficient is significant.
◆
In Example 4 we show what happens with the power function.
E XAMPLE 4
Find the power function that best fits the data on planetary motion and discuss how good
the fit is.
Solution The power regression equation for this data set is
Y  0.4079 X 1.4999,
or equivalently in terms of our variables here,
t  0.4079 D 1.4999.
The correlation coefficient is r  0.9999999, which indicates an almost perfect fit. The
power function superimposed over the scatterplot, as shown in Figure 3.64, certainly ap-
pears to be an excellent fit.
t

100,000
Length of year (days)

80,000

60,000

40,000

20,000

D
0 1000 2000 3000 4000
FIGURE 3.64 Average distance (million miles)

To verify the accuracy of this fit, we again look at the errors obtained when we use
the power function to predict the length of each planet’s year.

Distance Actual Predicted year Error 

Planet D Year t t  0.4079 D1.4999 actual  predicted
Mercury 36.0 88 88.1 0.1

Venus 67.2 225 224.6 0.4

Earth 92.9 365 365.1 0.1

Mars 141.5 687 686.2 0.8

Jupiter 483.3 4,329 4,331.2 2.2

Saturn 886.2 10,753 10,753.7 0.7

Uranus 1782.3 30,660 30,669.0 9.0

Neptune 2792.6 60,150 60,148.1 1.9

Pluto 3668.2 90,670 90,547.6 122.4

gord.3896.03.pgs 4/24/03 9:54 AM Page 233

3.6 How Good Is the Fit? 233

The predictions for five of the planets are accurate to within a fraction of a day. Even
the worst prediction—that for Pluto—is off by only 122.4 days; although that may seem
like a fairly large amount, consider it in terms of the magnitude of the correct value,
90,670 days. As a percentage, this result is off by only 0.00135  0.135% or about one
tenth of 1%.
◆

Finally, we compare how well each of the three functions created in Examples
2, 3, and 4 fit the data on planetary motion by considering the sum of the squares
of the deviations. For the linear function, we need the vertical distance from each
of the points to the line (the residuals shown in the last column of the table in Ex-
ample 2). We square these residual values and then add them. Therefore, for the
linear function, we have
Sum of squares  3807.12  3194.12  . . .  7078.32  192,035,629.
For the exponential function, we similarly find the vertical distance from each
point to the curve (the error value shown in the last column of the table in Exam-
ple 3), square each of these error values, and then add them. The result is
Sum of squares  1427.12 2  1318.42 2
 . . .  1174,024.82 2  3.0782  1010.
Finally, for the power function, we likewise sum the squares of the error values
from the last column of the table in Example 4 to obtain
Sum of squares  10.1 2 2  10.4 2 2  . . .  1122.42 2  15,081.48.
Summarizing these results, we have the following.
For the linear function: sum of the squares  192,035,629
For the exponential function: sum of the squares  3.0782  1010
For the power function: sum of the squares  15,081.48.
The value of the sum of the squares for the power function is quite small, especial-
ly considering the size of the numbers involved. (In fact, almost the entire value of
15,081.48 comes from the contribution of Pluto, for which the error is 122.4, so
that 122.42  14,981.8.) In contrast the sum of the squares for the linear function
is extremely large and that for the exponential function is astronomical. Therefore,
based on the sum of squares criterion, the power function is also by far the best fit
to these data.
Let’s look a little closer at the equation for the power function obtained in Ex-
ample 4, t  0.4079D 1.4999. The power 1.4999 is certainly greater than 1 as we ex-
pected. However, numerically it is virtually equal to 1.5, so we write the equation as
t  0.4079D 1.5.
This equation is known as Kepler’s third law of planetary motion.

Think About This In practice, measuring the length of the year for a planet is relatively easy, especial-
ly for the inner planets whose years are reasonably short. You do so by determining
how long it takes to make a complete revolution about the sun. However, measur-
ing the distance from the sun is much harder. Rewrite the equation for Kepler’s
third law to express the distance D as a function of the length of the year t. ❐
gord.3896.03.pgs 4/24/03 9:54 AM Page 234

234 CHAPTER 3 Fitting Functions to Data

Actually, there is a more interesting form for the relationship between the
length of the year and the average distance of the planets from the sun. The power
1.5 in the function can be written as the fraction 32 so that
t  0.4079D 3>2.
Squaring both sides of this equation, we get
t 2  10.4079D 3>2 2 2  0.1664D 3,
which is the form in which Kepler’s third law usually is expressed.
Astronomers discovered the planet Pluto in 1930 after observing some minor
perturbations in the orbit of Neptune. They hypothesized that these discrepancies
could be accounted for by the existence of a previously unknown outer planet.
Knowing the timing of the perturbation, the astronomers knew approximately
where to look for this unknown planet by using predictions based on Kepler’s
third law. Similarly, they have used the law to estimate how far newly discovered
planets are in their orbits about other stars.

Problems
1. The best-fit line is constructed for each of four sets Match each description with one of the two resid-
of nonlinear data, whose patterns can be roughly ual plots (i) and (ii) shown in the column at the left.
described as Explain your answer in each case.
a. increasing and concave up; 2. Three different types of functions are fitted to a set
b. increasing and concave down; of data based on the three residual plots (a)–(c)
c. decreasing and concave up; shown below and at the top of the next column.
d. decreasing and concave down. Decide which function best fits the data.
Residual Residual

0.1

–0.1

(a)
(i)
Residual
Residual

0.1

–0.1

(b)

(ii)
gord.3896.03.pgs 4/24/03 9:54 AM Page 235

3.6 How Good Is the Fit? 235

Residual cave up, increasing and concave down, decreasing

and concave up, decreasing and concave down,
or none of the above.
0.1 c. From among the families of functions that you
have studied so far, decide which types of func-
tions would be appropriate for each of the actual
data sets that you listed in part (b). State what you
would expect of the values of the parameters for
exponential functions and for power functions.
–0.1
5. The Statistical Abstract of the United States contains
data on the median starting salaries of people based
(c)
on their level of education in six categories: did not
3. Consider the points (0, 1), (1, 3), (2, 6), (3, 8) and complete high school, high school graduate, associ-
(4, 10). ate’s degree, bachelor’s degree, master’s degree, and
Ph.D. Those who didn’t complete high school had
a. The line y  2x  1 fits these points reasonably
approximately E  10 years of education; those who
well. Calculate by hand the residuals and the sum
have only a high school diploma have E  12 years
of their squares.
of education. Similarly, E  14 for an associate’s de-
b. The line y  2.5x  1 fits the data slightly bet-
gree, E  16 for a bachelor’s degree, E  18 for a
ter. Calculate the sum of the squares and com-
master’s degree, and E  20 for a Ph.D. The follow-
pare it to the value you found in part (a).
ing table relates the median starting salary S in
c. Find the equation of the regression line for the
1999 to the number of years E of education:
data. Then calculate the sum of the squares for
it. How does it compare to your results in parts
(a) and (b)? E 10 12 14
4. A line is fit to different sets of data, and the follow-
ing residual plots are obtained. Assume that in each S $16,053 $23,594 $32,468
case the residuals are all relatively small, so you are E 16 18 20
to consider only the patterns.
S $43,782 $52,794 $74,712

a. Find the equation of the regression line for the

data. What is the meaning of the slope in this
context?
b. Use the equation from part (a) to calculate the
residuals associated with each data value and
(i) (ii) construct the residual plot. Does the residual
plot indicate that the regression line fits the data
well? Explain.
c. Calculate the sum of the squares associated with
the regression line.
6. For the data in Problem 5
a. find the equation of the exponential function
(iii) (iv) that fits the data and state the meaning of the
growth factor in this context;
a. Decide which, if any, of the residual plots indicate b. calculate the sum of the squares associated with
that a linear fit is appropriate for the actual data. the exponential function you found in part (a);
b. For any residual plot that indicates that a linear and
fit is likely not appropriate, decide whether the c. the residuals for the exponential function S  Ac E
pattern in the original data is increasing and con- in part (a) are based on the linear equation,
gord.3896.03.pgs 4/24/03 9:54 AM Page 236

236 CHAPTER 3 Fitting Functions to Data

log S  log A  1log c2E. Calculate the residuals d. By comparing the residual plot for the linear
and draw the associated residual plot. Does the function from Problem 5, the exponential func-
residual plot indicate that the exponential func- tion from Problem 6, and the power function
tion fits the data well? Explain. from Problem 7, which of the three functions ap-
7. For the data in Problem 5: pears to be the best fit?
e. Based on the criteria in parts (a)–(d), which of
a. Find the equation of the power function that fits
the three functions do you conclude is the best
the data.
fit? Explain your decision.
b. Calculate the sum of the squares associated with
the power function you found in part (a); and 9. Assume that each of the planets from Mercury to
c. The residuals for the power function fit S  AE p Neptune revolves about the sun in a roughly circu-
in part (a) are based on the equation, log S  lar orbit.
log A  p . 1log E2. Calculate the residuals and a. Extend the first table in Example 2 to include the
draw the associated residual plot. Does the resid- average speed of each planet in its orbit. (Hint: If
ual plot indicate that the power function fits the
you know the average distance from a planet to
data well? Explain.
the sun, what is the approximate distance that it
8. Use the data in Problem 5 and the results from travels in its orbit around the sun?)
Problems 5–7: b. Find the best fit to this set of data on the speed of
a. By looking at the way the linear function from a planet as a function of the length of the planet’s
Problem 5, the exponential function from Prob- year from among linear, exponential, and power
lem 6, and the power function from Problem 7 fit functions.
the data, which of the three functions appears to c. Explain how the formula you found in part (b)
be the best fit? can be directly determined algebraically from
b. By comparing the correlation coefficients for the Kepler’s third law.
linear function from Problem 5, the exponential 10. Using the transformation approach of Section 3.5,
function from Problem 6, and the power function the power function in Example 4 can be found from
log D  log1t 0.667 2  0.2596.
from Problem 7, which of the three functions ap-
pears to be the best fit?
c. By comparing the sum of the squares for the lin- Show that you get the same result by using proper-
ear function from Problem 5, the exponential ties of logarithms when you write this function as
function from Problem 6, and the power function log D  log1t 0.667 2  0.2596.
from Problem 7, which of the three functions ap-
pears to be the best fit?

3.7 Linear Models with Several Variables

So far, we have studied situations in which one variable (y) depends on another (x).
In the real world, things aren’t always this simple and we often encounter situations
in which one quantity depends on two or more other independent variables. For ex-
ample, when the weather report gives the windchill factor during the winter, that
value depends on both the air temperature and the wind speed, so it is a function of
two independent variables. Similarly, when you take out a car loan, the monthly pay-
ment depends on the amount borrowed, the interest rate, and the length of the loan,
so there are three independent variables. A study conducted at a college found that
student performance in math classes could be accurately predicted from a combina-
tion of the student’s score on a placement test, the student’s age, gender, SAT score,
high school GPA, number of years since the previous math course, grade in that
course, and several other variables. Actually, each of a dozen independent variables
contributed valuable information, and a model based on all of them produced the
most accurate predictions.
gord.3896.03.pgs 4/24/03 9:54 AM Page 237

3.7 Linear Models with Several Variables 237

In general, suppose that we have a single variable y that depends on any num-
ber of independent variables x1 , x2 , . . . , xn . Statisticians often refer to independ-
ent variables as explanatory variables. To keep things simple, however, we focus on
the case of a single variable y that depends on two independent variables x1 and x2 ,
so that y  f 1x1 , x2 2.
Suppose that we want to determine whether a relationship exists between a per-
son’s serum cholesterol level and the person’s weight and systolic blood pressure.
(The systolic reading is the first, or higher, number in a blood pressure measure-
ment, such as 120 over 80; the smaller number is the diastolic reading.) We collect a
set of data on a random sample of individuals from some population group—say,
young males. The data in the accompanying table, for a sample of 11 apparently
normal males between the ages of 13 and 16, shows the weight of each of these indi-
viduals in kilograms (independent variable x1) and the systolic blood pressure of
each (independent variable x2), as well as their serum cholesterol level in mg/100 cc
(the dependent variable y). If there is a relationship between serum cholesterol level
and the other two variables, we can use it to predict serum cholesterol levels for
other members of this population group, based on weight and blood pressure.

Serum Cholesterol y Weight x1 Systolic Blood Pressure x2

152.2 59.0 108

158.0 52.3 111

157.0 56.0 115

155.0 53.5 116

156.0 58.7 117

159.4 60.1 120

169.1 59.0 124

181.0 62.4 127

174.9 65.7 122

180.2 63.2 131

174.0 64.2 125

Source: Wayne W. Daniel, Biostatistics: A Foundation for Analysis in the Health Sciences, 4th ed.,
New York: John Wiley & Sons, 1987.

Recall that we fit a line to a set of 1x, y2 data points by using the least squares
criterion and constructing the corresponding linear function y  ax  b. To do so
we minimize the sum of the squares of the vertical distances to the desired line to
obtain the line of best fit, as shown in Figure 3.65.
Here, we have a set of (x1 , x2 , y) data points in three dimensions and want to
construct the linear function that best fits the data. The equation of a linear func-
tion of two independent variables x1 and x2 can be written as
y  ax1  bx2  c,
where a, b, and c are three constants. The graph of any such equation is a plane in
three-dimensional space. Now, instead of fitting a line to a set of 1x, y2 data
gord.3896.03.pgs 4/24/03 9:54 AM Page 238

238 CHAPTER 3 Fitting Functions to Data

FIGURE 3.65

points in two dimensions, we need to fit a plane to a set of (x1 , x2 , y) data points
in three dimensions. We want to find the best-fit plane—the specific plane for
which the sum of the squares of the vertical distances to the desired plane is a
minimum, as shown in Figure 3.66.

FIGURE 3.66

It turns out that there is a single plane that satisfies this criterion of coming
closest to all the data points in three-dimensions; it is known as the regression
plane. The process of finding the equation of this plane is called multivariate re-
gression or multiple regression. (If there are more than two independent variables, a
natural extension is used, although it isn’t possible to visualize the “hyperplane” in
four or more dimensions that best fits these data points.)
The arithmetic involved in any multivariate regression tends to be extremely te-
dious, but the method is so widely used that a routine is available in almost any statis-
tical software package and in most spreadsheets. (It is not yet available on the
common hand-held calculators.) We aren’t concerned with the mechanics of calculat-
ing these quantities but simply cite the results obtained by using appropriate software.
In the problems at the end of this section, we similarly assume that you have access to
software that will perform the calculations for you.

E XAMPLE 1
(a) Find the multivariate regression equation expressing serum cholesterol level as a
function of both body weight and systolic blood pressure. (b) Interpret the coefficients
in the multivariate regression equation. (c) Use the regression equation to predict the
serum cholesterol level of a male in the 13- to 16-year-old age group who weighs 60 kg
and whose systolic blood pressure is 123.
gord.3896.03.pgs 4/24/03 9:54 AM Page 239

3.7 Linear Models with Several Variables 239

Solution

a. Using an appropriate software package, such as the Excel™ spreadsheet (we present
details on how to use it at the end of this section), we find that the multivariate re-
gression equation is
y  7.6419  0.6297x1  1.1315x2,
or equivalently
Cholesterol level  7.6419  0.6297 1weight2  1.1315 1systolic blood pressure2.
b. To interpret this equation, let’s see what happens to the values for the cholesterol level
when we change each of the independent variables, weight or systolic blood pressure.
In particular, suppose that a person’s weight x1 increases by 1 kg. According to the for-
mula, the cholesterol level y will increase by 0.6297 units. Alternatively, suppose that a
person’s systolic blood pressure x2 increases by 1 unit. In that case, the cholesterol level
y will increase by 1.1315 units. Thus an increase of 1 unit in blood pressure has a con-
siderably larger impact on cholesterol level than an increase of 1 unit in weight has on
cholesterol level. That is, the systolic blood pressure apparently has a greater effect on
cholesterol level than weight does for a 1-unit increase in each quantity.
c. According to this model, we predict that the serum cholesterol level for an individual
who weighs x1  60 kg and whose systolic blood pressure x2  123 is
y  7.6419  0.6297 1602  1.1315 1123 2  169.315,
or about 169 mg/100 cc.
◆
Note how the coefficients a  0.6297 and b  1.1315, play the same role in
this multivariate regression equation y  ax1  bx2  c that the slope m plays in
the equation of a line y  mx  b. We can think of the regression plane as having
two different slopes, one in the x1 direction and another in the x2 direction, and they
indicate how quickly the dependent variable increases for a given change in either x1
or x2 , respectively. Imagine balancing a book with one corner on a table by holding
it at the opposite corner. Suppose that the point on the table is the origin and that
x1- and x2-axes are drawn on the table. The cover of the book forms a plane in
three-dimensional space and you draw two lines on the cover of the book, one par-
allel to the x1-axis and the other parallel to the x2-axis, as shown in Figure 3.67. The
lines are inclined at two different angles and so have different slopes, one with re-
spect to the x1-axis and the other with respect to the x2-axis. For that reason, you
may want to think of the equation of the regression plane in the form
y  m1x1  m2x2  c,
z

y
FIGURE 3.67 x
gord.3896.03.pgs 4/24/03 9:54 AM Page 240

240 CHAPTER 3 Fitting Functions to Data

where m1 and m2 are the slopes in the x1 and x2 directions, respectively.

Also note that, in making the prediction in part (c) of Example 1, we took val-
ues for the independent variables x1 and x2 that were within the ranges of the data
set. Had we used values that were well outside the ranges of the data—say,
x1  70 kg and x2  140 —we would have considerably less confidence in the ac-
curacy of the prediction.

The Multiple Correlation Coefficient

In addition to finding the equation of the plane that best fits the set of data points
in three-dimensional space, we also need a way of measuring how good the fit is.
With 1x, y2 data points (where we think of y as a function of x), we used the linear
correlation coefficient r as such a measure. With two or more independent vari-
ables, we instead use a quantity known as the multiple correlation coefficient, denot-
ed by R. Like r, R also takes on values between 1 and 1, and the closer it is to
either 1 or 1, the stronger the linear relationship between y and the linear com-
bination of the independent variables.
Furthermore, the square of the multiple correlation coefficient, R 2, is known as
the coefficient of determination. The value of R 2 provides some extremely useful in-
formation about the extent to which the multivariate regression equation explains
the relationship between the dependent variable y and the independent variables x1
and x2 (if there are two) or x1 , x2 , . . . , xn (if there are more than two). In particular,
the value of R 2 indicates the percentage of the variation in the data that is explained
by the linear relationship. For instance, if R 2  0.85, say, then 85% of the variation
is explained by the linear function. (A similar interpretation applies to r 2, the square
of the usual correlation coefficient with 1x, y2 data points.)
As with the multivariate regression equation, both the multiple correlation co-
efficient and the coefficient of determination are typically calculated as part of the
output of both spreadsheets and statistical software.

E XAMPLE 2
For the data in the preceding table relating serum cholesterol level to an individual’s
weight and systolic blood pressure, find and interpret both the multiple correlation co-
efficient and the coefficient of determination.
Solution Using an appropriate software package, we find that the coefficient of deter-
mination is
R 2  0.8508
and so the multiple correlation coefficient is
R  0.9224.
The multiple correlation coefficient is reasonably close to 1, so we conclude that a high de-
gree of linear correlation exists between the dependent variable y and the two independent
variables x1 and x2 . (Statisticians have developed a table of critical values for R comparable
to the critical values for r, but we won’t go into that here.) Moreover, the value R 2  0.8508
indicates that about 85% of the variation in serum cholesterol levels in this population
group can be “explained” by these two variables.
◆
gord.3896.03.pgs 4/24/03 9:54 AM Page 241

3.7 Linear Models with Several Variables 241

In Example 2 the value for the coefficient of determination R 2  0.8508 indi-

cates that about 85% of the variation observed is explained by the two variables,
weight and systolic blood pressure. Thus, about 15% of the variation is not explained
by these two variables. We can explain a greater amount of the variation by introduc-
ing an additional variable, perhaps the number of hours that each individual in the
study spends exercising each week, as we might expect that cholesterol levels also de-
pend on the level of physical activity. In that case, we would be working with three in-
dependent variables x1 , x2 , and x3 and would obtain a linear regression equation that
relates serum cholesterol level to all three. Incidentally, when we do so, the coeffi-
cients of the original two independent variables will almost certainly change.
Each time we introduce an additional variable, we get a larger value for the coeffi-
cient of determination, although the increase might be minimal. At the same time, we
also increase the complexity of the model, and there are often drawbacks to doing so.

Performing Multivariate Regression in Excel

Because the Excel spreadsheet is so widely available, we briefly introduce its use for
performing a multivariate regression analysis. In Excel the dependent variable is al-
ways denoted by y, and the different independent variables are always denoted by x.
Think of the two independent variables as x1 and x2 . You begin by entering the given
data values in the first three columns labeled A, B, and C, as shown in Figure 3.68.

FIGURE 3.68

Once you have entered all the data values, click Tools on the top line and
scroll down to the last entry, Data Analysis. . . . (We indicate the computer
displays in a different font, for emphasis.) If Data Analysis . . . doesn’t ap-
pear, you will have to install the Excel Analysis ToolPak™ before proceeding. To in-
stall it, go to the Tools menu and click Add-ins. If Analysis ToolPak is listed,
just click it to permanently install it. If Analysis ToolPak isn’t listed in the
Add-Ins dialog box, click Browse and locate the drive and folder names and the
file name Analys32.xll for the Analysis ToolPak—it usually is located in
the Library\Analysis folder.
When you click Data Analysis . . . , you will see a long list of available
statistical procedures in alphabetical order. Scroll down until you reach
gord.3896.03.pgs 4/24/03 9:54 AM Page 242

242 CHAPTER 3 Fitting Functions to Data

Regression and then either double click it or single click and then click OK.
Doing so brings up the window shown in Figure 3.69.

FIGURE 3.69

In this window, you first have to enter the Input Y Range—the cells in which
the values of the dependent variable y have been entered. The simplest way to do so
is to click the icon at the right end of the box; that brings you back to the original
spreadsheet, and you can highlight the entries down the first column (the y values)
under A and then press Enter. You then have to enter the Input X Range—the
cells in which the values of the two (or more) independent variables have been en-
tered. Again, click the icon at the right end of the box and then highlight the entries
in the second and third columns under B and C and press Enter.
Finally, you have to enter the Output Range—where the results of the regres-
sion calculations will be printed on the spreadsheet. You don’t want them printed over
the data values in the first three columns, so you probably would want them printed
starting, say, in column E. Click the white circle to the left of Output Range and then
click the icon at the right end of the box. Designate a block of cells starting at the top
of column E and extending down and to the right by highlighting the first cell under
E; then press Enter. Finally, in the Regression window, press OK.
Excel will then print a large amount of information, as shown in Figure 3.70.
Only a few of these entries are of interest to us; the rest are used for more sophisti-
cated statistical analysis. In particular, note that the first block of output is called
Regression Statistics and that the first two numbers under it are labeled
Multiple R and R Square—the values for the multiple correlation coefficient R
and the coefficient of determination, R 2.
The third block of output starts with three lines labeled Intercept, X Vari-
able 1, and X Variable 2. The entries to their right give the coefficients. In
particular, if the regression equation is y  ax1  bx2  c, the intercept is the
constant coefficient c, the value corresponding to the first X Variable is the co-
efficient a for x1 , and the value corresponding to the second X Variable is the
coefficient b for x2 . Once you have these values, you can write the multivariate re-
gression equation.
gord.3896.03.pgs 4/24/03 9:54 AM Page 243

3.7 Linear Models with Several Variables 243

FIGURE 3.70

Problems
1. A study was conducted relating the heights of a. Use an appropriate software package to calcu-
teenage boys y to the length of their radius bones x1 late the coefficient of determination and the
and the length of their femur bone x2 , as shown in multiple correlation coefficient. How much of
the accompanying table. (All measurements are in the variation in height is explained by the
centimeters.) length of the two bones?
b. Find the equation of the plane that best fits the
data.
c. Use your equation from part (b) to predict the
y x1 x2 height of a teenage boy whose radius measures
25.50 cm and whose femur measures 49.90 cm.
149.0 21.00 42.50
How close does your prediction come to the
152.0 21.79 43.70 boy’s actual height of 172 cm?
155.7 22.40 44.75 2. The windchill factor is an adjustment made to tem-
perature readings to take into account the effects of
159.0 23.00 46.00 the wind and so indicate how cold it feels. The follow-
ing table gives the windchill factors associated with
163.3 23.70 47.00
different combinations of air temperature in degrees
166.0 24.30 47.90 Fahrenheit and wind speeds in miles per hour.
169.0 24.92 48.95 a. Use an appropriate software package to calculate
the coefficient of determination and the multi-
174.5 25.80 50.30 ple correlation coefficient. How much of the
176.1 26.01 50.90 variation in the windchill factor is explained by
the two independent variables?
176.5 26.15 50.85 b. Find the equation of the plane that best fits the
data. (Note: The actual formula used to calculate
179.0 26.30 51.10
the windchill factors isn’t a linear function.)
Source: Wayne W. Daniel, Bio- c. Use your equation from part (b) to predict the
statistics: A Foundation for Analy-
sis in the Health Sciences, 4th ed., windchill factor corresponding to a temperature
New York: John Wiley & Sons, 1987. of 18°F and a 22 mph wind.
gord.3896.03.pgs 4/24/03 9:54 AM Page 244

244 CHAPTER 3 Fitting Functions to Data

Temperature
Wind Speed 35 30 25 20 15 10 5 0

5 33 27 21 16 12 7 0 5

10 22 16 10 3 3 9 15 22

15 16 9 2 5 11 18 25 31

20 12 4 3 10 17 24 31 39

25 8 1 7 15 22 29 36 44

d. Suppose that the air temperature is 8°F and the plained by the two independent variables?
windchill factor is 30. Use the equation of the b. Find the equation of the plane that best fits the
plane to predict the wind speed. data.
e. Which variable, temperature or wind speed, has c. Use your equation from part (b) to predict the
a greater effect on the windchill factor? Explain. forced expiratory volume of an individual whose
3. A study was conducted on a group of people with vital capacity is 2.6 liters and whose total lung ca-
pulmonary function problems to relate their forced pacity is 3.9 liters?
expiratory volume, in liters per second, to their vital d. Which variable, vital lung capacity or total lung
lung capacity x1 , in liters, and to their total lung ca- capacity, has a greater effect on forced expiratory
pacity x2 , in liters. volume? Explain.
4. A study was conducted on the weight w, in kilo-
grams, of 14 patients with primary type II hyper-
y x1 x2 lipoproteinemia (a genetic disorder that leads to
1.6 2.2 2.5 massively high cholesterol levels) just prior to the
start of medical treatment. For each patient, the
1.0 1.5 3.2 total cholesterol level x1 and the triglyceride level
1.4 1.6 5.0 x2 were recorded, in milligrams per 100 cubic cen-
timeters, as shown in the table at the top of the
2.6 3.4 4.4 next page.
1.2 2.0 4.4 a. Use an appropriate software package to calculate
the coefficient of determination and the multi-
1.5 1.9 3.3 ple correlation coefficient. How much of the
1.6 2.2 3.2 variation in weight is explained by the levels of
cholesterol and triglyceride?
2.3 3.3 3.3 b. Find the equation of the plane that best fits the
2.1 2.4 3.7 data.
c. Use your equation from part (b) to predict the
0.7 0.9 3.6 weight of a patient with primary type II hyper-
Source: Wayne W. Daniel, Biostatis- lipoproteinemia who has a cholesterol level of
tics: A Foundation for Analysis 332 and a triglyceride level of 186. How close
in the Health Sciences, 4th ed., New
does your prediction come to the actual reading
York: John Wiley & Sons, 1987.
for this patient, who weighed 78 kg?
5. As part of a study to investigate the relationship be-
a. Use an appropriate software package to calculate tween stress y and several other variables, the ac-
the coefficient of determination and the multi- companying data were collected on a random
ple correlation coefficient. How much of the sample of 14 corporate executives. In this table, x1
variation in forced expiratory volume is ex- measures the size of the firm (the number of
gord.3896.03.pgs 4/24/03 9:54 AM Page 245

Chapter Summary 245

x1 302 336 220 300 382 379 331 332 426 399 279 410 389 302
x2 139 101 57 56 113 42 84 186 164 205 230 160 153 139

w 76 97 83 52 70 67 75 78 70 99 75 70 77 76
Source: Wayne W. Daniel, Biostatistics: A Foundation for Analysis in the Health Sciences, 4th ed., New York: John Wiley & Sons, 1987.

employees), x2 measures the number of years in the

present position, x3 is the annual salary in thou-
sands, and x4 is the person’s age. y x1 x2 x3 x4
a. Use an appropriate software package to calculate 101 812 15 70 38
the coefficient of determination and the multiple
correlation coefficient when y is considered as a 60 334 8 60 52
function of x1 and x2 only. How much of the vari- 10 377 5 60 27
ation in the measure of stress is explained by
these two variables only? 27 303 10 94 36
b. Find the equation of the plane that best fits the 89 505 13 92 34
data on y as a function of x1 and x2 only.
c. Use your equation from part (b) to predict the 60 401 4 67 45
stress level of a corporate executive whose com- 16 177 6 66 50
pany has 484 employees and who has been in the
present position for 8 years. 184 598 9 92 60
d. Repeat parts (a)–(c) if you now take into ac-
34 412 16 74 44
count the three independent variables, x1 , x2 ,
and x3 . In particular, how much more of the 17 127 2 68 39
variation in stress level is explained by includ-
78 601 8 82 41
ing the third variable? Predict the stress level for
the individual in part (c) if x3  81. 141 297 11 124 58
e. Repeat parts (a)–(c) if you now take into ac-
11 205 4 71 51
count all four independent variables. In partic-
ular, how much more of the variation in stress 104 603 5 78 63
level is explained by including the fourth vari-
Source: Wayne W. Daniel, Biostatistics: A Foundation for
able? Predict the stress level of the individual in Analysis in the Health Sciences, 4th ed., New York: John
parts (c) and (d ) if x4  40. Wiley & Sons, 1987.

Chapter Summary

In this chapter, we introduced you to regression analysis. Specifically, we showed

you how to find the best fit line or curve for the various families of functions and
interpret the results in the following ways.
◆ How to find the regression, or least squares, line that is the best linear fit to a set
of data.
◆ How to interpret the correlation coefficient as a measure of how good the
linear fit is.
◆ How to use the linear regression equation for making predictions.
◆ How to find the exponential function, power function, or logarithmic func-
tion that best fits a set of data.
◆ How to transform a set of data to linearize it if the underlying pattern is an
exponential function, a power function, or a logarithmic function.
gord.3896.03.pgs 4/24/03 9:54 AM Page 246

246 CHAPTER 3 Fitting Functions to Data

◆ How to undo the transformation to produce the best-fit exponential or

power function.
◆ How to interpret the residuals to assess how well a linear function fits the
original data or the transformed data.
◆ How to interpret the sum of squares to assess how well a function fits a set
of data.
◆ How to use the best fitting nonlinear function for making predictions.
◆ How to fit a linear function of several variables to a set of data.

Review Problems
1. The table shows the budget and attendance (both in 3. (a) Find the best linear, exponential, power, and
millions) at 15 zoological parks in the United logarithmic functions that can be used to model the
States. Find the best-fit function from among lin- total expenditures on public education in the Unit-
ear, exponential, power, and logarithmic functions ed States as a function of time since 1979. (b) Use
for the attendance as a function of the budget. How each model to predict the amount spent on public
good is the fit? education in 2004. (c) Of the four predictions,
which seems the most accurate and which the least
accurate. Explain your reasoning.
Budget 10.0 3.4 27.0 6.2 9.7 4. Consider the linear functions you created in Prob-
Attendance 1.0 0.5 2.0 0.6 1.3
lems 2 and 3.
a. Interpret the slope in each case.
Budget 7.0 4.8 18.0 6.5 13.0 b. Use the two functions to estimate when expen-
Attendance 1.0 1.1 4.0 0.6 3.0 ditures for health care first exceeded those for
public education.
Budget 9.0 15.7 7.0 3.2 14.7
5. Consider the exponential functions you created in
Attendance 0.5 1.3 1.0 0.5 2.7 Problems 2 and 3.
a. Interpret the base in each case.
Problems 2–6 are based on the data below showing total b. Use the two functions to estimate when expendi-
expenditures, in billions of dollars, for both health and tures for health care first exceeded those for public
public education in the United States for the years education.
shown. 6. Find the linear function that best fits health care ex-
2. (a) Find the best linear, exponential, power, and penditures as a function of education expenditures.
logarithmic functions that can be used to model the What is the meaning of the slope of this line?
total health expenditures in the United States as a 7. The table on the next page gives the relationship be-
function of the number of years since 1979. (b) Use tween the average longevity (in years) and the ges-
each model to predict the amount spent on health tation period (in days) for a sample of animals. The
expenditures in 2004. (c) Of the four predictions, data indicate that the animals’ average longevity
which seems the most accurate and which is the can be predicted reasonably well as a function of
least accurate. Explain your reasoning. the gestation period.

Year 1980 1985 1990 1992 1993 1994 1995 1996 1997 1998
Health Expenditures 247.3 428.7 699.4 836.5 898.5 947.7 993.7 1042.5 1092.4 1146.1

Public Education 345.1 378.7 486.0 506.7 517.6 527.9 541.9 554.4 569.5 583.8
Source: 2000 Statistical Abstract of the United States.
gord.3896.03.pgs 4/24/03 9:54 AM Page 247

Review Problems 247

a. Find the linear, exponential, power, or logarith- 8. The U.S. Postal Service charges 37¢ for first-class
mic function that best predicts longevity as a postage for the first ounce of mail and 23¢ for every
function of gestation period. additional ounce. What linear function, based on
b. Use your function grapher to graph the best-fit weights 1, 2, . . . , 10 oz, best models this situation?
function. Explain why this function does not give the exact
charge for an 8.5-oz letter.
Animal Gestation Longevity 9. Consumer credit data from 1970 through 1995
show the following amounts of outstanding con-
baboon 187 20
sumer credit (in billions of dollars) in the United
black bear 219 18 States at the end of each year.
house cat 63 12 a. Find the linear function that best fits the data.
b. Use the model from part (a) to predict the
dog 61 12 amount of outstanding consumer credit at the
cow 284 15 end of 1992.
c. Determine the year in which the outstanding
elephant 660 35 consumer credit first will exceed $1.5 trillion.
giraffe 425 10
Year 1970 1975 1980 1985 1987 1988
gorilla 258 20
Credit 133.8 207.5 355.4 601.6 692.0 742.1
horse 330 20
Year 1989 1990 1991 1993 1994 1995
kangaroo 36 7
Credit 791.8 809.3 796.7 863.9 988.8 1131.9
lion 100 15
Source: World Almanac and Book of Facts.
monkey 166 15
10. The number of airline flights generally has risen
mouse 19 3 during the past 20 years. Flights (in millions) are
given for the years 1983–2000 in the table below:
opossum 13 1
a. Find the exponential function that fits the data.
rabbit 31 5 b. How well does the exponential function fit the
data?
sea lion 350 12
c. Predict when the number of airline flights will
squirrel 44 10 first exceed 10 million per year.
human 270 76 11. The table on the next page shows the height H in
feet and the number of stories n of some notable
Source: The Universal Almanac.
buildings.
a. Find the linear, exponential, power, or logarith-
mic function that best relates a building’s height
to its number of stories.
b. What is the significance of the slope of the linear
function?

Year 1983 1984 1985 1986 1987 1988 1989 1990 1991

Number of flights 5.0 5.4 5.8 6.4 6.6 6.7 6.6 6.9 6.8

Year 1992 1993 1994 1995 1996 1997 1998 1999 2000

Number of flights 7.1 7.2 7.5 8.1 8.2 8.2 8.3 8.6 9.0
Source: 2000 Statistical Abstract of the United States.
gord.3896.03.pgs 4/24/03 9:54 AM Page 248

248 CHAPTER 3 Fitting Functions to Data

H
H n
9
Empire State Building (New York) 1250 102
7.2
John Hancock Tower (Boston) 788 61
5.76
Sears Tower (Chicago) 1450 110 4.61
NationsBank Plaza (Dallas) 921 72 3.69
2.95
TCBY Tower (Little Rock) 546 40 2.36

Peachtree Center (Atlanta) 374 31

Place Ville Marie (Montreal) 620 45 t

0.75 2.09 3.29 4.37 5.33 6.19 6.95
Source: World Almanac and Book of Facts. 1.42 2.69 3.83 4.85 5.76 6.57

c. On average, how many feet are allocated to each b. Note that the times when the ball reaches its
story? maximum heights also appear to follow a linear
12. Draw a scatterplot for each of the functions f, g, and pattern. Find the best linear fit to these times as a
h in the table. For each set of data, decide whether function of n, where now H  9 when n  0,
the pattern of data is linear, exponential, or loga- etc. What is the meaning of the slope?
rithmic. Explain your choices. c. Note that the maximum heights don’t follow a
linear pattern. Find the best nonlinear function
that fits these data values, as a function of n,
x 1 2 3 4 5 6 from among the families of functions you have
studied in this chapter. For the exponential fit,
f (x) 6 4 2 0 2 4
what is the significance of the base you obtain?
g(x) 5.4 4.86 4.374 3.937 3.543 3.189 d. Find the best fit for the maximum heights H as a
function of time t.
h(x) 2 0.194 0.863 1.612 2.194 2.669
e. Use the results you obtained in parts (a)–(d) to
predict the corresponding values for the time
13. An experiment is conducted in which a ball is and height on the next bounce of the ball.
dropped from an initial height of 9 feet and its sub-
14. In October 2002, astronomers reported the discov-
sequent height above floor level as a function of
ery of a new body half the size of Pluto in an orbit
time is recorded and displayed, as illustrated in the
that takes 105,120 days to complete a full revolution
accompanying figure. When the curve is traced, the
about the sun at a distance of about 4 billion miles
measurements shown on the graph indicate the
from the sun. Verify whether this object, named
times when the ball hits the floor, the times when
Quaoar after a Native American god, satisfies Ke-
the ball reaches its maximum heights, and the val-
pler's Third Law from Example 4 in Section 3.6.
ues of these maximum heights.
15. In October 2002, a pilot flying a small plane in south-
a. Note that the times when the ball hits the floor
west Alaska reported spotting a bird with a wingspan
appear to follow a linear pattern. Find the best
of 14 feet. Based on Problem 8 in Section 3.5, how
linear fit to these times as a function of the num-
much would this newly discovered bird weigh?
ber of the bounce; that is, bounce number n  1
occurs at time t  0.75, and so on. What is the
meaning of the slope?
Gord.3896.04.pgs 4/24/03 9:56 AM Page 249

4
Extended Families
of Functions
4.1 Introduction to Polynomial Functions
Samantha has been keeping track of the price of the stock of HyperTech Corpora-
tion since her grandmother gave her several shares as a gift. She has plotted the
stock values, as shown in Figure 4.1, and wants to construct a mathematical model
that represents the price of the stock. Clearly, a linear function, an exponential
function, a power function, or a logarithmic function is not a reasonable candidate
because none have this kind of behavior pattern. To better capture the trend in the
stock prices, Samantha needs a function that changes both its direction and its
concavity, as illustrated in Figure 4.2.
Note that the graph increases, then decreases, and finally increases again. Thus,
the graph has two turning points, one at the local maximum point and the other at
the local minimum point. Also, the curve initially is concave down and then is con-
cave up, so the graph has one point of inflection, where the concavity changes.
In this section, we introduce a new family of functions, the polynomial func-
tions, that possess this type of more complicated behavior. A polynomial function,
or polynomial, is any finite sum of power functions with nonnegative integer
powers. For instance,
y  3x  5; (1)
Stock price
Stock price

Time Time

FIGURE 4.1 FIGURE 4.2

249
Gord.3896.04.pgs 4/24/03 9:56 AM Page 250

250 CHAPTER 4 Extended Families of Functions

y  6x 2  x  7; (2)
y  6  8x  3x 2; (3)
y  4x 3  5x 2  7x  12; (4)
y  10x 8  7x 5  3 (5)

are all polynomials.

The degree of a polynomial is the highest power of the variable present.
Hence, the degree of Polynomial (1) is 1, the degree of Polynomials (2) and (3) is 2,
the degree of Polynomial (4) is 3, and the degree of Polynomial (5) is 8.
The constant multiples in each term in any polynomial are called its coefficients.
In particular, the coefficient of the highest power term is the leading coefficient.
Thus, in Polynomial (1), the coefficients are 3 and 5 and the leading coefficient is 3;
in Polynomial (2), the coefficients are 6, 1, and 7 and the leading coefficient is 6.
Note that, in Polynomial (3), the leading coefficient is 3 (it is not necessarily the
first coefficient). As we show in Section 4.2, the sign of the leading coefficient
determines the overall behavior of the polynomial.
Another way to describe a polynomial is to say that it is a linear combination of
power functions because, as we noted, it is made up of a sum of power functions.
In this sense, power functions are the basic building blocks we use to construct any
polynomial.
If a polynomial has degree 1, it is a linear function of the form, y  ax  b,
where a and b are constants and a  0. Its graph is a line with slope m  a and
vertical intercept b.
If a polynomial has degree 2, it is called a quadratic function and it has the
form
y  ax 2  bx  c,
where a, b, and c are constants and a  0. With three coefficients in the equation,
the set of all quadratic functions is a three parameter family of functions. The
graph of any quadratic function is a curve called a parabola. Such curves abound
in the real world—in the path of a fly ball in baseball, in the shape of the main sup-
port cable in a suspension bridge such as the Golden Gate Bridge or the George
Washington Bridge, or in the cross sections of a TV satellite dish, as depicted in
Figure 4.3.
If a polynomial has degree 3, it is called a cubic function and its graph is called
a cubic. In general, a cubic function has the form
y  ax 3  bx 2  cx  d,

FIGURE 4.3
Gord.3896.04.pgs 4/24/03 9:57 AM Page 251

4.1 Introduction to Polynomial Functions 251

y y
y = x 4 − 5x 3 − 2x 2 + 6x − 4
30 100
y = x 3 + 3x 2 − 8x − 4 80
20 60
40
10
20
x x
–5 –4 –3 –2 –1 1 2 3 4 –3 –2 –1 1 2 3 4 5 6
–20
–10 –40
–60
–20
–80

FIGURE 4.4 FIGURE 4.5

where a, b, c, and d are constants and a  0. For example, the graph of the cubic
y  x 3  3x 2  8x  4 is shown in Figure 4.4. This graph is typical of a cubic
function, having two turning points and one inflection point.
If a polynomial has degree 4, it is called a quartic function and its graph is
called a quartic. In general, a quartic polynomial has the form
y  ax 4  bx 3  cx 2  dx  e,
where a, b, c, d, and e are constants and a  0. The graph of the quartic
y  x 4  5x 3  2x 2  6x  4 is shown in Figure 4.5. This graph is typical of a
quartic polynomial. Notice that it has three turning points and two inflection points.

Think About This How many parameters are there in the family of cubic polynomials? In the family
of quartic polynomials? In the family of polynomials of degree n, for any n? ❐

The Zeros of a Polynomial

A key piece of information about any polynomial function is the value or values
of the variable x that make the function zero. These values are known as the
zeros of the polynomial. For instance, the zeros of the quadratic polynomial
P1x 2  x 2  6x  8 are x  2 and x  4 because
P12 2  22  612 2  8  0 and P142  42  614 2  8  0.
From a different point of view, if we set the expression for the polynomial func-
tion equal to zero, we have an equation and the solutions to this equation are called
the roots. So, corresponding to the quadratic polynomial P1x 2  x 2  6x  8, we
have the quadratic equation
x 2  6x  8  0.
Factoring this expression gives
1x  22 1x  4 2  0.
The two solutions of this equation, x  2 and x  4, are the roots of the quadratic
function.
Note that a function has zeros, that an equation has roots, and that there is a di-
rect correspondence between them. The zeros of a function f occur at precisely the
same points as the roots of the equation f 1x 2  0.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 252

252 CHAPTER 4 Extended Families of Functions

E XAMPLE 1
Find the zeros of the quadratic function P1x 2  x 2  5x  6 and the roots of the corre-
sponding quadratic equation x 2  5x  6  0, both graphically and algebraically.
Solution The graph of the quadratic function y  x 2  5x  6 is shown in Figure 4.6.
Note that the graph crosses the x-axis twice: once when x  2 and again when x  3. So,
graphically, we conclude that these are the zeros of the function. If we consider the asso-
ciated quadratic equation
x 2  5x  6  0,
its roots are x  2 and x  3.
y

y = x 2 − 5x + 6

x
1 2 3 4

FIGURE 4.6

Alternatively, we can solve this equation algebraically. We start with the associated
quadratic equation
x 2  5x  6  0.
We can factor the quadratic expression on the left as
1x  22 1x  3 2  0.
Recall that, when the product of two factors is zero, one or the other or both must be zero,
so we have either x  2  0 or x  3  0, leading to the roots x  2 and x  3. Because
they are the roots of the quadratic equation, they are also the zeros of the quadratic polyno-
mial P1x2  x 2  5x  6.
◆
If the coefficients in a quadratic are appropriately chosen, we may be able
to find the roots of a quadratic equation by algebraic factoring, as we did in
Example 1. If the coefficients are not just right — say, 4.35709x 2  15.46031x 
11.02013  0, or even 5x 2  3x  17  0 — the factoring approach won’t
work. The same principle applies to polynomials of higher degree, but the al-
gebra typically becomes much more complicated as the degree of the polyno-
mial increases. Consequently, factoring is far less likely to work when the
degree of a polynomial is 3 or higher.
The two roots for any quadratic equation
ax 2  bx  c  0, a  0,
always can be found from the quadratic formula.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 253

4.1 Introduction to Polynomial Functions 253

The Quadratic Formula

b  2b2  4ac
x
2a

This formula is derived in any algebra textbook.

E XAMPLE 2
Find the zeros of the quadratic polynomial P1x 2  x 2  3x  5.
Solution With a  1, b  3, and c  5, the quadratic formula gives the roots of
the associated quadratic equation x 2  3x  5  0 as
13 2  213 2 2  411 2 152
x
211 2
3  29  120 2

2
3  229
 .
2
The result is a pair of irrational numbers. Thus, the roots are
3  229 3  229
x  4.19258 and x   1.19258.
2 2
◆
The quadratic formula was essentially known to the ancient Babylonians, some
4000 years ago. However, not until about 1540 did Italian mathematicians Tartaglia
and Cardano discover a comparable, although considerably more complicated, for-
mula for the three roots of any cubic equation. Not long after that, another Italian
mathematician, Ferrari, discovered an even more complicated formula that gives
the four roots of any quartic equation. (These formulas are programmed into some
calculators and software packages.) Finally, in 1824, Danish mathematician Abel
proved that no general formula could exist that would give the roots of any polyno-
mial equation of fifth or higher degree. When we encounter polynomials of higher
degree, we usually have to resort to numerical methods to find the roots. We illus-
trate this approach in Example 3 for a polynomial of degree 3.

E XAMPLE 3
Find, correct to four decimal places, all the zeros of the cubic polynomial y  Q1x 2 
x 3  3x 2  8x  4.
Solution The graph of this polynomial is shown in Figure 4.7. Note that it crosses the
x-axis three times and that each of these points is a zero of the polynomial. By zooming
in on each point in turn, using a function grapher, we estimate that the zeros are located
at approximately x  4.56155, x  0.43845, and x  2.00000. This last value for x
suggests that the third zero may be x  2 precisely. To determine whether that is indeed
the case, we substitute x  2 into the formula for the cubic and find that
Q12 2  12 2 3  312 2 2  812 2  4  8  12  16  4  0,
Gord.3896.04.pgs 4/24/03 9:57 AM Page 254

254 CHAPTER 4 Extended Families of Functions

y
y = x 3 + 3x 2 − 8x − 4
80
60
40
20
x
–6 –4 –2 2 4 6
–20
–40
–60
–80

FIGURE 4.7

so x  2 is precisely the zero. Because we were asked to give the three zeros correct to
four decimal places, we conclude that x  4.5616, x  0.4385, and x  2.
◆
We found three zeros for the cubic polynomial in Example 3 based on its
graph in Figure 4.7. But, how do we really know that there are no additional zeros?
We could expand the viewing window and examine the graph from x  100 to
x  100, say, or perhaps from x  1000 to x  1000, and maybe other zeros will
come into view. Unfortunately, this kind of exploratory approach never completely
closes the door on the possibility that other zeros might exist if only we look further.
Instead, we need to know something about the behavior of polynomials in general,
which will give us information on how many zeros a particular polynomial has and
some indication of where to look for them. We discuss this in the next section.

Problems
1. What is the degree of each polynomial? d. P1x 2  8x 3  5x 2  6
a. P1x2  6x 3  5x 2  8 e. P1x 2  6x 3  5x 2  8
b. P1x2  5x 4  6x 3  7x  11 f. P1x 2  3x 2  5x  4
c. P1x2  6x  5x 2 g. P1x 2  4x 2  5x  8
d. P1x2  x 5  x 8 h. P1x 2  4  3x 2
e. P1x2  4x 3  9x 2  12 i. P1x 2  8  5x  4x 2
f. P1x2  10  4x  5x 3  3x 6 j. P1x 2  6x 4  5x 3  8x  3
k. P1x 2  3x 4  5x 3  8x  6
l. P1x 2  3  6x 4
2. What is the leading coefficient of each polynomial
in Problem 1?
For each polynomial in (a)–(l), indicate whether it is a
3. Which values of x are zeros of the polynomial
P1x2  x 3  2x 2  3x and which are not? i. quadratic polynomial,
ii. cubic polynomial,
a. x  3 b. x  2 c. x  1
d. x  0 e. x  1 f. x  2 iii. quartic polynomial,
g. x  3 iv. quadratic polynomial whose leading coefficient
is 4,
4. Consider these polynomials.
v. cubic polynomial whose leading coefficient is
a. P1x2  6x 3  5x 2  8 6, or
b. P1x2  6  5x 2  8x 3 vi. quartic polynomial whose leading coefficient is
c. P1x2  8  5x 2  6x 3 6.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 255

4.2 The Behavior of Polynomial Functions 255

5. The table gives some values for a polynomial P. points? Estimate, correct to two decimal places, all
Identify possible roots of the corresponding poly- the turning points.
nomial equation. 10. For the quartic polynomial in Problem 8, how
many turning points are there? how many inflec-
x 5 4 3 2 1 tion points? Estimate, correct to two decimal
places, all the turning points.
P(x) 227 21 0 8 0
y
x 0 1 2 3 4 5

P(x) 16 23 0 32 0 166

6. The figure to the right shows the graph of a polyno-

mial. How many zeros does it have?
7. Estimate, correct to three decimal places, all the zeros
of the polynomial P1x 2  2x 3  6x 2  5x  3. x

8. Estimate, correct to three decimal places, all the zeros

of the polynomial P1x2  x 4  4x 3  5x  1.
9. For the cubic polynomial in Problem 7, how many
turning points are there? how many inflection

Exercising Your Algebra Skills

Add or subtract each pair of polynomials by combining 9. x17  3x 2 10. x16  5x 2
like terms. 11. 1x  1 2 1x  3 2 12. 1x  2 2 1x  5 2
1. 16x 3  5x 2  82  16x  5x 2 2 13. 1x  2 2 1x  3 2 14. 1x  4 2 1x  3 2
2. 16x 3  5x 2  82  16x  5x 2 2 15. 1x  5 2 1x  5 2 16. 1x  3 2 1x  3 2
3. 15x 4  6x 3  7x  112  14x 3  9x 2  12 2 17. 1x  2 2 1x  2 2 18. 1x  21 2 1x  21 2
4. 15x 4  6x 3  7x  112  14x 3  9x 2  12 2 Raise each polynomial to the indicated power.
5. 110  4x  5x 3  3x 4 2  15x 4  6x 3  7x  11 2 19. 1x  1 2 2 20. 1x  3 2 2
6. 110  4x  5x 3  3x 4 2  15x 4  6x 3  7x  11 2 21. 1x  2 2 2
22. 12x  5 2 2
Multiply each pair of polynomials. 23. 12x  6 2 2 24. 1x  10 2 2
7. x13x  52 8. x14x  22

4.2 The Behavior of Polynomial Functions

The behavior of a polynomial depends on the ideas we introduced in Section 4.1:
the degree, the zeros, and the sign of the leading coefficient of the polynomial. Let’s
see how.

Quadratic Polynomials
We begin by analyzing the behavior of quadratic functions. The graph of any quad-
ratic function y  ax 2  bx  c is a parabola that opens either upward or down-
ward. The sign of the leading coefficient a in
y  ax 2  bx  c
determines whether the parabola opens upward or downward and so determines
the overall behavior of the parabola. When the leading coefficient is positive, the
Gord.3896.04.pgs 4/24/03 9:57 AM Page 256

256 CHAPTER 4 Extended Families of Functions

parabola opens upward and is concave up. When the leading coefficient is negative,
the parabola opens downward and is concave down. To understand why, think
about what happens when x gets very large—say, x  100 or x  1000. Then x 2 is
much larger, on the order of 10,000 or 1,000,000. Therefore the term ax 2 eventual-
ly overwhelms any contribution from the linear term bx or the constant term c.
Thus, when a is positive, the quadratic term is extremely positive and the parabola
opens upward. Similarly, when a is negative, the quadratic term is extremely nega-
tive and the parabola opens downward.
For instance, the parabola y  5x 2  20x  300 has a positive leading coeffi-
cient and so opens upward—when x becomes large, either positively or negatively,
the overall effect is positive. In contrast the parabola y  20  4x 2 has a negative
leading coefficient and so opens downward—when x becomes large, either posi-
tively or negatively, the overall effect is negative. Check the graphs of both func-
tions on your function grapher to convince yourself of the behavior in each case.
Moreover, whichever way the parabola opens, as x increases indefinitely in either
direction, the parabola either increases toward infinity or decreases toward nega-
tive infinity, as illustrated in Figure 4.8.

y y

x x

FIGURE 4.8 Leading coefficient > 0 Leading coefficient < 0

Every parabola has one turning point—also called its vertex. For instance, the
parabola y  x 2 has its vertex at the origin, because that is the location of the turn-
ing point. If a parabola opens upward, the turning point corresponds to the mini-
mum value of the function. If a parabola opens downward, the turning point
corresponds to the maximum value of the function. In addition, the parabola is al-
ways symmetric about the vertical line through its turning point, so the left and
right halves of the parabola are mirror images of one another. (See Appendix D for
a discussion of symmetry.)
Next, let’s examine the effects of the other two terms (the linear term and the
constant term) in the formula for a quadratic function. Figure 4.9 shows the graphs
associated with the quadratic functions y  x 2, y  x 2  6, y  x 2  5x  6, and
y  x 2  5x  6. The leading term determines the basic behavior of the quadrat-
ic function, so all four open upward. However, the other terms affect the location
of the graph. The constant term 6 in y  x 2  6 raises the parabola y  x 2 by
6 units (if the constant term is negative, the parabola would be lowered instead). Use
your function grapher to experiment with this effect on the graph of the parabola by
changing the constant term. For instance, how do the graphs of y  x 2  5x  7
and y  x 2  5x  2 compare to the graph of y  x 2  5x  6? Be sure to look at
Gord.3896.04.pgs 4/24/03 9:57 AM Page 257

4.2 The Behavior of Polynomial Functions 257

y y

y = x2 + 6

6
y = x2

x x

y y

y = x 2 − 5x + 6 y = x 2 + 5x + 6

x x
FIGURE 4.9 2 3 –3 –2

enough graphs to convince yourself of the effect of the constant term. We investigate
this effect in detail in Section 4.7.
The effect of the linear term is more complicated because it involves both ver-
tical and horizontal shifting of the parabola. We don’t go into that here but do so in
Section 4.7.
Furthermore, as we showed in Section 4.1, the two roots of any quadratic
equation
ax 2  bx  c  0
can always be found from the quadratic formula
b  2b2  4ac
x .
2a
These two roots could be real numbers (as in Examples 1 and 2 in Section 4.1) or
the roots could be a pair of complex numbers of the form x  a  bi and
x  a  bi where i  11 (a and b are the Greek letters alpha and beta, respec-
tively). A pair of complex numbers such as these is called a pair of complex conju-
gates. Complex numbers are discussed in Appendix E.

E XAMPLE 1
Find the roots of the quadratic equation x 2  2x  2  0.
Solution Using a  1, b  2, and c  2 in the quadratic formula, we get
2  212 2 2  411 2 12 2
x
211 2
Gord.3896.04.pgs 4/24/03 9:57 AM Page 258

258 CHAPTER 4 Extended Families of Functions

2  24  8

2
2  24 2  2i
  .
2 2
We now divide through by 2 and find that the two complex roots of the quadratic are
x  1  i and x  1  i.
◆
Because every quadratic equation has two roots, every quadratic function has ex-
actly two zeros. Let’s see what this means in terms of the graph of the quadratic func-
tion. Consider again the quadratic polynomial P1x2  x 2  5x  6, whose graph is
shown in Figure 4.10. This polynomial has zeros at x  2 and x  3 because we can
factor the quadratic as
x 2  5x  6  1x  2 2 1x  3 2 .

y = x 2 − 5x + 6

x
1 2 3 4

FIGURE 4.10

But the graph shows that the parabola crosses the x-axis at the points x  2
and x  3. Thus, just as the point at which a line crosses the x-axis gives the root of
a linear equation, the points at which a parabola crosses the x-axis give the real
roots of a quadratic equation, as illustrated in Figure 4.11 (a) and (b), respectively.
y y

Root
Root

x x

FIGURE 4.11 (a) (b)

If we know that a parabola crosses the x-axis at a point x  r, then x  r is a

zero of the associated quadratic function and x  r is a factor of the quadratic ex-
pression. You can locate the real roots of any quadratic to any desired level of accu-
racy with your graphing calculator or with the quadratic formula, so you can
always find the linear factors.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 259

4.2 The Behavior of Polynomial Functions 259

In general, for any quadratic function f 1x 2  ax 2  bx  c,

◆ the real roots of the quadratic equation ax 2  bx  c  0 corre-
spond graphically to the points where the associated parabola cross-
es the x-axis, and
◆ the real roots of the quadratic equation ax 2  bx  c  0 corre-
spond algebraically to the linear factors of the quadratic polynomial.

Depending on the orientation of the parabola (opening up or down) and the po-
sition of the turning point, a parabola may not touch the x-axis at all. This is the case
with the graph of y  x 2  6, as shown in Figure 4.12. For such a parabola, the cor-
responding quadratic equation still has two roots, but they are complex roots. If a
quadratic equation has complex roots, they must occur in conjugate pairs of the
form a  bi. This property follows directly from the quadratic formula for the case
where the term inside the radical, b2  4ac, is negative. The expression b2  4ac is
called the discriminant of the quadratic. When the discriminant is positive, the two
roots are real numbers. When the discriminant is negative, the two roots are complex
numbers. Finally, when the discriminant is zero, there is a double real root.

20

15

10

FIGURE 4.12 x
–4 –3 –2 –1 1 2 3 4

For instance, the discriminant for y  x 2  6 is 02  411 2 16 2  24  0, so

the two roots are complex and they occur in pairs. From the quadratic formula, the
roots are
0  202  411 2 16 2
x
211 2
 224

2
2 26
   26 i.
2

We have already demonstrated that a parabola can cross the x-axis at two
points (corresponding to two real roots) or that it may not ever cross the x-axis
(corresponding to a pair of complex conjugate roots). A third possibility is that
the parabola could be tangent to the x-axis; that is, it can touch the axis and
bounce back without ever crossing the axis. For instance, consider the quadratic
Gord.3896.04.pgs 4/24/03 9:57 AM Page 260

260 CHAPTER 4 Extended Families of Functions

10

x
FIGURE 4.13 –1 1 2 3 4 5

function y  x 2  4x  4. If you apply the quadratic formula, you will find that
the discriminant b2  4ac  14 2 2  411 2 14 2  0 and the two roots are x  2
and x  2, as shown in Figure 4.13. Use your function grapher to zoom in on this
point and note how the parabola just touches the x-axis at x  2. Thus, the quad-
ratic function has two roots, but because they are equal, x  2 is a double root.

Think About This Try changing the value of the constant term in y  x 2  4x  4 slightly from 4—
say, to 4.01 and then to 3.99. What happens to the graph in each case? What is the
value of the discriminant in each case? ❐

We summarize these ideas about quadratic polynomials as follows.

Characteristics of Quadratic Polynomials y  ax 2  bx  c

◆ A quadratic polynomial has degree 2 and has precisely 2 zeros.
◆ A parabola has precisely one turning point, its vertex.
◆ A parabola opens upward if the leading coefficient a is positive; it opens
downward if the leading coefficient a is negative, as shown in Figures 4.14
(a)–(f).
◆ The corresponding quadratic equation of degree 2 has precisely
2 roots. They may be real or complex.
◆ The complex roots occur in pairs of complex conjugates, x  a  bi,
where i  11, as illustrated in Figures 4.14 (a) and (d).
◆ The real roots correspond to the points where the parabola crosses the
x-axis, as illustrated in Figures 4.14 (b) and (e), or where the parabola
touches the x-axis, as illustrated in Figures 4.14 (c) and (f).
◆ You can always find the real roots graphically by using your function
grapher to zoom in on the points where the parabola crosses or
touches the x-axis.
◆ The real roots correspond to the linear factors of the quadratic
expression.
◆ You can always find the roots, real or complex, by using the quadratic
formula.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 261

4.2 The Behavior of Polynomial Functions 261

y y y

a>0 a>0 a>0

x x x

(a) (b) (c)

y y y

x x x

a<0 a<0 a<0

(d) (e) (f)

FIGURE 4.14

Cubic Polynomials
Next, we consider the characteristics of cubic polynomials having the form
y  ax 3  bx 2  cx  d,
where a, b, c, and d are constants and a  0. The graph of the cubic polynomial
y  x 3  3x 2  8x  4 is shown in Figure 4.15, which is typical of a cubic func-
tion. The cubic rises toward positive infinity in one direction and drops toward
negative infinity in the other. Also, this particular cubic has two turning points and
crosses the x-axis at three points, so it has three real zeros. Moreover, the curve has
one point of inflection, is concave down on one side of the point of inflection, and
is concave up on the other.

Turning point

Point of
inflection
x

Turning point

FIGURE 4.15

As with a quadratic function, the sign of the leading coefficient in a cubic al-
ways determines the overall behavior pattern of the function. If the leading coeffi-
cient is positive, the cubic increases as x increases (except possibly for a relatively
small dip between the two turning points), as shown on the left in Figure 4.16. If
Gord.3896.04.pgs 4/24/03 9:57 AM Page 262

262 CHAPTER 4 Extended Families of Functions

the leading coefficient is negative, the cubic decreases as x increases (except for a
possible rise between the two turning points), as shown on the right in Figure 4.16.

y y

a>0 a<0

x x

FIGURE 4.16

In general, a cubic function y  ax 3  bx 2  cx  d has three zeros, and the

corresponding cubic equation
ax 3  bx 2  cx  d  0
has three roots. The roots can all be real numbers or can consist of a single real
number and a pair of complex conjugate numbers. Each of the real roots corre-
sponds to a linear factor of the corresponding cubic expression. Any complex con-
jugate roots must occur in pairs and correspond to a quadratic factor of the cubic
polynomial.

The real roots of a cubic equation correspond graphically to the points at

which the associated cubic curve crosses the x-axis.
The real roots of a cubic equation correspond algebraically to the linear
factors of the cubic polynomial.

If a cubic has three real roots, its curve crosses the x-axis at the correspon-
ding three points. If it has only one real root, the curve crosses the x-axis only
once, as shown in Figure 4.17.

FIGURE 4.17
Gord.3896.04.pgs 4/24/03 9:57 AM Page 263

4.2 The Behavior of Polynomial Functions 263

E XAMPLE 2
Analyze the behavior of the cubic function
f 1x2  1x  1 2 1x  2 2 1x  5 2  x 3  6x 2  3x  10.

Solution The cubic has the three linear factors— 1x  1 2, 1x  2 2, and 1x  5 2 —so it
has three real zeros: at x  1, x  2, and x  5, corresponding to each of the three
factors. Consequently, its graph crosses the x-axis at x  1, 2, and 5, as shown in
Figure 4.18. Further, the leading term x3  1 # x3, being the highest power present, even-
tually dominates the other terms as x increases. Because the leading coefficient 1 is posi-
tive, the cubic must increase toward  as x S  and decrease toward  as
x S . Verify this result graphically by using your function grapher and numerically
by substituting some large positive and negative values for x.
y

y = x 3 + 6x2 + 3x − 10

FIGURE 4.18

◆
Although there is a formula for calculating the roots of a cubic equation, it is
considerably more complicated than the quadratic formula and is seldom used. If the
cubic polynomial happens to factor simply, you can find the zeros directly because
each factor corresponds to a zero. However, that is not likely to happen. Usually, the
simplest way to find the real roots of a cubic equation is to approximate them by
using your function grapher—just keep zooming in on the points where the curve
crosses the x-axis until you find the roots to whatever degree of accuracy you desire.
We summarize these ideas about cubic polynomials as follows.

Characteristics of Cubic Polynomials

y  ax3  bx2  cx  d
◆ A cubic polynomial of degree 3 has precisely 3 zeros.
◆ A cubic has at most two turning points.
◆ A cubic typically has one inflection point.
◆ A cubic increases (rises upward) to the right as x increases if the lead-
ing coefficient a is positive; it decreases (falls downward) to the right
as x increases if the leading coefficient a is negative.
◆ The corresponding cubic equation of degree 3 has precisely 3 roots.
The roots may all be real or one real and two complex.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 264

264 CHAPTER 4 Extended Families of Functions

◆ The complex roots occur in pairs of complex conjugates, a  bi,

where i  11 .
◆ The real roots correspond to the points where the cubic crosses the
x-axis.
◆ You can always find the real roots graphically by using your function
grapher to zoom in on the points where the cubic crosses the x-axis.
◆ The real roots correspond to linear factors of the cubic expression.

Figure 4.19 illustrates most of the possible cases for a cubic polynomial. In Fig-
ures 4.19 (a) and (b) there are three distinct real roots when the leading coefficient
a is either positive or negative. In Figures 4.19 (c) and (d) there are three real roots,
but one of them is repeated, so the x-axis is tangent to the cubic at the correspon-
ding point. These two graphs correspond to when the leading coefficient a 0;
similar graphs can be drawn when a  0. Figure 4.19 (e) shows a cubic with a
triple real root and a 0; note how the curve flattens as it crosses the x-axis. Think
about the graph of y  x 3 as it passes through the origin. Finally, in Figures 4.19
(f)–(h) there is one real root and a pair of complex roots, again when a 0. You
can draw similar graphs when a  0.
y y y y

a>0 a<0 a>0 a>0

x

x x x

(a) (b) (c) (d)

y y y y

a>0 a>0 a>0 a>0

x
x x

FIGURE 4.19 (e) (f) (g) (h)

Moreover, it turns out that every cubic is symmetric about its inflection
point. Imagine a cubic with a hinge at its inflection point—if you take either half
of the curve and rotate it about that hinge, it will eventually be a perfect fit to the
other half of the curve.

Think About This Prove that any cubic polynomial of the form f 1x 2  ax 3 is symmetric about its in-
flection point at the origin by showing that for any value of x—say, x  h 0—
then f 1h2  f 1h2 . ❐

Polynomials of Degree n
The ideas discussed for polynomials of degree 2 (quadratics) and degree 3 (cubics)
can be extended to polynomials of any degree n, where n is a positive integer. In
particular, they have the following characteristics.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 265

4.2 The Behavior of Polynomial Functions 265

Characteristics of Polynomials of Degree n

◆ A polynomial of degree n has precisely n zeros.
◆ A polynomial of degree n has at most n  1 turning points.
◆ A polynomial of degree n has at most n  2 points of inflection.
◆ If the leading coefficient is positive, the polynomial rises toward 
to the right as x S , and if the leading coefficient is negative, the
polynomial falls toward  to the right as x S .
◆ The corresponding polynomial equation of degree n has precisely n
roots, which may be real or complex.
◆ The complex roots occur in pairs of complex conjugates, a  bi,
where i  11 .
◆ The real roots correspond to the points where the curve crosses or
touches the x-axis.
◆ You can always find the real roots graphically to any desired level of
accuracy by using your function grapher to zoom in on the points
where the curve crosses the x-axis.
◆ The real roots correspond to linear factors of the polynomial expression.

Think About This Sketch the graph of a fifth degree polynomial with five real roots and a positive
leading coefficient. Sketch the graph of a fifth degree polynomial with three real
roots and a negative leading coefficient. ❐
E XAMPLE 3
Suppose that a polynomial has roots at x  4, 1, 1, 3, and 6. Find a possible formula
for it and describe its behavior.
Solution The polynomial has five real roots, so its degree must be at least 5; it might be
higher if there are complex roots or repeated roots. The five corresponding linear factors
are 1x  142 2  1x  4 2, 1x  1 2, 1x  1 2 , 1x  3 2, and 1x  6 2. If these are the
only roots, one possible formula for this polynomial is
P1x 2  1x  4 2 1x  1 2 1x  12 1x  3 2 1x  6 2,
although any constant multiple A of this expression would be an alternative formula.
We can determine the value of the multiple A if we know the vertical intercept of
the polynomial—or any other point on the curve. If the multiple A is positive, the graph
of the polynomial has the behavior shown in Figure 4.20. Note that P1x 2 rises toward 

x
–4 –1 1 3 6

FIGURE 4.20
Gord.3896.04.pgs 4/24/03 9:57 AM Page 266

266 CHAPTER 4 Extended Families of Functions

as x S  and that P1x2 falls toward  as x S . Alternatively, if the constant mul-
tiple A is negative, this behavior is reversed; the graph drops toward  as x S  and
rises toward  as x S . Can you explain why this is the case?
◆
What if a polynomial has a double or repeated factor? For instance,
P1x2  1x  1 2 1x2 2 1x4 2 2  0
has roots at x  1, 2, and 4, but x  4 is a double root because 1x  4 2 2 
1x  42 1x  4 2 is a repeated factor. Note that its graph, as shown in Figure 4.21,
falls to touch the x-axis at x  4 where it flattens and then rises again. Zooming in
on the curve about this point reveals that the x-axis is tangent to the graph at
x  4, just as the x-axis is tangent to the parabola y  x 2 at the origin.

P(x) = (x + 1)(x − 2)(x − 4)2

x
–2 –1 1 2 3 4

FIGURE 4.21

Think About This The polynomial y  1x  1 2 1x  2 2 3 has a triple factor. Examine its graph to see
what happens near that triple root. First, try to predict what will happen, based on
your knowledge of the behavior of y  x 3 near the origin. ❐

E XAMPLE 4
Use the fifth degree polynomial from Example 3 to demonstrate why it must have four
turning points and three inflection points.
Solution Let’s trace the polynomial’s curve in Figure 4.20 slowly from left to right. The
function starts rising as we move to the right and crosses the x-axis at the first root at
x  4. It must cross the x-axis again at x  1, so there must be a turning point be-
tween these two roots. Similarly, there must be a turning point between the roots at
x  1 and x  1, and in fact, there is a turning point between each successive pair of
roots. Because there are five real roots, there must be four turning points.
Now let’s consider inflection points. We begin with the first two turning points, one
near x  3 where the curve is concave down and the next near x  0.2 where the
curve is concave up. The change in concavity means that there must be an inflection
point between the successive turning points. In fact, between each successive pair of
turning points, there is a point of inflection. Because there are four turning points, there
must be three inflection points.
◆
Things may not be quite so simple if there are complex roots or multiple roots.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 267

4.2 The Behavior of Polynomial Functions 267

The End Behavior of a Polynomial

The end behavior of any polynomial depends on the sign of the leading term because,
as x approaches  or , the leading term eventually dominates all other terms.
For instance, consider the polynomial P1x 2  2x 4  6x 3  7x  10. When x is very
large, the term 2x 4 will overwhelm all other terms. The graphs of the functions
P1x 2  2x 4  6x 3  7x  10 and Q1x2  2x 4 for x between 3 and 4 and y be-
tween 25 and 100 are shown in Figure 4.22; the two curves look quite different.
Figure 4.23 presents a slightly larger view, where x is between 6 and 6 and y is be-
tween 50 and 950. Here the two curves look more similar than in the preceding
view. In the much larger view presented in Figure 4.24, where x is between 25
and 25 and y is between 0 and 500,000, there no longer is much difference be-
tween the two curves. The term 2x 4 dominates the behavior of the polynomial, and
the effect of the lower power terms is negligible.

y y y

100 950 500,000 Q(x) = 2x 4

Q(x) = 2x 4
Q(x) = 2x 4 400,000
75 750 P(x)

P(x) 300,000
50 550
P(x) 200,000
25 350
100,000
x 150
–2 3 x
x –25 –15 –5 5 15 25
–25 –4 –2 2 4

FIGURE 4.22 FIGURE 4.23 FIGURE 4.24

In general, for any polynomial, when x is large enough, the curve is indistin-
guishable from the curve corresponding to just the leading term. In other words, in
the large, the behavior of any polynomial is virtually identical to that of the power
function consisting of the leading term. You can see the end behavior easily on your
function grapher if you use a reasonably large viewing window.
On the one hand, if the viewing window is too large, the location of the turn-
ing points and the zeros of a polynomial is a local aspect of the graph and can be
easily missed. On the other hand, if the viewing window is too small, the overall
growth pattern of the polynomial is lost. For instance, by focusing too closely on
one particular turning point or root, you may lose sight of all the others. Rarely
does a single view suffice to show all the important details of a function. There-
fore, as a matter of routine, you should use the information given in several dif-
ferent views on your calculator or computer to sketch a rough hand-drawn graph
of the function, called the complete graph, which highlights the key information,
even if you intentionally do not draw it to scale.
We expect you to use your function grapher to produce the graph of a poly-
nomial, but you should interpret with care what the calculator or computer
shows. Usually, the important characteristics of any function—and a polynomial
in particular—are
Gord.3896.04.pgs 4/24/03 9:57 AM Page 268

268 CHAPTER 4 Extended Families of Functions

◆ the end behavior (is it increasing or decreasing as x S  or x S ?),

◆ the intervals over which the function is increasing or decreasing,
◆ the locations of the turning points,
◆ the intervals over which the function is concave up or concave down,
◆ the locations of the points of inflection, and
◆ the locations of the real zeros.

E XAMPLE 5
For the polynomial shown in Figure 4.25, answer the following questions.
a. How many real roots are there?
b. How many turning points are there?
c. How many inflection points are there?
d. What is the minimum degree of the polynomial?
e. How many complex roots does it have?
f. What is the sign of the leading coefficient?
y

FIGURE 4.25

Solution

a.–c. The graph shown in Figure 4.25 reveals five real roots that correspond to the five
points where the curve crosses the x-axis. It also shows six turning points and five
inflection points.
d. Because the number of turning points is typically 1 less than the degree of the
polynomial and the number of inflection points is 2 less than the degree, we con-
clude that the polynomial shown is at least a seventh degree polynomial.
e. Because there are five real roots and the degree of the polynomial is at least seven,
there must be at least two complex roots.
f. The graph eventually falls toward  as x S , so we conclude that the leading
coefficient must be negative.
◆
E XAMPLE 6
Factor the polynomial P1x2  x 4  5x 3  7x 2  8x  3.
Solution This polynomial is a quartic, so it has precisely four roots. We know that the
linear factors of the polynomial correspond to its real roots, and the graph shown in Fig-
Gord.3896.04.pgs 4/24/03 9:57 AM Page 269

4.2 The Behavior of Polynomial Functions 269

ure 4.26 reveals that there are four real roots. As a result, there cannot be any complex
roots. We can locate each of these real roots to any desired level of accuracy, using either
numerical or graphical methods. Correct to four decimal places, the roots are
x  1.6272, x  0.3105, x  1.0000, and x  5.9377. The third of these results,
x  1.0000, suggests that the root might be x  1 precisely. To verify that this is true, we
substitute into the formula for the polynomial and find that
P11 2  11 2 4  511 2 3  711 2 2  811 2  3  1  5  7  8  3  0.

x
–2 –1 1 2 3 4 5 6 7

–50

–100

–150
FIGURE 4.26

(If you do the same with the other three values, which are just approximations to the
roots, the value of the polynomial will only be close to, but not quite equal to, zero.)
The corresponding linear factors are therefore roughly 1x  1.6272 2, 1x  0.31052,
1x  12, and 1x  5.93772 , so the polynomial can be factored, approximately, as
P1x2  x 4  5x 3  7x 2  8x  3  1x  1.62722 1x  0.31052 1x  1 2 1x  5.93772.
◆
Problems
y y
1. The overall trend in the growth of the gross domes-
tic product (GDP) has been upward except for a
small dip. Sketch a graph representing the value of
the GDP as a function of time. What type of func-
x x
tion might model it? What can you conclude about
any of the coefficients?
2. The overall pattern in the growth of the Dow-Jones
(i) (ii)
Industrial Average over the past 10 years has been
one of increase except for three sharp, but relatively y y
short-term, drops. Sketch a graph representing the
value of the Dow as a function of time. What type of
function might model it? What can you conclude
about any of the coefficients? x x

3. Each graph represents a polynomial. For each one:

a. What is the minimum possible degree of the (iii) (iv)
polynomial? Why?
b. Is the leading coefficient of the polynomial posi- 4. Each table gives some values for a polynomial. What
tive or negative? Why? is the minimum degree of each polynomial? Based
Gord.3896.04.pgs 4/24/03 9:57 AM Page 270

270 CHAPTER 4 Extended Families of Functions

on the values given, what can you conclude about Based on your knowledge about roots and factors,
the sign of the leading coefficient in each case? sketch the graph of each polynomial function in Prob-
a. lems 6–9. Do not use your function grapher.
6. f 1x 2  1x  22 1x  1 2 1x  3 2
7. f 1x 2  51x 2  4 2 1x 2  25 2
x 3 2 1 0 1 2 3 4 5

y 145 16 5 9 12 3 21 2 48 8. f 1x 2  51x 2  4 2 1x 2  252

b. 9. f 1x 2  51x  4 2 2 1x 2  25 2
10. The polynomial P1x 2  2x 6  5x 5  8x 4 
x 3 2 1 0 1 2 3 4 5 21x 3  12x 2  22x  12 can be factored as P1x 2 
y 145 16 27 11 24 16 41 9 78 1x  2 2 1x  1 2 12x  1 2 1x  3 2 1x 2  2x  2 2.
a. What is the degree of the polynomial?
5. Match each polynomial expression a–f with its b. What are the real roots? The complex roots?
graph (i)–(vi). Use your knowledge about roots; do c. What happens as x S ? As x S ?
not use your function grapher. d. What is the maximum number of turning points
you expect? Explain.
e. What is the maximum number of points of in-
flection? Explain.
11. Determine cubic polynomials that represent the ac-
companying graphs.

a. b.
y y
(i) (ii)

2
1
x x
–3 1 3 –1 3

tangent

(iii) (iv)
12. Each graph represents a function. For each one,
(i) read off approximate intervals over which the
function is increasing and over which it is de-
creasing; (ii) estimate intervals over which the
function is concave up and concave down; and
(iii) find a possible formula for the function.

a. b.
(v) (vi) y y

a. f 1x2  1x  12 1x  3 2 1x  32
b. f 1x2  1x  12 1x  2 2 12  x2
c. f 1x2  1x  12 1x 2  42
f 1x2  1x  12 1x  12 1x  32 1x  3 2
x x
–4 1 4 –2 0 2 4
d.
e. f 1x2  3x 3  x 4
f. f 1x2  1x  22 1x  4 2 1x  3 2 2
Gord.3896.04.pgs 4/24/03 9:57 AM Page 271

4.2 The Behavior of Polynomial Functions 271

13. For each polynomial, (a) determine the number of b. f 1x 2  20x 4  3x 3  x 2  1000.
real roots and the number of complex roots; and c. f 1x 2  3x 4  20x 3  5x 2  x  20.
(b) find all real roots correct to three decimal places. d. f 1x2  x 4  x 5.
i. P1x2  x 4  8x 2  9 e. f 1x 2  4x 4  5x 5  6x 6.
ii. P1x2  x 5  4x 4  6x 3  6x 2  27x  27 17. Find the equation of a quadratic polynomial that
iii. P1x2  x 6  4x 5  6x 4  16x 3  has a real root at x  2 and a turning point at
11x 2  12x  6 11, 5 2.
iv. P1x2  x 6  9x 5  26x 4  41x 3  71x 2  18. A cubic polynomial P has turning points at 11, 42
and 15, 12 2 .
42x  6

a. What is the behavior of P1x 2 as x S ?

14. Determine which of the graphs suggest the end be-
havior for each polynomial.
b. Where is the point of inflection? (Hint: Recall that
a cubic is symmetric about its point of inflection.)
19. Suppose that a quadratic polynomial has roots at
x  6 and x  2.
a. Write a possible formula for the quadratic
function.
(i) (ii)
b. Use the fact that a quadratic is symmetric about
the vertical line through its turning point to de-
termine the x-coordinate of the turning point of
this quadratic function.
c. Suppose that the quadratic has a maximum
value of 20. What must be its equation?
(iii) (iv)
d. Suppose that the quadratic has a minimum value
a. y  5x 5  8x 4  2x 2  3x  4 of 20 instead. What must be its equation?
b. y  4x 6  3x 4  7x 3  8x 2  4x 20. An apple is tossed from ground level straight up at
c. y  3x 8  4x 5  6x 3  5x 2  6 time t  0 with velocity 64 ft>sec. Its height at time
d. y  x 7  4x 6  3x 4  6x 3  7x  9 t is f 1t2  16t 2  64t. Find the time when it hits
e. y  4x 9  6x 6  5x 3  35 the ground and the instant when it reaches its high-
f. y  100  x 4 g. y  19  6x 2 2 2 est point. What is the maximum height?
h. y  19  6x 2 3 2
i. y  19  6x 2 2 3 21. The height s (in cm) of an object above the ground
j. y  19  6x 2 3 3
at time t (in seconds) is given by
15. a. The graph of the polynomial P1x 2  2x 4 
1
6x 3  7x  10 in Figure 4.22 suggests that there s  v0t  gt 2,
are three turning points. Use your function gra- 2
pher to locate them to 3 decimal place accuracy where v0 represents the initial velocity and g is a
by zooming in on the graph. constant, the acceleration due to gravity.
b. Estimate all intervals over which P1x 2 is increas- a. At what height does the object start?
ing or decreasing. b. How long is the object in the air before it hits the
c. Estimate the locations of all points of inflection.
Estimate all intervals over which P1x 2 is concave
ground?
d. c. When will the object reach its maximum height?
up or concave down. d. What is that maximum height?
e. Estimate all real roots.
22. a. Sketch a smooth graph of today’s air tempera-
16. Describe the end behavior of each function. Specif- ture from midnight to midnight.
ically for the graph of each function f, (i) as x S , b. When is it a minimum? A maximum?
does f 1x2 S  or ? why? and (ii) as x S , c. When does it have a point of inflection?
does f 1x2 S  or ? why? d. What type of polynomial might be a good match
a. f 1x2  3x 3  70x 2  20. to the curve you drew?
Gord.3896.04.pgs 4/24/03 9:57 AM Page 272

272 CHAPTER 4 Extended Families of Functions

e. What function would be a better choice if you y

expand the domain to include the temperatures y = f (x)

for yesterday and tomorrow?
f (b) – f (a)
23. Factor the polynomial P1x2  x 3  5x 2  3x  7, m=
b–a
f (b)

using zeros that are correct to two decimal places.

∆y = f (b) – f (a)
24. a. Prove that any cubic polynomial of the form
f 1x2  ax 3  cx is symmetric about its inflection f (a)
point at the origin by showing algebraically that, ∆x = b – a
if you take any value of x—say, x  h 0 —
then f 1h2  f 1h2. x
a b

y
x 2 1 0 1 2 3

y  f (x)  x 3  4x 0 3 0 3 0 15

between x  2 and x  1, between x  1

d f (h) – d and x  0, and so on. What is the average value
d – f (–h) of all these slopes?
c. Extend the table to include the point x  4 and
x repeat parts (a) and (b). Does the same result
x = –h x=h hold?
d. Extend the table farther to include x  3.
Show that the same result holds.
e. Does the same result hold for any function and
b. Prove that any cubic polynomial of the form
f 1x2  ax 3  cx  d is symmetric about its
any set of points? State this result as a potential
inflection point at 10, d2 by showing alge-
theorem.
braically that, if you take any value of x—say, 26. Prove the result you conjectured in part (e) of
x  h 0—then f 1h2  d  d  f 1h2 , as Problem 25. Let f be defined on an interval from a
illustrated in the accompanying figure. (Note: to b. The average rate of change of f is
The same ideas apply to an arbitrary cubic f 1b 2  f 1a2
polynomial when the bx 2 term is present, but .
ba
the proof is considerably more complicated.)
Let
25. Recall that the average rate of change of a function f
over an interval x  a to x  b (see Section 2.8) is x0  a, x1 , x2 , . . . , xn  b
defined as the slope of the line segment connecting be any set of uniformly spaced points so that
the endpoints of the curve on that interval, or ba
f 1b2  f 1a2

x  .

y n
 ,

x ba 27. Find all polynomials p of degree  2 that satisfy
as illustrated in the accompanying figure. The table each set of conditions.
gives some values for the function f 1x 2  x 2  4x. a. p10 2  p11 2  p122  1
a. Find the average rate of change of f from b. p10 2  p11 2  1 and p12 2  2
x  2 to x  3. c. p10 2  p11 2  1
b. Calculate the average rate of change of f between d. p10 2  p11 2
each successive pair of points in the table; that is, (Hint: Think about the graphs.)
Gord.3896.04.pgs 11/21/03 10:35 AM Page 273

4.3 Modeling with Polynomial Functions 273

Exercising Your Algebra Skills

Factor each of the following polynomial expressions as 7. x2  6x  9 8. x 2  25
completely as possible. (Note that not all are factorable.) 9. x2  100 10. x 2  36
1. x 2  7x  12 2. x 2  4x  5 11. x3  x 2  20x 12. x 3  4x 2  3x
3. x 2  7x  12 4. x 2  x  12 13. x3  10x 2  25x 14. x 3  36x
5. x 2  x  12 6. x 2  4x  4

4.3 Modeling with Polynomial Functions

As we mentioned in Example 4 of Section 3.2, one of the most famous moments in
the history of science was Galileo’s reported experiment of dropping various ob-
jects from the top of the 180-foot high Leaning Tower of Pisa and discovering that
they fell at the same rate, regardless of their weight. Instead of looking at the speed
of a falling object, we now look at the height H, in feet, of an object falling from the
top of the tower at various times t, as given in the table.

Time 0 0.5 1.0 1.5 2.0 2.5 3.0

Height 180 176 164 144 116 80 36

Note how the object starts falling slowly and then accelerates. (Incidentally, these
values are considerably more accurate than anything Galileo could have measured
at the end of the fourteenth century.)
The ideas we introduced in Chapter 3 on fitting linear, exponential, power, and
logarithmic functions to a set of data can be extended to fitting polynomial func-
tions to data. All graphing calculators have the capability to fit quadratic, cubic,
and quartic polynomials to any set of data; spreadsheets such as Excel™ can fit
polynomials up to degree 6, and specialized software packages allow polynomials
of any finite degree. However, the approach used to determine a best-fit polynomi-
al is different from the types of transformations we used in Sections 3.4 and 3.5. In
fact, it is based on the idea of fitting a linear function of several variables to a set of
data, as we discussed in Section 3.7. As we also discussed there, the correlation co-
efficient does not apply directly. Instead, statisticians have developed a comparable
measure of the goodness of fit, known as the coefficient of determination, which is
denoted by R 2. Its value is provided by most calculators and software. It always lies
between 0 and 1, and the closer R 2 is to 1, the better is the fit; a value of 1 indicates
a perfect fit.

E XAMPLE 1
(a) Find an equation for the height of an object falling from the top of the 180-foot high
Leaning Tower of Pisa as a function of time. (b) Then use the formula to calculate how
long it takes for the object to hit the ground.
Solution

a. We show the scatterplot of the data for height H as a function of time t in Figure 4.27
and observe that the pattern in the data resembles the right half of a parabola with
Gord.3896.04.pgs 4/24/03 9:57 AM Page 274

274 CHAPTER 4 Extended Families of Functions

H H

180 180
150 150

Height (feet)

Height (feet)
120 120
90 90
60 60
30 30
t t
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Time (seconds) Time (seconds)

FIGURE 4.27 FIGURE 4.28

negative leading coefficient. Using the quadratic function regression routine on a cal-
culator, we obtain the equation
H  16t 2  180.
The corresponding value for the coefficient of determination is R 2  1.00, suggest-
ing that the parabola apparently is a perfect fit to the data, as shown in Figure 4.28.
b. The object hits the ground when H  0. To find how long it takes, we must find the
value of t for which
H  16t 2  180  0.
We can solve this quadratic equation graphically, with the quadratic formula, or by
direct algebraic means. Algebraically, we add 16t 2 to both sides of this equation to
obtain
16t 2  180
so that
180
t2   11.25.
16
When we take the square root of both sides, we get t  3.35. Because t  3.35 sec-
onds makes no real-world sense, we conclude that it takes about 3.35 seconds for the
object to hit the ground.
◆
Let’s look at the equation H  16t 2  180 for the height at any time when
the object is falling from the top of the 180-foot high tower. Note that the constant
term 180 equals the height of the tower. We rewrite the function as
H1t2  180  16t 2,
which indicates that the height starts at 180 feet, when t  0, and decreases there-
after. In general, if an object is dropped from any initial height H0 and is affected
only by the force of gravity, its height at any time t is given by
H1t2  H0  16t 2.
Now suppose that an object is not simply dropped but instead is tossed up-
ward with some initial velocity—say, 40 ft>sec. What do we expect? Obviously, the
object starts off rising until it reaches a maximum height and then falls back until
it hits the ground. The larger the initial velocity, the higher the object goes. In Ex-
ample 2, we construct a function to model such a situation.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 275

4.3 Modeling with Polynomial Functions 275

E XAMPLE 2
When an object is thrown vertically upward with an initial velocity of 40 ft>sec from the
top of the 180-foot high Tower of Pisa, the following set of measurements of its height as
a function of time are obtained.

t 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

H 180 196 204 204 196 180 156 124 84 36

a. Find an equation of a function that can be used to model the height of the object as a
function of time.
b. Estimate how long it takes for the object to reach its maximum height and what that
maximum height is.
c. How long does it take for the object to fall back to the ground?
Solution

a. The scatterplot of the data shown in Figure 4.29 indicates that the pattern for the
height H as a function of time t looks like a portion of a parabola with a negative
leading coefficient. Using a calculator to fit a quadratic function, we find that the
quadratic function that best fits the data is
H1t2  180  40t  16t 2.
Note that the coefficients of the constant and linear terms are essentially the same as
the initial height 180 feet and the initial velocity 40 feet per second, respectively.
Moreover, the coefficient of the quadratic term is the same, 16, as in Example 1.
This function superimposed over the scatterplot shown in Figure 4.30 reveals that it
is an excellent fit to the data. The associated coefficient of determination is R 2  1.0,
providing additional evidence that the fit is virtually perfect.

H H

210 210
180 180
Height (feet)

Height (feet)

150 150
120 120
90 90
60 60
30 30
t t
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (seconds) Time (seconds)

FIGURE 4.29 FIGURE 4.30

b. To estimate the time it takes for the object to reach its maximum height and the value
for that maximum height, we need merely trace along the curve to find the coordi-
nates of the turning point; or we can use the routine for locating the maximum for a
function that is on many calculators. Either way, the coordinates are t  1.25 seconds
and H  205 feet.
c. To find the time it takes for the object to return to the ground, we solve the equation
H1t2  180  40t  16t 2  0.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 276

276 CHAPTER 4 Extended Families of Functions

We can do this either graphically or by the quadratic formula. Using the quadratic
formula with a  16, b  40, and c  180 gives
40  2402  4116 2 1180 2
t
2116 2
40  21600  11520

32
40  213120
 .
32
Consequently, we get two possible values for t: t  4.83 seconds and t  2.33 seconds.
The second value makes no sense physically, so the realistic solution is t  4.83 seconds.
◆
In general, we can say the following.

The height of an object thrown vertically upward with initial velocity v0 from
an initial height H0 at any time t is
H1t2  16t2  v0t  H0 .

If there is no initial velocity, so that v0  0, this formula reduces to the expression

we had previously for the height of any object falling under the influence of gravity.
The questions that we would want to answer about any object thrown upward
into the air are:
1. How high does it go?
2. How long does it take to reach its maximum height?
3. How long does it take to return to the ground?

The Path of a Projectile

Picture the path of a long home run in baseball or the path of a perfect pass in foot-
ball or the arch of the high-pressure stream of water from a supershooter water
gun. In each case, the path looks something like the curve shown in Figure 4.31,
whose shape suggests a parabola or possibly some higher degree polynomial curve
with a negative leading coefficient. (If a strong wind is blowing, the path may not
be quite so symmetric and the analysis of the shape of the path is considerably
more complicated than that described here.)
y

x
FIGURE 4.31
Gord.3896.04.pgs 4/24/03 9:57 AM Page 277

4.3 Modeling with Polynomial Functions 277

Using various kinds of technology, such as time-lapse photography or a video

camera, we can capture a set of data on the path of such a projectile. For instance,
the following set of data consists of measurements for the path of a long fly ball in
baseball, where the height y of the ball depends on the distance x from home plate.
Both sets of measurements are in feet.

x 0 30 60 90 120 150 180 210 240 270 300 330 360 390

y 4 37 65 88 105 117 123 124 119 109 94 73 47 16

The ball rises to a maximum height of about 124 feet. More important, the ball
travels a horizontal distance of about 400 feet until it comes back down into the out-
fielder’s glove, hits the ground or fence, or lands in the stands. To determine what
happens, we need an equation for the path of the ball, which we find in Example 3.

E XAMPLE 3
(a) Determine the equation of a function that models the path of the baseball based on
the preceding data. (b) If the fence 400 feet from home plate is 8 feet high, will the ball
clear the fence to be a home run?
Solution

a. Because the shape of the data, as shown in the scatterplot in Figure 4.32, suggests a
parabola, we begin by fitting a quadratic function to the data. The result is the quad-
ratic function
y  0.003x2  1.202x  3.936,

125
100
Height (feet)

75
50
25
x
0 100 200 300 400
FIGURE 4.32 Distance (feet)

which is shown superimposed over the scatterplot and is an outstanding fit to the
data. As expected, the leading coefficient is negative. Moreover, the corresponding
value for the coefficient of determination is R 2  0.9999, which provides additional
evidence that the quadratic function is an excellent model to use.
b. In order for the ball to be a home run, it must clear the 8-foot high fence when it is
400 feet from home plate. Therefore we substitute x  400 into the equation of the
parabola and find that
y  0.0031400 2 2  1.2021400 2  3.936  4.736.
That is, when it reaches the fence, the ball’s height is somewhat less than 5 feet, so it
wouldn’t be a home run, as shown in the smaller view in Figure 4.33.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 278

278 CHAPTER 4 Extended Families of Functions

H=8
y = 4.736

FIGURE 4.33 x = 400

Fitting Polynomials to Data

The concept of fitting a polynomial function to data is one that applies in all walks
of life, not just in the physical situations we encountered in Examples 1–3. We il-
lustrate two other cases in Examples 4 and 5.

E XAMPLE 4
The table shows the accumulated total number of reported cases of AIDS in the United
States since 1983.

Year 1983 1984 1985 1986 1987 1988 1989 1990

Number of AIDS Cases 4589 10,750 22,399 41,256 69,592 104,644 146,574 193,878

Year 1991 1992 1993 1994 1995 1996 1997 1998

Number of AIDS Cases 251,638 326,648 399,613 457,280 528,144 594,641 653,084 701,353
Source: U.S. Centers for Disease Control and Prevention.

Determine a function that fits the data well and interpret the behavior of the function.
Solution In Example 4 of Section 3.3, we explored the possibility that the growth in the
total number of reported cases of AIDS in the United States follows an exponential pat-
tern. The resulting best-fit exponential function, found with a calculator, was
A  5413.511.3626 2 t,
where t is measured in years since 1980. The corresponding correlation coefficient
r  0.9483 is quite close to 1, suggesting that this function is a very good fit. But, when we
superimpose this exponential function over the data points, as shown in Figure 4.34, the
curve doesn’t fit the data well.

750
Number of AIDS cases

Exponential function
600
(thousands)

450
300
150
t
0 1983 1986 1989 1992 1995 1998
FIGURE 4.34 Year
Gord.3896.04.pgs 4/24/03 9:57 AM Page 279

4.3 Modeling with Polynomial Functions 279

Alternatively, suppose that we use the capability of the calculator to fit a polynomial
to this data. Most calculators allow us to fit polynomials of degree 2, 3, or 4 to a set of
data, and we can easily experiment with different degrees. When we do so, we find that a
cubic polynomial is an excellent fit to this set of data. The calculator gives the best cubic
function, rounded to one decimal place, as
A  221.9t 3  9261.8t 2  62275.9t  122988.9,
where t is again the number of years since 1980. When we superimpose this polynomial
over the AIDS data points shown in Figure 4.35, we get an exceptionally good fit, which
certainly is a far better fit than the exponential function shown in Figure 4.34.

750 Cubic polynomial

Number of AIDS cases 600

(thousands)
450
300
150
t
0 1983 1986 1989 1992 1995 1998
FIGURE 4.35 Year

This graph strongly suggests that the number of cases in the spread of AIDS fol-
lows a cubic pattern. (When scientists discovered this several years ago, they were ex-
cited because polynomial growth is much slower than exponential growth, which is
the trend that they too had expected.) The corresponding coefficient of determination,
R2  0.99996, provides further evidence of how well the cubic function fits the data.
We know from the formula for the cubic that the leading coefficient is negative, so
the cubic will eventually approach . The larger view in Figure 4.36 suggests that the
cubic passed its inflection point in about 1995 or 1996 and that the growth in AIDS has
begun to slow somewhat since then. The graph also shows that the function will reach a
turning point in about 2003.

A
Number of AIDS cases

1200
(thousands)

800

400

t
0 1973 1979 1985 1991 1997 2003
FIGURE 4.36 Year

◆
However, recall that the data represent the total number of AIDS cases reported in
the United States, so the cubic can’t actually turn and begin to decline; it can only slow
and, at best, eventually level off. Thus we demonstrate again how dangerous extrapo-
lation with a mathematical model can be. The model only describes the situation
based on the data points; it is not a guarantee of the actual process, especially for ex-
trapolating into the future or the past.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 280

280 CHAPTER 4 Extended Families of Functions

Let’s look at another example of fitting polynomials to data. Figure 4.37 shows
a picture of the famous Gateway Arch in St. Louis. Its shape suggests a portion of a
downward opening parabola. Let’s see if we can determine a specific function that
best models the arch.

FIGURE 4.37

E XAMPLE 5
Determine a polynomial function that fits the Gateway Arch well.
Solution To find an appropriate function, we need some measurements for the arch.
Overall, the arch stands 630 feet tall, and the distance between its two legs also is 630 feet.
We superimpose a grid on the arch, as shown in Figure 4.38, and choose the coordinate
system so that the vertical axis passes through the center of the arch. We then construct
the following table of estimates of the height H corresponding to various horizontal dis-
tances x. We make our estimates from the middle of the arch; slightly different results
might occur if we use values from the inner edge or the outer edge. We ask you to inves-
tigate these possibilities in the problems at the end of this section.
H

650

550

450
Height (feet)

350

250

150

50
x
–300 –200 –100 0 50 100 200 300
FIGURE 4.38 Horizontal distance (feet)

x 325 300 250 200 150 100 0 100 150 200 250 300 325

H 0 100 330 500 570 610 630 610 570 500 330 100 0
Gord.3896.04.pgs 4/24/03 9:57 AM Page 281

4.3 Modeling with Polynomial Functions 281

The first thing we notice from both the figure and the table is that the measurements
are symmetric about the vertical axis x  0. As a result, we would expect that the best-fit
parabola has no x term. When we enter the data into the quadratic regression routine of
a calculator, we find the quadratic function that best fits the data is
H  0.0064x 2  0x  699.01.
We plot this function over the data points, as illustrated in Figure 4.39, and conclude
that it is a reasonably good fit, though certainly not a great one. Among other things, the
curve rises much too high above the central data point and the pattern of data points
flattens out far more than the parabola does near the center.
H

700
Best quadratic
fit to the Arch
600
500
400
300
200
100
x
–350 –250 –150 –50 0 50 150 250 350
FIGURE 4.39 Horizontal distance (feet)

In our discussion of power functions with integer powers in Section 2.7, we pointed
out that the higher the power, the flatter the curve as it passes through the origin. This
result suggests that we should use a higher degree polynomial than a quadratic. From the
basic shape of the arch, we know that a cubic would not be appropriate—it doesn’t have
the correct behavior. How about a quartic polynomial? When we try it, the calculator re-
sponds with the equation
H  13.27  108 2x 4  0x 3  0.00282x 2  0x  644.25.
When we superimpose this function over the data points, as shown in Figure 4.40, it ap-
pears visually to be an exceptionally good fit to the shape of the arch. The coefficient of
determination for this fit is R 2  0.9953, which also indicates that it is a very good fit.
(Actually, the true shape of the arch is a curve known as a hyperbolic cosine, which you
may encounter in calculus.)
H

700 Best quartic fit

to the Arch
600
500
400
300
200
100
x
–350 –250 –150 –50 0 50 150 250 350
FIGURE 4.40 Horizontal distance (feet)

◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 282

282 CHAPTER 4 Extended Families of Functions

Problems
1. We showed in the text that the cubic function
y  221.9t 3  9261.8t 2  62275.9t  122988.9
is an excellent fit to the total number of reported
cases of AIDS in the United States from 1983 to
1998, where t is the number of years since 1980.
a. Based on this model, what is the prediction for
the total number of cases through 2000?
b. Check a recent copy of the Statistical Abstract of
the United States or an almanac to see how accu-
rate the prediction in part (a) is.
c. If this cubic pattern continues, how many total
cases would you expect by 2004?
d. When would you expect a total of 850,000 cases
of AIDS, based on this model?
2. Find the equations of the best quadratic and quar-
tic functions to fit measurements taken at the outer
edge of the Gateway Arch instead of at the middle.
3. Repeat Problem 2 with measurements taken at the 6. The table gives the horsepower generated on a
inner edge of the arch instead of at the middle. Chevy 383 car engine at different rpm.
4. The table shows the percentage of the U.S. popula-
tion that is foreign born in various years. Horsepower 138 172 203 216
a. What is the minimum degree polynomial that
Revolutions
you would use to model this data?
per Minute 2000 2500 3000 3500
b. Find that polynomial and use it to estimate the
time when the percentage of foreign-born peo- Horsepower 209 182 144 98 42
ple in the United States was a minimum. What
was that minimum percentage? Revolutions
per Minute 4000 4500 5000 5500 6000
Source: Student project.
Year 1950 1960 1970 1980 1990 2000
Percentage 6.9 5.4 4.8 6.2 7.9 10.4 a. Which variable is the independent variable and
which is the dependent variable?
Source: 2000 Statistical Abstract of the United States.
b. What is the equation of the quadratic function
that relates these two quantities?
5. The accompanying figure shows a grid superim- c. What does your model predict for the horse-
posed on the image of the McDonald’s arches. power generated by this engine at 4800 rpm?
a. Decide on a scale that you can use to estimate d. If the engine puts out 165 horsepower, what is
measurements on the arches. (Hint: Think about the possible value for the rpm according to this
where you want to set up your coordinate axes.) model?
b. Use your estimated measurements to determine 7. Car enthusiasts know that it’s not horsepower that
the equation of a polynomial that best fits one of is significant, but rather the amount of torque that
the arches. (Hint: Think again about where you an engine puts out that really matters in how quick-
want to set up your coordinate axes.) ly a car moves forward. The table gives the torque,
c. Can you use the formula you obtained for one in foot-pounds, generated at different rpm values
of the arches to construct a formula for the for a Chevy 383 engine. From among the usual
other arch? Explain. families of functions (linear, exponential, power,
Gord.3896.04.pgs 4/24/03 9:57 AM Page 283

4.3 Modeling with Polynomial Functions 283

quadratic, and cubic), find the one that seems to be 11. While approaching the Verrazano Bridge in New
the best fit to these data. York City, Ken noticed that the main cable looks
like a parabola, as illustrated in the accompanying
Torque 363 361 355 324 figure. As his car crawled across the bridge in heavy
traffic, he estimated the following heights, in feet, of
Revolutions the cable above the road and the distance, in feet,
per Minute 2000 2500 3000 3500 starting from one of the vertical support columns.
Torque 275 213 151 93 36

Revolutions
per Minute 4000 4500 5000 5500 6000
Source: Student project.

8. a. Create a single table based on the information

given in Problems 6 and 7 relating the amount of
Distance from
torque generated to the horsepower for the
Support Column 0 1000 2150
Chevy 383 engine.
b. From among the usual families of functions (lin- Estimated Height 500 150 20
ear, exponential, power, quadratic, and cubic), find
the one that seems to be the best fit to this data. Distance from
Support Column 3000 4000 4300
9. The table shows the number of 18- to 24-year-olds
in the United States in recent years. Find the quad- Estimated Height 100 400 500
ratic function that best fits this data set and use it to
predict the number of people in this age range in Find an equation of the parabola that best fits Ken’s
(a) 2000 and (b) 2005. Which prediction would you estimates. (Think how to set up the coordinate axes.)
have more confidence in?
12. The table shows the price of a barrel of oil, in dol-
lars, in different years.
Year 1970 1975 1980

Population (millions) 24.71 28.76 30.35

Year 1960 1970 1975 1980
Year 1985 1990 1995
Price 11 9 37 64
Population (millions) 29.48 26.14 24.85
Year 1985 1990 1995 2000
Source: 2000 Statistical Abstract of the United States.
Price 40 28 20 32
10. According to the theory of relativity, the mass M of
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
an object increases as its velocity v increases so that
M  f 1v2. Suppose that the mass of an object is 1 unit
That Are Shaping Our Future.

when it is at rest 1v  02. The table gives the mass a. What type of function is reasonable to use as a
of the object at different speeds that are expressed model for the price of oil as a function of time?
as fractions of c, the speed of light (about 186,280 b. Find the equation of the polynomial function of
miles per second). Find the best quadratic fit to this appropriate degree to fit the data.
set of data. c. What does your model predict for the price of a
barrel of oil in 2005?
Velocity (fraction of c) 0 0.1 0.2 d. Use the graph of your function to estimate the lo-
cation of the turning points for the function. Ac-
Mass 1 1.0050 1.0206 cording to this model, what was the maximum
Velocity (fraction of c) 0.3 0.4 0.5 price of a barrel of oil between 1960 and 2000
and when did it occur? What was the minimum
Mass 1.0483 1.0911 1.1547 price and when did it occur?
Gord.3896.04.pgs 4/24/03 9:57 AM Page 284

284 CHAPTER 4 Extended Families of Functions

13. The table shows the trend in worldwide grain pro- b. Let t be the number of years since 1890. Deter-
duction (wheat, rice, and corn, primarily), in kilo- mine the best polynomial function of the degree
grams per person. The pattern in the data suggests that you decided was appropriate in part (a) to
that a quadratic function is an appropriate model for model the number of high school graduates as a
grain production per person as a function of the year. function of time t.
c. Use this function to predict the number of high
school graduates in 2010.
Year 1965 1970 1975 1980
d. Use this function to predict the year in which
Amount there will be 5 million high school graduates.
per Person 270 291 303 321 15. The table, collected from a chemistry lab experi-
Year 1985 1990 1995 1999 ment, gives the density D of water, in grams per
milliliter, at various temperatures T, in °C.
Amount
per Person 339 335 301 309
Temperature, T 0° 4° 10°
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends
That Are Shaping Our Future. Density, D 0.99987 1.00000 0.99973

a. Find the equation of the quadratic that best fits Temperature, T 20° 30° 40°
these data.
Density, D 0.99823 0.99567 0.99224
b. Based on the model, what was the maximum
level of grain production per person worldwide? Temperature, T 60° 80° 100°
c. What does the model predict for the amount of
Density, D 0.98324 0.97183 0.95838
grain produced per person in 2010?
d. Write a paragraph describing the possible rea- Source: John R. Holum, Elements of General and Biological Chemistry,
8th ed. New York: John Wiley & Sons, 1991.
sons for this trend and the implications if the
trend continues.
a. Find a quadratic function that fits these data.
14. The table gives the total number, in thousands, of b. Use your function from part (a) to find the den-
high school graduates in the indicated years since sity of water at 70°C.
1900. In Problem 12 of Section 3.3, we asked you to c. Find the temperature at which the density of
find the best linear, exponential, and power func- water is 0.99100 grams per milliliter.
tions to fit these data. If you examine the data care-
fully, you should expect that a polynomial function 16. The height of an object falling from an initial height
would be a better fit. of y0 is given by the formula
y  y0  16t 2,
with units of feet and seconds. What is the equivalent
Year 1900 1910 1920
formula based on the metric system of units with me-
High School Grads 95 156 311 ters and seconds? (Hint: 1 foot  0.3048 meters.)
17. Galileo conducted his famous experiment in
Year 1930 1940 1950 1960
which he dropped objects from the top of the
High School Grads 667 1221 1200 1858 180-foot high Leaning Tower of Pisa in about
1590. His goal was to obtain experimental data to
Year 1970 1980 1990 2000
show that all bodies fall with equal velocities.
High School Grads 2889 3043 2586 2839 How long did it take for the objects that he
Source: Digest of Education Statistics 2000, U.S. Department of Education.
dropped from the tower to hit the ground?
18. The Eiffel Tower is 300 meters tall. How long would
a. What degree polynomial function is a good can- it take an object dropped from its top to hit the
didate to fit these values? Explain. ground?
Gord.3896.04.pgs 4/24/03 9:57 AM Page 285

4.4 The Roots of Polynomial Equations: Real or Complex? 285

4.4 The Roots of Polynomial Equations: Real or Complex?

The Roots of Quadratics
In Section 4.1, we stated that, for any quadratic equation,
ax 2  bx  c  0, a  0,
we can always find its roots by using the quadratic formula
b  2b2  4ac
x .
2a
Further, the roots may be two distinct real numbers, a repeated real root, or a pair of
complex conjugate numbers of the form a  bi, where i  21 , as illustrated in
Figure 4.41.
y y y

x
x
x

tangent

y y y
tangent

x x

FIGURE 4.41

Most students think that complex roots occur very rarely. In this section we in-
vestigate how frequently they do arise. To do so, we consider many different quad-
ratic equations and find the percentage of them that do have complex roots. A
quadratic equation ax 2  bx  c  0 has complex roots when its discriminant,
b2  4ac, is negative. The quadratic formula then requires taking the square root of
that negative discriminant to produce two complex numbers. For instance, for the
quadratic equation x 2  2x  2  0, the discriminant is 12 2 2  411 2 12 2  4,
so the roots will be complex. The quadratic formula gives the roots as
12 2  24  8 2  24 2  2i
x    1  i,
2 2 2
or x  1  i and x  1  i. Thus we can use the sign of the discriminant as the
criterion to decide whether any particular quadratic has complex roots.
To come to any meaningful conclusions about the percentage of quadratics that
have complex roots, we must examine a very large number of quadratics. Doing so
requires using a computer or calculator program rather than hand computation.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 286

286 CHAPTER 4 Extended Families of Functions

Even the simplest case—when the quadratic has integer coefficients—has infinitely
many possible quadratics, so the best we can do is examine a finite selection of
them. Let’s examine all possible quadratics y  ax 2  bx  c where the coeffi-
cients a, b, and c are integers from 0 to 5, say, but a  0. We write this in interval no-
tation as 30, 5 4. We then use a computer program that considers all possible integer
values for a, b, and c in this interval and keeps track of how many of the quadratics
have complex roots, using the discriminant criterion. Similarly, we can investigate
all possible integer coefficients in various other intervals, the results of which are
shown in Table 4.1.

TABLE 4.1

Interval for a, b, Percentage with

and c, a  0 Complex Roots
All in [0, 5] 70
All in [0, 10] 73
All in [0, 20] 74
All in [0, 50] 74.5
All in 33, 34 37.4
All in 35, 54 37.5
All in 310, 104 37.8
All in 320, 204 37.7
All in 350, 504 37.5
[0, 5], [0, 5], 35, 04 0
[0, 5], 35, 0 4, [0, 5] 70

Therefore, rather than being a rarity, complex roots actually occur with sur-
prising frequency. In fact, almost three-fourths of quadratics whose coefficients are
all nonnegative integers have complex roots. Even allowing for negative values al-
most 40% have complex roots.

Think About This There is one exception in Table 4.1. If the constant coefficient c is negative while a
and b are both positive, the quadratic apparently always has two real roots. Can you
explain why? Can you give another example where the quadratic always has two
real roots? Look at the discriminant. (Note that we have checked only specific inte-
ger values for a and b between 0 and 5 and c between 5 and 0, so we can’t gener-
alize to what may happen over all similar intervals of values.) ❐

We suggest that you conduct your own investigations of these ideas if an ap-
propriate program is available or if you want to write a fairly short program for
your calculator. Think about the following questions.
◆ With integer coefficients, what happens as the size of the interval increases?
Does the frequency of complex roots stay roughly the same or does it in-
crease or decrease significantly?
◆ What happens if you use different ranges of values for each coefficient?

Don’t be too generous in your choices when you begin; such systematic
processes tend to take a long time. For example, if you want to check all quadrat-
ics where a, b, and c are integers between 0 and 10, say, you are actually having the
computer or calculator investigate 1210 different equations. (There are 10 possible
Gord.3896.04.pgs 4/24/03 9:57 AM Page 287

4.4 The Roots of Polynomial Equations: Real or Complex? 287

values for a since the equation would not be quadratic if a were zero. There are 11
possible values for b and 11 for c, which leads to 10  11  11  1210 different
cases.) If you ask for all integers from 0 to 100 on each of the coefficients, the com-
puter or calculator will investigate 100  1012 different quadratics. It may take all
night to complete this study of more than one million cases.
We should also find out what happens when the quadratic has noninteger
coefficients, either rational numbers or irrational numbers. In such cases, we
can’t simply check all possible equations because there are infinitely many pos-
sibilities, even for any finite interval. Instead, we use a random selection process
to generate large numbers of quadratics with randomly selected (noninteger)
coefficients in desired intervals, test each for the nature of its roots, and keep
track of how many of the roots are complex. (We perform just such an analysis
in Supplementary Section 11.3 as part of our study of probability.)

The Roots of a Cubic Function

We next consider an arbitrary cubic equation
ax 3  bx 2  cx  d  0,
where a, b, c, and d are any four real numbers and a  0. Recall that, just as a
quadratic equation has two roots, a cubic equation has three. They can be either
real or complex roots. Recall also that any complex roots must occur as a pair of
complex conjugates, a  bi and a  bi. Thus, for any cubic equation, the three
roots may be either three real numbers, or a single real number and a pair of com-
plex conjugate numbers.
Moreover, we know that the real roots correspond geometrically to points
where the cubic crosses the x-axis. If there are three distinct real roots, the cubic
crosses the x-axis in three places, as illustrated in Figures 4.42(a) and 4.42(b). If
there is a double real root and a separate real root, the x-axis is tangent to the cubic
at the point corresponding to the double root and the curve crosses the x-axis at

y y y y

x x x x

(a) (b) (c) (d)

y y y y

x x x x

FIGURE 4.42 (e) (f) (g) (h)

Gord.3896.04.pgs 4/24/03 9:57 AM Page 288

288 CHAPTER 4 Extended Families of Functions

the point corresponding to the other real root, as depicted in Figures 4.42(c) and
4.42(d). If there is a triple real root (as with y  x 3), the cubic flattens as it crosses
the x-axis at the single point, as shown in Figures 4.42(e) and 4.42(f). Finally, if
there is a single real root and a pair of complex conjugate roots, the cubic crosses
the x-axis once, as illustrated in Figures 4.42(g) and 4.42(h). Thus a cubic can have
either three real roots or one real root.
We have demonstrated that quadratic equations are likely to have complex
roots. How likely is it for a cubic equation to have complex roots? To answer this
question, we again use a computer program to investigate many different cu-
bics. First, though, we must devise a test comparable to using the sign of the dis-
criminant in the quadratic formula to decide whether a particular cubic has
complex roots.
Suppose that a cubic has three real roots. In that case, the curve crosses the
x-axis at three points if the three roots are distinct, it crosses the axis at two
points if there is a double real root, and it crosses the axis at one point if there is
a triple real root. The cubics shown in Figure 4.43 all have the same shape; the
only difference is the height of the turning points. The cubic on the left has its
first turning point above the x-axis and its second below; therefore it has three
real roots. The second cubic has both turning points above the x-axis and so
must have one real root and a pair of complex roots. The third cubic has both
turning points below the x-axis, so it also must have one real root and a pair of
complex roots.

y y y

x x

x
FIGURE 4.43

A further case occurs when the x-axis is tangent to the curve at one of the turn-
ing points; such a cubic has a double real root, so it cannot have a pair of complex
roots and its third root must be real. The final case is when the two turning points
coincide along the x-axis; this case corresponds to a triple real root. Therefore, in
order to have two complex roots, a cubic must have both turning points above the
x-axis or both below it.
When you study calculus, you will be able to determine that the two turning
points of the cubic y  ax3  bx2  cx  d  0 are located at
b  2b2  3ac
x ,
3a
provided that b2  3ac 0. (This formula clearly resembles the quadratic formula.)

Think About This Verify graphically that this formula gives the approximate location of the turning
points of the cubic y  x 3  4x 2  4x  5. ❐
Gord.3896.04.pgs 4/24/03 9:57 AM Page 289

4.4 The Roots of Polynomial Equations: Real or Complex? 289

Call these two x-values x1 and x2 . Because we know the equation of the cubic
curve,
y  f 1x 2  ax 3  bx 2  cx  d,
we can determine the heights of the two turning points:
y1  f 1x1 2 and y2  f 1x2 2 .
Once we have calculated these values, we need only check whether both are posi-
tive or both are negative to conclude that the cubic has complex roots, as illustrat-
ed in Figure 4.44. If the two y-values have opposite signs or if either is zero, the
cubic has three real roots. We use this criterion in our investigation.
We apply this criterion to cubics with integer coefficients a, b, c, and d with-
in various intervals of values. In Supplementary Section 11.3 we investigate cases
with randomly generated noninteger values for a, b, c, and d within any desired
intervals of values, provided that a  0.
In Table 4.2 we list the results of performing this investigation with all possible
integer coefficients in the indicated intervals of values. This table indicates that a
cubic with integer coefficients seems even more likely to have complex roots than a
quadratic does.

TABLE 4.2
y1 = f (x 1)
Intervals for a, b, c, and d, a  0 Percentage of Complex Roots
All in [0, 5] 94.54
All in 33, 34 78.43
All in 34, 44 78.74
y2 = f (x 2 )
All in 35, 54 78.93
[0, 4], [0, 4], [0, 4], 34, 04 88.4
[0, 4], [0, 4], 34, 04, [0, 4] 74.8
[0, 4], 34, 0 4, [0, 4], [0, 4]
x1 x2
88.4
FIGURE 4.44 [0, 4], [0, 4], 34, 04, 3 4, 04, 44
[0, 4], 34, 0 4, 34, 04, [0, 4] 44
[0, 4], 34, 0 4, 34, 04, 34, 04 74.8

Think About This In intervals of the form 3k, k4 for all four coefficients, the proportion of complex
roots seems to be essentially the same regardless of the value of k. Does that make
sense? Imagine what would happen if you have a particular cubic and multiply each
coefficient by 10, say. Wouldn’t you expect the same type of roots? In fact, wouldn’t
you expect the identical roots? ❐

Think About This When we studied the nature of the roots of quadratics, we saw that the two roots are
always real whenever c  0 and a 0. Are there any simple combinations of values
for the coefficients a, b, c, and d in a cubic that likewise guarantee real roots? (What
about d  0, c  0, and a 0?) ❐

It turns out that for polynomials of higher degree, the likelihood of complex
roots is even greater than for quadratics or cubics, but we won’t investigate these
cases.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 290

290 CHAPTER 4 Extended Families of Functions

Using Information on the Nature of the Roots

We next turn to an application for which knowing the nature of the roots of a poly-
nomial is crucial. Home thermostats and automobile cruise controls are examples
of control systems that engineers use to control a process. In such devices, when the
system deviates slightly from the specified level, it should return to that level auto-
matically—the temperature shuts off or the car stops accelerating. Such a system is
called stable. Often, control systems are described mathematically by a polynomial.
A control system is stable if
1. all the real roots are negative, and
2. all the complex roots have negative real parts.
A control system having any positive real roots or having complex roots whose real
parts are positive is unstable. That is, the system does not return to the specified
level when small changes are introduced.

E XAMPLE
A control system is described by the cubic polynomial P1s2  s 3  3s2  4s  2. Deter-
mine whether the system is stable or unstable.
Solution The graph of this cubic polynomial is shown in Figure 4.45. Its associated
cubic equation s 3  3s2  4s  2  0 has only one real root, so it must therefore have a
pair of complex conjugate roots. Moreover, it is evident that the real root is negative. If
we zoom in on the point where the curve crosses the s-axis, we find that the root appears
to be located near s  1. We can determine whether the root is s  1 exactly by
evaluating
P11 2  11 2 3  3112 2  411 2  2
 1  3  4  2  0,
which shows that the root is precisely s  1.
The problem we now face is to determine the complex roots. We know the real root
s  1, so the corresponding linear factor is 1s  12 . We can therefore factor the poly-
nomial by dividing it by 1s  1 2 , using the technique of long division for polynomials
from algebra:
s2  2s  2
1s  12 s3  3s2  4s  2
s3  s2
2s2  4s
2s2  2s
2s  2
2s  2
0
Thus, 1s2  2s  22 is the quadratic factor, so that the original cubic polynomial is
P1s2  s3  3s2  4s  2  1s  12 1s 2  2s  2 2.
We now apply the quadratic formula to find the complex roots of the quadratic factor:
2  222  4 . 1 . 2
s
2.1
Gord.3896.04.pgs 4/24/03 9:57 AM Page 291

4.4 The Roots of Polynomial Equations: Real or Complex? 291

2  24 2  2i
  .
2 2
The two complex roots are therefore s  1  i and s  1  i. Because the real parts
of both complex roots are negative, and the real root is negative also, the control system
is stable.

P(s)

100

75

50

25

s
–4 –3 –2 –1 1 2 3 4
–25

FIGURE 4.45 –50

◆
Problems
1. Consider all quadratics with a  1 and the coeffi- have four real roots? four complex roots? Ex-
cient of the linear term b  0 so that they take the plain your answers.
form y  x 2  c. What percentage of these quad- b. How would your answers change if e  0?
ratics should have two real roots?
5. For each of the following cubic equations, use your
2. Consider all quadratics of the form y  ax 2  bx function grapher to produce the graph and zoom in
with c  0. What percentage of them should have to estimate where the two turning points are locat-
real roots? ed. Then apply the formula
3. Show that, if each coefficient in the quadratic
y  ax 2  bx  c is multiplied by 10, the resulting b  2b2  3ac
x
discriminant is multiplied by 100. What would you 3a
expect to happen to the discriminant if each coeffi-
from the text (based on y  ax 3  bx 2  cx  d)
cient were multiplied by the same number k? How
to verify that the values given by the formula match
do the roots of the two quadratics compare?
the points you found graphically.
4. Consider all fourth degree polynomials of the form
a. y  x 3  4x 2  8x  3
y  ax 4  bx 3  cx 2  dx  e, b. y  x 3  7x 2  2x  6
where, for simplicity, you may consider a 0. c. y  5x 3  3x 2  6x  8
a. Based on the general graph shown without axes, d. y  4x 3  3x 2  5x  4
how likely do you think it is (roughly 10%, 25%, e. y  4x 3  3x 2  5x  4
50%, 75%, or 90%) for such a polynomial to 6. a. Determine the location of the turning points for
the cubic y  x 3  3x 2  2x  10. What are
the maximum and minimum values for this
function?
b. Use the fact that a cubic is symmetric about its
point of inflection to determine the location of
the point of inflection of the cubic in part (a).
7. If a different control system is described mathemati-
cally by each polynomial, determine whether it is sta-
ble or unstable.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 292

292 CHAPTER 4 Extended Families of Functions

a. P1s2  s2  6s  8 d. P1s2  s3  4s2  12s

b. P1s2  s2  5s  12 e. P1s2  s3  3s2  7s  5
c. P1s2  s2  5s  3

4.5 Finding Polynomial Patterns

In Section 2.2, we developed a criterion for determining whether a set of m points
1x1 , y1 2, 1x2 , y2 2, . . . , 1xm , ym 2 follows a linear pattern when the x-values are uni-
formly spaced.

A set of points lies on a line if the differences between successive y-values are
all equal when the x-values are uniformly spaced. The slope of that line

y
m

x
is the constant difference between successive y-values divided by the uniform
spacing between successive x-values.

We now consider the related problem of determining whether a set of points

x y y follows a quadratic, a cubic, or a higher degree polynomial pattern. Suppose that
0 1 we have the points 10, 1 2, 11, 22 , 12, 52, 13, 10 2, 14, 172, and 15, 26 2, which actually
1 lie on the parabola y  x 2  1. We construct the table at the left of differences of
1 2 the y-values. Obviously, the
y values are not constant. In fact, they clearly follow a
3
2 5 linear pattern because the differences between successive
y values (the differences
5 of the differences) are all constant. The differences of the differences,
1
y2, are
3 10
7 called the second differences and are written
2y. If we extend the previous table to
4 17 include the second differences of the y-values, as shown in the table below, we get a
9 constant value for all the second differences.
5 26 In general, we have the following criterion based on uniformly spaced x-values.

A set of points 1x1 , y1 2, 1x2 , y2 2 , . . . , 1xm , ym 2 lies on a quadratic y  ax 2 

bx  c if the second differences of the y-values are all constant when the x-values
are uniformly spaced.

x y y 2y
0 1
1
231
1 2 3
2
2 5 5
2
3 10 7
2
4 17 9
5 26
Gord.3896.04.pgs 4/24/03 9:57 AM Page 293

4.5 Finding Polynomial Patterns 293

In the problems at the end of this section we ask you to explore the significance of
this constant second difference.

E XAMPLE 1
Show that the points 10, 2 2 , 11, 02 , 12, 4 2, 13, 14 2, 14, 30 2, and 15, 522 lie on a parabola.
Then find the equation of the parabola by using regressions methods.
Solution We construct a table of second differences.

x y y 2y
0 2 2  0  2
1 0 6  4  122
4
2 4 6
10
3 14 6
16
4 30 6
22
5 52

Because the differences of the differences are constant, the points follow a quadratic pat-
tern of the form
y  ax 2  bx  c,
where the coefficients a, b, and c must be determined.
Thinking of the points as data values and using the curve fitting routines of a calcu-
lator, we find that the quadratic function that best fits the data is
y  3x 2  5x  2.
The corresponding coefficient of determination is R 2  1, which suggests a perfect fit to
the data. Figure 4.46 shows that the graph of this parabola apparently passes through all
six points. Test this result by substituting each value in the formula that we created.

60
50
40
30
20
10
x
FIGURE 4.46 0 1 2 3 4 5

Alternatively, we could find the equation of the quadratic function that fits
these points by using algebraic methods, as we demonstrate in Example 2.

E XAMPLE 2
Find the equation of the parabola that passes through the points 10, 2 2, 11, 0 2, 12, 4 2,
13, 14 2, 14, 30 2, and 15, 522, using algebraic methods.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 294

294 CHAPTER 4 Extended Families of Functions

Solution As in Example 1, we have to find the three coefficients a, b, and c in the equa-
tion of the quadratic function y  ax 2  bx  c. Substituting the coordinates from the
first point x  0 and y  2 gives
2  a # 10 2  b # 10 2  c,
so c  2 and therefore the equation of the parabola becomes y  ax 2  bx  2. Using
the second point 11, 0 2 , we get
0  a # 112 2  b # 11 2  2  a  b  2,
and so
a  b  2 (1)
Using the third point 12, 4 2, we get
4  a # 122 2  b # 12 2  2
 4a  2b  2,
or
4a  2b  2.
Dividing both sides of this equation by 2 yields
2a  b  1. (2)
Equations (1) and (2) are a system of two linear equations in two unknowns.
We can solve for a and b by using the usual algebraic methods. We subtract Equa-
tion (1) from Equation (2) to get
a  3.
Substituting this value into Equation (1) gives
3  b  2 or b  5.
So, as before, the desired quadratic is
y  3x 2  5x  2.
You can easily verify that the last three points satisfy this function.
Alternatively, we can solve this system of two equations in two unknowns by using
the matrix methods described briefly in Appendix C and also find that a  3 and
b  5. Thus the equation of the parabola again is y  3x 2  5x  2.
◆
We can extend these ideas to develop similar criteria for deciding when a set of
m points 1x1 , y1 2 , 1x2 , y2 2, . . . , 1xm , ym 2 follow a polynomial pattern of degree n for
any n. For instance, we have the following criterion for n  3.

A set of m points 1x1 , y1 2, 1x2 , y2 2, . . . , 1xm , ym 2 lies on a cubic y 

ax 3  bx 2  cx  d if the third differences (differences of the differences of
the differences) of the y-values are all constant when the x-values are uni-
formly spaced.

Think About This Show that the points 13, 17 2 , 12, 0 2, 11, 5 2, 10, 4 2, 11, 3 2, 12, 8 2 , and 13, 252
lie on a cubic polynomial by creating a difference table that extends to the third dif-
ferences. ❐
Gord.3896.04.pgs 4/24/03 9:57 AM Page 295

4.5 Finding Polynomial Patterns 295

Sums of Integers
We use the preceding ideas on differences and polynomial patterns to develop a
number of formulas involving sums of numbers that arise frequently in mathe-
matics. Among them are the sum of the first n integers
1  2  3  ...  n
and the sum of the squares of the first n integers
12  22  32  . . .  n2.
Let’s begin with the expression for the sum of the integers. We let Sn denote the
sum of the first n integers:
Sn  1  2  3  . . .  n.
For instance, S4  1  2  3  4  10. We want a formula for Sn for any value of
n. We derive it in two ways.
The first is a particularly simple way that involves a nice trick. If
Sn  1  2  3  . . .  1n  2 2  1n  1 2  n,
we can also write this sum in the reverse order as
Sn  n  1n  1 2  1n  2 2  . . .  3  2  1.
We now add these two equations together term by term in the following way:
Sn  Sn  31  n4  32  1n  1 2 4  33  1n  2 2 4  . . .  3 1n  1 2  2 4  3n  1 4
 1n  1 2  1n  1 2  1n  1 2  . . .  1n  1 2  1n  1 2.
166666644666626664664666663
n times
Because there are n of these terms on the right side, we have
2Sn  n1n  1 2.
Dividing both sides by 2, we obtain
n1n  1 2
Sn  ,
2
which gives the following general result.

The sum of the first n integers is

n1n  1 2
1  2  3  ...  n  . (3)
2

E XAMPLE 3
Find the sum of the first 100 integers: 1  2  3  . . .  100.
Solution Using Formula (3) with n  100, we get
1001101 2
1  2  3  . . .  100   5050.
2
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 296

296 CHAPTER 4 Extended Families of Functions

We can also write Formula (3) in summation notation (see Appendix A3):
n n1n  12
a k123
...  n  .
k1 2
Alternatively, we can derive this result by using either the ideas on fitting func-
tions to data or algebraic methods, as shown in Example 4. The advantage of deriv-
ing this formula in other ways is that it demonstrates techniques that can be applied
to more complicated cases for which the trick we used previously doesn’t work.

E XAMPLE 4
Derive the formula for the sum of the first n integers by using (a) curve fitting methods
and (b) algebraic methods.
Solution We again write the sum of the first n integers as Sn  1  2  3  . . .  n.
Thus the sum of the first integer is S1  1; the sum of the first two integers is
S2  1  2  3; the sum of the first three integers is S3  1  2  3  6, and when
we continue S4  10, S5  15, S6  21, and so on. If we form a table of second differ-
ences with these entries, we get the following.

n Sn Sn 2Sn
1 1
2
2 3 132
3
3 6 1
4
4 10 1
5
5 15 1
6
6 21

The second differences

2Sn are all constant, so the desired pattern is a quadratic func-
tion of n. Thus Sn  an2  bn  c, where a, b, and c are constants that we must now
determine.
a. Using the regression features of a calculator, we find that the quadratic function that
best fits these points is
Sn  0.5n2  0.5n  0
1 1
 n2  n
2 2
1
 n1n  12 ,
2
as before. All the points lie on this curve, as we can verify by substituting the coordi-
nates of the points into the equation.
b. Using the point n  1, S1  1 in the quadratic function Sn  an2  bn  c, we get
S1  1  a # 112 2  b # 11 2  c,
and so
a  b  c  1.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 297

4.5 Finding Polynomial Patterns 297

When n  2, we have S2  3 so that

S2  3  a # 122 2  b # 12 2  c,
and hence
4a  2b  c  3.
Similarly, when n  3 and S3  6, we have
S3  6  a # 132 2  b # 13 2  c,
so that
9a  3b  c  6.
We therefore have a system of three linear equations in three unknowns:
a  b  c  1;
4a  2b  c  3;
9a  3b  c  6.
Using the matrix techniques from Appendix C, we find that
1 1
a , b  , and c  0,
2 2
which is the same set of coefficients we found in part (a).
◆
There is one difficulty with both derivations in Example 3. We used both meth-
ods to derive a formula for the sum of the first n integers, which is supposed to be true
for any n. But, in fact, we based both derivations on just the first six values that we cal-
culated for S1 , S2 , . . . , S6 , which we showed followed a quadratic pattern by looking
at a table of second differences. The catch is that we can’t know for sure, just by look-
ing at examples, that all subsequent values for Sn continue to follow a quadratic pat-
tern. So the “proof” really isn’t legitimate unless we can demonstrate that it applies to
every value of n, not just the first six. We do so in Example 5.

E XAMPLE 5
Show that all the values for Sn  1  2  3  . . .  n, for all values of n, fall in a
quadratic pattern.
Solution To show that all values of Sn fall in a quadratic pattern, we must demonstrate
that the second differences are always constant for any value of n. Let’s consider any
value of n, so that the sum of the first n integers is
Sn  1  2  . . .  n.
If we take the next integer, n  1, and form the sum of the first n  1 integers, we get
Sn1  11  2  . . .  n 2  1n  1 2 .
The difference between Sn and Sn1 is

Sn  Sn1  Sn  n  1,
because all other terms cancel.
Similarly, the sum of the first n  2 integers is
Sn2  11  2  . . .  n2  1n  1 2  1n  2 2.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 298

298 CHAPTER 4 Extended Families of Functions

The difference between this total and Sn1 is

Sn2  Sn 1 
Sn1  n  2,
because, again, all other terms cancel. As a result, the second difference, or difference of
the differences, is just

Sn1 
Sn  1n  22  1n  1 2  1,
for any value of n. Therefore all values of Sn have a constant second difference and con-
sequently, no matter what value of n we select, the sum of all the differences must follow
a quadratic pattern.
◆

Sums of Squares of Integers

We now find a formula for the sum of the first n squares,
Sn  12  22  32  42  . . .  n2.
We have
S1  1, S2  12  22  5, S3  12  22  32  14,
S4  30, S5  55, and S6  91,
and so on. For simplicity, we also use the sum of the squares of the first zero terms,
S0  02  0. Arranging these values in a table, we obtain the following.

n Sn Sn 2Sn 3Sn

0 0
110
1 1 341
4 253
2 5 5
9 2
3 14 7
16 2
4 30 9
25 2
5 55 11
36
6 91

The third differences

3Sn are all constant, so these data values follow a cubic
pattern; that is, the formula for the sum of the squares of the first n integers is a
cubic function
Sn  an3  bn2  cn  d.
Using polynomial regression, we find the cubic polynomial that fits the points
10, 02, 11, 1 2, 12, 5 2, 13, 142, 14, 30 2, 15, 55 2 , and 16, 912 has coefficients a 
0.33333333 (or 13), b  0.5  12 , c  0.16666667 (or 16) and d  3.5E 1122 
3.5  1012  0.0000000000035, which essentially is 0. Therefore the cubic
function that fits the data is

Sn  a b n3  a b n2  a b n  0.
1 1 1
3 2 6
Gord.3896.04.pgs 11/21/03 10:35 AM Page 299

4.5 Finding Polynomial Patterns 299

We factor out the common factors n and 16 and then factor the resulting quad-
ratic to get

Sn  n c a b n2  a b n  a b d
1 1 1
3 2 6
 a b n32n2  3n  14
1
6
 a b n12n  12 1n  12,
1
6
which is more commonly written as
n1n  1 2 12n  1 2
Sn  .
6
Alternatively, we could solve for the coefficients of the cubic polynomial
Sn  an3  bn2  cn  d algebraically.
Using n  0 and S0  0, we get 0  d.
Therefore we have
Sn  an3  bn2  cn.
Further
when n  1 and S1  1: a b c1
when n  2 and S2  5: 8a  4b  2c  5
when n  3 and S3  14: 27a  9b  3c  14
These results give a system of three equations in the three unknowns a, b, and c; we
have already determined that d  0. Using matrix methods to solve this system of
equations, we again get a  13 , b  12 , and c  16 .
In general, we have the following formula.

The sum of the squares of the first n integers is

n n1n  1 2 12n  12
. . .  n2 
a k 1 2 3 
2 2 2 2
. (4)
k1 6

E XAMPLE 6
Find the sum of the squares of the first 100 integers: 12  22  . . .  1002.
Solution Using Formula (4) with n  100, we get
10011012 1201 2
12  22  . . .  1002   338,350.
6
◆
Note that, although Formula (4) is true for all values of n, we have only estab-
lished it for n  0, 1, . . . , 6 by using both of these approaches. As with the sum of
the first n integers, we must prove that the sum of the squares of the first n integers
follows a cubic pattern for every possible value of n. We do so in Example 7.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 300

300 CHAPTER 4 Extended Families of Functions

E XAMPLE 7
Prove that the sum of the squares of the first n integers, for any n, follows a cubic pattern.
Solution To do so, we have to show that the third differences of Sn are all constant, for
any value of n. We write
Sn  12  22  . . .  n2
so that
Sn1  12  22  . . .  n2  1n  1 2 2;
Sn2  12  22  . . .  n2  1n  1 2 2  1n  22 2;
Sn3  12  22  . . .  n2  1n  1 2 2  1n  22 2  1n  3 2 2.
We begin by forming the first differences of each successive pair. In each case, all terms
but one cancel, leaving us with
Sn1  Sn 
Sn  1n  1 2 2  n2  2n  1;
Sn2  Sn1 
Sn1  1n  2 2 2  n2  4n  4;
Sn3  Sn2 
Sn2  1n  3 2 2  n2  6n  9.
Each of these first differences is a quadratic function of n. We now form the second dif-
ferences by taking the difference of each successive pair of first differences:

Sn1 
Sn 
2Sn  1n2  4n  4 2  1n2  2n  1 2  2n  3;

Sn2 
Sn1 
2 Sn1  1n2  6n  9 2  1n2  4n  4 2  2n  5.
Finally, we find the third differences by forming the difference between these last two ex-
pressions and get

2Sn1 
2Sn 
3Sn  12n  5 2  12n  3 2  2,
which is constant for all values of n. That is, the sum of the squares of the first n integers
follows a cubic pattern for every value of n.
◆
E XAMPLE 8
When cannonballs are stacked in a pyramidal pile, as shown in the accompanying figure,
they are organized from the top layer down as follows: A single ball is at the top of the
pile; four balls are in the second layer, arranged in a square to support the single ball on
top; nine balls are in the third layer, arranged in a square of size 3 by 3 that supports the
second layer; and so on. How many cannonballs are in a pile that is 10 layers high?
Solution The number of cannonballs is
12  22  32  . . .  102.
We can evaluate this total using Formula (4) for the sum of the squares of the first n in-
tegers with n  10. Thus
10110  1 2 32110 2  1 4
12  22  32  . . .  102 
6
10111 2 1212

6
 385 cannonballs.
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 301

4.5 Finding Polynomial Patterns 301

Example 9 illustrates some additional applications of these ideas to find the

total for a quantity when the individual amounts are known. In it we use the fol-
lowing basic properties of sums of numbers:
n n n

a 1ak  bk 2  a ak  a bk (5)
k1 k1 k1

and
n n

a 1m ak 2  m a ak ,
. . for any constant m. (6)
k1 k1

We ask you to prove these two results in the problems at the end of this section.

E XAMPLE 9
A study of the financial records of a company finds that its monthly revenues, in thou-
sands of dollars, are modeled by the function R1x 2  0.001x 2  0.02x  32, where x is
the number of months since the start of the study and x 1. Find the total revenue for
this company over its first 10 years of operation.
Solution The 10-year period is equivalent to 120 months. We need to add the rev-
enues R11 2 in month 1, R12 2 in month 2, R13 2 in month 3, . . . , R1120 2 in month 120.
Doing so, we get
120
R  R112  R12 2  . . .  R1120 2  a 10.001k 2  0.02k  32 2 ,
k1

where the variable k takes on all values between 1 and 120. Using Property (5) of sums,
we simplify the preceding equation and get
120 120 120
R  a 0.001k 2  a 0.02k  a 32,
k1 k1 k1

Using Property (6), we get

120 120 120
R  0.001 a k 2  0.02 a k  32 a 1.
k1 k1 k1

The first term involves the sum of the squares of the first 120 integers, so
120 11202 1120  1 2 12 . 120  1 2
a k 2
  583,220.
k1 6
The second term involves the sum of the first 120 integers, so
120 1120 2 1121 2
a k 2
 7260.
k1

The third term involves the sum of 120 ones, so

120

a 1  120 11 2  120.
.
k1

Therefore the total revenue for this company over the 10-year period is
R  0.0011583,2202  0.02172602  3211202  4568.42
thousand dollars, or about $4.568 million.
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 302

302 CHAPTER 4 Extended Families of Functions

Problems

1. In Examples 1 and 2, we found the parabola that a. Find a formula for the terms in the third diago-
passes through the points 10, 22, 11, 02 , 12, 4 2, nal: 1, 3, 6, 10, 15, . . . , in terms of the row num-
13, 142, 14, 302, and 15, 522 . Suppose now that the ber n.
points are 10, 22, 11, 0 2, 12, 42, 13, 152, 14, 302 , and b. Find a formula for the terms in the fourth diago-
15, 522 instead. nal 1, 4, 10, 20, . . . , in terms of the row number n.
a. Show that these points do not lie on a parabola. 5. Construct the quadratic polynomial that passes
b. Attempt to repeat the procedure used in Exam- through the points 10, 1 2, 11, 4 2, and 12, 9 2. Use it
ple 2 to see what goes wrong. to estimate the value of the underlying function
2. Determine which sets of values come from a quad- when x  0.5 and when x  3.
ratic function and which come from a cubic func- 6. The main support cable of a suspension bridge is a
tion. For those that come from a quadratic function, parabola. For the Golden Gate bridge, suppose that
determine the equation of the quadratic. the cable’s lowest point is 15 feet above the roadway.
Use the dimensions shown in the accompanying
figure to find an equation of the cable for the Gold-
x f (x) g(x) h(x) k(x)
en Gate bridge.
0 0 1 1 3

1 2 6 0 1

2 2 13 5 3
546′
15′
3 12 22 22 9

4 28 33 57 19 4200′ 200′

5 50 46 116 33
7. Find (a) the sum of the first 25 integers, (b) the sum
3. The following measurements were taken on a quan-
of the first 100 integers, and (c) the sum of the first
tity that follows a cubic pattern. However, one of
1000 integers.
the values was recorded in error. Find the incorrect
entry and correct it. (Hint: It isn’t necessary to actu- 8. Find (a) the sum of the squares of the first 25 integers
ally determine the formula for the cubic.) and (b) the sum of the squares of the first 50 integers.
9. Suppose that the produce manager in a supermar-
ket receives a delivery of 1000 large grapefruit,
x 0 1 2 3 4 5 6 7
which he wants to display in a pyramid with a
y 40 34 24 22 40 90 184 344 square base. How many layers are needed?
10. a. Find the sum of the integers from 83 through
4. Consider the array of numbers known as Pascal’s tri-
225, inclusive.
angle in which each row begins and ends with 1 and
b. Find the sum of the squares of these integers.
each intermediate entry is simply the sum of the two
numbers diagonally above it in the previous row. 11. The annual rainfall R, in inches, in a particular region
in year t since the start of the last century can be mod-
1 1 eled by the formula R1t2  0.02t 2  1.8t  42.
1 2 1 Find the total rainfall from 1900 (when t  0)
1 3 3 1 through 2000 in that region.
1 4 6 4 1
12. Cannonballs are sometimes stacked in rectangular
1 5 10 10 5 1
piles. The accompanying figure shows the fourth
1 6 15 20 15 6 1
layer of a stack of n rectangular layers.
The rows are numbered n  1, 2, . . . The second di- a. Suppose that such a stack ends with a single row
agonal consists of the entries 1, 2, 3, 4, 5, and 6, . . . of two balls at the top. Devise a formula in sum-
Gord.3896.04.pgs 4/24/03 9:57 AM Page 303

4.5 Finding Polynomial Patterns 303

conjecture that the sum of the cubes of the first n

integers
n
Sn  a k 3  13  23  33  . . .  n3
k1
follows a quartic polynomial pattern.
a. Calculate values for Sn for n  0, 1, 2, . . . , 7.
b. Use a table of differences to show that these val-
Fourth layer ues follow a quartic pattern.
mation notation for the number of balls in a stack c. Find a formula for the sum of the cubes of the
n layers high. first n integers.
b. Use the properties of summations to expand the
17. Find the sum of the cubes of the first 25 integers.
formula you found in part (a).
c. Suppose that a stack of cannonballs ends with a 18. By writing out
n
single row of three balls as the top layer. Devise a
a ak  a1  a2 
. . .  an ,
formula for the number of balls in a stack n lay- k1
ers high. n n

a 1m ak 2,
.
a bk  b1  b2 
. . .  bn , and
d. Use the summation formulas from parts (a) and
k1 k1
(c) to predict the result if the top layer consists of show that
a single row of four balls. n n n

13. a. Consider the function y  ax 2. Construct a a 1ak  bk 2  a ak  a bk

k1 k1 k1
table of values for the function if x  2, 1, 0, and
1, 2, 3 and extend it to a table of differences until n n

a 1m ak 2  m a ak ,
you can construct a formula for
2y for this . for any constant m.
function. k1 k1

b. Repeat part (a) for the function y  ax 3 to de- 19. A Pythagorean triple is a set of three integers a, b, and
vise a formula for
3y. c that satisfy the Pythagorean theorem a2  b2  c2
c. Repeat part (a) for the function y  ax 4 to de- and hence represent the sides of a right triangle.
vise a formula for
4y. The following is a list of the first five Pythagorean
d. Based on your results in parts (a)–(c), predict a triples 1an , bn , cn 2.
formula for
5y when y  ax5.
For the sequence of numbers 5 y0 , y1 , y2 , . . . , yn ,
n an bn cn
14.
yn1 , yn2 , . . . 6, show that 1 3 4 5
a.
2y0  y2  2y1  y0 . 2 5 12 13
b.
2yn  yn2  2yn1  yn , for any n.
Suppose that a set of data values 1x0 , y0 2 , 1x1 , y1 2,
3 7 24 25
15.
1x2 , y2 2, . . . has uniformly spaced x-values 1
x 2 and 4 9 40 41
constant second differences
2y  k so that the
5 11 60 61
points follow a quadratic pattern y  ax 2  bx  c.
Use the result of Problem 14 to show that the
(There are infinitely many Pythagorean triples.) No-
leading coefficient is
tice that, for any n, an  2n  1 and cn  bn  1.
1
2y
a
2 1
x 2 2
. a. Construct a table of differences to determine the
pattern in the bn terms.
(Hint: Write x1  x0 
x and x2  x0  2
x and b. Find a formula for bn for each value of n, based
use the first three points to construct a system of on the pattern from part (a).
linear equations in a, b, and c.) c. Show that the resulting triple 1an , bn , cn 2 , forms a
16. Because the sum of the first n integers follows a Pythagorean triple for any integer n.
quadratic pattern and the sum of the squares of the d. What is the next Pythagorean triple following
first n integers follows a cubic pattern, you might the ones shown in the table?
Gord.3896.04.pgs 4/24/03 9:57 AM Page 304

304 CHAPTER 4 Extended Families of Functions

4.6 Building New Functions from Old:

Operations on Functions
The functions that we’ve considered so far, such as y  1x, y  x , y  10 , and
5 x

y  log x, can be thought of as building blocks from which we can construct other,
more complicated functions. In a simple case, we can take the power functions
y  x and y  x 2 and the constants 3, 4, and 5 to create the quadratic function
f 1x2  3x 2  4x  5 as a linear combination of power functions. In fact, we can
think of any polynomial as a linear combination of power functions. In this sec-
tion, we investigate how to generate larger classes of functions by applying simple
operations (e.g., addition, subtraction, multiplication, and division) to the basic
families of functions that we already have discussed.

Sums and Differences

Let’s begin with the sum of two functions. The function
f 1x 2  x 2  2x
is the sum of the two functions y  x 2 and y  2x  1>2x  A 12 B x. Their individual
graphs are shown in Figure 4.47. If we “pile” one set of y-values on top of the other
and add, we get the graph of the sum as shown in Figure 4.48. You can verify that
this result is indeed the case by plotting the sum function on your function grapher.

y y

y = x2

y = x 2 + 2 –x

y = 2 –x
x x

FIGURE 4.47 FIGURE 4.48

Note the behavior of the sum y  x 2  2x. Because y  2x  A 12B x decays rap-
idly as x increases, its contribution becomes less and less significant, and the quadrat-
ic term eventually dominates in the sum when x is large. As a result, toward the right,
the graph quickly becomes indistinguishable from a parabola. For negative values of
x, both functions become large, but 2x grows much faster than x 2 does and so the ex-
ponential term dominates on the left.
In general, we write
S1x 2  f 1x 2  g 1x 2
for the sum of two functions f 1x 2 and g 1x 2 . That is, for each value of x, we add the
values of f 1x 2 and g 1x2 to produce the value of S1x 2. For instance, if f 13 2  15
and g 132  4, then S13 2  f 132  g 13 2  15  4  19.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 305

4.6 Building New Functions from Old: Operations on Functions 305

Similarly, we construct the difference of two functions by taking the difference

between their values for each possible value of x. In general, we write
D1x 2  f 1x 2  g 1x 2
for the difference of two functions. Thus, if f 13 2  15 and g 13 2  4, then
D13 2  f 132  g 13 2  15  4  11. Graphically, if we subtract g 1x 2 from f 1x 2
and f 1x 2 g 1x 2, the difference D1x 2 is just the difference in height between the
two curves for each value of x.

Products of Functions
For the product of the two functions f 1x2  x 2 and g 1x 2  2x, we use the same
interpretation as with sums and differences of functions. Thus the product of the
two functions
P1x 2  f 1x 2 . g 1x 2
means that, for each permissible value of x, we multiply the corresponding func-
tion values. So, if f 13 2  15 and g 13 2  4, then P13 2  f 13 2 . g 13 2  15 . 4  60.
What does the graph of the product function look like? Unlike the sum and
difference of two functions, there is rarely a direct graphical interpretation of the
product of two functions. However, you can produce the graph of the product of
two functions on your function grapher and then analyze the behavior of that
graph. For instance, consider
P1x 2  f 1x2 . g 1x 2  x 2 . 2x.
We know that, for large positive x, y  x 2 grows ever larger and y  2x approach-
es zero. We also know that an exponential function with a positive exponent grows
much faster than a power function does. Similarly, an exponential function with a
negative exponent decays much faster than a power function with a negative
power. Together, these facts indicate that, in the product x 2 . 2x, the exponential
term drives the product toward zero as x increases. For values of x  0, both func-
tions grow without bound, so their product becomes infinitely large. By using your
function grapher, you can obtain the result shown in Figure 4.49.
Let’s look at a real-life example of a product of two functions. Lyme disease is
caused by a bacterial infection transmitted by blood-sucking ticks. When a person is
infected, the body produces antibodies to fight the bacteria. Figure 4.50 shows the
level of concentration of the antibody in the bloodstream as a function of the num-
ber of weeks since the first infection. Note that the pattern is remarkably similar to

y
y
Antibody level

y = x 2 . 2 –x

x
FIGURE 4.49 FIGURE 4.50 Weeks since infection
Gord.3896.04.pgs 4/24/03 9:57 AM Page 306

306 CHAPTER 4 Extended Families of Functions

the behavior of the function f 1x 2  x 2 . 2x for x 0. This pattern suggests that
such functions, known as surge functions, might be appropriate as mathematical
models to describe the antibody level, both for Lyme disease and possibly other in-
fections. We explore some applications of surge functions in Section 4.9.

Quotient of Functions
When we consider the quotient of two functions
f 1x 2
Q1x 2 
g 1x 2
,

there is a complication that we must take into account. The quotient is undefined
at any point where the denominator g 1x 2 is zero, and typically a vertical asymptote
occurs there. We illustrate this behavior in Example 1.

E XAMPLE 1
Sketch the graph of the function
x2  1
Q1x 2  .
x2  1
Solution We begin analyzing the behavior of this function by looking at what happens
when the denominator x 2  1  1x  1 2 1x  12 is zero. That occurs when x  1 and
x  1, so the quotient Q1x2 is not defined there. When you take values of x very close to
either of these two points, the corresponding values for the quotient Q1x 2 become ex-
tremely large, positively or negatively. To see this result, first consider points near x  1.
Suppose that x is slightly larger than 1. So
if x  1.001, then y  Q11.0012  1000.5;
if x  1.0001, then y  Q11.0001 2  10000.5;
if x  1.000001, then y  Q11.0000012  1,000,000.5.
Hence, as x approaches 1 from the right (or from above) through values of x that are
slightly larger than 1, y becomes ever larger and approaches .
Now suppose that x is slightly smaller than 1. So
if x  0.999, then y  Q10.9992  999.5;
if x  0.9999, then y  Q10.9999 2  9999.5;
if x  0.999999, then y  Q10.9999992  999,999.5.
Hence, as x approaches 1 from the left (or from below) through values of x that are
slightly smaller than 1, y approaches .
By a similar analysis around the point x  1, you can verify that, as x approaches
1 from the right, the function approaches , whereas, if x approaches 1 from the
left, the function approaches . Therefore it is not surprising that this quotient func-
tion has vertical asymptotes at x  1 and x  1.
We next analyze the end behavior—what happens to this function as x becomes large,
both positively and negatively. Suppose, for instance, that x  1000. The value of the
function then is

Q11000 2 
1,000,001
 1.000002,
999,999
Gord.3896.04.pgs 4/24/03 9:57 AM Page 307

4.6 Building New Functions from Old: Operations on Functions 307

which is extremely close to 1. Actually, adding 1 to x 2 in the numerator and subtracting

1 from x 2 in the denominator really has little effect on the value of the function when x
is 1000. If x were even larger—say, 1,000,000—adding or subtracting 1 from x 2 would
have a negligible effect. Thus for large values of x, the numerator is dominated by the x 2
term and the denominator is dominated by the x 2 term, so the quotient behaves like
x2
Q1x 2  2  1
x
when x is large. As a result, this quotient function gets closer and closer to a height of 1,
so that it has a horizontal asymptote of y  1 as x approaches .
What happens as x approaches  ? Again, for large negative values of x, the 1 in the
numerator and the 1 in the denominator are negligible and x 2 again dominates both
the numerator and the denominator. Thus there is also a horizontal asymptote of y  1
as x approaches .
We next look for the points where the curve crosses the two axes. It crosses the y-axis
when x  0, so Q10 2  1>11 2  1. Where does the curve cross the x-axis? For that
to happen, y must equal 0, so the numerator has to be 0. Because the numerator for Q1x 2,
x 2  1, is never 0 for real values of x, the quotient cannot be 0 anywhere. Hence the curve
never crosses the x-axis. The complete graph of this function is shown in Figure 4.51.
y

x2 + 1
y=
x2 − 1

x
–1 1

FIGURE 4.51

Rational Functions
Example 1 illustrates most of the ideas involving the behavior of quotients of func-
tions in general and quotients of polynomials in particular. The quotient (or ratio)
of two polynomials is called a rational function. We assume that any common fac-
tors in the numerator and the denominator have been canceled and therefore that
the rational function is expressed in simplest form.
The following are some of the important facts about rational functions.

Behavior of Rational Functions R(x)  P(x)>Q(x)

◆ The zeros of the numerator P1x 2 correspond to zeros of the rational
function R1x 2 ; its graph crosses the x-axis at these points.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 308

308 CHAPTER 4 Extended Families of Functions

◆ The zeros of the denominator Q1x 2 correspond to the points where

the rational function R1x2 is not defined; its graph usually has a verti-
cal asymptote at these points.
◆ The highest power term in the numerator P1x 2 dominates the numer-
ator for large values of x, either positive or negative.
◆ The highest power term in the denominator Q1x 2 dominates the de-
nominator for large values of x, either positive or negative.
◆ For large values of x, either positive or negative, the rational function
R1x 2 behaves like the highest power term of the numerator divided by
the highest power term of the denominator. The result may be a hori-
zontal asymptote or the values may approach  or  as x increases ei-
ther positively or negatively.

We illustrate these ideas in Examples 2 and 3.

E XAMPLE 2
Analyze the behavior of the rational function
x2  1
R1x 2  .
x2
Solution Here, R1x2 has zeros when its numerator x2  1  0, so that x   1, and
the graph crosses the x-axis at these two points. Also, the denominator is zero when
x  2, which creates a vertical asymptote there. Suppose that x approaches 2 from the
right (with values slightly larger than 2); for instance,
if x  2.01, then y  R12.012  304.01;
if x  2.001, then y  R12.0012  3004.001;
if x  2.00001, then y  R12.000012  300004.00001.
Thus, when x approaches 2 from the right, R1x 2 approaches . Similarly, when x app-
roaches 2 from the left, R1x 2 approaches  (try some values of x slightly less than 2—
say, x  1.99 or x  1.9999).
You can locate the vertical asymptotes of a rational function by finding the roots of the
denominator, but you must check what happens on either side (in this case at x  2.001
and x  1.999, for example) to determine the sign of the function on each side of the verti-
cal asymptote. Doing so lets you decide whether the curve rises toward  or drops toward
 on each side of the vertical asymptote.
Next, consider the end behavior of R1x2 . For large values of x, the numerator is domi-
nated by the leading x 2 term and the denominator is dominated by the leading x term. As a
result, for large values of x, the quotient behaves like y  R1x 2  x2>x  x. For instance,
if x  10, then R110 2  12.375;
if x  100, then R1100 2  102.0306;
if x  1000, then R11000 2  1002.003006.
The larger x is, the closer R1x 2 is to x and, for large positive values of x, the graph in-
creases toward .
Similarly, for large negative values, the quotient R1x 2 behaves like y  x 2>x  x and
the graph tends toward .
Gord.3896.04.pgs 4/24/03 9:57 AM Page 309

4.6 Building New Functions from Old: Operations on Functions 309

Figure 4.52 displays all this behavior in the complete graph of R.

x2 − 1
y=
x−2

x
0 2
FIGURE 4.52

◆
Think About This Graph both the original quotient function R1x 2  1x2  1 2> 1x  2 2 and the
limiting function y  x in the same fairly large viewing window—say, from 1000
to 1000 for both x and y—on your function grapher. What do you observe? ❐

Getting all the important details on the behavior of a rational function from a
single view in your function grapher is often almost impossible. Try it for the func-
tion R1x2 in Example 2 and see what types of information may be lost because of
the scale you use for the domain and range.

E XAMPLE 3
Analyze the behavior of the rational function
x2
S1x 2  .
x2  1

Solution Here, S1x 2 was formed by interchanging the numerator and denominator of
the rational function R1x 2 in Example 2, but the behavior of the two functions is quite
different.
Note that S1x 2 has only one zero at x  2 when the numerator is zero. It has two ver-
tical asymptotes, one at x  1 and the other at x  1 when the denominator is zero.
Let’s see what happens on either side of the asymptotes. When x  1.001, say, we have
S11.0012  499.25, so we conclude that the curve drops toward  as x approaches 1
from the right. Similarly, when x  0.999, we have S10.9992  500.75 and the curve
rises toward  as x approaches 1 from the left. Similarly, when x  1.001, we have
S11.0012  1499.75 and the curve drops toward  as x approaches 1 from the
left. Also, when x  0.999, S10.9992  1500.25, and the curve rises toward  as x
approaches 1 from the right. Use your calculator to check these conclusions numeri-
cally with other values of x on either side of x  1 and on either side of x  1.
Further, the numerator is dominated by x and the denominator is dominated by x 2, so
for large values of x, the rational function behaves like y  x>x 2  1>x. Therefore, for
large positive values of x, the function is positive and decays toward the x-axis as a hori-
zontal asymptote. Similarly, for large negative values of x, the function is negative and rises
toward the x-axis as a horizontal asymptote.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 310

310 CHAPTER 4 Extended Families of Functions

The complete graph of S1x2 is displayed in Figure 4.53.

y

x−2
y=
x2 − 1

x
–1 1 2

FIGURE 4.53

◆
As before, though, we urge you to examine the behavior carefully with your
function grapher to see how viewing the overall characteristics depends on the
window you use.

Think About This Examine the graphs of the quotient function S1x 2 and the limiting function
y  1>x in the same large viewing window. What do you observe? ❐

We next consider a real-world application that involves rational functions.

E XAMPLE 4
According to the law of universal gravitation, the gravitational force between any two
objects of mass m1 and m2 is
Gm1m2
F ,
r2
where r is the distance between the objects and G is the gravitational constant. Envision
a spacecraft traveling from the Earth to the moon, a distance of about 240,000 miles. Be-
cause the mass of the Earth is roughly 81 times that of the moon, the Earth’s gravita-
tional effect on the spacecraft will be greater than that of the moon’s until the spacecraft
is quite close to the moon, when it’s gravity becomes dominant. Determine the distance
from the Earth when the two gravitational forces exactly balance each other.
Solution We begin with a sketch of the situation, as shown in Figure 4.54, where r rep-
resents the distance, in thousands of miles, from the Earth to the spacecraft. Hence

240 – r

FIGURE 4.54
Gord.3896.04.pgs 4/24/03 9:57 AM Page 311

4.6 Building New Functions from Old: Operations on Functions 311

240  r is the distance from the moon to the spacecraft. Let m0 be the mass of the space-
craft, m1 be the mass of the Earth, and m2 be the mass of the moon. The Earth’s gravita-
tional force on the spacecraft Fe is
Gm0m1
Fe  ,
r2
and the moon’s gravitational force on the spacecraft Fm is
Gm0m2
Fm 
1240  r2 2
.

Both Fe and Fm are rational functions of r. Because the Earth is 81 times as massive as the
moon, m1  81m2 . We rewrite Fe as
Gm0 181m2 2
Fe  .
r2
The two gravitational forces are equal when
81Gm0m2 Gm0m2

1240  r2 2
2 .
r
Dividing both sides of this equation by Gm0m2 (because none of these quantities are zero)
gives
81 1
2 
1240  r2 2
.
r
Cross-multiplying yields
r 2  811240  r2 2.
We expand the expression on the right by squaring the binomial term and obtain
r 2  8112402  480r  r 2 2  811240 2 2  811480 2r  81r 2.
Collecting like terms and simplifying, we have the quadratic equation
80r 2  38,880r  4,665,600  0.
Dividing through by the common factor 80, we get
r 2  486r  58,320  0.
Using the quadratic formula, we find that the roots of this quadratic equation are
r  216 and r  270. These answers are distances in thousands of miles from the Earth.
Because the moon is about 240 thousand miles from the Earth, the only reasonable an-
swer is the first. Therefore the two forces balance at a point about 216 thousand miles
from the Earth and about 24,000 miles this side of the moon. The second solution,
270,000 miles from the Earth, corresponds to a point beyond the moon where the effects
of the moon’s gravity and the Earth’s gravity are numerically the same, though both
forces are in the same direction.
◆
A Function of a Function
There is yet another way in which we can construct new functions from simpler
functions. In Example 5 of Section 2.2, we showed that the rate R at which a snow
tree cricket chirps is a function of the temperature T, and we found a mathematical
Gord.3896.04.pgs 4/24/03 9:57 AM Page 312

312 CHAPTER 4 Extended Families of Functions

model for this relationship as the function R  f 1T2  4T  160. However, the air
temperature doesn’t remain constant, but actually varies with the time of the day, so
the temperature T is really a function of time t: T  g 1t2 . As a result, the chirp rate,
though a function of the temperature T, is actually a function of time t. That is, we
have two functions:
R  f 1T2  4T  160 and T  g 1t2.
If we substitute T  g 1t2 into the expression R  f 1T2, we get
R  f 1T2  f 1g 1t2 2.
We call this type of situation a function of a function or a composite function.
Let’s look at this notion from a different perspective. Consider the function
f 1x2  2x 3  1 . To see what it means, suppose that x  1. Then

f 112  213  1  22 .
For x  2,

f 12 2  223  1  29  3.
To evaluate this function in each case, we actually performed two successive steps:
(1) for each value of x, we evaluated the expression x 3  1; and (2), we took the
square root of the result. The reason is that we are really working with two func-
tions successively: first the “inner” function x 3  1 and then the “outer” function
1u, where u  x 3  1. The final function f is therefore a function of a function.
Let’s set up the mathematical framework for this concept. Suppose
that we let y  F1u2 , where u  G1x 2. Here, y  F1u2  1u, where in turn
u  G1x 2  x 3  1. Consequently,

y  F1u2  F1G1x 2 2  F1x 3  1 2  2x 3  1 .

Our original function f is the result of applying the functions G and F
successively. This composite function y  F1G1x 2 2 is sometimes written as F  G
and read “F of G”.
In general, for two functions F and G, the composite function F1G1x 2 2 is the
result of evaluating the two functions successively, as depicted in Figure 4.55. We
start with a value of x, which is carried into a value u by the first, or inner, function
G, which in turn is carried to a value y by the second, or outer, function F. For this
method to make sense mathematically, the domain of the outer function F must
include the range of the inner function G.

G F

y = F(u)
x u = G(x)
= F(G(x))
FIGURE 4.55

Using composite functions, we can construct many other types of functions by

using the basic functions as building blocks.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 313

4.6 Building New Functions from Old: Operations on Functions 313

E XAMPLE 5
Find two functions f and g so that y  f 1g 1x 2 2  103x.
Solution Think about how you would evaluate this function for any value of x—first,
triple the x value because of the 3x term and then take that power of 10. That is, the lin-
ear function 3x is used as the exponent for the exponential function with base 10. So the
first, or inner, function is g 1x 2  3x followed by the second, or outer, function
y  f 1x 2  10x. The result gives f 1g 1x 2 2  f 13x 2  103x, as required.
◆

E XAMPLE 6
Find two functions f and g so that y  f 1g 1x 2 2  log1x 2  5x  22.
Solution Here, the quadratic function x 2  5x  2 is used as the argument of the log
function. So the first, or inner, function is the quadratic g 1x 2  x 2  5x  2 and the sec-
ond, or outer, function is the log function y  f 1x 2  log x. Using the same approach as
in Example 5, we get f 1g 1x 2 2  f 1x2  5x  22  log 1x2  5x  2 2, as required.
◆

Are F  G and G  F the same?

Is the order important in forming the composition of two functions? That is, is
F  G the same as G  F ? Again consider

f 1x 2  2x3  1  F1G 1x 2 2  F  G 1x 2,
where
u  G1x 2  x 3  1 and y  F1u2  2u .
If we interchange the order to form G1F1x 2 2 , we get

G1F1x2 2  G1 2x 2  1 2x 2 3  1  x3>2  1,
which clearly is not the same as F1G1x 2 2  2x 3  1. By substituting a couple of
values for x—say, x  1 or x  2, you can see that the results are numerically dif-
ferent. In general, except in rare cases,
G1F1x 2 2  F1G1x 2 2 .
However, if F and G are inverse functions, the equality does hold.

E XAMPLE 7
In Example 6 we chose f 1x 2  log x and g 1x 2  x2  5x  2. Find f 1g 1x 2 2 and
g 1 f 1x2 2.
Solution We have
f 1g 1x 2 2  f 1x 2  5x  2 2  log 1x 2  5x  22 ,
whereas
g 1 f 1x 2 2  g 1log x 2  1log x 2 2  5 log x  2.
Clearly, they are very different functions.
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 314

314 CHAPTER 4 Extended Families of Functions

Applications of Composite Functions

We next consider a real-world application of composite functions in Example 8.

E XAMPLE 8
When a kicker punts a football, it’s path can be modeled by the quadratic function
y  f 1x2  x 2>27  1.92x  1, where the height y and the horizontal distance
downfield x from the point where the ball is kicked are measured in yards. Furthermore,
the horizontal distance x from the kicker is given by x  g 1t2  12t, where t is measured
in seconds.
a. Find an equation giving the height of the football as a composite function of time t.
b. Determine the hang-time for the football—how long it remains in the air after being
punted.
Solution

a. The path of the ball is the parabola shown in Figure 4.56, where y  f 1x 2  x 2>27 
1.92x  1. The graph shows that the ball carries somewhat more than 50 yards from
the point where it is kicked, which is usually about 10 yards behind the line of scrim-
mage. Using the formula x  g 1t2  12t for x as a function of t, we can form the
composite function giving the height y as a function of t:
112t2 2
y  f 1x 2  f 1g 1t2 2    1.92112t2  1
27
144t 2
  23t  1
27
 5.33t 2  23t  1.
Note that this is also a quadratic function of t with a negative leading coefficient.
y

30
Height (yards)

20

10

x
0 10 20 30 40 50
FIGURE 4.56 Horizontal distance (yards)

b. The hang-time for the football is the value of t when the ball comes back to the ground.
It is the zero of the composite function, so we must solve the quadratic equation
5.33t 2  23t  1  0.
Equivalently, if we multiply both sides by 1, we get
5.33t 2  23t  1  0.
Using either graphical methods or the quadratic formula, we find that t 
4.36 seconds.(A second solution to the quadratic equation gives a negative value for
t, which makes no sense in this context.)
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 315

4.6 Building New Functions from Old: Operations on Functions 315

E XAMPLE 9
Two functions f and g are defined in the following table. Use the values given in the table
to complete it. (If any operations are not defined, write “UNDEF.”)

x f (x) g(x) f (x)  g(x) f (x) # g(x) f (x)>g(x) f (g(x)) g( f (x))

0 1 3

1 0 1

2 3 0

3 2 2

Solution The values of the functions for four specific values of x—namely, x  0, 1, 2,
and 3—are defined in the table. The first open column asks for the difference between
the two functions for each value of x. For instance, when x  0, the first entry for this
column is f 10 2  g 10 2  1  3  2, and so on down that column. The second open
column asks for the product of the two functions for each value of x. When x  0, we
get f 102 . g 10 2  1 . 3  3, and so on down the column.
The third open column asks for the quotient of the two functions. When x  0, we
have f 10 2 >g 10 2  1>3 , and so on. However, because g 12 2  0, the quotient is not de-
fined when x  2, so we enter UNDEF in the corresponding position in the table.
The fourth and fifth open columns ask for values for the composite functions
f 1g 1x2 2 and g 1 f 1x 2 2 . In the fourth column, the function g is applied first and then the
function f is applied. When x  0, we need to form f 1g 10 2 2 . To do so we evaluate
g 10 2  3 first and then take f 1g 10 2 2  f 13 2  2, so the first entry in the fourth column
is 2. For the next entry, we start with x  1 and form f 1g 112 2. Because g 11 2  1, we get
f 1g 112 2  f 11 2  0. Similarly, we get the remaining two entries in this column.
To fill in the entries in the last column, we reverse the order of operations of the two
functions and apply first f, followed by g. Starting with x  0, we now need g 1 f 10 2 2. Be-
cause f 10 2  1, we have g 1 f 10 2 2  g 11 2  1. Similarly, when x  1, we need g 1 f 11 2 2.
Because f 112  0, we have g 1 f 11 2 2  g 10 2  3. Incidentally, for each of the four values
of x, f 1g 1x 2 2  1g 1 f 1x 2 2 .
We now have the completed table.

x f (x) g(x) f (x)  g(x) f (x) # g(x) f (x)>g(x) f (g(x)) g( f (x))

0 1 3 2 3 1
3 2 1

1 0 1 1 0 0 0 3

2 3 0 3 0 UNDEF 1 2

3 2 2 0 4 1 3 0
◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 316

316 CHAPTER 4 Extended Families of Functions

Problems
1. For f 1x2  3x  4 and g 1x2  1x , find f  g, g  f , f . g, f >g, and g> f . Decide which is
a. f 152  g 152. b. f 152  g 15 2 . which and give reasons for your answers.
f 15 2
c. f 15 2 . g 15 2.
g 15 2
d. . y y

e. f 1g 152 2. f. g 1 f 15 2 2. 2 2

g. f 1 f 152 2. h. g 1g 152 2.
1 1

i. f 1x2  g 1x2. j. f 1x2  g 1x 2 .

x x
1 2 3 4 5 1 2 3 4 5
–1 –1
f 1x2
k. f 1x2 . g 1x2.
–2 –2

g 1x2
l. . (a) (b)

m. f 1g 1x2 2. n. g 1 f 1x2 2.
y y

o. f 1 f 1x2 2. p. g 1g 1x2 2. 2 10

2. Repeat Problem 1 for f 1x2  x 2  4 and g 1x 2  1x .

1 5
x x

3. Repeat Problem 1 for f 1x2  10x and g 1x 2  log x.

1 2 3 4 5 1 2 3 4 5
–1 –5
–2 –10
4. Two functions f and g are defined in the table at the (c) (d)
bottom. Use the values given to complete the table. y y

If any of the entries are not defined, write “UNDEF.” 10 2

5. The functions f and g have the values f 12 2  10,

5 1

f 14 2 20, f 162  35, g 122  8, g 14 2  4, and

x x
1 2 3 4 5 1 2 3 4 5
–5 –1
g 162  2. Which expressions, (a)–(g), are correct, –10 –2
(e) (f )
which are incorrect, and which, if any, are not
defined?
a. f 16 2  f 142  2 b. f 1g 162 2  35 7. For the pairs of functions f and g shown, sketch the
c. g 1g 162 2  8 d. f 122  g 16 2  8 graph of the function y  f 1x 2  g 1x 2.
e. f 142  g 142  0 f. f 14 2 . g 14 2  16
g. f 142>g 142  5
y y
f
6. Two functions f and g are given in the accompany- g
f
ing figure. The six graphs (a)–(f) represent f  g,
g

1.5
g(x) x x
1
0.5
x (a) (b)
1 2 3 4 5
–0.5
–1 f (x) 8. For the pairs of functions f and g shown, sketch the
–1.5 graph of the function y  f 1x 2  g 1x 2.

x f (x) g(x) f (x)  g(x) f (x)>g(x) g(x)> f (x) f (x) # g(x) f (g(x)) g( f (x)) f 1
(x)
0 1 0
1 2 3

2 3 1

3 0 2
Gord.3896.04.pgs 4/24/03 9:57 AM Page 317

4.6 Building New Functions from Old: Operations on Functions 317

y y y y

x x x
g
(a) (b) (c)
x
y

(a)

f g x

(d)

11. Often in technical books and articles, graphs are

x
shown for log y as a function of x, as in the follow-
ing graphs. In each case, given the graph of log y as
(b) a function of x, sketch the graph of y as a function
of x.
9. The graphs of three functions, (a)–(c), are shown in
the accompanying figure.
y y

y y

x x

x x (a) (b)

y
(a) (b)
y

x (c)

(c)
12. Match each function with its graph.
x2  1 x2  1
Sketch a rough graph of (i) 2f, (ii) log f, and (iii) f 2. a. y  2 b. y  2
If any portion of a graph is not defined, mark it on x x6 x x6
the x-axis. 9x 2
x2  x  6
c. y  2 d. y 
10. For each function (a)–(d), sketch the graph of log f. x 4 x2  1
If any portion of a graph is not defined, mark it x3  x 1x  1 2 1x  42
e. y  2 f. y 
along the x-axis. x 4 x2  4
Gord.3896.04.pgs 4/24/03 9:57 AM Page 318

318 CHAPTER 4 Extended Families of Functions

y y For Problems 15–18, determine functions F and G such that

h1x 2  F1G1x 2 2. There are different correct answers to this
question; however, do not use F1x 2  x or G1x 2  x.
15. h1x 2  x 4  5 16. h1x 2  1x  5 2 4
x 17. h1x 2  log1x  3 2 18. h1x 2  3  log x
2 –2 –1 1 2
x 19. The time t that a traffic light should remain yellow
–2 1 4
depends on the speed limit s on the road. The func-
tion t  1  s>20  70>s, where t is measured in
seconds and s is the speed in feet per second, is used
(i) (ii)
to determine the length of the yellow cycle. Note that
y y 30 mph  44 ft>sec.
a. How long is the yellow cycle if the posted speed
limit is 30 mph?
b. How long is the yellow cycle if the posted speed
–1 1 limit is 50 mph?
x c. What are reasonable values for the domain and
–1 1 –2 3
x
–2 3 range of this function?
d. Suppose that the traffic department using this
formula wants to increase the length of the yel-
low cycle somewhat. Should it increase or de-
(iii) (iv)
crease the values of each of the two parameters
y y 20 and 70 to do so?
e. Rewrite the formula for t as a rational function
by combining all the terms over a common de-
nominator.
–2 2
x
20. According to Einstein’s theory of relativity, the
–3 3 mass M of an object increases as its speed increases
x according to the formula
–2 3
v2 1>2
M  f 1v2   M0 a1  b
M0
,
(v) (vi) 21  1v2>c2 2 c2
where M0 is the mass of the object when it is at
13. For the two functions f and g that are defined by the rest 1v  0 2 and c is the speed of light (about
graphs shown, find (a) f 1g 11 2 2; (b) g 1 f 11 2 2;
(c) f 1g 112 2 ; and (d) g 1 f 112 2.
186,282 miles per second). Suppose that an object has
a rest mass of M0  1 unit.
y y a. Construct a table of values for the mass of the
object for each of the following speeds expressed
3 3
as a fraction of the speed of light: v  0, 0.5c,
2
g(x) 0.9c, 0.95c, 0.99c, and 0.999c.
1 1 b. Sketch a graph showing the behavior of the mass
–3 –1 1 2 3
x
–3 –2 –1 1 2 3 of an object as its speed approaches the speed of
–1 –1 light.
f (x)
–2 –2 c. What is the mathematical significance of the
–3 –3 speed of light? What is the physical significance
of the speed of light in the context of the speeds
14. For the functions f and g that are defined by the of moving objects?
graphs in Problem 13, sketch the graph of 21. Some physicists hypothesize the existence of parti-
(a) f 1g 1x2 2 and (b) g 1 f 1x2 2. cles called tachyons that exist only at speeds greater
Gord.3896.04.pgs 4/24/03 9:57 AM Page 319

4.6 Building New Functions from Old: Operations on Functions 319

than that of light. The slower that a tachyon moves, 27. The volume of a sphere is given by V  43 pr3 and its
the greater is its mass; the speed of light is a lower surface area is given by S  4pr 2.
limit on the possible speed of a tachyon. Sketch a a. Find a formula for the volume as a function of
graph of the mass as a function of speed for all pos- the surface area. Interpret the result in terms of a
sible values of v 0. Indicate which region corre- composite function.
sponds to normal particles and which to tachyons. b. Find a formula for the inverse function of the
22. According to Newton’s laws of motion, the speed of function you found in part (a). What does it tell
an object can be changed only by applying a force. you?
Also, the greater the mass of an object, the more 28. The degree of polynomial P is m and the degree of
force is needed to accelerate it to a given velocity in polynomial Q is n, where m  n.
a fixed amount of time. Suppose that an object is to a. What is the degree of P  Q?
be accelerated from speed 0 to almost the speed of b. What is the degree of P  Q?
light. c. What is the degree of P . Q?
a. Sketch the graph of the force needed to acceler- d. What is the end behavior of P>Q?
ate it as a function of the velocity v. Pay careful e. What is the end behavior of Q>P ?
attention to concavity. 29. In Problem 20 of Section 1.3, we introduced a func-
b. Sketch the graph of the velocity as a function of tion f that represents a simple replacement code in
the force needed, paying careful attention to which each letter of the alphabet is replaced by a
concavity. different letter according to f 1A2  M, f 1B2  D,
23. a. Graph the two functions y  2x 2  25 and f 1C2  K, f 1D2  V, f 1E 2  X, f 1F2  B,
x  5. Are they the same? f 1G2  P, f 1H2  T, f 1I2  J, f 1J2  S,
b. Repeat part (a) with y  2x 2  4 and f 1K 2  Z, f 1L2  Q, f 1M2  H, f 1N2  O,
y  x  2. Are they the same? f 1O2  A, f 1P2  L, f 1Q2  W, f 1R2  C,
c. Can you find any value for a for which f 1S2  F, f 1T2  Y, f 1U 2  R, f 1V2  G,
2x 2  a2 x  a? f 1W 2  I, f 1X2  U, f 1Y2  N, and f 1Z2  E.
Suppose that we now have a second such code de-
1
24. a. Graph the two functions y  and fined by the function g:
g 1A2  P, g 1B2  K, g 1C2  T, g 1D 2  E,
x4
1 1
y   . Are they the same?
x g 1E 2  L, g 1F2  U, g 1G2  H, g 1H2  N,
g 1I2  Y, g 1J 2  C, g 1K2  R, g 1L2  W,
4
g 1M2  G, g 1N2  Z, g 1O2  B, g 1P2  J,
1 1 1
b. Repeat part (a) with y  and y   .
 g 1Q2  A, g 1R2  X, g 1S2  Q, g 1T2  D,
x 5 x 5
Are they the same?
g 1U2  S, g 1V2  M, g 1W2  V, g 1X2  I,
c. Can you find any value for a so that g 1Y2  O, and g 1Z2  F.
a. Find g 1 f 1A2 2. b. Find f 1g 1A2 2.
1 1 1
  ?
xa
c. Find f 1 f 1P2 2. d. Find g 1g 1K2 2.
x a

25. For the function f 1x2  x x 1 , (a) what is f 11 2? e. Find f 1 1g 1 1A2 2.

(b) f 1 f 112 2? (c) f 1 f 1 f 112 2 2? (d) Continue to apply 30. The algebraic method of elimination for solving a
the function f repeatedly to the previous result, ex- system of linear equations involves adding a multi-
pressing all your answers as fractions. Do you ob- ple of one equation to another equation to eliminate
serve any pattern in the values for the numerators one of the variables. Consider the system of two
and denominators of the fractions that you’re gen- equations in two unknowns:
erating? (e) Now look at the decimal representa-
y  4x  3 (1)
tions of the fractions that you generated in parts
y  7  x. (2)
(a)–(d). Do they appear to be approaching a fixed
value? a. Plot the two lines carefully on a sheet of graph
26. For any two linear functions f 1x 2  ax  b and paper and determine the point of intersection.
g 1x2  cx  d, is f  g the same as g  f ? b. Solve the two equations algebraically.
Gord.3896.04.pgs 4/24/03 9:57 AM Page 320

320 CHAPTER 4 Extended Families of Functions

c. Add two times Equation (2) to Equation (1) to e. Add four times Equation (2) to Equation (1) and
get a new linear function. Plot its graph on the plot that function on the same graph. What can
same graph you created in part (a). What do you you conclude from this result?
observe about the three lines? f. Find an appropriate multiple of Equation (2)
d. Add three times Equation (2) to Equation (1) that, when added to Equation (1), will eliminate
and plot that function on the same graph. What the x term. What will the graph of the resulting
do you observe about the four lines? line look like when x is eliminated?

4.7 Building New Functions from Old: Shifting,

Stretching, and Shrinking
In Section 4.6, we created new functions from known functions by extending the
standard arithmetic operations of addition, subtraction, multiplication, and division
to functions. We also created new functions by using composition of functions. In
this section we introduce several other ways in which we can build new functions
from a single function. Suppose that we have the function y  f 1x 2. We can form a
related function by changing either the independent variable x or the dependent vari-
able y by multiplying it by a constant or by adding or subtracting a constant from it.

Shifting Functions
We can shift functions up and down or left and right. The former involves trans-
forming the y-variable, and the latter involves transforming the x-variable.

Shifting Up and Down We first investigate the effect on any function y  f 1x 2

of adding a constant to y or subtracting a constant from y.

E XAMPLE 1
Consider y  f 1x 2  x 2 and the related functions y  x 2  1, y  x 2  3, y  x 2  2,
and y  x 2  5. What is the effect of the constant in each case?
Solution All these functions are shown in Figure 4.57. Clearly, each constant shifts the
basic parabola y  x 2 up or down by the corresponding amount that is added or sub-
tracted. For instance, the curve y  x 2  1 lies 1 unit above y  x 2 for each value of x,
whereas y  x 2  2 lies 2 units below it.

y
y = x2 + 3

y = x2 + 1
y = x2

y = x2 − 2
3
y = x2 − 5
1
x

–2

FIGURE 4.57 –5

◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 321

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 321

In general, the following principle holds for any function of x, assuming b 0.

Vertical Shift
Replacing f 1x 2 with f 1x2  b shifts the graph of f 1x 2 up by the amount b.
Replacing f 1x 2 with f 1x 2  b shifts the graph of f 1x 2 down by the
amount b.

We can get a different feel for what is happening if we rewrite each of these ex-
pressions by moving the constant term to the left side. For instance, y  x 2  1 is
equivalent to y  1  x 2, which emphasizes the fact that it is the variable y, or the
height, which is being affected by the constant.
We can therefore rephrase the vertical shift principle for any function of x,
assuming b 0, as follows.

Vertical Shift
Replacing y with y  b shifts the graph of f 1x2 up by the amount b.
Replacing y with y  b shifts the graph of f 1x2 down by the amount b.

Shifting Left and Right Next we investigate the effect on y  f 1x 2 of adding a

constant to x or subtracting a constant from x.

E XAMPLE 2
Consider y  f 1x 2  x 2 and the related functions y  1x  12 2, y  1x  3 2 2, and
y  1x  2 2 2, where we replace x by 1x  1 2, 1x  32 , or 1x  2 2, respectively. What is
the effect of the constant in each case?
Solution The resulting graphs are shown in Figure 4.58. Each of these changes causes a
horizontal shift. For instance, y  1x  1 2 2 has a double zero at x  1, so the graph of
y  x 2 is shifted to the right by 1 unit. Similarly, y  1x  2 2 2 has a double zero at
x  2, so the graph of y  x 2 is shifted to the left by 2 units.
y

y = (x + 2) 2 y = (x − 1) 2 y = x2 y = (x − 3) 2

x
FIGURE 4.58 –2 1 3

◆
Gord.3896.04.pgs 4/24/03 9:57 AM Page 322

322 CHAPTER 4 Extended Families of Functions

In general, the following principle holds for any function of x.

Horizontal Shift
Replacing x with x  a shifts the graph of f 1x 2 to the right by the amount a 0.
Replacing x with x  a shifts the graph of f 1x 2 to the left by the amount a 0.

Thus, for instance, the graph of y  10x2 has the identical shape as the graph
of y  10x, but is shifted to the right by 2 units. Similarly, the graph of
y  1x  3 has the same shape as the graph of y  1x, but is shifted to the left
by 3 units. Check these and other graphs on your function grapher.
In summary, when we replace x by x  a or x  a, we are changing x and so
produce a horizontal effect. When we replace y with y  b or y  b, we produce a
vertical effect.
When we combine a horizontal shift (replace x by x  a) and a vertical shift
(replace y by y  b), we effectively have a diagonal shift. For example, consider the
graph of y  1x  4 2 2  7, or equivalently y  7  1x  4 2 2. It involves a
change in x (x is replaced by x  4) and a change in y (y is replaced by y  7). So,
y  1x  4 2 2  7 corresponds to shifting the parabola y  x 2 four units to the
right and seven units up. This produces a parabola whose vertex is at 14, 7 2, as
shown in Figure 4.59.
Similarly,
x 2  y2  r 2
is the equation of a circle with radius r centered at the origin (see Appendix A6).
We should then expect that
1x  5 2 2  1 y  3 2 2  r 2
is the graph of a circle with radius r that has been shifted 5 units to the right and 3
units up. It is therefore the equation of a circle with radius r centered at the point
15, 32. The new circle is produced from the original circle by a combination of a
horizontal shift (5 units to the right) and a vertical shift (3 units up), as shown in
Figure 4.60.

y = (x − 4) 2 + 7
y
(x − 5) 2 + (y − 3) 2 = r 2
(4, 7)

P(5, 3)
x +
2 y2 = r2
7 3

x
y = x2 (0, 0) 5

x
(0, 0) 4

FIGURE 4.59 FIGURE 4.60

Gord.3896.04.pgs 4/24/03 9:57 AM Page 323

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 323

Stretching and Shrinking Functions

We can stretch and shrink functions vertically and horizontally. Vertical stretching
and shrinking involves multiplying a function by a constant, whereas horizontal
stretching and shrinking involves multiplying the independent variable by a constant.

A Constant Multiple of a Function We can also create a new function from a

given function by multiplying the function, or equivalently the y-value, by a con-
stant. For example, consider the two functions y  2x and y  5 . 2x. Both are
exponential decay functions, as shown in Figure 4.61. The function y  2x passes
through the point 10, 1 2 , whereas the transformed function y  5 . 2x passes
through the point 10, 5 2, so you might be tempted to think of the second function
as resulting from a vertical shift of the first. However, think about what each looks
like for large values of x; both curves have the x-axis as a horizontal asymptote.
Therefore the relationship between them cannot be a vertical shift. In particular, the
height for every point on the curve y  5 . 2x is five times the height of the corre-
sponding point (with the same value for x) on the curve y  2x. The effect of the
constant multiple 5 is to increase the height all along the curve by a factor of 5. If we
multiply the original function by 20, the curve will be stretched to a new curve that
is everywhere 20 times as tall.
y

y = 5⋅ 2 –x

y = 2 –x 5

1
x
FIGURE 4.61

If instead we multiply the original function by 14 , the curve will shrink to a new
curve that is one fourth the original height for each value of x. Finally, if we multi-
ply the function by a negative constant, such as 3, the curve is stretched by a fac-
tor of 3, but it is also flipped upside down across the horizontal axis. Figure 4.62
shows the graphs of y  1x and y  3 1x . Not only is the graph of the second
function flipped upside down across the x-axis, but it also moves downward much
faster (three times as fast) than the first function rises. Verify this result on your
function grapher with some other functions.
y

2 y = √x
1
x
1 2 3 4
–1
–2
–3
–4 y = –3√x
–5
FIGURE 4.62 –6
Gord.3896.04.pgs 11/21/03 10:35 AM Page 324

324 CHAPTER 4 Extended Families of Functions

In general, we have the following principle.

Vertical Stretching and Shrinking

Multiplying a function by a constant changes the height of its graph by that
multiple, but it does not change the general shape.
If the multiple is greater than 1, the height is increased.
If the multiple is a number between 0 and 1, the height is decreased.
If the multiple is negative, the curve is flipped over across the horizontal
axis.

We illustrate an application of some of these ideas in Example 3.

E XAMPLE 3
Suppose that a chicken is taken from the freezer at 0°F and put directly into an oven kept
at a constant temperature of 350°F. After 30 minutes, the temperature of the chicken is
110°F. Construct a function to model the temperature of the chicken as it cooks in the
oven.
Solution The temperature of the chicken rises rapidly at first and then increases ever
more slowly the closer the chicken’s temperature comes to the oven temperature of 350°.
Eventually, the temperature of the chicken levels off at the temperature setting for the
oven. The temperature T, in °F, plotted against time t, in hours, looks like the graph
shown in Figure 4.63. (This description is actually an oversimplification because the
temperature rise will temporarily stop at the freezing point of 32° while the ice melts.
Also, the chicken should be removed from the oven when its temperature reaches about
180°, or it will begin to burn.)

350°
Temperature (°F)

t
0 1 2 3
FIGURE 4.63 Time (hours)

The horizontal line representing the temperature of 350° is a horizontal asymptote

because the curve gets ever closer to this line as time goes by, but never quite reaches it.
The rate at which the temperature of the chicken increases slows as it approaches 350° (if
we left the chicken in the oven that long), so the curve is concave down.
We want to model this process by creating a formula giving the temperature T as a
function of the time t. Simplistically we will find a mathematical model by inspecting
Gord.3896.04.pgs 4/24/03 9:57 AM Page 325

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 325

the graph of the process and deciding which type of function has the right shape. In Sec-
tion 5.4 we demonstrate how to construct such a function directly.
The graph in Figure 4.63 appears to be an exponential decay function turned upside
down so that it rises toward the oven temperature 350° instead of dropping asymptoti-
cally toward the horizontal axis. We can form such a function from a pure exponential
function y  Ac x by using a negative coefficient (to turn the curve upside down) and a
vertical shift so that the curve approaches 350 instead of 0. Thus a formula for T might
look like
T  350  Ac t,
where t is in hours and 0  c  1. As t increases, the term ct approaches 0, and the en-
tire expression 350  Ac t approaches 350.
What might be possible values for A and c? We know that at time t  0, the chicken’s
temperature is T  0 when it comes out of the freezer, so
T10 2  350  Ac 0  350  A  0.
Thus A  350 and the formula becomes
T  350  3501ct 2.
Furthermore, the temperature of the chicken after half an hour is T1 12 2  110°. This
value yields
T1 12 2  350  3501c1>2 2  110.
So we have
3501c1>2 2  350  110  240;
c1>2  240>350  0.686.
Squaring both sides of this equation gives
c  0.47.
Consequently, our formula for the temperature becomes
T  350  35010.47 2 t  350 31  10.47 2 t 4,
where t is measured in hours.
This function is an upside down exponential: As t increases, 10.472 t gets ever small-
er, so 1  10.47 2 t increases and gets ever closer to 1. That is, 1  10.47 2 t S 1 as t S .
Consequently,
T  350 31  10.472 t 4 S 350, as t S ,
confirming that the graph has a horizontal asymptote at T  350.
◆
Think About This Verify the behavior of the preceding function on your function grapher. Look at
the overall shape and then zoom in to verify the height of the asymptote. Estimate
by eye from the graph when T reaches 180°, when it reaches 250°, and when it
reaches 300°, 340°, and 349°. ❐

In general, consider the function y  f 1t2  L  Ac t, where 0  c  1. We

know that as t increases, ct decays toward zero so that the function approaches a
limiting value of L. The question is: How does it approach L—from above or from
Gord.3896.04.pgs 4/24/03 9:58 AM Page 326

326 CHAPTER 4 Extended Families of Functions

below? First, whenever A  0, the values of the function are less than L. As the
term Act decreases, the amount subtracted from L decreases, and the values of the
function increase toward L in a concave down manner, as illustrated by the upper
curve in Figure 4.64. This was the case with the temperature of the chicken in the
oven. Second, whenever A 0, the values of the function are greater than L and
so decrease toward it in a concave up manner, as illustrated by the lower curve in
Figure 4.64.

A>0
L
A<0

FIGURE 4.64 t

This type of function is used as the mathematical model for many different
real-world processes.

A Constant Multiple of the Independent Variable Finally, we investigtate the ef-

fects of multiplying the independent variable x by a constant.

E XAMPLE 4
Consider the cubic function y  f 1x 2  x 3  12x and the related functions y  f 12x2,
y  f 14x 2, and y  f 1 12 x 2 . What is the effect of the constant multiple in each case?
Solution Figure 4.65 shows the graphs of y  f 1x 2  x 3  12x and y  f 12x 2 
12x2  1212x2. The cubic y  f 1x 2  x 3  12x passes through the origin and has
3

two turning points. If you trace along the curve, you will find that one turning point
is at x  2 and the other at x  2. (We could also locate the turning points by using
the formula presented in Section 4.4.) The corresponding local maximum (at
x  2) is at a height of y  16 and the local minimum (at x  2) is at a height of
y  16.
The cubic y  f 12x 2  12x 2 3  1212x 2  8x 3  24x also passes through the ori-
gin and has two turning points, one at x  1 and the other at x  1. The correspon-
ding local maximum is at y  16, and the local minimum is at y  16. Hence the
heights are the same; they just occur sooner. In fact, the curve for y  f 12x 2 traces out
the identical vertical values as f 1x 2, but does so twice as fast.
Figure 4.66 shows the graphs of y  f 1x 2  x 3  12x and y  f 14x 2  14x 2 3 
1214x2  64x 3  48x. The local maximum for y  f 14x 2 now occurs at x   12 and the
local minimum occurs at x  12 . Again, the same heights are achieved, but the curve
y  f 14x 2 is traced out four times as fast as y  f 1x 2 .
Figure 4.67, shows the graphs of y  f 1x 2  x 3  12x and y  f 1 12 x 2  1 12 x 2 3 
121 2 x2 18 x 3  6x, but we had to extend the window to show the details. The function
1

y  f 1 12 x2 achieves its local maximum at x  4 and its local minimum at x  4. The

curve y  f 1 12 x2 traces out the identical heights, but does so half as fast as y  f 1x 2.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 327

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 327

y y y

50 50 50 f (x)
40 f (2x) 40 f (4x) 40
(–1, 16) 30 (– 1
2, )
16 (–2, 16) f ( 12 x)
(–2, 16) 20 f (x) (–2, 16) f (x) (–4, 16) 20
10 10
x x x
–4 –3 –2 –1 1 2 3 4 –4 –3 –2 –1 1 2 3 4 –8 –6 –4 –2 2 4 6 8
–10 –10 –10
–20 (2, –16) –20 (2, –16) –20 (4, –16)
–30 (1, –16) –30 –30 (2, –16)
–40 –40
( 1
2, 16 ) –40
–50 –50 –50

FIGURE 4.65 FIGURE 4.66 FIGURE 4.67

◆
We summarize the ideas on stretching and shifting functions horizontally as
follows.

Horizontal Stretching
Multiplying the independent variable x by a constant k changes the speed at
which the graph is traced out, but it does not change the general shape.
If the multiple k is greater than 1, the graph of y  f 1kx 2 is traced out k
times faster than y  f 1x2.
If the multiple k is between 0 and 1, the graph of y  f 1kx2 is traced out
more slowly than y  f 1x 2 .
If the multiple k is negative, then the curve y  f 1kx 2 is reflected across
the y-axis.

E XAMPLE 5
For the function f 1x 2  x 3  12x, draw the graph of f 13x 2 and locate its turning
points.
Solution Figure 4.68 shows the graphs of the two functions. The graph of y  f 13x 2
has the same basic shape as the graph of y  f 1x 2, but is flipped upside down across the

25
(–2, 16) 20 ( 23 , 16)
15
10
f (–3x) f (x)
5
x
–4 –3 –2 –1 1 2 3 4
–5

(– 23 , –16)–20 (2, –16)

FIGURE 4.68 –25
Gord.3896.04.pgs 4/24/03 9:58 AM Page 328

328 CHAPTER 4 Extended Families of Functions

x-axis and is traced out 3 times as fast. As before, the turning points for y  f 1x 2 are at
12, 16 2 and 12, 16 2. The turning points for y  f 13x 2 are at 1 23 , 16 2 , a local
minimum, and at 1 23 , 16 2 , a local maximum.
◆

E XAMPLE 6
A function f is defined in the following table. Use the values given in the table to com-
plete it. If any entries are not defined, mark them “UNDEF.”

x f (x) f (x)  1 f (x  1) f (2x) 3f (x)

0 1

1 0
2 3

3 2

Solution The values of the function for x  0, 1, 2, and 3 are defined in the table. The first
open column asks for a vertical shift when the function’s values are reduced by 1 for each
value of x. For instance, when x  0, the first entry is asking for f 10 2  1  1  1  0,
and so on down that column.
The second open column asks for a horizontal shift of 1 unit to the right, because x
is replaced by x  1. Thus, when x  0, we want f 10  1 2  f 11 2 , but there is no
way to determine this value from the information given in the table; that is, the function
is not defined for x  1, so we record it in the table as “UNDEF.” However, when x  1,
we want f 11  1 2  f 10 2  1, and so on down the column.
The third open column asks for values when the independent variable is doubled. So,
when x  0, we need f 12 . 02  f 10 2  1; similarly, when x  1, we need
f 12 . 12  f 122  3. However, when x  2, f 12 . 2 2  f 14 2, which is not defined. When
x  3, f 12 . 32  f 16 2 is also not defined.
Finally, the last open column asks for 3 times the value of the function. When x  0,
we need 3 . f 102  3 . 1  3, and so on down the column. The completed table follows.

x f (x) f (x)  1 f (x  1) f (2x) 3f (x)

0 1 0 UNDEF 1 3

1 0 1 1 3 0

2 3 2 0 UNDEF 9

3 2 1 3 UNDEF 6
◆
Gord.3896.04.pgs 4/24/03 9:58 AM Page 329

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 329

Problems

x f (x) 5f (x) f (x)  3 f (x  1) [ f (x)]2

3 5

4 2

5 1

6 3

7 8

1. A function f is defined in the table above. Use the 3 3

values given to complete the table. If any of the en-
tries are not defined, write “UNDEF.” x x
2. A function y  f 1x2 is defined by the accompany-
–4 –2 2 4 –4 –2 2 4

ing graph. Match each transformation of f with one –3

of the graphs (i)–(vi).
a. y  f 12x2 b. y  2f 1x 2
(v) (vi)

c. y  f 1x2  2 d. y  f 1x  2 2 3. For the functions f and g that are defined by the

e. y  f 1x2  2 f. y  f 1x  2 2 graphs shown, sketch the graph of
a. 2g1x2 . b. g12x 2.
c. f 1x  1 2 . d. f 1x  1 2 .
y

f. g a xb .
3 1
e. f(x)+1.
2 2
1 y = f (x) y
x 3
–4 –3 –2 –1 1 2 3 4
–1

–2 1
x
–3 –3 –1 1 2 3
–1
f (x)
y y –2
–3
3 3
y

x x 3
–4 –2 2 4 –4 –2 2 4
2
–3 –3 g(x)
1
(i) (ii) x
y y –3 –2 –1 1 2 3
–1
3 3 –2
–3
x x

4. Consider the function y  f 1x 2  x 2.

–4 –2 2 4 –4 –2 2 4

–3 –3
a. Write an equation for the function that you get
(iii) (iv) when you stretch the graph of f by a factor of 2
Gord.3896.04.pgs 4/24/03 9:58 AM Page 330

330 CHAPTER 4 Extended Families of Functions

and then shift it up 3 units. Call this new func- c.

tion F and sketch its graph. y
b. What is the equation you get if you reverse the
10
order of the two operations in part (a)? Call this
y = f (x)
new function G and sketch it.
c. What is F  G?
5. a. Translate the line y  mx to a line with slope m
that passes through the point 15, 122.
2
x
b. Repeat part (a) if the new line passes through the
point 1x0, y0 2. What do you call this new equation? 9. Consider the function f in the table. If this function
6. a. Translate the parabola y  x 2 to a parabola with is shifted 4 units to the right and 7 units upward,
vertex at 15, 122. construct the corresponding table for the trans-
b. Repeat part (a) if the new parabola has its vertex formed function.
at the point 1x0, y0 2.
7. For the function f shown, sketch the graph of x 1 0 1 2 3
a. y  f 1x2 b. y  2f 1x2
c. y  f 1x2  1 d. y  f 1x  1 2
y 24 16 11 8 9
e. y  f 1x  12 f. y  f 1x2  1 x 4 5 6 7

y 15 27 39 35

10. a. Use the graphs of f 1x 2  x and g 1x 2  log x to

y

1
sketch a rough graph of the product P1x 2  x log x.
y = f (x) b. Estimate the values of x for which log x  x and
the values for which log x x.
x
0 1 2 3 4 c. Because log x grows exceedingly slowly, the prod-
uct x log x grows only slightly faster than x does.
Use your function grapher to decide whether
–1 x log x ever grows faster than x 1.5, than x 1.1, than
x 1.05. What does this investigation suggest to you
about the rate of growth of x log x compared to
8. The graphs of three functions (a)–(c) are shown in power functions x p?
the accompanying figures. Sketch the graph of 11. If f 1x 2  x 2  3x  4 and h is a constant, find
(i) y  f 1x2 (ii) y  2f 1x 2 a. f 1x 2  h b. f 1x  h2
(iii) y  2f 1x2 (iv) y  f 1x  2 2 f 1x  h2  f 1x 2
c. f 1x  h2  f 1x 2
(v) y  f 1x2  2 (vi) y  f 1x2  2
d.
h
(vii) y  f 1x  22 e. What is the value of the expression in part (d)
if x  5 and if h  0.1? if h  0.01? if
h  0.0001?
a. b.
12. a. An unbaked apple pie is taken from the counter
y y
in a kitchen where the temperature is 70°F and
10 placed in an oven. Suppose that, after 60 min-
y = f (x) y = f (x) utes, the temperature of the pie is 180°F. Sketch
a graph of the temperature of the pie as a func-
tion of time.
2 b. The pie is removed from the oven and placed
(2, 1)
x
back on the counter. Suppose that it takes anoth-
er 60 minutes for its temperature to come back
Gord.3896.04.pgs 4/24/03 9:58 AM Page 331

4.7 Building New Functions from Old: Shifting, Stretching, and Shrinking 331

down to 70°F. Sketch a graph of the temperature a. Based on this graph, sketch the graph of the total
of the pie as a function of time. national debt as a function of time.
c. When the first pie is removed from the oven, a b. Does your graph have any points of inflection? If
second, unbaked pie is put in the oven to bake. so, what do they represent?
Sketch a graph of the sum of the temperatures c. Do you agree or disagree with the president’s as-
of the two pies as a function of time over the sertion that the war has been won? Explain.
60-minute period. 15. Use your function grapher to graph the functions
d. Find a formula that models the temperature of f 1x 2  x n 10.5 2 x, for n  1, 2, 3, 4, 5, and estimate
the pie, while it cools, as a function of time. the location of the turning point for each curve for
13. A Thanksgiving turkey is taken from the refrigera- x 0. Then perform a linear regression analysis on
tor at a temperature of 40°F and placed in a hot the x-values of these turning points, as functions of
oven at 350°F to cook. After 1 hour, the internal n. Is the linear fit appropriate? What does it predict
temperature of the bird is 124°F. Write a possible for n  1.5? Is it accurate compared to the actual
formula for the temperature of the turkey as a func- graph?
tion of time, in minutes. 16. Use your function grapher to graph the functions
14. In an attempt to claim responsibility for winning f 1x 2  x 2ax, for a  0.3, 0.4, 0.5, 0.6, and 0.7. Esti-
the war against the growing national balance of mate the location of the turning point for each
trade deficit, the president presented a graph simi- curve by zooming in on it. Then determine the func-
lar to the one shown to illustrate the trend in the tion from among the usual families of functions—
annual deficit. linear, exponential, and power—that best fits these
data as a function of the base a.
17. Describe how you might use the results of Problems
y
15 and 16 to find a function of the form
f 1x 2  x pax that matches the function for the level
of Lyme disease antibody in the bloodstream dis-
cussed in Section 4.6 (see Figure 4.50).
18. Find conditions on the coefficients a, b, and c in
Deficit

P1x 2  ax 2  bx  c if P is to satisfy each equation

for all values of x.
a. P1x 2  P1x2 b. P1x 2  P1x2
c. P12x 2  2P1x 2
x
1990 1991 1992 1993 1994
Year

Exercising Your Algebra Skills

For the function f 1x2  x 2  5x  3, find a simplified 3. f 14x2. 4. f a xb .
1
expression for 2
1. f 12x2. 2. f 13x 2. 5. f 1x  12. 6. f 1x  22.
7. f 12x  12. 8. f 1x 2 2.

4.8 Using Shifting and Stretching to Analyze Data

The ideas on shifting and stretching functions in Section 4.7 can be applied to cre-
ate functions that fit sets of data that do not quite fall into the standard behavior
patterns, such as exponential growth or decay, that we have discussed.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 332

332 CHAPTER 4 Extended Families of Functions

Analyzing a Cooling Experiment

Suppose that an experiment is conducted to study the rate at which temperature
changes. A temperature probe (a thermometer connected to a calculator) is first
heated in a cup of hot water and then removed and placed in a cup of cold water, as
illustrated in Figure 4.69. The temperature of the probe, in °C, is measured every
second for 36 seconds and recorded in Table 4.3; the data are also displayed in the
scatterplot in Figure 4.70.

TABLE 4.3 Experimental Data: Temperature 1°C2 versus Time

Time Temperature Time Temperature Time Temperature
1 42.3 13 12.51 25 9.29
Hot Cold 2 36.03 14 11.91 26 9.16
3 30.85 15 11.54 27 9.16
4 26.77 16 11.17 28 9.04
FIGURE 4.69 5 23.58 17 10.67 29 8.91
6 20.93 18 10.42 30 8.83
7 18.79 19 10.17 31 8.78
8 17.08 20 9.92 32 8.78
9 15.82 21 9.8 33 8.78
10 14.77 22 9.67 34 8.78
11 13.82 23 9.54 35 8.66
12 13.11 24 9.42 36 8.66

45
40
Temperature (°C)

35
30
25
20
15
10
5
t
0 4 8 12 16 20 24 28 32 36
Time (seconds)

FIGURE 4.70

E XAMPLE 1
Find a function that fits the data from the temperature cooling experiment.
Solution The pattern depicted in Figure 4.70 is that of a decreasing, concave up func-
tion, so we might consider either a decaying exponential function or a power function
with p  0. However, a power function is not a good model for the process because it has
a vertical asymptote at time t  0, whereas the function we want must have a finite value
when t  0. So the more appropriate model would be an exponential decay function.
But there is a catch. Any exponential decay function decreases to zero, but the tem-
perature readings decay to the temperature of the cold water (which cannot be 0°C, for
then the water would be frozen). From the experimental data, the temperature of the
Gord.3896.04.pgs 4/24/03 9:58 AM Page 333

4.8 Using Shifting and Stretching to Analyze Data 333

cold water is about 8.6°C. How do we construct a function that decays to about 8.6
rather than to 0? Probably the most reasonable approach is to subtract 8.6 from each of
the temperature readings to obtain a new set of data that decays to zero. This approach is
equivalent to performing a vertical shift downward of 8.6 (i.e., replacing the temperature
T with T  8.6) to produce the transformed data shown in Table 4.4.

TABLE 4.4 Transformed Data: 1T  8.62 versus Time

Time T  8.6 Time T  8.6 Time T  8.6
1 33.7 13 3.91 25 0.69
2 27.43 14 3.31 26 0.56
3 22.25 15 2.94 27 0.56
4 18.17 16 2.57 28 0.44
5 14.98 17 2.07 29 0.31
6 12.33 18 1.82 30 0.23
7 10.19 19 1.57 31 0.18
8 8.48 20 1.32 32 0.18
9 7.22 21 1.20 33 0.18
10 6.17 22 1.07 34 0.18
11 5.22 23 0.94 35 0.06
12 4.51 24 0.82 36 0.06

The scatterplot of the transformed data, shown in Figure 4.71, looks like an expo-
nential decay pattern that tends toward 0. Using a calculator, we find that the exponen-
tial function that best fits the transformed data is
y  T  8.6  35.439410.8482 t.
y

45
40
35
T – 8.6 °C

30
25
20
15
10
5
t
0 4 8 12 16 20 24 28 32 36
FIGURE 4.71 Time (seconds)

The graph of this function is shown superimposed over the transformed data in Figure 4.71,
and there appears to be extremely close agreement. The corresponding correlation coeffi-
cient is r  0.9948, which is very close to 1.
Having found the exponential function that best fits the transformed data, we now
have to undo the transformation. We simply add the same amount, 8.6, to the function
y  T  8.6 to create the final expression
T1t2  8.6  35.439410.8482 t.
This function is shown superimposed over the original temperature data in Figure 4.72,
and it is an exceptionally good fit to the temperature readings. In particular, note how
this function approaches the limiting value of about 8.6 for the temperature readings as
t increases.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 334

334 CHAPTER 4 Extended Families of Functions

45
40

Temperature (°C)
35
30
25
20
15
10
5
t
0 4 8 12 16 20 24 28 32 36
FIGURE 4.72 Time (seconds)

Analyzing the Challenger Data

At the beginning of Chapter 3, we considered data relating the number of incidents
involving O-ring problems on space shuttle launches to the air temperature at
launch. These data eventually were used to identify the O-rings as the likely cause
of the Challenger disaster. We now use this set of data as a case study to illustrate
the process of data analysis when it is necessary to shift the data values.
Recall that the dependent variable was the number N of O-ring problems or
“incidents” as a function of launch temperature T. The data are shown in the fol-
lowing table.

T 53 57 58 63 66 67 67 67 68 69 70 70

N 3 1 1 1 0 0 0 0 0 0 1 1

T 70 70 72 73 75 75 76 76 78 79 80 81

N 0 0 0 0 2 0 0 0 0 0 0 0

Figure 4.73 shows the scatterplot for this data along with a curve superimposed
over the data points to indicate the nature of the relationship, which appears to be
a decaying exponential. However, this curve is only an artist’s rendering of the ap-
parent relationship. We want to obtain a formula for such a function.

4
Number of O-rings affected

0
30 40 50 60 70 80 90
FIGURE 4.73 Temperature (°F)
Gord.3896.04.pgs 4/24/03 9:58 AM Page 335

4.8 Using Shifting and Stretching to Analyze Data 335

E XAMPLE 2
Find a function that can model the data on the number of O-ring incidents as a function
of the air temperature.
Solution The decreasing, concave up pattern in the scatterplot in Figure 4.73 suggests
either a decaying exponential function or a power function with p  0. The power func-
tion does not make sense, however, because there is no vertical asymptote at T  0. Fit-
ting an exponential function to the set of data by using the transformation approach
used by calculators and spreadsheets involves plotting the logarithm of the number of
incidents log N versus the temperature T. But because the values for N include N  0,
we cannot take the logarithm of 0—it is not defined!
One way to circumvent this problem is to shift the data values up to avoid the zeros.
The simplest approach is to increase each value of N by 1, replacing N by N  1 and
then comparing N  1 to T. We first construct the exponential function that best fits the
resulting set of data to obtain the exponential regression equation relating N  1 to T.
We then shift back down to obtain an expression for N in terms of T. The data values
that we work with are given in the following table, and the associated scatterplot of
N  1 versus T is shown in Figure 4.74.

T N N1 T N N1
53 3 4 70 0 1

57 1 2 70 0 1

58 1 2 72 0 1

63 1 2 73 0 1

66 0 1 75 2 3

67 0 1 75 0 1
67 0 1 76 0 1

67 0 1 76 0 1

68 0 1 78 0 1
69 0 1 79 0 1

70 1 2 80 0 1

70 1 2 81 0 1

The resulting exponential regression equation giving N  1 as a function of T is

N  1  13.4110.9672 T,
which is shown superimposed on the scatterplot, in Figure 4.74. Finally, we solve for N by
subtracting 1 from both sides to get the exponential function that can be used to model N
as a function of T:
N  13.4110.9672 T  1,
which is shown superimposed over the original scatterplot in Figure 4.75. It appears to
capture the trend in the data reasonably well.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 336

336 CHAPTER 4 Extended Families of Functions

N+1
4
N
3.5
3
3

Number of O-rings affected

2.5
2.5
2
2
1.5
1.5
1
1
0.5
0.5
0 T
50 55 60 65 70 75 80
T
0 50 55 60 65 70 75 80 85 –0.5
Temperature (°F) Temperature (°F)

FIGURE 4.74 FIGURE 4.75

◆
The graph certainly suggests that the likelihood of trouble with the O-rings will
increase dramatically with falling temperature. However, we know that there is a dan-
ger in extrapolating far beyond the range of data values. But the overall trend is so dra-
matic and the potential loss in terms of both human life and hardware is so extreme
that there shouldn’t have been a launch if the data had been analyzed in this way.

Terminal Velocity in Skydiving

Matthew is a skydiving enthusiast. He knows, from his reading and from first-hand
experience, that the faster he is falling, especially in a spread-eagled position, the
greater the air resistance, so that eventually his speed reaches a maximum, known
as the terminal velocity. He also found the following set of data on the downward
velocity v, in feet per second, of a skydiver at different times t, in seconds, after
jumping out of a plane.

t 0 1 2 3 4 5 6 7 8 9 10 11 12

v 0 16 46 76 104 124 138 148 156 163 167 171 174

Source: Student project.

E XAMPLE 3
(a) Find a function that fits these data on velocity as a function of time from among the
usual candidates. (b) Improve on the fit by using an appropriate shift.
Solution

a. The data falls in an increasing, concave down pattern, as shown in Figure 4.76. The
potential candidates for a function having such a pattern are either a power function
with 0  p  1 or a logarithmic function. However, a log function is not defined at
t  0. Also, both functions grow indefinitely, while the values for the skydiver’s ve-
locity approach terminal velocity, which is a horizontal asymptote. What’s worse, we
don’t know what this limiting value for the terminal velocity is. Thus neither function
can be a good fit.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 337

4.8 Using Shifting and Stretching to Analyze Data 337

Velocity (feet per second)

180
150
120
90
60
30
t
0 2 4 6 8 10 12
FIGURE 4.76 Time (seconds)

Moreover, we cannot use a calculator or spreadsheet program to fit a power func-

tion to the data because of the first point 10, 0 2 —their regression routines all use the
transformation approach, which involves having to take the log of 0. However, if we
delete the point 10, 0 2, we can fit a power function to the remaining data. Figure 4.76
shows the graph of the best-fit power function v  23.2t 0.908 (obtained using a cal-
culator) superimposed over the scatterplot of the data. The corresponding correla-
tion coefficient is r  0.962, which is fairly close to 1. The power function is a
reasonable fit, but it clearly becomes less good when extended to the right.
b. The pattern in the data suggests an upside down exponential decay function that
rises toward a horizontal asymptote. Suppose that we conjecture a value for the ter-
minal velocity by mentally extending the preceding table. We might extrapolate that
the limiting value is about 180 ft>sec. We then subtract each velocity value from this
supposed limit (replacing v with 180  v) to obtain the transformed data shown in
the following table. Effectively, this transformation is equivalent to a vertical shift
with a flip across the horizontal axis due to the multiple of 1.

t 0 1 2 3 4 5 6 7 8 9 10 11 12
180  v 180 164 134 104 76 56 42 32 24 17 13 9 6

180 – v
180 – v (feet per second)

180
150
120
90
60
30
t
0 2 4 6 8 10 12
FIGURE 4.77 Time (seconds)

The scatterplot of these transformed data is shown in Figure 4.77. The decreas-
ing, concave up pattern in this transformed data suggests either a decaying exponen-
tial function or a power function with p  0; however, the latter has a vertical
asymptote at zero, so it is not an appropriate candidate. A calculator gives the expo-
nential function that best fits this transformed data as
y  180  v  226.2510.74922 t
with a correlation coefficient of r  0.9963. Figure 4.78 shows this function super-
imposed over the transformed data, and it is a very good fit.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 338

338 CHAPTER 4 Extended Families of Functions

We undo the transformation algebraically by solving for the velocity v to get

v  180  226.2510.74922 t.
The graph of this function is shown superimposed over the original data in Figure 4.79,
demonstrating a much better fit than the power function in Figure 4.76. This conclu-
sion is further borne out by the correlation coefficient r  0.9963, which is consid-
erably closer to 1 than the correlation coefficient r  0.962 for the power function
was to 1.

180 – v v

Velocity (feet per second)

180 – v (feet per second)

180 180
150 150
120 120
90 90
60 60
30 30
t t
0 2 4 6 8 10 12 0 2 4 6 8 10 12
FIGURE 4.78 Time (seconds) FIGURE 4.79 Time (seconds)

The value of 180 ft>sec that we chose for the terminal velocity was reasonable,
but it was just an intelligent guess. Had we chosen a somewhat different value, we
would have obtained a somewhat different function. With a little experimentation,
you should be able to get a still better fit.
In Example 3, an exponential function was a very good fit to the transformed
data, although the values for 180  v did not fall precisely in an exponential decay
pattern. Sometimes, a set of values fall precisely in an exponential pattern as they
grow or decay toward a horizontal asymptote. The problem we face in such cases is
not knowing exactly what that horizontal asymptote is, as was the case in Example 3.
If the transformed data do fall in an exponential pattern, we can determine the
limiting value precisely.
Suppose that a set of values x0 , x1 , x2 , x3 , x4 , . . . is such that the values either
fall toward an unknown limiting value L or rise toward L in a purely exponential
manner, as shown in Figure 4.80. In particular, suppose that each of the values is
below the unknown horizontal asymptote L, so that the set of transformed values
L  x0 , L  x1 , L  x2 , L  x3 , L  x4 , . . .
xn

n
FIGURE 4.80

decays toward zero in an exponential decay pattern. As a result, we know that the
successive ratios should be a constant, say k. That is,
Gord.3896.04.pgs 4/24/03 9:58 AM Page 339

4.8 Using Shifting and Stretching to Analyze Data 339

L  x1 L  x2 L  x3
   . . .  k,
L  x0 L  x1 L  x2
where k is the constant, although unknown, ratio. Consider the first of these
equalities:
L  x1 L  x2
 .
L  x0 L  x1
We can solve this equation algebraically for the unknown limiting value L by
first cross-multiplying to get
1L  x1 2 2  1L  x0 2 1L  x2 2 .
Expanding these terms gives
L2  2x1L  x12  L2  x0L  x2L  x0 x2 .
Subtracting L2 from both sides of this equation and then collecting like terms yields
1x0  2x1  x2 2 L  x0x2  x1 2 ,
so that
x0 x2  x12
L , (1)
x0  2x1  x2
provided that the denominator x0  2x1  x2  0. In fact, if the numbers x0 , x1,
x2 , x3 , x4 , . . . approach L in an exponentially decaying manner precisely, the com-
parable expression—using any three successive values of the x’s, not just the first
three—gives the same value for L. If the values are not exact, however—even if the
discrepancies are due to rounding—quite different values could arise with every
group of three successive values for the x’s.
E XAMPLE 4
Prozac is prescribed for individuals suffering from depression. Typically, a patient takes
a dose of Prozac once a day and, for extreme depression, the dosage is 80 mg. The levels
of Prozac in the blood on successive days following the start of treatment are given in the
following table. (Note that the last two values are rounded to four decimal places.) It
turns out (we investigate this result in detail in Section 5.1) that the level of Prozac P
rises toward a horizontal asymptote in a precisely upside down exponential decay man-
ner as a function of the number of days n. Find the value of this horizontal asymptote,
assuming that the course of treatment continues.

n 0 1 2 3 4 5 6 ...

P 80 140 185 218.75 244.0625 263.0469 277.2852 ...

Solution We start with a scatterplot of the data, as shown in Figure 4.81, where the
points appear to be approaching a horizontal asymptote at a level somewhat above
300 mg. We call this level L.
These values fall in an upside down decaying exponential pattern as they rise toward
the horizontal asymptote, so we can use Equation (1) with the first three values x0  80,
x1  140, and x2  185 to find that
x0 x2  x12 801185 2  1402
L   320.
x0  2x1  x2 80  21140 2  185
Gord.3896.04.pgs 4/24/03 9:58 AM Page 340

340 CHAPTER 4 Extended Families of Functions

P
320

Level of Prozac (mg)

280
240
200
160
120
80
40
t
0 5 10 15 20
FIGURE 4.81 Number of Days

If instead we use the second, third, and fourth values, so that x0  140, x1  185, and
x2  218.75, we obtain
x0 x2  x12 1401218.752  1852
L   320.
x0  2x1  x2 140  21185 2  218.75
If we use the last three values shown, so that x0  244.0625, x1  263.0469, and
x2  277.2852, we obtain, in the same way, L  320.0001. As you will see in Section 5.1
when we develop a complete mathematical model for the level of Prozac in the blood,
the limiting value is 320 mg.
◆

Horizontal Shifts
We next consider some applications involving horizontal shifts. As Example 5
demonstrates, that’s just what we’ve been doing when we changed the scale in the
independent variable.

E XAMPLE 5
The following data fall in a linear pattern. Determine the line that passes through the
points (a) when t represents the number of years since 1980; (b) when t represents the
number of years since 1900; (c) when t represents the number of years since year 0.
(d) Explain how the three expressions compare by using ideas on shifting functions.

t 1980 1985 1990 1995 2000

y 30 40 50 60 70

Solution

a. We rescale the values of the independent variable so that t represents the number of
years since 1980.

t 0 5 10 15 20

y 30 40 50 60 70

Note that each 5 years, the value of y increases by 10, so we have a line with slope

y 10
m   2.

t 5
Gord.3896.04.pgs 4/24/03 9:58 AM Page 341

4.8 Using Shifting and Stretching to Analyze Data 341

The equation of the line then is

y  30  21t  0 2, t  number of years since 1980.

b. We now rescale the values in the table so that t represents the number of years since
1900.

t 80 85 90 95 100

y 30 40 50 60 70

These data values also lie on a line whose slope is m  2, so the equation of the line is
y  30  21t  80 2, t  number of years since 1900.

c. Finally, we use the original values given in the table where t represents the number of
years since the year 0. The slope is still m  2, so the corresponding equation of the
line is
y  30  21t  19802, t  number of years since year 0.

d. We now compare the three equations. In each case, the slope is m  2 because all three
lines increase at the same rate. If we expand all the equations to put them in slope–
intercept form, we get
y  2t  30, y  2t  130, and y  2t  3930,

respectively. Note the great differences in the vertical intercepts for the three lines.
Let’s focus on the equation in part (a), y  30  2t, as a baseline, where t rep-
resents the number of years since 1980. We first compare it to the equation in part (b),
y  30  21t  80 2 . The second equation is the result of a horizontal shift to the
right of 80 years—moving from a “starting point” of t  0 in 1980 to a “starting
point” of t  0 in 1900. Similarly, compare the first equation to the third equation in
part (c), y  30  21t  1980 2, which involves a horizontal shift of 1980 years to
the right. So scaling the values of the independent variable is equivalent to a horizon-
tal shift by the amount of the scaling.
◆
Let’s look at a more realistic example to see how these ideas on horizontal
shifts apply when we fit an exponential function to data.

E XAMPLE 6
The following table shows the growth, in millions, of cellular phone users since 1985.
Find the exponential function that best fits these values (a) when t represents the
number of years since 1985; (b) when t represents the number of years since 1900;
(c) when t represents the number of years since year 0. (d) Explain how the three ex-
pressions compare, using ideas on shifting functions.

t 1985 1988 1990 1991 1992 1993 1994 1995 1996 1997 1998
C 1 4 11 16 23 34 55 91 142 215 319
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That Are Shaping Our Future.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 342

342 CHAPTER 4 Extended Families of Functions

Solution

a. We first scale the years so that t represents the number of years since 1985, giving the
transformed set of data.

t 0 3 5 6 7 8 9 10 11 12 13

C 1 4 11 16 23 34 55 91 142 215 319

A calculator gives the exponential function that best fits the data as
C  1.06311.5552182 t, t  number of years since 1985.
The corresponding correlation coefficient is r  0.99925.
b. We next scale the years in the original data so that t represents the number of years
since 1900.

t 85 88 90 91 92 93 94 95 96 97 98

C 1 4 11 16 23 34 55 91 142 215 319

Again, a calculator gives the exponential function that best fits the modified data as
C  5.2999  1017 11.5552182 t, t  number of years since 1900.
The corresponding correlation coefficient again is r  0.99925.
c. Finally, the exponential function that best fits the original data where t represents the
number of years since the year 0 is
C  2.092  10381 11.5552182 t, t  number of years since year 0.
The corresponding correlation coefficient once more is r  0.99925.
d. The growth factor, 1.555218, is the same in all three expressions. It indicates that
the use of cellular phones is growing, on average, by 55.5% per year, whichever
model we construct. The correlation coefficient r  0.99925 is also the same in all
three models. It indicates that the fit in all three cases is equally excellent. Only the
constant coefficient changes from one expression to the next, and it reflects the
vertical intercept of each curve.
We now look at the equation for the exponential function C  1.06311.5552182 t in
part (a), where t represents the number of years since 1985. If we perform a horizontal
shift of 85 years to the right so that t represents the number of years since 1900, the for-
mula for the function becomes
C  1.06311.5552182 t85  1.06311.5552182 t # 11.5552182 85 buv  bubv
 31.063  11.5552182 85 4 11.5552182 t
 5.299988  1017 11.5552182 t,
which is virtually identical to the expression in part (b).
Similarly, if we perform a horizontal shift of 1985 to the right in the equation in
part (a), so that t represents the number of years since the year 0, the formula for the
exponential function becomes
C  1.06311.5552182 t1985  1.06311.5552182 t # 11.5552182 1985 buv  bubv
 31.063  11.5552182 1985 4 11.5552182 t
 2.0934  10381 11.5552182 t,
Gord.3896.04.pgs 4/24/03 9:58 AM Page 343

4.8 Using Shifting and Stretching to Analyze Data 343

which again is virtually identical to the expression C  2.092  10381 11.5552182 t in

part (c). (The differences are due to rounding.)
◆
This principle—scaling the values of the independent variable is equivalent to
a horizontal shift in the function—applies to the linear, exponential, and polyno-
mial families of functions, but it does not apply to power functions. Let’s see why.
Recall that an increasing power function always passes through the origin, and
that a decreasing power function always has a vertical asymptote at 0. When we fit
a power function to a set of increasing data, the origin is automatically added as an
extra point. Suppose that we now scale the values of the independent variable x to
form a new independent variable X. When we then attempt to fit a power function
to the transformed values, a different “origin” is added automatically. This new
“origin” for X is much closer to the shifted data set than the origin for the original
data was, as illustrated in Figure 4.82.

y y

x X

FIGURE 4.82 (a) Original x's (b) Shifted x's

Consider, for instance, the following data.

x 1 2 3 4
y 1 4 9 16

Clearly, these are points on the curve y  x 2 and, if we applied a power function
regression routine, that is precisely the equation we would get. This curve certainly
passes through the origin 10, 0 2 for the original variable x. It also passes through
each of the data points, as shown in Figure 4.83.

20

15

10

x
FIGURE 4.83 0 1 2 3 4 5

Let’s now shift the data horizontally to the right by 10 units to get the corre-
sponding table of values for the new variable X  x  10.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 344

344 CHAPTER 4 Extended Families of Functions

X 11 12 13 14

y 1 4 9 16

If we apply a power function regression routine, we get the function y 

1.496  1012X 11.43. This function passes through the new origin for X, as shown
in Figure 4.84. But it misses all the data points. The resulting curve has been flat-
tened enormously to force the new origin to become a point on the graph. As a re-
sult, the power for this transformed function is 11.43 rather than 2 for the original
function y  x 2. Recall that, for power functions y  x p, the higher the power p,
the flatter the curve as it passes through the origin. Thus the resulting new power
function is distorted compared to the original power function, reflecting the differ-
ent origin.
y

20

15

10

x
FIGURE 4.84 0 3 6 9 12 15

As we mentioned in Chapter 3, if we use different scales for the independent vari-

able with a power function, not only does the appearance of the function change, but
also, and far more important, the predictions based on using the different forms differ.
Let’s look at what happens when we use the two preceding functions to predict the next
value in each table. In the first case, y  x 2. When x  5, we get y  25. In the second
case, y  1.496  1012X 11.43. The corresponding value of X is X  x  10  15,
and we get y  41.462, a dramatically different prediction. If we shifted x by other
amounts, we would get still different predictions each time. The farther we shift from
the original data, the worse this difference gets.
Thus, although power functions are useful, you must use them with extreme
caution and careful thought.

Using Stretches
The ideas on stretching functions from Section 4.7 also have direct application
when we fit functions to data. In Example 1 in Section 3.3, we created the function
P1t2  3.06911.3212 t
to model the growth of the U.S. population from 1780 to 1900, where P is meas-
ured in millions and t is measured in decades since 1780. Actually, we measured P
in millions for convenience. If we count the number of people, the corresponding
function would then be
P1t2  3,069,00011.321 2 t.
Clearly, these two expressions differ by a factor of 1,000,000, and one function is
therefore stretched into the other by this constant multiple of the function.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 345

4.8 Using Shifting and Stretching to Analyze Data 345

Moreover, when we created the function for the U.S. population, it was con-
venient to use t to represent the number of decades since 1780. However, it might
be more meaningful to have a function in which the independent variable repre-
sents the number of years since 1780 instead. If we use the data values for P from
Section 3.3 but count the years t  0, 10, 20, . . . rather than decades, we get the
function
P1 1t2  3.06911.028232 t, t  number of years since 1780,
compared to
P2 1t2  3.06911.321 2 t, t  number of decades since 1780.
How do these two expressions compare? We know that each decade consists of
10 years, which suggests a constant multiple of 10 for the number of years. So, if we
start with the first expression P1 1t2 for the population where t is measured in years
and multiply the independent variable t by 10, we get
P1 110t2  3.06911.028232 10t
 3.0693 11.028232 10 4 t apu  1ap 2 u
 3.06911.3212 t,
which is the same expression as P2 1t2  3.06911.3212 t. Whenever we convert the
units for the independent variable, from years to centuries, from hours to days,
from inches to centimeters, and so on, we actually are stretching or shrinking the
function horizontally.

Problems
1. In the analysis of the data on the cooling experi- 4. While watching his VCR, Ken noticed that the
ment, we assumed that the water temperature was counter seems to move much faster near the begin-
8.6°C and so subtracted 8.6 from each of the data ning of the tape than toward the end of the tape, so
values. Assume that the water temperature is 8.65°C he knows that the readings are not linear. To find
instead. Find the corresponding function to fit the the actual pattern, he records the counter reading
original data. Does it appear to be a better or worse every 15 minutes and obtains the following set of
fit to the data? data relating the counter reading to the elapsed
2. Instead of adding 1 to each value of N, as we did time, in hours.
with the Challenger data in this section, suppose
that you add some other quantity (say, 2) to each Time 0 0.25 0.50 0.75 1.0
value. How do the results compare to those ob-
tained earlier? Reading 0 445 817 1162 1448
3. A cup of hot coffee at 200°F is left on the table in a Time 1.25 1.5 1.75 2.0 2.25 2.5
70°F room to cool. The temperature readings on
the coffee at different times as it cools to 70°F are as Reading 1732 2005 2260 2503 2721 2942
follows.
a. From among exponential, power, and logarith-
Time, t 0 5 10 15 20 mic functions, find the function that best fits the
data giving the VCR counter reading in terms of
Temperature, T 200 163 139 118 108 the elapsed time.
b. Using the function from part (a), what would
Find the exponential function that best fits the data. you predict the reading to be after 3 hours?
Gord.3896.04.pgs 4/24/03 9:58 AM Page 346

346 CHAPTER 4 Extended Families of Functions

c. Suppose that the label on a VCR tape indicates a. From a scatterplot of the data of T versus P, it ap-
that a certain program Ken recorded runs from pears that the boiling point of water approaches a
1600 through 3400 on the counter. How long horizontal asymptote as the pressure P increases.
will that program run? This behavior might suggest an upside down ex-
d. Suppose that the VCR tape is a 6-hour tape. Pro- ponential function of the form y  A  Bc t,
grams already recorded end at a counter reading with c  1. Assume that the horizontal asymp-
of 4200. How much time is left on the tape for tote is at T  110°. Use this value to transform
the next recording? the data and find the corresponding exponential
5. In Problem 23 of Section 3.3 we looked at how the function.
boiling point (the temperature T) of water in a con- b. Use your function from part (a) to find the boil-
fined space (say, in a pressure cooker) depends on ing point of water when the vapor pressure is
the pressure of the vapor water. The table gives the 6.2 kilo-pascals.
boiling point of water, in degrees Celsius, at various c. What vapor pressure is needed if the boiling
vapor pressures, in kilo-pascals. point of the water is 105°C?

Pressure, P 0.61 1.22 2.34 4.25 7.38 12.34 19.93 31.18 47.37 70.12 101.32

Temperature, T 0° 10° 20° 30° 40° 50° 60° 70° 80° 90° 100°
Source: CRC Handbook of Chemistry and Physics, 1996.

4.9 The Logistic and Surge Functions

In this section, we consider two other families of functions—the logistic and the
surge functions—that frequently arise as mathematical models in a wide variety of
applications.

The Logistic Function

A great many processes start out growing exponentially, but eventually other factors
come into play to slow the rate of growth, causing a leveling off, as shown in Fig-
ure 4.85. Most populations grow in this manner, and many diseases spread in a
comparable pattern. The use of new technological innovations, be they new elec-
tronic devices such as microwave ovens, cellular phones, or DVD players and new
medical products, also spread this way. Such a pattern is called a logistic curve, and
the associated function is called a logistic function.

Logistic growth model

Leveling off

Slowing down

Exponential growth

FIGURE 4.85 Time

Gord.3896.04.pgs 4/24/03 9:58 AM Page 347

4.9 The Logistic and Surge Functions 347

Logistic processes can be modeled mathematically in several different ways,

and we look at one in considerable detail in Section 5.3. For now, we consider the
family of functions of the form

f 1t2 
C
,
1  AeBt
where A, B, and C are positive constants and e  2.71828 . . . is the base of the natu-
ral logarithm system that we introduced in Section 2.6. In most practical situations,
the constant A typically is very large, the constant C is fairly large, and the constant B
is usually between 0 and 1. Let’s first analyze the behavior of this family of functions.
This function actually is the quotient of two functions, so we have to reason in
the same way that we analyzed the behavior of rational functions in Section 4.6. In
particular, because the numerator is a positive constant, the function has no real
roots and thus never crosses the t-axis. Also, in the denominator, both the constant
A and the exponential decay function eBt are positive, so the denominator is never
zero and the function has no vertical asymptotes. Furthermore, when t is negative
or when t is positive and relatively small, the term AeBt is extremely large com-
pared to 1. Thus the denominator behaves like 1  AeBt  AeBt, and therefore
the function f 1t2 behaves like
CeBt
f 1t2 
C C
  .
1  AeBt AeBt A
At first (when t is small) this function grows like an exponential function: To
the left, it approaches 0 as t S , and to the right, as t increases, it is increasing
and concave up. As t gets larger, however, the term eBt decays toward 0, so that the
function behaves as if

f 1t2 
C C
Bt   C,
1  Ae 1  A.0
which is a constant. That is, the function eventually (when t is larger) grows more
slowly, so there is an inflection point. Beyond that point, the curve levels off and
approaches a limiting value at the height of C. Thus this type of function has the
shape shown in Figure 4.85 and so is called a logistic function. In Figure 4.86 we
show the graph of the function

f 1t2 
500
.
1  200e0.5t
y

600

400

200

x
FIGURE 4.86 0 5 10 15 20 25

It has the shape of a logistic curve, eventually leveling off at a height of about 500.
In Example 1 of Section 3.3, we found that the growth in the U.S. population
from 1780 to 1900 closely followed an exponential growth pattern with a growth
Gord.3896.04.pgs 4/24/03 9:58 AM Page 348

348 CHAPTER 4 Extended Families of Functions

rate of 32.1% per decade. Corresponding to the best fit exponential curve, we had
a correlation coefficient of r  0.998. However, we pointed out that this exponen-
tial pattern doesn’t apply during the twentieth century because the growth rate has
slowed dramatically for various reasons. This behavior suggests that a logistic func-
tion may be a better choice than an exponential function for modeling the U.S.
population over the entire time period since 1780.
E XAMPLE 1
(a) Find a logistic function to fit the following data on the growth of the U.S. population,
in millions, since 1780. Let t represent the number of decades since 1780. (b) What does
the function predict about the eventual maximum population of the United States?
(c) Use the function to predict the U.S. population in 2020.

Year Population Year Population

1780 2.8 1900 76.0

1790 3.9 1910 92.0

1800 5.3 1920 105.7

1810 7.2 1930 122.8

1820 9.6 1940 131.7

1830 12.9 1950 150.7

1840 17.1 1960 179.3

1850 23.2 1970 203.3

1860 31.4 1980 226.5

1870 39.8 1990 248.7

1880 50.2 2000 281.4

1890 62.9

350
Population (millions)

300
250
200
150
100
50
t
0 1750 1800 1850 1900 1950 2000
FIGURE 4.87 Year

Solution

a. We begin with the scatterplot of the data shown in Figure 4.87, which indicates that
population growth appeared to slow during the latter part of the twentieth century.
The successive ratios of the population values also indicate that the rate of population
Gord.3896.04.pgs 4/24/03 9:58 AM Page 349

4.9 The Logistic and Surge Functions 349

growth slowed from over 20% per decade at the beginning of the twentieth century to
about 10% per decade at the end.
We now want to fit a logistic curve to these data. Some calculators have the capa-
bility of fitting the best logistic function of the form discussed here to a set of data in
the least squares sense. When we use this routine on the U.S. population values, we
get the function
659.45
y  P1t2  .
1  92.05e0.198t
This function, superimposed over the population data in Figure 4.88, appears to be
an excellent fit to the data.
P

350

Population (millions)
300
250
200
150
100
50
t
0 1750 1800 1850 1900 1950 2000
FIGURE 4.88 Year

b. To find the limiting population predicted by this logistic function, we have to deter-
mine what happens as t S . As t increases, the term e0.198t approaches 0, so that
the quotient approaches 659.45 million people.
c. Based on this model, the population in 2020, when t  24 decades, will be

P124 2 
659.45
 367.42 million people.
1  92.05e0.1981242
◆

The Surge Function

Picture what happens when a medication is first administered to a patient. The ef-
fective level of the drug in the bloodstream initially is zero. The drug level then
rises rapidly toward a maximum as it is absorbed into the blood. Finally, the drug
level decays slowly as it is washed out of the body by the kidneys that filter impu-
rities from the blood. The overall pattern has the shape shown in Figure 4.89.
Similarly, a new advertising campaign produces an immediate jump in sales, but

t
FIGURE 4.89
Gord.3896.04.pgs 4/24/03 9:58 AM Page 350

350 CHAPTER 4 Extended Families of Functions

then the effects of the ad campaign tend to die out slowly over time. The resulting
pattern can also be represented by a curve like that shown in Figure 4.89.
Both of these processes are examples of a surge function, which can be writ-
ten as
S1t2  At pbt,
where A, p, and b  1 are three parameters. For realistic situations, we consider
only t 0. This formula for a surge function actually is the product of a power
function t p and an exponential decay function bt because b  1. For instance, Fig-
ure 4.90 shows the graph of the surge function S1t2  100t 2.5 10.752 t.

2000
1600 S(t) = 100t 2.5(0.75)t
1200
800
400
t
FIGURE 4.90 0 5 10 15 20 25 30

The coefficient A determines the maximum height of the curve. For the surge
function shown in Figure 4.90, this maximum is slightly more than 1800. The
power function term t p reflects the initial impetus and, in fact, the power p deter-
mines the location of the maximum value of the function. For this surge function,
the maximum occurs at about t  8.5. The decaying exponential term bt is re-
sponsible for the eventual slow decay. Also, remember that an exponential function
dominates any power function for large t so that, in the product of the two func-
tions, the exponential decay eventually overwhelms the growth in the power func-
tion term.

E XAMPLE 2
The drug L-dopa is administered to people suffering from Parkinson’s disease to relieve
symptoms such as extreme tremors and rigidity. To be effective, fairly high doses are re-
quired because only a small portion of a dose actually lasts in the body long enough to
be effective. The side effects of the large doses can be reduced by administering another
drug in conjunction with L-dopa. The following table shows the level of L-dopa L in the
blood, in nanograms per milliliter, as a function of time t, in minutes.

t 0 20 40 60 80 100 120 140 160 180 200 220 240 300 360

L 0 300 2700 2950 2600 1550 1100 900 725 600 510 440 300 250 225

A plot of these points is shown in Figure 4.91, which suggests the pattern for a surge
function. Find the equation of a surge function that models the data.
Solution The plot of the data indicates that the surge function reaches its maximum at
about t  60, where the maximum value is approximately 3000. The function also has
two points of inflection, one on either side of the peak. From the table of data, the great-
est increase in L occurs between t  20 and t  40, so we estimate that one inflection
Gord.3896.04.pgs 4/24/03 9:58 AM Page 351

4.9 The Logistic and Surge Functions 351

3000

Level of L-Dopa
2500
2000
1500
1000
500
t
0 60 120 180 240 300 360
FIGURE 4.91 Time (minutes)

point occurs at t  30, say. The greatest decrease in L occurs between t  80 and
t  100, so we estimate that the other inflection point occurs at about t  90.
We write L1t2  At pbt as the general equation of a surge function, where A, p, and b
are the three parameters whose values we have to determine. Unfortunately, routines to
find these parameters are not built into any calculators or directly into any software
packages, so we have to find an indirect way of estimating their values. To do so we
apply a transformation approach similar to that used in Sections 3.4 and 3.5. Thus, if
L1t2  At pbt, when we take logs of both sides, we get
log L  log1At pbt 2  log A  log t p  log bt log1u . v . w2  log u  log v  log w
 log A  p log t  t log b. log1up 2  p log u
Therefore, if L is a surge function of t, log L is a linear function of t and of log t. Thus, we
extend the preceding table to include values for log t and log L.

t 0 20 40 60 80 100 120 140 160 180 200 220 240 300 360

L 0 300 2700 2950 2600 1550 1100 900 725 600 510 440 300 250 225

log t UNDEF 1.3 1.60 1.78 1.90 2.00 2.08 2.1 2.2 2.2 2.3 2.3 2.3 2.4 2.56

log L UNDEF 2.4 3.43 3.47 3.41 3.19 3.04 2.9 2.8 2.7 2.7 2.6 2.4 2.4 2.35

Note that the first entries for log t and log L are marked UNDEF because the logarithmic
function is not defined.
Because log L is a linear function of both t and log t, we can use the values from this
table in a program that performs multivariate linear regression, as discussed in Sec-
tion 3.6. The linear function that best fits these data is
Y  1.5004  0.00667X1  1.1591X2,
where X1  t and X2  log t. The regression equation is equivalent to
log L  1.5004  0.00667t  1.1591 log t.
We undo the transformation, as we did in Sections 3.4 and 3.5, by taking powers of 10 on
both sides of this equation:
10log L  L  101.50040.00667t1.1591 log t 10log u  u
 1101.5004 2 1100.00667t 2 1101.1591 log t 2 10uv  10u . 10v
 131.652 1100.00667 2 t # 110log t 2
1.1591
log up  p log u
 31.6510.98482 tt1.1591. 10log u  u
Gord.3896.04.pgs 4/24/03 9:58 AM Page 352

352 CHAPTER 4 Extended Families of Functions

Therefore our model for the surge function is

L1t2  31.65t 1.1591 10.98482 t.
The graph of this function superimposed over the original data set for the level of L-dopa
in the blood is shown in Figure 4.92. It is not a particularly good fit to the data, especial-
ly for t between 0 and about 100 minutes. It reflects the overall pattern in the data but
not very accurately. Moreover, the corresponding coefficient of multiple determination
is R 2  0.6876, so R  0.8292, which is reasonably close to 1. It is close enough for there
to be a significant level of correlation.

3000
Level of L-Dopa

2500
2000
1500
1000
500
t
0 60 120 180 240 300 360
FIGURE 4.92 Time (minutes)

Problems
1. The growth pattern in human height or weight devel-
opment from birth through age 18, say, usually follows 4 102.9 101.6
a logistic growth pattern. The table gives the typical 5 109.9 108.4
height, in centimeters, of a male and a female in the
50th percentile for height at different ages, in years. 6 116.1 114.6
a. From the table, estimate the typical height of full 7 125.0 120.6
grown males and females in the 50th percentile
8 127.0 126.4
(assuming full growth occurs by age 18).
b. If you have access to a calculator or software 9 132.2 132.2
package that fits a logistic function to a set of
data, find a pair of logistic functions that can be 10 137.5 138.3
used to model the heights of both males and fe- 11 143.3 144.8
males as a function of age t for those in this 50th
percentile group. 12 149.7 151.5
c. What do the formulas from part (b) predict 13 156.5 157.1
about the typical heights of full grown males and
females in the 50th percentile? 14 163.1 160.4

15 169.0 161.8
Age Males Females
16 173.5 162.4
0 50.5 49.9
17 176.2 163.1
1 76.1 74.3
18 176.8 163.7
2 87.6 86.5 Source: U.S. Department of Health, Education, and Welfare, NCHS
Growth Curves for Children, Vital and Health Statistics, National
3 96.5 95.6 Health Survey. Washington, D.C.: U.S. Government Printing Office.
Gord.3896.04.pgs 11/21/03 11:53 AM Page 353

4.9 The Logistic and Surge Functions 353

2. Sweden has one of the longest collections of census

records of any country. The table to the right shows t Population
the Swedish population, in millions, starting in 0 1.763
1750 when t  0 through 1920 when t  170.
10 1.893
a. From the table, estimate the population of Swe-
den at the point of inflection when t  110. 20 2.030
b. If you have access to a calculator or software
30 2.118
package that fits a logistic function to a set of
data, find a logistic function that can be used to 40 2.158
model the population of Sweden as a function of
time t. 50 2.347
c. What does the formula from part (b) predict 60 2.378
about the maximum possible population of
Sweden? 70 2.585
d. Consult the population table in Appendix G to 80 2.888
see how well the logistic function you found in
part (b) predicts the actual population in 2002. 90 3.139
3. Consider the surge function S1t2  100t 2.5 10.752 t 100 3.483
(see Figure 4.90). Without using your function gra-
pher, predict how the graph of each surge function 110 3.800
(a)–(d) compares to this function in terms of the
120 4.168
location of the turning point and the rate at which
the function decays to 0. 130 4.566
a. f 1t2  100t 3 10.752 t
140 4.785
b. f 1t2  100t 2 10.752 t
c. f 1t2  100t 2.5 10.702 t
150 5.136

d. f 1t2  100t 2.5 10.802 t 160 5.522

170 5.9004
Source: Raymond Pearl, The Biology of
Population Growth. New York: Knopf,
1925.

Chapter Summary

In this chapter, we introduced several additional families of functions and ways to

build new functions out of old functions. More specifically, we described:
◆ How quadratic, cubic, quartic, and higher degree polynomials behave.
◆ How the real roots of a polynomial equation relate to the linear factors.
◆ How the real roots of a polynomial equation relate to the graph.
◆ How the number of turning points and the number of inflection points re-
late to the degree of a polynomial.
◆ How the end behavior of a polynomial depends on the sign of the leading
coefficient.
◆ How to find the real roots of a polynomial graphically, numerically, and—in
the case of quadratic functions—algebraically.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 354

354 CHAPTER 4 Extended Families of Functions

◆ How polynomial functions arise as models in the real world.

◆ How to fit polynomial functions to sets of data.
◆ The relative frequency with which complex roots occur.
◆ How to interpret the higher order differences of a set of numbers to deter-
mine polynomial patterns in sets of data.
◆ How to find the sum of the first n integers and the sum of the squares of
the first n integers.
◆ What it means to add, subtract, or multiply functions to form new func-
tions.
◆ The behavior of rational functions.
◆ What it means to have a function of a function.
◆ The effects of shifting, stretching, and shrinking on a function.
◆ How to interpret shifting and stretching of functions in terms of fitting
functions to data.
◆ How logistic and surge functions behave.
◆ How to use logistic and surge functions as models.

Review Problems
Sketch the graph of each function without using your and end behavior as x approaches  and . Esti-
function grapher. mate all turning points graphically.
1. f 1x2  1x  32 1x  22 1x  42 x2  4 x2  4
a. R1x2  2 b. Q1x 2  2
2. g1x2  12  x2 1x  32 1x  12 x 9 x 9
3. F1x2  1x  22 1x  32 1x  42 1x  12
x2  4 x2  4
c. S1x 2  2 d. T1x2  2
x 9 x 9
4. G1x2  1x  32 1x  22 1x  42 2
14. For each function shown, sketch the graph of
i. f 1x 2 ii. 3f 1x 2
Factor each polynomial to determine its roots alge-
iii. f 1x 2  4 iv. f 1x  3 2
braically.
5. P1x2  x 2  x  6 6. Q1x2  2x 2  9x  5
v. f 1x  32 vi. f 1x  4 2
7. R1x2  x 3  3x 2  2x
a. y b. y
8. Use the quadratic formula to verify your answers to f
Problems 5–7.
9. A quadratic function f has its vertex at the point f

15, 19 2, and f 182  4. What is f 122?

10. A cubic function f has its inflection point at 16, 4 2,
and f 122  12. What is f 110 2? x x
11. Estimate the location of the turning points of the
graph of the function y  x 3  4x 2  5. c. y

12. Determine the graphs of each pair of functions f

and g and use them to draw the graph of f  g. f
a. f 1x2  x 2  5, g 1x2  3x  2
b. f 1x2  2x 3  4, g 1x2  x 2 x
13. Analyze the behavior of each rational function in-
cluding identifying all zeros, vertical asymptotes,
Gord.3896.04.pgs 4/24/03 9:58 AM Page 355

4.9 The Logistic and Surge Functions 355

15. Suppose that f 1x2  2x 2  1 and g 1x 2  18. The return in dollars on an investment seems to
1x  12> 1x  22. Find the following. be well approximated by the function F1t2 
a. f 132  g 132 b. f 1 f 132 2 2t 2  t  4.2, whereas the return on another in-
c. g 1 f 13 2 2 d. g 1g 13 2 2 vestment is modeled by G1t2  7.8t  3.5. Deter-
mine for which values of t 0 the second
f 13 2
e. g 13 2 f 13 2 investment is better than the first.
g 13 2
f.
19. Evaluate the sum
g. f 1g 1x2 2 h. f 1 f 1x2 2
3  6  9  12  15  . . .  300.
i. g 1 f 1x2 2 j. g 1g 1x 2 2
f 1x 2
20. A polynomial has four turning points.
k. g 1x2 f 1x2
g 1x 2
l. a. How many inflection points must it have?
Explain.
16. Suppose that f 102  2, f 11 2  2, f 12 2  3, b. What is the minimum degree of the polynomial?
f 132  0 and that g102  1, g11 2  0, g12 2  2, c. What is the minimum number of real roots that
g132  3. Find the following quantities for x  0, the polynomial can have? Explain your answer
1, 2, and 3. with a sketch of a polynomial to illustrate what
a. f 1g1x2 2 b. g1 f 1x2 2 can happen.
f 1x 2 d. What is the maximum number of real roots that
c. f 1x2  g1x2
g1x 2
d. the polynomial can have? Explain your answer
with a sketch of a polynomial to illustrate what
17. Repeat Problem 16(a)–(d) for the functions f and g can happen.
shown in the graphs below for x  1, 2, 3, and 4. e. Are there any other values for the number of real
roots between the minimum number in part (c)
y and the maximum number in part (d) that the
polynomial can have? Explain your answer with
4 a sketch of a polynomial to illustrate what can
happen.
f (x)
3 21. The accompanying figure shows the graph of a
fourth degree polynomial. Use regression methods
2 to find a possible formula for this polynomial.

1
y

x 16
0 1 2 3 4
12
(i) 8
4
y x
–3 –2 –1 1 2 3
–4
4 –8
g(x) –12
3 –16
–20
2

1
22. The table that follows gives some values, rounded
to the nearest integer, for a rational function.
x a. Sketch a possible graph of this rational function
0 1 2 3 4
R1x 2 .
(ii) b. Find a possible formula for this rational function.
Gord.3896.04.pgs 4/24/03 9:58 AM Page 356

356 CHAPTER 4 Extended Families of Functions

24. The table gives the total number of cell phone sub-
x 4 3 2 1 0 1 scribers, in millions, in the United States since 1990
R(x) 10 0 UNDEF 0 3 0 and the average local monthly bill, in dollars, for
cell phone service.
x 2 3 4 5 6 7
a. Find the exponential growth function that best
R(x) UNDEF 0 18 UNDEF 0 21 fits the data on the number of subscribers as a
function of time since 1990.
b. Find the exponential decay function that best fits
23. Each function shown in the accompanying figure the data on the average monthly bill as a func-
can be interpreted as a shift applied to an exponen- tion of time since 1990.
tial, a power, or a logarithmic function. c. The total industry revenue each year is the pro-
a. Identify which is which. duct of the number of subscribers and the aver-
b. Write a possible formula for each function. age monthly bill for service. Use the results of
parts (a) and (b) to write a function that models
y y y the total cell phone revenue as a function of time
since 1990. What is the growth or decay factor
for this revenue function?
d. Extend the table to include a row that gives the
5
total annual revenue in the cell phone industry.
x
x x Then find the exponential function that best
–10 –4 fits the data on the annual revenue as a func-
(i) (ii) (iii)
tion of the number of years since 1990. How
y y y
does this result compare to the one you found
in part (c)?

7 9
x x x
5
(iv) (v) (vi)

Year 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

Subscribers 4.37 6.38 8.89 13.07 19.28 28.15 38.20 48.71 60.83 76.28

Average bill 83.94 74.56 68.51 67.31 58.65 52.45 48.84 43.86 39.88 40.24
Source: 2000 Statistical Abstract of the United States.
gord.3896.05.pgs 4/24/03 10:02 AM Page 357

5
Modeling with
Difference Equations
5.1 Eliminating Drugs from the Body
Prozac is one of the most widely prescribed drugs used to combat extreme depression.
Typically, a patient takes a Prozac tablet once a day. As we discussed in Section 2.5,
once a medication has been absorbed into the bloodstream, it is washed out of the sys-
tem by the kidneys, which purify the blood by filtering out foreign chemicals.
For now, let’s assume that a person takes a single 80 mg dose of Prozac and
that it has been completely absorbed into the blood. During any 24-hour time pe-
riod, the kidneys eliminate approximately 25% of the Prozac in the bloodstream,
so that 75% of the drug remains. As we showed in Section 2.5, the amount of
Prozac in the bloodstream based on a single 80 mg dose can be modeled by the ex-
ponential decay function
D1t2  8010.752 t,
where t measures the number of 24 hour periods since the Prozac was taken. The
graph of this function is shown in Figure 5.1.

80
Drug Level (milligrams)

70
60
50
40
30
20
10
t
0 2 4 6 8 10 12
Time (days)

FIGURE 5.1

In particular, D10 2  80. After one 24 hour period, assuming no additional

Prozac is taken, D11 2  8010.752  60. After 2 days, D12 2  8010.75 2 2  45, and
so on. These particular points are highlighted in Figure 5.1.

357
gord.3896.05.pgs 4/24/03 10:02 AM Page 358

358 CHAPTER 5 Modeling with Difference Equations

The collection of values so obtained,

580, 60, 45, 33.75, 25.3125, . . . 6,
is called a sequence—it is a set of numbers in a particular order. Thus, a sequence is
a function that is defined on a set of nonnegative integers such as n  0, 1, 2, . . .
The corresponding range is some set of real numbers, such as the levels of Prozac,
determined by the rule for the sequence. In other words, for each positive integer,
n  0, 1, 2, . . . , there corresponds a real number D1n2 that is the nth term in the
sequence. We can write the sequence for the levels of Prozac in the blood as
D102, D11 2 , D12 2 , . . . to emphasize that it is a function.
Although sequences are functions, a special notation is used for them instead
of the usual functional notation. For instance, with the Prozac model, we write
D0  80 instead of D10 2  80, D1  60 instead of D11 2  60, and so on. The gen-
eral, or nth, term in the sequence is then written as Dn instead of D1n 2, where
n  0, 1, 2, . . . Unless we’re working in a context where some other letters are
more appropriate (such as Dn for the level of a drug), we denote the nth term in a
sequence by xn , which represents 5x0 , x1 , x2 , . . . 6.
Keep in mind that, when you work with a sequence such as Dn for the level of
Prozac in the system, you are looking only at the value of the function at the start
of each 24-hour period. What happens in between isn’t considered because the
process is basically discrete. Graphs such as the one shown in Figure 5.1 with a
smooth curve superimposed over the points in the sequence are convenient for vi-
sualizing a trend, but can sometimes be misleading. Such a curve can completely
miss anything else that may happen from one value of n to the next.

Repeated Drug Dosages

In practice people often use a drug such as Prozac on a maintenance basis—they
take a fixed dose of the medication every time period, rather than a single initial
dose. It might be a repeated daily dose of Prozac or some high-blood-pressure
medication or a repeated dose of a cold medication every 4 hours. In such cases,
the exponential decay model is not realistic. What can we then say about the
amount of medication in the body as a function of time?
Suppose that a person takes 80 mg of Prozac each day, starting on some partic-
ular day. After the first 24 hours, 25% of the Prozac, or 20 mg, is eliminated, leav-
ing 75%, or 60 mg and then the next day’s dose adds 80 mg to that amount.
Therefore, after the first 24 hours, the amount of Prozac in the body is
D1  60  80  140 mg,
or considerably more than the original dose!
During the second 24 hour period, the kidneys eliminate 25% of the 140 mg of
Prozac present, or 35 mg, leaving 75% of the 140 mg of Prozac present, or 105 mg.
Then the person takes the next dose of 80 mg. Thus, after two 24 hour periods, the
amount of Prozac in the body is
D2  0.751140 2  80  105  80  185 mg.
After 3 days, the level of Prozac is
D3  0.751185 2  80  218.75 mg,
gord.3896.05.pgs 4/24/03 10:02 AM Page 359

5.1 Eliminating Drugs from the Body 359

and so on, indefinitely. The corresponding sequence of Prozac levels is

580, 140, 185, 218.75, 244.0625, 263.0469, . . . 6.
Note how the level of the drug keeps rising, but in a concave down manner, as illus-
trated in Figure 5.2.

Think About This Does the foregoing explanation suggest that the level of Prozac in the bloodstream
keeps rising indefinitely? If so, is that reasonable? ❐

Let’s look at a slightly different way to describe this process. The initial dose is
D0  80 mg. During the first 24 hour time period, 25% of this amount is removed
from the blood, leaving 75% of D0 , and the person then takes the next dose of 80 mg.
Thus
D1  0.75D0  80.
Similarly, during the second day, 25% of the Prozac is eliminated, leaving 75% of
D1 , and the person then takes another 80 mg dose, so that
D2  0.75D1  80.
Again, after the third day,
D3  0.75D2  80.
In general, at the end of n  1 days, for any value of n,
Dn1  0.75Dn  80.
This equation shows the relationship between the level of Prozac on any two succes-
sive days. In particular, if we know the amount of Prozac in the body after n days,
this equation allows us to calculate the amount of Prozac present the following day.
We call an equation such as Dn1  0.75Dn  80 that relates the successive
terms in a sequence a difference equation. As you will see, it is an effective way to
model drug concentrations in the body. (Note that some authors reserve the term
difference equation only for equations that explicitly involve the difference
xn1  xn between successive values in a sequence; we adopt the more common,
broader use of the term for any relationship between xn1 and xn.)
In the drug model, the initial dose of Prozac was D0  80 mg. Therefore, if we
set n  0 in the difference equation Dn1  0.75Dn  80, we determine D1 to be
D1  0.75D0  80  0.751802  80  140 mg.

320
Drug Level (milligrams)

280
240
200
160
120
80
40
n
0 2 4 6 8 10 12
FIGURE 5.2 Time (days)
gord.3896.05.pgs 4/24/03 10:02 AM Page 360

360 CHAPTER 5 Modeling with Difference Equations

When n  1 we obtain,
D2  0.75D1  80  0.751140 2  80  185 mg.
When n  2 we get,
D3  0.75D2  80  0.751185 2  80  218.75 mg.
Continuing in this way, we get
D4  0.75D3  80  244.0625 mg;
D5  0.75D4  80  263.0469 mg;
D6  0.75D5  80  277.2852 mg;
and so on. Over time, the amount of the drug in the body is given by the sequence of
numbers
580, 140, 185, 218.75, 244.0625, 263.0469, 277.2852, . . . 6.
Figure 5.3 shows the graph of these points, connected by a smooth curve. This se-
quence of numbers is the solution to the difference equation; we refer to it as the
solution sequence.

350 Limit = ?
Drug Level (milligrams)

300
250
200
150
100
50
n
0 2 4 6 8 10 12 14
Time (days)

FIGURE 5.3

Note that each successive term, although larger than the preceding value, has
grown by somewhat less than the term before. That is, the change from D0 to D1 is 60
mg; the change from D1 to D2 is 45 mg; the change from D2 to D3 is 33.75 mg, and so
on. The curve drawn through the points in Figure 5.3 is concave down, and the suc-
cessive values seem to be leveling off. That is, the drug level continues to rise but at a
less steep rate. The overall pattern is characteristic of an upside-down exponential
decay process. Rather than gradually dying out toward the horizontal axis as a hori-
zontal asymptote (as an exponential decay function does), this process gradually rises
toward the limiting amount of drug in the body as a horizontal asymptote. By contin-
uing this process numerically, we find that the limiting amount L appears to be very
close to 320 mg. We say that the terms of the sequence converge to this limiting value
in the sense that they get closer and closer to L the farther we go in the sequence.
Generating the successive terms of this type of iteration process on either a cal-
culator or a spreadsheet is quite simple. On a calculator, simply enter the starting
value for the sequence and press Enter. Then enter the iteration formula, using 2nd
ANS as the variable. For instance, using the Prozac model, you would enter the initial
dose 80 and then the formula for the difference equation Dn1  0.75Dn  80 in
the form
gord.3896.05.pgs 4/24/03 10:02 AM Page 361

5.1 Eliminating Drugs from the Body 361

0.75*ANS+80.

When you press Enter, you get the next value of the sequence, or 140. Then each time
you press Enter, you get the following value. Try this process to see how simple it re-
ally is.
In terms of the original problem, the limiting value L represents the maximum
level of Prozac that will be reached in the body. This horizontal asymptote is
known as the maintenance level for the drug. Once that level of Prozac has been
reached, the amount in the body will remain constant at that level every 24 hours,
so long as the same dose is taken repeatedly every 24 hours.

Think About This The curve shown in Figure 5.3 is actually incorrect; it was obtained by simply con-
necting the points to demonstrate the overall pattern. However, it completely ig-
nores what happens during the 24 hours between successive doses. Sketch a more
detailed curve that accurately reflects what happens. ❐
In practice, researchers determine the specific level L of a medication that is most
effective, considering factors of both safety and effectiveness. An initial dose of, say, 80
mg of the medication means that for some period of time, the amount in the blood-
stream is below the optimal level. As a result, doctors often prescribe an initial dose
higher than the normal dose so that the drug level approaches the maintenance level L
more rapidly. For instance, an initial dose of 240 mg followed by daily doses of 80 mg
will achieve the desired level quickly. However, the safety of prescribing such a large
dosage, especially as the first dose of the drug, must be considered.

EXAMPLE 1
Suppose that the initial dose of Prozac is 160 mg (instead of 80 mg) but that all subse-
quent doses are 80 mg.
a. Find the first six terms of the solution sequence.
b. How does this solution compare to the one we had before?
Solution

a. We still have the same difference equation model,

Dn1  0.75Dn  80,
but now the initial dose is D0  160 mg. Therefore
if n  0: D1  0.75D0  160  0.751160 2  80  200 mg;
if n  1: D2  0.75D1  80  0.751200 2  80  230 mg;
if n  2: D3  0.75D2  80  0.751230 2  80  252.5 mg.
Continuing in this manner,
D4  0.75D3  80  269.375 mg;
D5  0.75D4  80  282.031 mg;
D6  0.75D5  80  291.523 mg;
and so on. Thus the levels of Prozac in the bloodstream are now given by the solution
sequence
5160, 200, 230, 252.5, 269.375, 282.031, 291.523, 298.643, 303.982, . . . 6.
b. By changing the initial dose D0 , we get a different sequence of values. That is, the dif-
ference equation has a different solution sequence that depends on the initial value
gord.3896.05.pgs 4/24/03 10:02 AM Page 362

362 CHAPTER 5 Modeling with Difference Equations

we choose. However, the behavior of this solution is essentially the same—it is an in-
creasing, concave down function that rises toward the same limiting value of
L  320 mg, as shown in Figure 5.4.

350
Drug Level (milligrams) 300
250
200
150
100
50
n
0 2 4 6 8 10 12 14
Time (days)

FIGURE 5.4

What happens if the level of drug in the system gets very high? We explore this
situation in Example 2.

EXAMPLE 2
Suppose that a person takes an initial dose of 500 mg of Prozac and thereafter takes the
usual 80 mg daily.
a. Find the corresponding solution sequence.
b. Discuss the behavior of the solution.

Solution

a. We again use the same difference equation,

Dn1  0.75Dn  80,
but now the initial dose is D0  500, so that

D
Drug Level (milligrams)

500
400
300
200
100
n
0 2 4 6 8 10 12 14
Time (days)

FIGURE 5.5
gord.3896.05.pgs 4/24/03 10:02 AM Page 363

5.1 Eliminating Drugs from the Body 363

if n  0: D1  0.75D0  80  455 mg;

if n  1: D2  0.75D1  80  421.25 mg;
if n  2: D3  0.75D2  80  395.938 mg;
and so on. Thus the levels of the drug in the bloodstream are now given by the solution
sequence
5500, 455, 421.25, 395.938, 376.953, 362.715, 352.036, 344.027, 338.020, . . . 6.

b. Figure 5.5 shows these points, which fall in a decreasing, concave up pattern that ap-
parently approaches the same limiting value L  320 mg, but from above rather than
from below.

Consequently, if the drug level rises too high, some counteracting effects re-
duce the level. Thus the process that we have described can’t lead to an infinite drug
level in the body. In fact, our difference equation model predicts that, if the level
ever exceeds the maintenance level for the drug, the drug level will decrease as sub-
sequent daily doses are taken.
Again, note that by changing the initial condition D0 in the difference equa-
tion, we obtain a different solution sequence.

Determining the Maintenance Level L

We estimated the maintenance level for Prozac as being about 320 mg by looking at
the successive values we calculated in the solution sequence. We now determine the
limiting value L for Prozac precisely by using the following argument.
Suppose that, for some value of n, Dn reaches the limit L so that all successive
levels of Prozac are the same. Thus for a large enough n, we assume that both
Dn1  L and Dn  L. Substituting these values into the difference equation for
the Prozac drug model,
Dn1  0.75Dn  80,
we obtain
L  0.75L  80 so that 0.25L  80.
Hence the limiting value is
80
L  320 mg.
0.25
We developed all these ideas in the context of a single drug whose level in the
bloodstream decreases at a particular rate. In actuality, different drugs are “washed
out” of the blood at different rates. For example, aspirin is removed quite rapidly: Its
level is reduced by about 50% every 29 minutes. In fact, all drugs are rated by how
long it takes for 50% to be eliminated from the body, which is known as the biological
half-life or simply the half-life of the drug. Thus the half-life of aspirin is 29 minutes.

Think About This In the usual 4-hour time period between successive 325 mg doses of aspirin, how
much of the aspirin in the blood would be removed by the kidneys? ❐
gord.3896.05.pgs 4/24/03 10:02 AM Page 364

364 CHAPTER 5 Modeling with Difference Equations

Finding a Formula for the Solution

In the preceding situations, we worked with the terms in the solution sequences of
the difference equation
Dn1  0.75Dn  80
for different values of the initial dose D0 . Thus, if the initial dose is D0  80, the
corresponding solution sequence is Dn  580, 140, 185, . . . 6. But we did not find
a formula for this solution as a function of n. (In Supplementary Chapter 12 we de-
velop a powerful technique for finding formulas for the solution sequences to such
difference equations. For now, we construct such a solution by using some of our
previous ideas about functions.)
EXAMPLE 3
Find a formula for Dn , the solution sequence to the drug model difference equation for
Prozac, based on an initial dose D0  80.
Solution We can think of the values in the solution sequence as a set of data values.

n 0 1 2 3 4 5 6
Dn 80 140 185 218.75 244.0625 263.0469 277.2852

(We could use more values, but these will suffice.) Recall that the shape of the curve shown
in Figure 5.2 is like an upside-down exponential decay function. Because the values for Dn
approach a limiting value of L  320 mg as a horizontal asymptote, we can shift each Dn
value to obtain the corresponding values of 320  Dn , as shown in the following table and
in Figure 5.6.

320 – Dn

320
280
320 – Drug Level

240
(milligrams)

200
160
120
80
40
n
0 1 2 3 4 5 6
FIGURE 5.6 Number of Days

n 0 1 2 3 4 5 6
Dn 80 140 185 218.75 244.0625 263.0469 277.2852
320  Dn 240 180 135 101.25 75.9375 56.9531 42.7148

Note how the values of 320  Dn in the bottom row decay as n increases. If we continue
the process further, these differences approach 0 as n gets larger (because Dn approaches
320 as n increases). It therefore makes sense to fit a decaying exponential function to
gord.3896.05.pgs 4/24/03 10:02 AM Page 365

5.1 Eliminating Drugs from the Body 365

320  Dn as a function of n. (We would not use a decaying power function because the
data start with a finite value of 240 when n  0.)
Using a calculator, we find that the exponential function that best fits these data is
320  Dn  239.9997810.75000072 n.
The corresponding correlation coefficient is r  1, which suggests a virtually perfect
fit. If we solve for the level of Prozac Dn, we obtain
Dn  320  239.9997810.75000072 n.
The numbers in this expression suggest that the “correct” formula for the solution might
be
Dn  320  24010.752 n.
We later show that this formula holds for every possible value of n, not just for the few
particular values of n we used in constructing the best-fitting exponential function.
◆
Constructing the Solution in General
We now extend the preceding solution formula to solve the comparable difference
equation for any medication with any fixed periodic dose. Suppose that the kid-
neys remove a fixed percentage of a medication every time period, leaving a frac-
tion a, 0  a  1, in the bloodstream. Also, suppose that the repeated dose is an
amount B. The corresponding difference equation then is
Dn1  aDn  B.
(In our previous development with Prozac, we had a  0.75 and B  80.) For any
value of a between 0 and 1 and any positive value of B, the successive terms in the
solution sequence for Dn have behavior comparable to that shown in Figure 5.2
assuming D0 is less than the maintenance level: The solution sequence is an in-
creasing concave down function, and approaches a horizontal asymptote. If your
graphing calculator displays solutions of difference equations, select some typical
values for a and B and check out the behavior of the solution.
Because the level of the medication in the blood rises toward the maintenance
level L, we can solve for L in this general case by realizing that, should this level ac-
tually be achieved, then both Dn  L and Dn1  L. Therefore
L  aL  B, so that L  aL  L . 11  a2  B.
Hence the maintenance level is
B
L .
1a
The formula for Dn that we constructed previously for Prozac was Dn  320 
24010.752 n, based on a  0.75 and B  80, with an initial value D0  80. In this
expression, 320 is the limiting value L  B>11  a2 and 0.75 is the decay factor a.
The coefficient 240 is 320  80, the difference between the limiting value L and
the initial dose D0 . These results suggest that the general formula for the solution
sequence is
Dn  L  1L  D0 2 an
for any value of n, with parameters a, B, L  B> 11  a2, and D0 .
gord.3896.05.pgs 4/24/03 10:02 AM Page 366

366 CHAPTER 5 Modeling with Difference Equations

EXAMPLE 4
Verify that the preceding expression for Dn is indeed a formula for the solution sequence
of the difference equation Dn1  aDn  B for every value of n.
Solution To verify that the expression for Dn is actually a formula for the solution se-
quence, we must show that it satisfies the difference equation
Dn1  aDn  B
for every value of n. Thus we substitute Dn  L  1L  D0 2 an and the corresponding
expression Dn1  L  1L  D0 2an1 when n is replaced by n  1, into the difference
equation. The left-hand side of the difference equation becomes
Dn1  L  1L  D0 2an1.
The right-hand side becomes
aDn  B  a3L  1L  D0 2an 4  B
 aL  a . 1L  D0 2 an  B
 aL  1L  D0 2an1  B.
However,

so that B  L . 11  a2 .
B
L
1a
Therefore the right-hand side becomes
aDn  B  aL  1L  D0 2 an1  B
 aL  1L  D0 2 an1  L . 11  a2
 aL  1L  D0 2 an1  L  aL
 L  1L  D0 2an1,
which is identical to the left-hand side. Thus the expression
Dn  L  1L  D0 2 an
is a formula for the solution sequence Dn and it holds for all possible values of n.
◆

We summarize the preceding results as follows.

Level of Medication in the Bloodstream

Assumptions
◆ The kidneys remove a fixed proportion, 1  a, of a medication from
the bloodstream every time period.
◆ The repeated dosage of this medication every time period is B.

Mathematical Model
◆ Difference equation: Dn1  aDn  B
◆ Maintenance level for the medication: L  B> 11  a2
◆ Solution: Dn  L  1L  D0 2 an
gord.3896.05.pgs 4/24/03 10:02 AM Page 367

5.1 Eliminating Drugs from the Body 367

We developed this mathematical model by simplifying the situation considerably.

First, when a person takes a medication, a certain amount of time is needed for it to be
completely absorbed into the blood, as well as to reach the intended part of the body.
Second, the rate at which a particular drug is washed out of the blood depends on
many factors, including a person’s weight, metabolism, and the state of the kidneys
and liver, all of which are involved in the elimination process. You definitely should
not make any medical judgments about the effectiveness of a drug based on the sim-
plified models of this section.

Summary
Let’s summarize some of the fundamental ideas about difference equations.
1. A difference equation relates the successive terms—say, xn and xn1 —of a
sequence. For instance, it might be
xn1  1.2xn  3.5 or xn1  0.6xn  2n.
2. The solution to a difference equation is a sequence (which is a function
of n) that satisfies the difference equation. For a given initial value x0 for
the sequence, we can always calculate every successive value of the solu-
tion sequence 5x0 , x1 , x2 , . . . 6 by using the difference equation directly.
3. For every possible initial value x0 , there is a different solution sequence to
the difference equation. Figure 5.7 shows the graphs of several different so-
lution sequences to the difference equation xn1  axn  B for the drug
model, based on different initial doses. Note that, whenever the initial
dose x0 is less than the limiting value L, the pattern is increasing and concave
down, approaching L as a horizontal asymptote. Note also that, whenever
the initial dose x0 is greater than L, the pattern is decreasing and concave up
and approaches L as a horizontal asymptote.
xn

500
Drug Level (milligrams)

400

300

200

100

n
0 2 4 6 8 10 12
FIGURE 5.7 Number of Days

4. Although the solution sequence can always be expressed in terms of the

specific numbers in the sequence, as determined from the difference equa-
tion, it is desirable, whenever possible, to write the solution as a formula
for xn in terms of n. Such a formula is called a closed form expression for
the general term xn . Thus a solution sequence xn can be defined
a. in closed form with a formula for xn in terms of n,
b. by the actual sequence of numbers 5x0 , x1 , x2 , . . . , xn, . . . 6, or
c. by a difference equation that relates successive terms of the sequence.
Think of these ideas in terms of the drug-level model we developed. The
same ideas apply to each of the difference equation models we develop
throughout the remainder of this chapter.
gord.3896.05.pgs 4/24/03 10:02 AM Page 368

368 CHAPTER 5 Modeling with Difference Equations

Finally, the formula Dn  L  1L  D0 2 an we developed for the solu-

tion sequence to the difference equation Dn1  aDn  B for the drug
model can be applied to many other situations. To do so, we rewrite it in a
somewhat more general way. Using xn as the dependent variable gives the
difference equation
xn1  axn  B.
Also, not every difference equation of this form has solutions with a hori-
zontal asymptote. The solutions may grow toward  , so we write
B>11  a2 in place of L. With this terminology, the solution to the differ-
ence equation becomes

 a  x0b an
B B
xn 
1a 1a

 ax0  b an.
B B

1a 1a
When we multiply out the first equation and collect like terms in a differ-
ent manner, we get

11  an 2  x0an .
B B B
xn   an  x0an 
1a 1a 1a
In summary we have the following result.

The complete solution to the difference equation

xn1  axn  B,

where a and B are constants, is

11  an 2  x0an.
B
xn 
1a
If 0  a  1, the solution can be written as
xn  L  1x0  L2an,

where
B
L
1a
is the limiting value for the solution as n S .

This formula for the solution applies only to difference equations of the particular
form xn1  axn  B. Thus, it applies to the difference equation
xn1  1.05xn  3,
but it does not apply to the difference equation
xn1  1.05xn  3n
because 3n is not a constant.
gord.3896.05.pgs 4/24/03 10:02 AM Page 369

5.1 Eliminating Drugs from the Body 369

Furthermore, a difference equation of the form

xn1  axn  B, n  0,
starting at x0 when n  0, is also known as a forward difference equation because it ex-
presses the next value xn1 in a sequence in terms of the current value xn . You can
think of it as “looking forward” to where you are going. We can also write a backward
difference equation in which the current value xn in a sequence is written in terms of
the previous value xn1 . Think of it as “looking backward” where you came from. The
comparable difference equation is
xn  axn1  B, n  1.
You can convert a forward difference equation to an equivalent backward differ-
ence equation by replacing n with n  1 everywhere that n appears, and vice versa.
Either way, you get the identical solution sequence, either as a collection of values
or as a formula in terms of n.
Many calculators have the capability to calculate successive terms in the solu-
tion sequence for a difference equation and to plot the solution as part of their
Sequence Mode. Check your calculator manual for specific instructions.

Problems

1. Suppose that the kidneys remove 30% of a drug from is taken. How do the two graphs compare? Use the
the bloodstream every 4 hours. If a person takes a sin- two graphs—drawn on the same set of axes—to con-
gle dose of 16 mL, find the amount of the drug in the struct a graph showing the actual level of the medica-
body after 12 hours and after 24 hours. How long does tion in the bloodstream at all times, not only at the
it take for the level to drop below 1 mL? below 0.01 mL? times just before or just after the medication is taken.
2. Suppose that the person in Problem 1 takes repeated Dn
doses of 16 mL of the same drug every 4 hours.
What is the drug level after 12 hours? after 24 hours? 70
What is the limiting value for the dosage? 60

3. Suppose that an initial dose of Prozac is D0  320 mg 50

Drug Level

followed by daily doses of 80 mg. 40

a. Calculate the first six terms of the solution se- 30

quence. What do you observe about them? 20
b. Can you explain this result? 10
4. Suppose that the kidneys remove 25% of a drug in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
n
the bloodstream every time period and that the ini- Time
tial dose is 48 mL followed by 16 mL doses every
time period thereafter. How long will it be until the 7. The drug dosage for a certain drug is 10 mg per day,
drug level in the bloodstream exceeds 60 mL? and the initial dose is also 10 mg. If the kidneys re-
5. Recall that Figure 5.3 ignores the removal by the move 60% of the drug every 24 hours, find the
kidneys of the Prozac during each 24-hour time maintenance level for the medication.
period. Sketch a more accurate graph of the level of 8. Suppose that the daily dosage of the drug in Prob-
Prozac in the bloodstream versus time. lem 7 is halved to 5 mg per day. Find the mainte-
6. The accompanying graph shows the level of a med- nance level. Is the maintenance level also halved?
ication in the bloodstream just after each repeated 9. Suppose that the person in Problems 7 and 8 de-
dose of 16 mL is taken. Sketch the graph of the drug cides to take 10 mg every second day instead of 5
level of the same medication just before the next dose mg each day. Does the patient achieve the same
gord.3896.05.pgs 4/24/03 10:02 AM Page 370

370 CHAPTER 5 Modeling with Difference Equations

maintenance level for the medication? Explain why f. How would the difference equation and the lim-
or why not. iting value change if you fill up when the tank is
10. Two 5-grain aspirin tablets contain 650 mg of the 40% full instead of 50% full?
drug. With aspirin’s half-life of 29 minutes, how g. How would the limiting value change if your gas
much is left in the bloodstream after 2 hours? How tank holds 16 gallons instead of 14 gallons and
long does it take for the level to be equivalent to 10 you fill up when the tank is half full?
mg of aspirin? If an individual takes two tablets every Write the first six terms of each sequence whose general
4 hours, what is the maintenance level of the aspirin? term is given.
11. Some studies have shown that taking one aspirin 15. xn  4n 16. xn  3n  5
tablet per day significantly reduces a person’s risk of
heart attack or stroke. If a person follows this regimen, 17. xn  12 n 18. xn  n2  5
find the maintenance level of the aspirin in the blood. n21
19. xn  n3  10 20. xn 
12. The maintenance level for a certain drug is 600 mg. A n22
patient starts with an initial dose of 100 mg and re- 2n n2
peats it daily. Sketch the graph of the level of the drug 21. an  22. an  n
3n 2
in the bloodstream as a function of time. Suppose 1 log n
that D5  400 and D10  520. Use the graph to de- 23. yn  , n  1 24. yn  , n1
n n
termine which values are possible for the drug level
and which are impossible. 25. pn  1  10.2 2 n 26. pn  1  10.2 2 n
a. D7  440 b. D7  460
27–38. Decide which sequences in Problems 15–26 seem
c. D12  540 d. D12  560 to converge and which clearly do not. Give reasons
13. The daily dosage for a certain medication is 200 mL for your decision. For those that you’re not sure
and the maintenance level is 500 mL. A person taking about, what could you do to come to a decision?
this medication reaches a level of 450 mL in 10 days. 39–50. Plot the points 1n, xn 2 for each of the sequences in
Let r1 represent the average daily rate of increase of Problems 15–26. Decide which appear to be strictly
the drug level over the full 10-day period, let r2 be the increasing or strictly decreasing and which are con-
average daily rate of increase over the first 5-day peri- cave up or concave down.
od, and let r3 be the average daily rate of increase over
51. Consider the sequence E whose general term is
the last 5 days. Without calculating their values, list
1 n
these three rates in increasing order. (See Problem 24 en  a1  b , for n  1, 2, . . .
of Section 4.1.) n
14. Suppose that your car has a 14-gallon gas tank that a. Calculate the first 10 terms of this sequence and plot
you fill as soon as the level drops to half-full. Also, them. Does the graph suggest an eventual limit?
every time you fill up, you add one quart 1 14 gallon) b. Calculate e100 , e500 , e1000 , e10,000 , e100,000, and
of an additive that mixes thoroughly with the gas e1,000,000. What does the limit of the sequence en
and is then used up along with the gas. appear to be if you let n increase indefinitely?
a. Write a difference equation that models the 52. Consider the sequence en  11  1>n2 n again. Use
amount of the additive An in the tank from one your calculator to evaluate e1000 , e1,000,000 , e10,000,000 ,
fill-up to the next. and e100,000,000 . Keep track of all the results obtained.
b. Use the difference equation to calculate the amount What do you observe about the terms of this se-
of additive in the tank over the first 10 fill-ups. quence? What is your best estimate for the limiting
c. Sketch the graph of An as a function of n based value? Continue the process of taking larger and larg-
on the values from part (b). What does the be- er values for n—say, up to n  1015. You will find, de-
havior of the function suggest? pending on your calculator, that the terms eventually
d. Find the limiting value for the amount of the ad- jump to 1 instead of continuing as you would expect,
ditive in the tank as n increases indefinitely. owing to calculator round-off. By trial and error, can
e. Find the closed form solution of the difference you find the point where that occurs on your calcula-
equation. tor? If so, what is it?
gord.3896.05.pgs 4/24/03 10:02 AM Page 371

5.2 Modeling with Difference Equations 371

53. Consider the sequence f n  11  1>n2 n. What is 55. What is the limit of the sequence hn  11  n 2 1>n as
the limiting value for the sequence as n S ? How n S ?
is this limiting value related to the one in Problem 56. Suppose that the successive terms of a sequence are
51? (Hint: There is a simple arithmetic relationship.) increasing and that the graph drawn through the
54. Repeat Problem 53, using the sequence gn  corresponding points is concave up for all n. Can the
11  2>n2 n . How is the limiting value related to the sequence converge to a limit? Explain your answer.
one in Problem 51? Based on this result, conjecture
what the limiting value is for 11  5>n 2 n as n S .

5.2 Modeling with Difference Equations

As we have previously discussed, a mathematical model for a process is an expres-
sion or an equation that represents that process. Sometimes we simply create a
formula to fit a set of data. Often, especially in the sciences, we try to determine
the underlying principles or assumptions on which a process is based and then
attempt to find a relatively simple mathematical relationship that reflects those
principles or assumptions. For instance, in football, when a quarterback throws a
long pass down the field, the path of the ball can be represented by a simple
mathematical formula involving a quadratic function relating the height y to the
horizontal distance x—a model based on one of Newton’s laws of motion.
Throughout this chapter, the models we develop are difference equations such
as Dn1  0.75Dn  80 that models the level of Prozac in the bloodstream. The solu-
tion sequences to such difference equations can be given in two ways: either as a
collection of numbers such as 580, 140, 185, 218.75, 244.0625, 263.0469, 277.2852, . . . 6
based on an initial value D0  80, say, or as a formula such as
Dn  320  24010.752 n,

a function of n that holds for every value of n.

Exponential Growth and Decay Models

The population of the world passed 6 billion people in 1999 and was growing at
a rate of about 1.5% per year, so the growth factor was 1.015. Assuming that this
trend continues, we can model the earth’s population with the exponential
growth function P1t2  611.0152 t, where t is the number of years since 1999 and
the population is in billions of people. Alternatively, if we use n to represent the
number of years since 1999, we can write this exponential function in sequence
notation as Pn  611.0152 n. We know that a set of numbers follows an exponential
pattern if the successive ratios are all constant. That is, for any n,
Pn1
 c  1.015,
Pn
or when we multiply by Pn ,
Pn1  cPn  1.015Pn ,
starting with P0  6 when n  0 in 1999. This difference equation is a model for
the world’s population. It tells us that each term in the population sequence P  5Pn 6
is precisely the same multiple, 1.015, of the preceding term.
gord.3896.05.pgs 4/24/03 10:02 AM Page 372

372 CHAPTER 5 Modeling with Difference Equations

Let’s now look at the reverse of the preceding argument. Suppose that each
term in a sequence X  5xn 6 is a constant multiple c of the preceding term. That is,
x1  cx0 , x2  cx1 , x3  cx2 , . . . , or, in general xn1  cxn , which is a difference
equation for the sequence for any value of n. What does this result mean?

EXAMPLE 1
Find a formula for the solution sequence to the difference equation
xn1  cxn
for any constant c  0.
Solution Starting with the initial term x0 corresponding to n  0, we have
x1  cx0 ;
x2  cx1  c . 1cx0 2  c2x0 ;
x3  cx2  c . 1c2x0 2  c3x0 ;
x4  cx3  c . 1c3x0 2  c4x0 ;
and so on. This process leads to an obvious formula for the general term of the solution
sequence,
xn  cnx0 , for any n  0, 1, 2, . . .
◆
Thus the solution to any difference equation of the form xn1  cxn is an expo-
nential function xn  cnx0 or xn  x0cn. Such a sequence is known as a geometric
sequence or an exponential sequence in which the ratio of successive terms is a
constant, equal to c, because each term is the same multiple of the preceding term.

The difference equation for exponential growth or decay

xn1  cxn

has as its solution the exponential sequence

xn  x0cn,

where x0 is the starting value for the sequence.

Recall that, whenever the constant multiple c  1, the values for xn get succes-
sively larger; whenever c is between 0 and 1, the subsequent values become succes-
sively smaller. We presented some illustrations of this property in Chapter 2.

Think About This What happens if c  0? ❐

From a somewhat different perspective, the equation Pn1  cPn is a difference

equation for exponential behavior and the formula Pn  P0cn is the solution se-
quence. Whenever c  1, we have exponential growth, and whenever 0  c  1,
we have exponential decay. Thus in any sequence in which each term is a constant
multiple of the preceding term,
xn1  cxn ,
xn is an exponential function of n and xn  x0cn.
gord.3896.05.pgs 4/24/03 10:02 AM Page 373

5.2 Modeling with Difference Equations 373

Alternatively, suppose that Pn1  cPn . If we subtract Pn from both sides of the
difference equation, we get
Pn1  Pn  cPn  Pn  1c  12 Pn  aPn ,
where a  c  1. Note that, if c  1, then a  0 and it is the growth rate; if
0  c  1, then a  c  1  0 and this gives the decay rate. The expression on the
left, Pn1  Pn , is simply the difference between successive values, so we can write
Pn  Pn1  Pn  aPn .
Because we can rewrite any such equation as a difference in this way, we call it a
difference equation.) This difference equation indicates that, for any exponential
process, the successive differences Pn are always a fixed multiple aPn of the quan-
tity itself.
For reference, we recast the previous difference equation xn1  cxn for expo-
nential growth or decay in this alternative format.

The difference equation for exponential growth or decay

xn  axn
has as its solution the exponential sequence
xn  x0 . 11  a2 n  x0cn,

where x0 is the starting value for the sequence.

Writing difference equations in this form is often helpful because the xn for-
mulation lets us think of the change in a quantity from one stage to the next instead
of how one value depends on the preceding value.
We now consider several examples that demonstrate the use of these ideas re-
lating to difference equations for growth and decay.

EXAMPLE 2
Suppose that you deposit $1000 in a bank account paying 5% interest, compounded an-
nually. Write a difference equation to represent the balance in your account after any
number of years and find an expression for the balance at any time from the difference
equation.
Solution From the discussion of exponential growth in Chapter 2, we know that the
balance in the account after any number of years is given by b1t2  100011.052 t. We now
look at this situation from the point of view of difference equations.
Let the original balance be b0  1000. After 1 year, the original $1000 balance has
earned 5% of $1000, or 0.05110002  $50 in interest, so the new balance is
b1  1000  0.0511000 2  11  0.05 21000  11.05 21000  1050.
Symbolically,
b1  b0  0.05b0  1.05b0 .
By the end of the second year, the balance has grown to
b2  b1  0.05b1  1.05b1 .
gord.3896.05.pgs 4/24/03 10:02 AM Page 374

374 CHAPTER 5 Modeling with Difference Equations

Similarly, by the end of third year, the balance is

b3  b2  0.05b2  1.05b2 .
In general, by the end of the 1n  1 2st year, for any n, the balance in the account is
bn1  bn  0.05bn  1.05bn ,
which is the difference equation relating the balance in any year to the balance the next year.
Because this is a difference equation for exponential growth, we immediately know
that the solution after n years is given by
bn  b0 . 11.052 n.
For the initial deposit b0  1000, the balance after n years is
bn  100011.05 2 n,
which is identical to the expression for the exponential growth model.
◆
Before going on, let’s recall the formula at the end of Section 5.1.

For the difference equation

xn1  axn  B,

a formula for the solution sequence is

 ax0  b an,
B B
xn 
1a 1a
where x0 is the initial value for the solution sequence.

If the constant B is 0 in the difference equation xn1  axn  B, then this

equation reduces to xn1  axn , which is the difference equation for an exponen-
tial process. The formula for the solution sequence similarly reduces to xn  x0an,
or the exponential function that we would expect.

Modeling an IRA Account We next consider how to use a difference equation to model
the growth of an IRA account in which a fixed amount of money is invested each year.

EXAMPLE 3
At age 25, Alison sets up an IRA retirement savings account. She invests $3000 annually
into an account that earns 5% interest per year.
a. Write a difference equation that models the balance Bn in her account after n years.
b. Find a formula for the solution sequence to the difference equation.
c. Find the balance in her account on her 60th birthday. How much of this balance is at-
tributed to contributions and how much to accrued interest?
Solution

a. We denote the balance in the account after n years by Bn . During the succeeding year,
the balance grows by 5% of Bn and is then augmented by the next contribution of
$3000. Therefore the balance after n  1 years is
Bn1  Bn  5% of Bn  3000  Bn  0.05Bn  3000  1.05Bn  3000,
gord.3896.05.pgs 4/24/03 10:02 AM Page 375

5.2 Modeling with Difference Equations 375

which is a difference equation model for the balance in the account.

b. Note that the difference equation Bn1  1.05Bn  3000 has the same form as the
difference equation xn1  axn  B, with a  1.05, B  3000, and initial condition
x0  3000  B0 . The solution to this difference equation is

 a3000  b 11.05 2 n
3000 3000
Bn 
1  1.05 1  1.05
 a3000  b 11.052 n
3000 3000

0.05 0.05
 60,000  13000  60,0002 11.052 n
 63,00011.05 2 n  60,000.
Figure 5.8 shows the graph of this solution sequence, with the continuous exponen-
tial function superimposed.

Bn
IRA Account Balance

100,000
80,000
(dollars)

60,000
40,000
20,000
n
0 4 8 12 16 20
Time (years)

FIGURE 5.8

c. At age 60, Alison’s IRA account will have been in existence for n  35 years. There-
fore the balance in the account will be
B35  63,00011.05 2 35  60,000  287,508.97.
Of this total, the amount contributed at $3000 per year is 35130002  105,000,
so the account actually earned about $182,509 in interest.
◆
Modeling a Population with Harvesting We now introduce a population
growth model in which part of the population is removed each time period.
EXAMPLE 4
A poultry farm has 30,000 chickens whose growth rate is 20% per month. Suppose that
5000 chickens are killed (harvested) each month for shipment to stores.
a. Write a difference equation to model this situation.
b. Write a formula for the solution sequence for the difference equation.
c. Discuss the behavior of the solution function and explain the long-term population
pattern of the chickens at the farm.
d. What would happen if 8000 chickens are harvested monthly?
e. Determine the number of chickens that should be harvested monthly to maintain a
constant population from one month to the next.
gord.3896.05.pgs 4/24/03 10:02 AM Page 376

376 CHAPTER 5 Modeling with Difference Equations

Solution

a. Let Cn be the chicken population in the nth month. During that month, the population
grows by 20% and 5000 chickens are harvested. Thus the difference equation model is
Cn1  1.20Cn  5000, C0  30,000.
b. We use the formula for the solution sequence with a  1.20 and B  5000 to get
5000 5000
Cn   a30,000  b 11.20 2 n
1  1.20 1  1.20
5000 5000
  a30,000  b 11.20n 2
0.20 0.20
 25,000  130,000  25,0002 11.202 n
 25,000  5,00011.202 n.
Cn

250,000
Chicken Population

200,000
150,000
100,000
50,000
n
0 4 8 12 16 20
FIGURE 5.9 Time (months)

c. The graph of this function with the continuous function superimposed is shown in
Figure 5.9. Note how it grows in a concave up pattern. The solution function is a mod-
ified exponential function with a vertical shift of 25,000 and a growth factor of 1.20, so
the chicken population at the farm is increasing at an increasing rate. Clearly, this
growth pattern can’t continue indefinitely.
d. If 8000 chickens are harvested each month instead of 5000, the formula for the solution
becomes
8000 8000
Cn   a30,000  b 11.20 2 n
1  1.20 1  1.20
 40,000  130,000  40,0002 11.202 n
 40,000  10,00011.20 2 n.
This solution is an upside-down exponential growth function that has been shifted
up by 40,000, as shown in Figure 5.10. Note that it drops ever more quickly, indicat-
ing that the chicken population will die out very rapidly.
Cn

30,000
Chicken Population

25,000
20,000
15,000
10,000
5000
n
0 1 2 3 4 5 6 7 8
FIGURE 5.10 Time (months)
gord.3896.05.pgs 4/24/03 10:02 AM Page 377

5.2 Modeling with Difference Equations 377

e. To maintain a constant chicken population from one month to the next requires
harvesting the number of chickens that will exactly counterbalance their monthly
growth. If the farm starts with 30,000 chickens and they grow at a monthly rate of
20%, 6000 new chickens will be hatched. Therefore, if 6000 older chickens are killed
monthly, there is no net increase or decrease.

◆
Modeling the Level of Pollutants In Examples 5(a)–(e) we consider the level of
contaminants in a lake under a variety of circumstances to illustrate how models
with difference equations can arise. We also show how the solution to the differ-
ence equation model can be used to determine the behavior pattern for the level
of contamination over time.
EXAMPLE 5(a)
Initially, 600 lb of a contaminant are dumped into a lake, and 10% of it is washed away
each year. Find a formula for the level of contaminant present after any number of years.
Solution Let Cn represent the level of contaminant present after n years; we know that
C0  600. Because 10% of the contaminant present is washed out of the lake during any
year, 90% of the amount present at the start of any year will be left a year later. Therefore
the situation is modeled by the difference equation
Cn1  0.9Cn , C0  600.
This is a difference equation for exponential decay, so we know that the solution is
Cn  60010.9 2 n.
This exponential decay function tells us that the level of contaminant will slowly fall over
time, as shown in Figure 5.11.
Cn
Contaminant Level (pounds)

600
500
400
300
200
100
n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.11 Time (years)

◆
EXAMPLE 5(b)
Initially, there are 600 lb of the contaminant in the lake, 10% of it is washed out each
year, and a manufacturing plant annually dumps 100 lb of the contaminant into a river
that feeds into the lake. Find the level of contaminant present after any number of years.
Solution The amount of contaminant present in the lake is reduced by 10% during a
year, so 90% of the amount present each year remains in the following year. However,
this amount is then increased by an additional 100 lb each year. This situation is mod-
eled by the difference equation
Cn1  0.9Cn  100, C0  600.
gord.3896.05.pgs 4/24/03 10:02 AM Page 378

378 CHAPTER 5 Modeling with Difference Equations

Note that the form of the difference equation is identical to the equations that we devel-
oped previously for the Prozac drug level model and the IRA account balance. Using the
formula for the solution to such a difference equation with a  0.9 and B  100—start-
ing with an initial level of C0  600—gives a formula for the solution sequence

 a600  b 10.9 2 n,
100 100
Cn 
1  0.9 1  0.9
or
Cn  1000  40010.9 2 n.
Its graph is shown in Figure 5.12. To understand the graph, note that the exponential term
40010.9 2 n dies out as n increases. But, because this term is subtracted from 1000, the solution
rises toward 1000 as a horizontal asymptote, which is the limiting value for the contaminant.
Cn
Contaminant Level (pounds)

1000
800
600
400
200
n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.12 Time (years)

EXAMPLE 5(c)
The situation is the same as in Example 5(b), but now the plant increases its annual pro-
duction and thus increases the amount of the contaminant it dumps by 50 lb each year,
starting with the initial level of 600 lb.
Solution The 50-pound per year increase in the amount of contaminant dumped into
the river means that 100 lb are dumped the initial year, 150 lb the following year, 200 lb
the year after that, and so on, following this pattern of linear growth. Thus during the
nth year, the company will dump 100  50n pounds of the contaminant into the river
that feeds the lake. As before, 90% of the contaminant in the lake at the start of any year
remains at the end of that year and is then augmented by the additional amount
dumped into the river. The difference equation that models this situation is

Cn
Contaminant Level (pounds)

7000
6000
5000
4000
3000
2000
1000
n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.13 Time (years)
gord.3896.05.pgs 4/24/03 10:02 AM Page 379

5.2 Modeling with Difference Equations 379

Cn1  0.9Cn  100  50n, C0  600.

Note that this difference equation has a form different from the preceding two: The term
on the right, 100  50n, is no longer a constant but is now a linear function of n, so the
formula that we previously developed doesn’t apply. We show in Supplementary Section
12.3 that a formula for this solution sequence is
Cn  460010.9 2 n  4000  500n.
Note that this formula consists of a decaying exponential term and a linear term with pos-
itive slope. Although the exponential term slowly dies out, the linear term continues to in-
crease as time passes. Over the long term, the level of contaminant eventually increases in a
roughly linear pattern with a slope of 500 lb per year, as shown in the graph in Figure 5.13.
◆
EXAMPLE 5(d)
The situation is the same as in Example 5(b), except that the plant now increases the amount
of contaminant dumped into the river by 20% per year starting with the initial level of 100 lb.
Solution The yearly increase in contaminant dumped into the river is now given by the
exponential function 10011.202 n, so the difference equation modeling this situation is
Cn1  0.9Cn  10011.20 2 n, C0  600.
We show in Supplementary Section 12.3 that the solution sequence is

Cn  a b 10.9 2 n  a b 11.2 2 n.
800 1000
3 3
The first term is an exponential decay function that eventually dies out, while the second
term is an exponential growth function. Early on, the decay term makes a contribution,
but because its coefficient 1266.72 is smaller than the growth term’s coefficient
1333.32 , the contribution is minimal and rather quickly diminishes. The overall be-
havior pattern is one of roughly exponential growth in the amount of contaminant, as
illustrated in Figure 5.14. The eventual exponential growth factor is about 1.2 (verify this
result by calculating a pair of successive terms in the solution for moderately large values
of n—say, n  20 and n  21), so the annual growth rate is eventually about 20%.

Cn
Contaminant Level (pounds)

14,000
12,000
10,000
8000
6000
4000
2000
n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.14 Time (years)

◆
EXAMPLE 5(e)
The situation is the same as in Example 5(b), with the plant initially dumping 100 lb of the
contaminant, but EPA regulations require that it reduce the level of dumping by 25% per year.
gord.3896.05.pgs 4/24/03 10:02 AM Page 380

380 CHAPTER 5 Modeling with Difference Equations

Solution The amount dumped into the river is now represented by the decaying expo-
nential function 10010.752 n, so the corresponding difference equation is
Cn1  0.9Cn  10010.75 2 n, C0  600.
In Supplementary Section 12.3 we show that the solution sequence is

b 10.9 2 n  a Cn  a
b 10.752 n.
3800 2000
3 3
Both terms decay to zero exponentially, so the level of contaminant will eventually die
out. The graph of the solution sequence in Figure 5.15 shows a surprising pattern in
which the level of contaminant first increases and then dies out. Let’s see why. Suppose
that we calculate the first few terms of the solution sequence:
Cn
Contaminant Level (pounds)

700
600
500
400
300
200
100
n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.15 Time (years)

C1  a b 10.9 2 1  a b 10.75 2 1
3800 2000
 1140  500  640;
3 3
C2  a b 10.9 2 2  a b 10.75 2 2
3800 2000
 1026  375  651;
3 3
C3  a b 10.9 2 3  a b 10.75 2 3
3800 2000
 923.4  281.25  642.15;
3 3
C4  a b 10.9 2 4  a b 10.75 2 4
3800 2000
 831.06  210.94  620.12.
3 3
Obviously, both 10.9 2 n and 10.75 2 n decrease as n increases, but the second term de-
creases more rapidly because its decay factor 0.75 is considerably smaller than the
decay factor 0.90 of the first term. As a result, when n is small (n  1 or 2), the second
term has a much greater effect on the result than it does when n gets larger—the
amount subtracted from the first term is relatively large at first but then quickly di-
minishes. As less is subtracted away, the solution increases somewhat at first; however,
eventually the second term has minimal effect and the first term decays toward zero, as
we expect. In particular, C10  404.12, C20  151.88, and C30  53.58.
◆
The Fibonacci Model for Population Growth We next consider one of the earliest
mathematical models for a biological process, developed by Italian mathematician
Fibonacci, who lived about 1200 A.D. Fibonacci constructed a simple model for pre-
dicting the local rabbit population based on the following assumptions.
1. Newborn rabbits mature in 1 month.
2. Once they have matured, rabbits have litters monthly.
3. Each litter consists of one male and one female.
gord.3896.05.pgs 4/24/03 10:02 AM Page 381

5.2 Modeling with Difference Equations 381

The first two assumptions are fairly accurate, but we can argue about the third
assumption on a variety of grounds. First, rabbit litters tend to be considerably
larger than 2. However, there is a certain mortality rate for newborn rabbits
(they can’t run fast enough to escape from the predators in the neighborhood)
that lowers the number per litter that survive to maturity. Second, expecting one
male and one female to survive from each litter is unreasonable. However, if we
consider a large population of rabbits, the numbers of males and females for the
entire population average out to about a 50–50 split per litter.
Let’s start with one pair of newborn rabbits—one male, the other female—on
January 1 of some year, By February 1, the original pair are now mature and ready
to do what rabbits do best: produce new rabbits. On March 1, there are two pairs
of rabbits, the original pair and their first set of offspring. By April 1, the original
pair has produced another litter while their first litter has matured and is ready to
enter the family business. Thus there are now three pairs of rabbits, two mature
and one newborn. However, by this stage, things are starting to get complicated,
so we use the diagram shown in Figure 5.16 to keep track of the rabbits. Let the
symbol denote an immature pair and represent a mature breeding pair.
Date Rabbit pairs Number of pairs

January 1 1

February 1 1

March 1 2

April 1 3

May 1

or 5

June 1

or 8

July 1

or 13

FIGURE 5.16

Consider the rabbit population on July 1, say, which consists of 8 mature pairs
and 5 immature pairs. Note that the number of mature pairs, 8, is equal to the
total number of pairs in June, the preceding month. Also, the number of imma-
ture pairs, 5, is equal to the population 2 months earlier on May 1. Therefore the
rabbit population on July 1 is equal to the population on June 1 (mature) plus the
population on May 1 (immature). This pattern is not coincidental, and it persists
indefinitely—in each month, the rabbit population is equal to the sum of the
population values the preceding two months. Let’s see why.
The population in the current month Pn consists of breeding pairs and new-
born pairs. The number of breeding pairs this month is equal to the total popula-
tion Pn1 last month––all are still alive and all are now mature. Now think about
gord.3896.05.pgs 4/24/03 10:02 AM Page 382

382 CHAPTER 5 Modeling with Difference Equations

the number of newborn pairs this month. Every rabbit alive two months ago, Pn2,
was mature last month and so gave birth to a new litter this month. Therefore the
number of newborn pairs this month must be equal to the total population two
months ago. So for any n  2,
Pn  Pn1  Pn2 .
This is a difference equation relating the successive population values.
Alternatively, instead of considering the current population Pn in terms of
the preceding two months’ populations (a backward difference equation), we
look at the population two months ahead Pn2 , which is determined by the cur-
rent population Pn and next month’s population Pn1 . As a result, we can rewrite
the difference equation in the equivalent forward form
Pn2  Pn  Pn1 , n  0,
starting with P0  P1  1. It is known as the Fibonacci difference equation or the
Fibonacci model. Note that with either form we get the population values
1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, . . .
This particular sequence of numbers is called the Fibonacci sequence.

Fibonacci Sequence
51, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, . . . 6

These numbers arise in a surprising variety of ways—in nature (e.g., the arrange-
ment of petals on sunflowers and the number of rings in seashells), in economics,
in human psychology, and in art (see Example 7 of Section 3.2). However, here we
continue to focus on the constantly growing rabbit population of old Italy.
EXAMPLE 6
Construct a table based on Fibonacci’s model for the rabbit population over the first
30 months and discuss the growth in this population.
Solution Starting with the initial values P0  P1  1, the Fibonacci difference equa-
tion gives the values presented in the table and the graph shown in Figure 5.17 for the
first 30 months.

Fibonacci's rabbit model

Pn

1.4
Pairs of rabbits (millions)

1.2
1
0.8
0.6
0.4
0.2
n
0 5 10 15 20 25 30
FIGURE 5.17 Time (months)
gord.3896.05.pgs 4/24/03 10:02 AM Page 383

5.2 Modeling with Difference Equations 383

Month Pairs Month Pairs Month Pairs

0 1 11 144 21 17,711

1 1 12 233 22 28,657

2 2 13 377 23 46,368

3 3 14 610 24 75,025

4 5 15 987 25 121,393

5 8 16 1,597 26 196,418

6 13 17 2,584 27 317,811

7 21 18 4,181 28 514,229

8 34 19 6,765 29 832,040
9 55 20 10,946 30 1,346,269

10 89

Both the table and Figure 5.17 show a population explosion among the rabbits very
quickly. After the first few entries in the table, the ratio of successive terms is approxi-
mately 1.618. For example, 233>144  1.618, 377>244  1.618, and so on. These out-
comes suggest that, eventually, the growth pattern is roughly exponential with a growth
rate of 0.618  61.8% per month.
◆
These numbers are, if anything, conservative because each litter will likely con-
tain more than two rabbits. However, deaths among the rabbits have been ignored.
Consequently, the Fibonacci model for the rabbit population may not be a partic-
ularly good match to the actual population.
Based on Fibonacci’s model, Italy clearly would have had a major overpopula-
tion problem with rabbits in his time, let alone by now. Because that hasn’t hap-
pened, something is wrong either with the mathematical model or the assumptions
on which it is based. Actually, the mathematical model is fairly accurate—at least up
to a point. So long as the rabbit population remains relatively small, the model gives
numbers that reasonably estimate the population. However, it shouldn’t be carried
too far because no process can continue to grow exponentially indefinitely. Instead,
other factors that act to curb the growing population must be taken into account.
For example, as the number of rabbits increases, so too will the number of foxes and
other predators that live off them. In turn, the larger numbers of predators eventu-
ally reduce the rabbit population. Also, when the rabbit population grows too large,
they quickly consume most of the available food supply and there won’t be enough
food to sustain such a large population. The result is starvation until the population
decreases to a more sustainable size. We discuss the mathematical details of this
more realistic type of scenario in Section 5.3.
A difference equation that relates one term of a sequence xn1 to the preceding
term xn is called a first order difference equation. It is the primary type of difference
equation we cover in this chapter and in Supplementary Chapter 12. A difference
equation such as Fibonacci’s that relates one term xn2 to the preceding two terms xn
and xn1 is called a second order difference equation.
gord.3896.05.pgs 4/24/03 10:02 AM Page 384

384 CHAPTER 5 Modeling with Difference Equations

Problems

1. In Example 3 about Alison’s IRA account, (a) what a. If she plans to withdraw $10,000 yearly, write a
would the account be worth at age 65 instead of age difference equation for the balance in the account.
60? and (b) what would it be worth at age 65 if she b. How long will it take for the balance in the ac-
started the account at age 20 instead of age 25. count to be depleted if she withdraws $10,000
every year?
2. Repeat the calculations in Example 3 and in Problem
c. Suppose that she plans to withdraw 20% of the
1 if the rate of return for the IRA account is 6% in-
account balance every year. Write a difference
stead of 5%.
equation for the balance in the account.
3. In Example 4 about the population of chickens, we d. How long will it take for the balance to be de-
assumed that there were 30,000 chickens on the pleted with this withdrawal plan?
farm and a growth rate of 20% per month. We e. Is there a fixed amount she can withdraw from
showed that, if the owner kills 8000 chickens a the account every year without diminishing the
month, the population eventually will die out. As- balance? If so, find it.
suming that the owner doesn’t notice this develop-
7. Marine biologists estimate that there were about
ing problem, how long will it be until no chickens
8000 bowhead whales in the waters near Alaska in
are left?
1992 and that they were growing at an annual rate
4. Suppose that the population of a certain species of of 3%. Alaskan Eskimos are allowed to catch about
fish in a lake grows exponentially with a growth rate a. 50 whales per year.
Write a difference equation to model each situation. a. Write a difference equation giving the popula-
a. The fish population grows exponentially. tion of the whales from one year to the next.
b. Fishermen catch and remove 100 fish from the b. Calculate the projected whale population each
lake each year. year until 2005.
c. Fishermen catch and remove 40% of the fish
c. Determine the largest number of whales that the
from the lake each year.
Eskimos could catch each year without the whale
d. The number of fish that fishermen remove from
population going into decline.
the lake each year is proportional to the square
root of the number of fish in the lake. d. Suppose that the Eskimos are petitioning the
e. Fishermen remove 40% of the fish each year, and government to increase their annual whale har-
the state’s wildlife department restocks the lake vest and you are acting as their representative
with 500 fish each year. before the panel making the decision. What ar-
guments would you use to justify increasing the
5. Jack and Jill are setting up plans for a retirement
annual harvest?
fund. Write a difference equation for the balance bn
in this account for each of the scenarios they are e. Suppose that you were representing a conserva-
contemplating. tion group opposed to increasing the annual
a. They deposit $2000 in an account guaranteed to whale harvest. What arguments would you use
pay 6% interest per year. to request a denial of the petition?
b. They deposit $2000 initially in an account that 8. A company expects the productivity of new em-
pays 6% interest per year and then deposit an ployees to increase each day as they gain experience.
additional $1000 each year. When a new person starts “cold,” the company ex-
c. They deposit $2000 initially in the account pay- pects the employee to produce P0 items per hour.
ing 6% per year and then increase their contri- The following day, hourly production should in-
bution by $1000 each year. crease by one item per hour to P0  1; the day after
d. They deposit $2000 initially into the account that, production should increase by 2 items per
paying 6% per year and then increase their con- hour; then by 3 items per hour; and so on.
tribution by 10% each year. a. Write a difference equation to model this situation.
6. Claire has $80,000 in a retirement fund that pays b. Find an expression for the solution to this differ-
6% interest per year. ence equation for any number of days n.
gord.3896.05.pgs 4/24/03 10:02 AM Page 385

5.3 The Logistic or Inhibited Growth Model 385

c. Write a paragraph discussing whether this ex- 12. Tribbles are adorable, furry little creatures. The only
pectation for continued improvement seems to trouble with tribbles is that they breed like tribbles.
be sensible. Specifically, suppose that a tribble matures in 3 days
9. Psychologists have found that, when a person learns a and then reproduces asexually daily by splitting off
new body of knowledge, the amount of new knowl- a new tribble on the fourth day and every day there-
edge gained in any time period is proportional to the after.
amount that the person does not know. That is, it is a. Construct a difference equation for the tribble
easier to improve when you know a little than it is to population, starting with one newborn tribble,
improve when you know a lot. Suppose that Greg, by expressing Tn3 in terms of Tn , Tn1 , and Tn2 .
while preparing for the SAT vocabulary test, is trying (Hint: Let ^  newborn tribble,   day old
to learn 400 new words from a set of flash cards. immature tribble and   mature tribble and
a. Write a difference equation for this learning keep track of the number of each over the first
model based on the number of words Wn that 10 days.)
Greg knows out of the 400 total on the nth pass b. Use the difference equation to calculate the trib-
through the deck. ble population during the first 15 days, based on
b. Is the constant of proportionality in the differ- an initial newborn tribble the first day.
ence equation positive or negative? Is it less than c. Examine the ratio of successive terms to deter-
1 or greater than 1? mine whether the tribble population appears to
c. Write a paragraph explaining why it is reason- be growing exponentially. If it does, what is the
able to expect Greg to learn more words during exponential growth rate?
the first few passes through the deck of cards d. Using the result of part (c), estimate the tribble
than through later passes through it. population after a full year.
d. Sketch a graph of the possible number of new e. Assume that the human population of the Earth
words that Greg learns as a function of the num- is currently 6 billion and growing exponentially
ber n of passes through the deck based on this at an annual rate of 1.7%. Estimate when every
learning model. human being alive will have a tribble of his or
10. Repeat the calculations associated with Fibonacci’s her own.
rabbit model for the first year, assuming an initial 13. Consider the difference equation xn1  xn  kn,
population of 10 pairs of newborn rabbits. How do for any constant multiple k. Show that the solution
your values relate to those presented in the text? is always a quadratic function of n. (Hint: Use a re-
11. Suppose that a particular breed of rabbits take sult from Section 4.5.)
2 months to mature instead of 1 month, but that 14. Consider the difference equation xn1  xn  kn2,
Fibonacci’s other assumptions still hold. Calculate the for any constant multiple k. Show that the solution
rabbit population each month during the first year. is always a cubic function of n.

5.3 The Logistic or Inhibited Growth Model

In Section 5.2 we demonstrated that a population Pn growing exponentially satis-
fies the difference equation
Pn1  cPn
for all values of n. A formula for Pn is
Pn  cnP0 ,
where c is the growth (or decay) factor.
Alternatively, if we subtract Pn from both sides of the difference equation, we get
Pn1  Pn  cPn  Pn  1c  12 Pn  aPn ,
or
Pn  Pn1  Pn  aPn ,
gord.3896.05.pgs 4/24/03 10:02 AM Page 386

386 CHAPTER 5 Modeling with Difference Equations

where a is the growth (or decay) rate. In this form, the difference equation tells us
that, for any exponential process, the successive differences are always a fixed mul-
tiple—the growth (or decay) rate—of the quantity itself.

The Logistic Growth Model

The exponential growth model is effective in predicting the growth of most popula-
tions over the short run. However, if any growth process were to continue indefinitely,
other factors come into play to slow down, or inhibit, the rate of growth. We now
modify the exponential growth difference equation Pn  aPn to model a situation in
which a population starts growing exponentially, then slows, and eventually levels off
as the population reaches the maximum size that can be sustained by the environ-
ment. This type of behavior is illustrated in Figure 5.18.

Leveling off
Population

Exponential growth

Time

FIGURE 5.18

In the 1830s, the Belgian biologist Verhulst developed a simple extension of the
exponential growth model Pn  aPn to reflect this situation. He introduced an
extra term in the difference equation—one that serves to decrease the rate of growth
in the population—to account for this leveling-off effect. Verhulst and other scien-
tists who have since studied such processes have observed that these processes can be
modeled by subtracting a term that is proportional to the square of the population.
Thus we use the expression for the change in Pn ,
Pn  Pn1  Pn  aPn  bPn2 , ˇ

where b is the inhibiting constant, which typically is much smaller than the growth
rate a. Alternatively, if we add Pn to both sides, we can rewrite this difference equa-
tion as
Pn1  11  a2Pn  bPn2 .
ˇ

In either form, the resulting process is known as the logistic growth model or the
inhibited growth model. Note that, if b  0, the logistic growth model reduces to
the exponential growth model.

The Logistic or Inhibited Growth Model

Pn  Pn1  Pn  aPn  bPn 2 or Pn1  11  a2Pn  bPn2 ,

where b is much smaller than a.

gord.3896.05.pgs 4/24/03 10:02 AM Page 387

5.3 The Logistic or Inhibited Growth Model 387

To work with this logistic growth model, we must get some feel for appropriate
values for the inhibiting constant b. You may want to experiment, using your
graphing calculator (if it has difference equation capabilities) or appropriate com-
puter software. For instance, suppose that the exponential growth rate is a  1 for a
species such as rabbits that breeds rapidly and that the inhibiting constant is
b  0.04. These values give a graph similar to the one shown in Figure 5.18, and a
curve with this shape is called a logistic curve. However, if you try b  3 instead,
say, the result will be quite different. (We discuss the resulting type of chaotic
behavior in Supplementary Section 12.8.) In fact, with a little experimentation,
you will find that the model is an effective description of inhibited population
growth when b is much smaller than a.
EXAMPLE 1
Consider the logistic model with a  1 and b  0.0004 for a rabbit population, where n
represents the number of months. If the initial rabbit population is P0  1 pair, use the
difference equation to calculate the population over the first 16 months. What does the
maximum sustainable population appear to be?

Solution With a  1 and b  0.0004, the logistic difference equation is

Pn1  11  a2 Pn  bPn2  2Pn  0.0004Pn2 .
Therefore, using n  0 and P0  1, we get
P1  2P0  0.0004P02  211 2  0.0004112 2  2.
Continuing with n  2 and n  3, we get
P2  2P1  0.0004P12  212 2  0.0004122 2  4;
P3  2P2  0.0004P22  214 2  0.0004142 2  8;
and so on. Continuing this process, we obtain the following table of values and the graph
shown in Figure 5.19.

Month Number of Pairs Month Number of Pairs

Pn 0 1 9 463

2500 1 2 10 840

2000 2 4 11 1398
Number of pairs

1500 3 8 12 2014

1000 4 16 13 2406

500 5 32 14 2496

n 6 63 15 2500
0 2 4 6 8 10 12 14 16
Time (months) 7 125 16 2500
FIGURE 5.19 8 243

Note that the population increases to a maximum of 2500 and seems to remain at that
level. Note also that the population grows rapidly to about half this maximum during the
first 11 or so months and then grows more slowly thereafter until it reaches the 2500 level.
◆
gord.3896.05.pgs 4/24/03 10:02 AM Page 388

388 CHAPTER 5 Modeling with Difference Equations

This behavior pattern is typical of the logistic model: An initial spurt in the
population is followed by a slower rate of growth and then an eventual leveling off
to a constant fixed population known as the maximum sustainable population or
the limit to growth. (This type of horizontal asymptote is similar to what happens
with the maintenance level for a drug discussed in Section 5.1.)
The logistic curve changes from concave up to concave down, so it has one
point of inflection. We know that one characteristic of a point of inflection is that
the function is growing most rapidly or decreasing most rapidly there. To deter-
mine roughly where the point of inflection occurs, we can examine the table to es-
timate where the largest increase in the population occurs by looking at the
differences in successive values of the population. In the preceding table the popu-
lation has its largest increase during the 11th month when it jumps from 1398 to
2014, an increase of 616.

EXAMPLE 2
In illustrating the logistic model, we used a  1 and b  0.0004. How does the behavior
of the solution sequence change if we use a  1 and b  0.000032 instead?
Solution The corresponding results (rounded to the nearest integer) based on the lo-
gistic difference equation Pn1  2Pn  0.000032Pn2 over the first 20 months are shown
in the following table.
Pn
Month Number of Pairs Month Number of Pairs
35
Number of pairs (in thousands)

30 0 1 11 1,982
25 1 2 12 3,839
20
2 4 13 7,206
15
10 3 8 14 12,751
5
4 16 15 20,299
n
0 2 4 6 8 10 12 14 16 18 20 5 32 16 27,412
Time (months)
6 64 17 30,779
FIGURE 5.20
7 128 18 31,243

8 255 19 31,250

9 508 20 31,250

10 1,007

The resulting pattern, shown in Figure 5.20, has the same logistic shape as that
shown in Figure 5.19. In fact, the first few population values are the same as in the
previous table because we start with the same initial population P0  1, the initial
growth rate a is the same, and the inhibiting term has minimal impact while the pop-
ulation is small. Also, because b is now smaller than before, the population grows at a
faster rate for a longer time and so the population now has a higher maximum sus-
tainable level—31,250—than in Example 1. Further, the point of inflection now oc-
curs during the 14th month when the population grows from 12,751 to 20,299.
◆
gord.3896.05.pgs 4/24/03 10:02 AM Page 389

5.3 The Logistic or Inhibited Growth Model 389

The Maximum Sustainable Population

To understand the behavior of the solution of the logistic difference equation, let’s
choose a constant b that is much smaller—say, by a factor of 1>1000—than a. So
long as the population Pn remains relatively small, the inhibiting term, bPn2 , in
the logistic difference equation
Pn  aPn  bPn2
is negligible compared to the exponential term, aPn . Therefore, initially, the equa-
tion is essentially equivalent to the difference equation
Pn  aPn
for exponential growth. As Pn grows larger, the inhibiting term bPn2 becomes
larger at a faster rate and so has an ever-greater impact on reducing the value of
Pn . Hence the change in Pn begins to decrease. Of course, the fact that the
change decreases does not necessarily mean that Pn itself gets smaller; it simply
doesn’t grow as fast. Moreover, the values for Pn eventually approach a horizon-
tal limit, with no further population growth.
Now let’s find a formula for the maximum sustainable population. At this max-
imum level, there should be no change in Pn , so that the difference between succes-
sive terms, Pn  Pn1  Pn , should be zero. Therefore we set the right-hand side of
the logistic equation Pn  aPn  bPn2 to 0 and obtain
aPn  bPn2  Pn 1a  bPn 2  0.
Because Pn
0, we must have a  bPn  0, so that
a
Pn  .
b

The maximum sustainable population occurs at the level of P  a>b  L.

This ratio represents the maximum possible population and is the value, or
height, of the horizontal asymptote where the population stabilizes forever. In Exam-
ple 1 we used a  1 and b  0.0004 so that L  a>b  2500. In Example 2 we
used a  1 and b  0.000032 so that L  a>b  31,250. These values agree with
what we got numerically.
What if the original population P0 is larger than the maximum sustainable pop-
ulation? For instance, in Example 1, if the initial population were greater than 2500,
what would the model predict? In general, for any n, suppose that Pn  L  a>b.
When we multiply both sides of the inequality Pn  a>b by the positive constant b,
we get bPn  a so that a  bPn  0. Because Pn  0,
Pn  aPn  bPn2  1a  bPn 2 Pn  0
and the population is decreasing. Thus the phrase maximum sustainable population
means just what it says: There are too many organisms for the environment to sup-
port, and the population will decline due to starvation, predators, and other inhibiting
factors. The graph for this scenario is shown in Figure 5.21. You may want to explore
this situation further, using a graphing calculator or computer program.
gord.3896.05.pgs 4/24/03 10:02 AM Page 390

390 CHAPTER 5 Modeling with Difference Equations

Pn

P0

n
FIGURE 5.21

The Point of Inflection

We next locate the point of inflection, at which the logistic curve changes concavi-
ty. It represents the point at which the population is growing most rapidly, so we
want to find where the change in the population, Pn  Pn1  Pn , is greatest, as
shown in Figure 5.22. To the left of this point, the logistic curve is concave up, so
not only is the population increasing, but it is also increasing at an increasing rate
3 1 Pn 2  0 4 . To the right of this point, the logistic curve is concave down, so the
population is increasing at a decreasing rate 3 1 Pn 2  0 4 . At the point of inflec-
tion, the population is growing most rapidly, so Pn must be a maximum. Biologi-
cally, the most vigorous growth in a population occurs just before decline sets in.

∆Pn

Pn ∆Pn = aPn − bPn2

∆(∆Pn ) < 0

P0 ∆(∆Pn ) > 0
Pn
n L L
FIGURE 5.22 FIGURE 5.23 2

Let’s now find a formula for the location of the point of inflection for the solu-
tion sequence of the logistic difference equation,
Pn  aPn  bPn2 .
Although we typically think of Pn as a function of time n, this equation also express-
es the change, Pn , in the population as a quadratic function of the population size
Pn , as shown in Figure 5.23. Note that the parabola opens downward because the
leading coefficient, b, is negative. Furthermore, the quadratic equation
aPn  bPn2  Pn 1a  bPn 2  0
has two real roots, one when Pn  0 and the other when Pn  L  a>b, which is the
limit to growth. At both of these extremes, Pn is zero and the population isn’t
growing. Moreover, because a parabola is symmetric about its vertex, the maximum
value for Pn occurs at the midpoint of this interval, which is when Pn is half of L, or
2 1a>b2. This maximum value for Pn corresponds to the point of inflection.
1
gord.3896.05.pgs 4/24/03 10:02 AM Page 391

5.3 The Logistic or Inhibited Growth Model 391

The point of inflection occurs at the level of P  12 1a>b2  12 L.

EXAMPLE 3
Find the point of inflection for the rabbit population in Example 2. How does it match
what we observed by comparing values in the table?
Solution In Example 2, based on a  1 and b  0.000032, the maximum sustainable
population was L  31,250, so the point of inflection must occur at a height of
2 L  15,625. We previously observed that the point of inflection occurred during the 14th
1

month when the population grew most rapidly, jumping from 12,751 to 20,299. Hence the
value obtained from the formula agrees well with our previous observation on the data.
◆
The solution sequence of any difference equation consists of a discrete set of
numbers, so it isn’t reasonable to expect that any one of them will precisely equal 12 L.
Thus the best we can usually do is to estimate the location of the point of inflection
from a table of data values.
We can summarize the logistic growth model as follows.

Summary of the Logistic Growth Model

The logistic difference equation is
Pn  aPn  bPn2 or Pn1  11  a2Pn  bPn2 ,
where a is the initial growth rate and b is the inhibiting constant.
The maximum sustainable population is L  ab .
The point of inflection occurs when

Pn  L  a b .
1 1 a
2 2 b

We’ve looked at only one specific application of the logistic model concerned
with population growth for a single species. The same mathematical model applies
to most other species—only the values of the constants a and b change.

EXAMPLE 4
A bacterial culture grows according to the logistic model with a  0.4 and b  0.00008. If
there are initially 500 bacteria in the culture, find (a) the number present for n  1, 2, . . . , 6,
(b) the limiting population for the culture, and (c) the location of the point of inflection.
Solution

a. Because the bacterial culture satisfies the logistic model, we know that

Pn1  11  a2Pn  bPn2

 1.4Pn  0.00008Pn2 ,
gord.3896.05.pgs 4/24/03 10:02 AM Page 392

392 CHAPTER 5 Modeling with Difference Equations

starting with P0  500. Therefore we find successively that

P1  1.415002  0.000081500 2 2  700  20  680;
P2  1.416802  0.000081680 2 2  952  37  915;
P3  1.419152  0.000081915 2 2  1214;
P4  1.4112142  0.0000811214 2 2  1582;
P5  1.4115822  0.0000811582 2 2  2015;
P6  1.4120152  0.0000812015 2 2  2496.
We rounded each successive entry to the nearest whole number because we’re dealing
with the number of bacteria in the culture.
b. The maximum sustainable population is
a 0.4
  5000.
b 0.00008
c. The inflection point occurs at

a b  15000 2  2500,
1 a 1
2 b 2
or when the population passes the 2500 level, as depicted in Figure 5.24. Note that the
first six terms of the sequence grew to almost one-half the maximum value. Howev-
er, considerably more than another six terms will be required to get close to the lim-
iting value. For example, P12  4680 and P18  4984.

Pn

5000

4000
Population

3000

2000

1000

n
0 2 4 6 8 10 12 14 16 18 20
FIGURE 5.24 Time

◆
This type of mathematical modeling is the foundation for most of the projections of
limits on world population growth. Moreover, because the growth factor 1  a contains
the exponential growth rate a, you might also want to interpret some of these ideas in
the context of the values given in the population table in Appendix G. In particular, you
might explore the results of using some of the growth rates shown for different countries
and assume different values for the inhibiting constant b to see the effects on the limits to
growth. Remember, though, that b should be much smaller than a.We discuss how to es-
timate values for a and b, based on actual population data, later in this section.

Behavior of the Logistic Function

Let’s now examine the behavior of the solution to the logistic difference equation in
greater detail. Suppose that the maximum sustainable population for an environ-
ment is L  1000. The point of inflection then occurs at the level where Pn  500.
gord.3896.05.pgs 4/24/03 10:02 AM Page 393

5.3 The Logistic or Inhibited Growth Model 393

Case 1: If the initial population P0 is less than 500, the population will begin growing
in a concave up manner. When it passes a height of 500, the concavity changes
and the population continues to grow, but in a concave down manner as it ap-
proaches its horizontal asymptote at the level of 1000, as shown in Figure 5.25.
Pn

1500
Case 3
L = 1000
Case 2
500
Case 1

n
FIGURE 5.25

Case 2: If the initial population P0 is between 500 and 1000, it starts above the
point of inflection. The logistic curve grows toward the limit to growth,
1000, in a concave down manner, as shown in Figure 5.25.
Case 3: Suppose that the initial population P0 is greater than the maximum
sustainable level of 1000. This can occur if there is a sudden influx of
immigrants, an unexpected baby boom, or a change in conditions such
as a drought or famine that reduces the maximum sustainable level. The
resulting logistic curve starts above the limiting value of 1000 and de-
creases toward 1000 in a concave up manner, as shown in Figure 5.25.
Case 4: Finally, if the initial population value P0 precisely equals the limiting
value L, all subsequent values remain the same. To see this, we consider the
logistic difference equation
Pn  aPn  bPn2
when Pn  L  a>b. Then

a 2 a2 a2 a2 a2
Pn  a . a b  b . a b   b . a 2b 
a
  0.
b b b b b b

There is no change, so the population remains constant thereafter. For this

reason, the limiting value is also considered the equilibrium because the
population remains balanced at that level. (The maintenance level for a
drug is also an equilibrium.)
Although we typically think of a logistic curve as the S-shaped curve corre-
sponding to initial population values below the inflection point, the other cases are
also logistic curves because they are solutions to the logistic difference equation.
More generally, suppose that we have the logistic model
Pn  aPn  bPn2  1a  bPn 2Pn ,
with limiting value L  a>b and point of inflection at height 12 L. If we factor out
the coefficient b from the right-hand side of the difference equation, we get

Pn  b . a  Pnb Pn  b1L  Pn 2 Pn .
a
b
gord.3896.05.pgs 4/24/03 10:02 AM Page 394

394 CHAPTER 5 Modeling with Difference Equations

Now suppose that, for some value of n, Pn  L; then L  Pn  0. Therefore

Pn  b1L  Pn 2Pn  0. Because Pn  Pn1  Pn , we see that Pn1 must be smaller
than Pn . That is, the solution must be decreasing at such a point. We illustrate this
behavior in the top region shown in Figure 5.26 where the arrows are all pointing
downward to indicate the direction for the solutions.
Pn

1
2L

FIGURE 5.26

Next, suppose that, for any value of n, 0  Pn  L, so L  Pn  0. Therefore

Pn  b1L  Pn 2Pn  0 and so Pn1 must be larger than Pn . That is, the solution
must be increasing. We show this behavior in the two middle regions of Figure
5.26 where the arrows are all pointing upward to indicate the direction of the so-
lutions. However, the slopes of the arrows crossing the level for the inflection
point are steeper than those below or above that level because the logistic curve
increases most rapidly at its inflection point.
Finally, although this case is not realistic as a model for population growth, suppose
that Pn  0 for some n.Then L  Pn  0 and therefore Pn  b1L  Pn 2 Pn  0. Con-
sequently, Pn1 must be smaller than Pn and the solution must be decreasing.We show this
behavior in the bottom region of Figure 5.26 where all the arrows are pointing downward.
The cases we consider here with the logistic difference equation model are some-
what simplistic. For instance, if the values for a and b are larger than the ones we have
used, the solution can overshoot the equilibrium level L for some value of n. Howev-
er, once there, the succeeding term must decrease toward (or actually even past) the
equilibrium. Thus it is possible to obtain a solution that oscillates above and below
the equilibrium level while still converging to it. In many ways, this scenario is more
realistic than the ideal one we portrayed, in which the solution strictly increases to-
ward L. A real population is likely to grow beyond its limits, then decrease below the
maximum sustainable level, and oscillate in this manner thereafter.

Applications of Logistic Growth

The mathematical ideas used in the logistic model for population growth can be
applied to many other situations, as Examples 5 and 6 illustrate.
EXAMPLE 5
Suppose that 4000 people are in a university dormitory complex when one student starts
a rumor at 9 A.M. Saturday morning. If each person who hears the rumor passes it on to
two other people every hour, how long will it be until everyone has heard it?
Solution At first thought, this situation might seem like an exponential growth model
because, seemingly, the number of people who have heard the rumor triples every hour.
gord.3896.05.pgs 4/24/03 10:02 AM Page 395

5.3 The Logistic or Inhibited Growth Model 395

In practice, the number of people who hear the rumor does grow roughly exponentially
for a while, but eventually the people who are passing the rumor on will find it difficult,
if not impossible, to encounter two people who haven’t yet heard it.
We must change the underlying exponential formulation to introduce an inhibiting
term. The number of new people who hear the rumor after n  1 hours (i.e., the change
in the number who have heard it during the past hour) depends not just on how many
people Pn have already heard it, but also on the number of people left of the original 4000
people, 14000  Pn 2 , who haven’t heard it. That is, Pn , the change in Pn , is proportional
to both Pn and 14000  Pn 2 , which leads to the difference equation
Pn1  Pn  mPn . 14000  Pn 2  4000mPn  mPn2 ,
where m is the constant of proportionality. Adding Pn to both sides of the equation, we get
Pn1  11  4000m2 Pn  mPn2  11  a2Pn  bPn2 ,
where we have written a  4000m and b  m to emphasize that this is a logistic model.
Initially, P0  1 and, because each person tells two new people, a  2 and 1  a  3.
The limiting value is L  a>b  4000m>m  4000, so
a 2
b   0.0005.
L 4000
The resulting difference equation is
Pn1  3Pn  0.0005Pn2 .
The following table shows the number of people (rounded to the nearest whole person)
who have heard the rumor after each hour, based on this logistic model. The graph of
these solution points is shown in Figure 5.27. Thus we conclude that the 4000 students
will have heard the rumor in about 9 hours.
Pn
Time Number
5000
0 1
Number of People

4000

3000 1 3
2000 2 9
1000
3 27
n
0 1 2 3 4 5 6 7 8 9 4 80
Time (hours)
5 238
FIGURE 5.27
6 686
7 1823

8 3807

9 4174
◆
The final value displayed, 4174 when n  9, actually overshot the limiting
value of 4000. As we said before, once a value of a logistic sequence exceeds L, the
following value determined by the difference equation will be smaller because
Pn  0 and, in fact, will usually overshoot downward to produce a value below L.
gord.3896.05.pgs 4/24/03 10:02 AM Page 396

396 CHAPTER 5 Modeling with Difference Equations

Think About This Continue the preceding process repeatedly to verify that the successive values oscillate
above and below the limiting value 4000 and eventually converge to it as n increases. ❐
One problem with the logistic difference equation model is that no one has
ever been able to discover a closed form solution for Pn . However, the fact that no
explicit solution is known isn’t a major handicap. Other than the aesthetic pleas-
ure of having the solution expressed as a formula, a closed form solution would-
n’t likely contribute much to this model. The formula would be used primarily to
calculate the values for the population at each time. However, the logistic differ-
ence equation itself allows us to calculate the values for the solution recursively
by using the previous value at each time. There is a slightly different way to cre-
ate a continuous logistic model using calculus. That model can be solved to give
a closed form solution for a logistic curve of the form

f 1t2 
C
,
1  AeBt
where A, B, and C are positive constants and e  2.71828 . . . is the base of the nat-
ural logarithm system, as we discussed in Section 4.9.
Applications of the logistic growth model include the spread of technological
innovations—how fast a new product penetrates the marketplace. For example,
when vacuum cleaners were first introduced, they were purchased by relatively few
people. Based on word of mouth and advertising, more and more people bought
them and the number in use grew rapidly. Eventually, the market for vacuum
cleaners became saturated as virtually every household had one, so that the rate of
growth diminished, and new sales were basically for replacements. The same logis-
tic pattern applies to items such as televisions and stereos. Newer products such as
home computers, VCRs, CD players, DVD players, and cellular phones are now at
various stages of the logistic growth process.
Wal-Mart reportedly uses this application of logistic growth to maintain its
profit levels. It tracks the sales of lines of merchandise, say, a particular style of
jeans. Once sales of that item pass the point of inflection on the logistic curve, Wal-
Mart stops ordering that item and orders a different one instead. Thus the compa-
ny only stocks and sells an item while it is “hot.” Hence few, if any, items are sold at
clearance prices, which would substantially reduce Wal-Mart’s profits.

EXAMPLE 6
The first diesel locomotive was put into service in the United States in 1925. In the early
years, the use of diesels among the 25 major railroads then in existence grew at about 25%
per year. It took about 30 years for all the railroads to use diesel locomotives. Construct a
logistic model for the spread of their use.
Solution Because the early annual growth rate was 25%, we assume that a  0.25 so
that 1  a  1.25. Further, because the limiting value must be L  25 railroads,
a
L  25  .
b
We find that

a 0.25
b   0.01.
L 25
gord.3896.05.pgs 4/24/03 10:02 AM Page 397

5.3 The Logistic or Inhibited Growth Model 397

Thus our logistic model is the difference equation

R n1  1.25R n  0.01R n2 .
We begin in 1925 when n  0 and R 0  1 for the one railroad using a diesel locomotive.
We then use the difference equation to obtain the values shown in the table and in Figure 5.28.

Rn
Year Number of Diesels
1.2511 2  0.0111 2 2
25
Number of Railroads

20
1 1.24

15 2 1.53 1.2511.24 2  0.0111.242 2

10
3 1.89 1.2511.53 2  0.0111.532 2
5
4 2.33
n
0 5 10 15 20 25 30
Time (years)
5 2.86

10 7.31
FIGURE 5.28
o o
15 14.62
o o
20 21.03
o o
25 23.88
o o
30 24.72

The model agrees with the historical fact that it took about 30 years after 1925 (or until
about 1955) for all 25 of the major railroads to introduce diesels. Incidentally, the other
values from the model match the actual spread in the use of diesels quite well. Incidentally,
we didn’t round the values calculated to the nearest integer because the numbers involved
were so small, even though obviously a fraction of a diesel locomotive makes no sense.
◆

Estimating a and b
Suppose that a set of data points fall in a pattern such as the one shown in Figure
5.29 and that we want to fit a logistic curve to it. The hardest part of constructing a
logistic model is determining appropriate values for the growth rate a and the in-
hibiting constant b. To do so, we transform the data so that the resulting points are
in a linear pattern, analogous to what we did in Chapter 3.
Suppose that a set of data points 10, P0 2, 11, P1 2, 12, P2 2, . . . appear to follow a
logistic pattern so that Pn satisfies the difference equation
Pn  aPn  bPn2  1a  bPn 2Pn .
gord.3896.05.pgs 4/24/03 10:02 AM Page 398

398 CHAPTER 5 Modeling with Difference Equations

yn

n
FIGURE 5.29

Dividing both sides by Pn , we obtain

Pn
 a  bPn .
Pn
Therefore, if Pn follows a logistic pattern, the quantity 1 Pn 2>Pn is actually a lin-
ear function of Pn . We plot 1 Pn 2 >Pn against Pn (not against n) and find that it falls
in a roughly linear pattern, as shown in Figure 5.30. Then we can find the best lin-
ear fit to the transformed data. Note that the pattern shown is decreasing, which is
the usual case, so the slope of the regression line will be negative. The correspon-
ding regression equation is Y  a  bX, which is equivalent to
Pn
 a  bPn .
Pn
When we multiply both sides by Pn , we get the logistic difference equation
Pn  aPn  bPn2 .
Once we have this equation, we have our estimates for the logistic coefficients a
and b. We illustrate how to apply these ideas in Examples 7 and 8.

∆Pn
Pn

Pn
FIGURE 5.30

EXAMPLE 7
German biologist R. Carlson studied the growth of yeast under controlled conditions
and obtained the following set of measurements for the weight of yeast, in milligrams, as
a function of time, n.
gord.3896.05.pgs 4/24/03 10:02 AM Page 399

5.3 The Logistic or Inhibited Growth Model 399

Estimate the parameters a and b to create a logistic model to fit the growth of the yeast population.

Pn
n Pn
700
1 9.6 600

Weight of Yeast
500
2 29.0
400
3 71.1 300
200
4 174.6 100
n
5 350.7 0 2 4 6 8 10
Time (n)
6 513.3

7 594.4 FIGURE 5.31

8 640.0
Solution By inspecting either the entries in the table or the corresponding scatterplot
9 655.9
of the points shown in Figure 5.31, we see that they appear to fall in a logistic pattern. To
10 661.8 linearize the data, we add two extra columns to the table, one for the differences Pn of
the successive terms and the other for the ratio 1 Pn 2 >Pn .
Figure 5.32 shows the plot of the values of 1 Pn 2>Pn versus Pn (not n). They seem to
fall in a roughly linear pattern with negative slope. Using the linear regression routine on
a calculator, we find the corresponding regression equation:
Pn
 1.66  0.00271Pn .
Pn

∆Pn

(Pn)>Pn
Pn n Pn
Pn
119.4>9.6 2
1 9.6
2.5 19.4 —— 2.021
2 29.0
42.1 —— 1.452
2
3 71.1
103.5 —— 1.456
4 174.6
176.1 —— 1.009
1.5 5 350.7
162.6 —— 0.464
6 513.3
1 81.1 —— 0.158
7 594.4
45.6 —— 0.077
0.5 8 640.0
15.9 —— 0.025
9 655.9
5.9 —— 0.009
0 100 200 300 400 500 600 700
Pn 10 661.8

FIGURE 5.32

The corresponding correlation coefficient is r  0.967, which suggests a very high degree
of negative correlation between 1 Pn 2 >Pn and Pn . Multiplying both sides of the regression
equation by Pn yields the logistic difference equation
Pn  1.66Pn  0.00271Pn2 ,
giving a  1.66 and b  0.00271.
gord.3896.05.pgs 4/24/03 10:02 AM Page 400

400 CHAPTER 5 Modeling with Difference Equations

The sequence of points based on this difference equation superimposed over the orig-
inal data is shown in Figure 5.33. Note that the curve connecting the points from the logis-
tic model fits the points of the original data reasonably well, especially the first five points.
However, the values of the logistic model level out somewhat below the limiting value for
the actual yeast population data.

Pn

700
600

Weight of Yeast
500
400
300
200
100
n
0 2 4 6 8 10
FIGURE 5.33 Time (n)

◆
In Example 1 of Section 3.3, we found that the growth in the U.S. popula-
tion from 1780 to 1900 closely followed an exponential growth pattern with a
growth rate of 32.1% per decade. However, we pointed out that this exponential
pattern didn’t apply during the twentieth century and, in fact, the rate of
growth has slowed considerably since 1900, suggesting a logistic model. In Ex-
ample 8 we examine the growth in the U.S. population over the entire period
since 1780.
EXAMPLE 8
Find a logistic model that fits the data on the U.S. population since 1780 and find the
limiting value for the population using this logistic model.
Solution The data on the U.S. population are presented in the table on the next page
and in the scatterplot shown in Figure 5.34.

Pn

300
Population (millions)

250
200
150
100
50
n
FIGURE 5.34 0 1750 1800 1850 1900 1950 2000
Year

Both the table and the scatterplot show that the rate of population growth
slowed considerably during the twentieth century. Successive ratios also slowly de-
creased, indicating that the rate of population growth slowed from over 20% per
decade at the start of the century to slightly over 10% per decade by the end of the
century.
gord.3896.05.pgs 4/24/03 10:02 AM Page 401

5.3 The Logistic or Inhibited Growth Model 401

Year Population
Ratio
1780 2.8
1.39
1790 3.9
1.36
1800 5.3
1.36
1810 7.2
1.33
1820 9.6
1.34
1830 12.9
1.33
1840 17.1
1.36
1850 23.2
1.35
1860 31.4
1.27
1870 39.8
1.26
1880 50.2
1.25
1890 62.9
1.21
1900 76.0
1.21
1910 92.0
1.15
1920 105.7
1.16
1930 122.8
1.07
1940 131.7
1.14
1950 150.7
1.19
1960 179.3
1.13
1970 203.3
1.11
1980 226.5
1.10
1990 248.7
1.13
2000 281.4

To fit a logistic curve to this data, we must first calculate the differences Pn 
Pn1  Pn and then the ratios
Pn Pn1  Pn
 ,
Pn Pn
the results of which we show in the table on the next page.
When we plot these transformed values 1 Pn 2>Pn against Pn , as shown in Figure 5.35,
we see that they fall in a predominantly decreasing, roughly linear pattern. However, there
does seem to be a fair amount of variation about the regression line. Nevertheless, the
corresponding plot of the residuals, as shown in Figure 5.36, indicates that the fit is rea-
sonably accurate (the residuals are small and roughly half are below and roughly half are
above the baseline), though there does seem to be a pattern to the residuals. This pattern
suggests that the logistic model may not fully explain all the variation in the data. The
corresponding correlation coefficient, r  0.8803, represents a high degree of negative
correlation for the 22 data pairs. Incidentally, we expect a negative correlation between
the transformed values and the actual population values because the slope of the regres-
sion line is negative. The equation of the regression line for the transformed data is equiv-
alent to
Pn
 0.3314  0.0011Pn .
Pn
Multiplying both sides of this equation by Pn gives us the logistic model
Pn  0.3314Pn  0.0011Pn2 ,
gord.3896.05.pgs 4/24/03 10:02 AM Page 402

402 CHAPTER 5 Modeling with Difference Equations

(Pn 2>Pn
Year Pn
Pn
0.393 11.1>2.82
1780 2.8
1.1
1790 3.9
1.4 0.359
1800 5.3
1.9 0.358
1810 7.2
2.4 0.333
1820 9.6
3.3 0.344
1830 12.9
4.2 0.326
1840 17.1
6.1 0.357
1850 23.2
8.2 0.353
1860 31.4
8.4 0.268
1870 39.8
10.4 0.261
1880 50.2
12.7 0.253
1890 62.9
13.1 0.208
1900 76.0
16.0 0.211
1910 92.0
13.7 0.149
1920 105.7
17.1 0.162
1930 122.8
8.9 0.072
1940 131.7
19.0 0.144
1950 150.7
28.6 0.190
1960 179.3
24.0 0.134
1970 203.3
23.2 0.114
1980 226.5
22.2 0.098
1990 248.7
32.7 0.131
2000 281.4

∆Pn
Pn

0.4
0.1
0.3
Residuals

Pn
0.2 0 100 150 200 250 300

0.1 –0.1

Pn
0 50 100 150 200 250 300 –0.2
Population (millions) Population (millions)

FIGURE 5.35 FIGURE 5.36

where a  0.3314 and b  0.0011. Figure 5.37 shows the values predicted by this logis-
tic model, starting from P0  2.8; they are the points connected by the smooth curve.
For comparison, the original data points for the U.S. population are also shown. The
model that we constructed seems to be an excellent fit until about 1930, but the accura-
cy then seems to break down.
Based on this logistic model, the limiting population for the United States is
a 0.3314
L 
b 0.0011
 301.3 million.
gord.3896.05.pgs 4/24/03 10:02 AM Page 403

5.3 The Logistic or Inhibited Growth Model 403

Pn

300

Population (millions)
250
200
150
100
50
n
0 1750 1800 1850 1900 1950 2000
FIGURE 5.37 Year

◆
This prediction is unrealistically low because the U.S. population was fairly
close to this level in 2000 and was growing at a rate of about 20 or 30 million peo-
ple per decade—it isn’t reasonable to expect that population growth will slow that
much over the next decade. However, recall that, based on the residual plot, we
pointed out that the logistic model doesn’t seem to account for all the variation in
the data. It provides a relatively good fit to the growth of the population, but it isn’t
an outstanding fit and perhaps a more sophisticated model is needed to give a bet-
ter match to the population data.

Problems
Calculate the first 10 values predicted for the logistic so- formation to estimate the size of the population
lution subject to each set of values and plot the points. after 11 time periods. Is the actual value higher or
What is the limiting value for the population? How close lower than your estimate? How do you know?
do you come to it during the first five time periods? c. Suppose that the population is 900 after 30 time
periods and 910 after 31 time periods. Use this
1. a  0.02, b  0.0005, P0  10
information to estimate the population after 35
2. a  0.02, b  0.0005, P0  3 time periods. Is the actual value higher or lower
3. a  0.02, b  0.0001, P0  5 than your estimate? How do you know?
4. a  0.02, b  0.002, P0  5 8. Suppose that the deer population in a wildlife
5. A population grows according to the logistic refuge follows a logistic growth pattern with an an-
growth model Pn  0.05Pn  0.00002Pn2 . Sketch nual growth rate of 40% and an inhibiting rate of
the behavior of the population if (a) P0  500, 0.02%. Write difference equations to model each
(b) P0  1500, and (c) P0  3500. situation.
6. Consider again the rumor spreading through the a. The deer population grows according to the lo-
dorms in Example 2. Suppose that each person re- gistic model.
peats it to three new people each hour instead of b. The population grows according to the logistic
two people. How long will it be until all 4000 stu- model, but hunters are allowed to eliminate 120 deer
dents have heard it? from the region each year.
7. Suppose that a certain population grows according c. The logistic model applies, and hunters elimi-
to the logistic model from an initial size of 100 to a nate 30% of the deer in the region each year.
final size of 1000. d. The logistic model applies, and the number of
a. Sketch the graph of the population as a function deer that hunters eliminate in the region each year
of time. is proportional to the square root of the number
Use the concavity of your graph from part (a) to of deer living there.
answer the following questions. e. The logistic model applies, hunters eliminate
b. Suppose that the population is 250 after 10 time 40% of the deer, and the state’s wildlife depart-
periods and 300 after 12 time periods. Use this in- ment moves 75 new deer into the area each year.
gord.3896.05.pgs 4/24/03 10:02 AM Page 404

404 CHAPTER 5 Modeling with Difference Equations

9. The growth of a population follows a logistic 14. Consider the growth model based on the difference
trend. The limiting value seems to be about 16,000 equation
and the first few values are 100, 105, 110.2, 115.17, Pn  aPn  bPn3 ,
and 121.5. Use this information to estimate both where b is smaller than a.
coefficients in the logistic equation and determine a. Sketch the graph of Pn as a function of Pn , for
how closely the terms you calculate match these Pn  0.
observed values. b. Determine the maximum sustainable popula-
10. The values for a and b in a logistic model are typical- tion L in terms of a and b for a species modeled
ly estimated based on a set of observations. As such, by this difference equation.
they are likely to be somewhat inaccurate. It is im- c. Use your graph from part (a) to determine the
portant to know how sensitive the results of the lo- sign of Pn if Pn is between 0 and L. What does it
gistic model are to slight changes (or errors) in tell you about the behavior of Pn ? Explain.
either a or b. Suppose that you estimate a  0.05 d. Use your graph from part (a) to determine the
and b  0.00002 so that L  a>b  2500. What sign of Pn if Pn is greater than L. What does it
would be the effect on L if a were actually 10% larger tell you about the behavior of Pn ? Explain.
than 0.05 (0.055 instead of 0.05) while b remains e. Use your graph from part (a) to show that the
fixed? 20% larger? 30% larger? 10% smaller? 20% solution to this difference equation must have a
smaller? How does the value of L depend on the esti- point of inflection if P0 is small enough.
mate for a if b remains fixed? f. The turning points of the general cubic curve
11. Repeat Problem 10 by considering the effect on L of y  Ax 3  Bx 2  Cx  D occur at
changes in b if a is fixed. In particular, what would B  2B2  3AC
be the effect on L if b were actually 10% larger than x .
3A
0.00002? 20% larger? 30% larger? 10% smaller?
Use this fact to determine the location of the
20% smaller? How does the value of L depend on
point of inflection for the solution to this differ-
the estimate for b if a is fixed?
ence equation.
12. Based on your results in Problems 10 and 11, does g. You know that the point of inflection for the lo-
the logistic model seem more sensitive to errors or gistic model occurs at half the height of the
changes in the estimates for a or for b? In estimat- limiting value. For the model in this problem,
ing values for these two parameters based on a set does the point of inflection occur at a compara-
of data, which do you expect to be more accurate? ble, a higher, or a lower level? What does that
Explain. tell you about the kind of behavior for a popu-
13. The population of a region can be modeled by the lation that can be well-modeled by this differ-
logistic growth model with coefficients a and b. A ence equation?
major drought hits the region. 15. Repeat parts (a)–(e) of Problem 14 for the differ-
a. Which coefficient, a or b, is more likely to change ence equation model
because of the drought? Does it become larger or Pn  aPn  bPn4 .
smaller?
16. The table gives the average number of words (both
b. Depending on how close the population was to
spoken and understood) in the vocabulary of the
the maximum sustainable population level be-
typical child aged 1–6. Find the best fit to these data
fore the drought, write a short paragraph to de-
from among each of the logistic, linear, exponen-
scribe the effects of the drought on the long-term
tial, and power functions. From among the best fit
behavior of the population.
in each family, which seems to be the overall best fit
to the pattern in the data?

Age 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6

Vocabulary 50 200 350 600 880 1200 1450 1800 2150 2425 2750
Source: B. A. Moskowitz, Acquisition of Language. Scientific American, 1978, vol 279, pp. 92–108.
gord.3896.05.pgs 4/24/03 10:02 AM Page 405

5.3 The Logistic or Inhibited Growth Model 405

17. The density of human bones increases through child- Find the best fit among each of the logistic, linear, ex-
hood and adolescence. By age 18, 95% of bone densi- ponential, and power functions for this data. From
ty has been achieved. The table shows the percentage among the best fit in each family, which seems to be
of maximum bone density achieved at different ages. the overall best fit to the pattern in the data?

Age 2 4 6 8 10 12 14 16 18

Percentage of Bone Density 43 49 51 56 63 71 82 91 95

Source: Student project.

18. The table shows the accumulated total number of re- c. If the trend is indeed logistic, find the total num-
ported cases of AIDS in the United States since 1983. ber of deaths from AIDS that the model predicts
a. Determine the logistic fit to this data. in the limit.
b. Use the model to predict the number of AIDS
cases in 1990.

Year 1983 1984 1985 1986 1987 1988 1989 1990

Number of AIDS Cases 4589 10,750 22,399 41,256 69,592 104,644 146,574 ?

Year 1991 1992 1993 1994 1995 1996 1997 1998

Number of AIDS Cases 251,638 326,648 399,613 457,280 528,144 594,641 653,084 701,353
Source: U.S. Centers for Disease Control and Prevention.

19. Biologists Reed and Holland studied the growth of sun- heights pass their point of inflection.
flower plants. They measured the heights H of a num- b. Estimate the parameters for a logistic difference
ber of sunflowers on the same day of successive equation to fit these data.
weeks t and averaged the readings, which are given c. Use the difference equation to calculate the pre-
in centimeters in the table. dicted heights for these sunflowers. How close
a. Estimate the maximum height to which these do the predictions come to the actual values?
sunflowers will grow and the point at which the

Week t 1 2 3 4 5 6

Mean Height, H 17.9 36.4 67.8 98.1 131.0 169.5

Week t 7 8 9 10 11 12

Mean Height, H 205.5 228.3 247.1 250.5 253.8 254.5

Source: H. S. Reed and R. H. Holland, The Growth Rate of an Annual Plant He-
lianthus, Proceedings of the National Academy of Sciences, vol 5, 1919.

20. The table on the next page gives measurements on b. Estimate the parameters for a logistic difference
the weight W in grams of a pumpkin on different equation to fit this data.
days t while it is growing. c. Use the difference equation to calculate the pre-
a. Estimate the maximum weight that this pump- dicted weight for this pumpkin. How close do
kin will reach and the point at which the weight the predictions come to the actual values?
passes its point of inflection.
gord.3896.05.pgs 4/24/03 10:02 AM Page 406

406 CHAPTER 5 Modeling with Difference Equations

t 5 6 7 8 9 10 11 12 13 14 15

W 267 443 658 961 1498 2200 2920 3366 3758 4092 4488

t 16 17 18 19 20 21 22 23 24 25

W 4720 4864 4980 5114 5176 5242 5298 5352 5360 5366
Source: Raymond Pearl, The Biology of Population Growth. New York: Alfred A. Knopf, 1925.

21. When Penicillium chrysogenum, a strain of penicillin 24. In Figure 5.32 on the yeast experiment data in Ex-
antibiotic, is grown in a batch fermentation process ample 7, the scatterplot of 1 Pn 2 >Pn versus Pn may
under carefully controlled conditions, the concen- suggest a power function with a negative exponent
tration of the cell as measured by its dry weight at rather than a linear function.
various time intervals is given in the table at the bot- a. Using the transformed data, find the power
tom left of this page. Determine the coefficients for function that best fits the data.
the logistic fit to the data. b. Detransform the results to obtain a first order,
22. From the two tables on the U.S. population in Ex- nonlinear difference equation relating Pn to Pn .
ample 8, compare the entries in the ratio column c. Use the difference equation from part (b) togeth-
and the entries in the logistic transformation col- er with the starting value of P0  9.6 grams of
umn 1 Pn 2>Pn when rounded to two decimal yeast to calculate the amount of yeast present for
places. What interesting fact do you notice about n  1, 2, . . . , 10. Extend the table you construct-
the two columns of values? Explain how you can ac- ed in Problem 23 to include these values. How
count for it mathematically. close do these values come to matching the exper-
23. In Example 7 on the yeast experiment, we construct- imental data shown in Example 7.
ed the best logistic model to fit the data based on 25. The scatterplot shown in Figure 5.32 may also suggest
the difference equation Pn  1.66Pn  0.00271Pn2 a decaying exponential function. Repeat Problem 24,
or Pn1  2.66Pn  0.00271Pn2 . using the exponential function that best fits the trans-
The initial measurement on the yeast was formed data.
P1  9.6. Use this value and the difference equation 26. The growth pattern in human height or weight de-
to construct a table showing the amount of yeast velopment from birth through age 18, say, usually
present over the first 10 time periods. Compare follows a logistic growth pattern. The table below
these predictions to the actual values measured in gives the typical height, in centimeters, of a male
the experiment. and a female in the 50th percentile for height at dif-
ferent ages, in years. Use the data to construct a pair
of logistic functions that model the heights of boys
t Concentration
and girls as a function of age t for people in this
0 0.40 50th percentile group.
22 0.99

46 1.95 Age 0 1 2 3 4 5 6 7 8 9

70 2.52 Boys 50.5 76.1 87.6 96.5 102.9 109.9 116.1 125.0 127.0 132.2

94 3.09 Girls 49.9 74.3 86.5 95.6 101.6 108.4 114.6 120.6 126.4 132.2

118 4.06 Age 10 11 12 13 14 15 16 17 18

142 4.48 Boys 137.5 143.3 149.7 156.5 163.1 169.0 173.5 176.2 176.8

166 4.25 Girls 138.3 144.8 151.5 157.1 160.4 161.8 162.4 163.1 163.7
Source: NCHS Growth Curves for Children. Vital and Health Statistics, National Health Survey, U.S. Department of
190 4.36 Health, Education and Welfare.
gord.3896.05.pgs 4/24/03 10:02 AM Page 407

5.4 Newton’s Laws of Cooling and Heating 407

27. The table below shows the worldwide electric gen- c. From the table, estimate when the worldwide
erating capacity of nuclear power plants, measured electric generating capacity of nuclear power
in gigawatts, over time. plants was growing most rapidly.
a. Estimate the parameters for the logistic model d. From your model in part (a), when was the
that fits these data. worldwide electric generating capacity of nu-
b. What is the limiting value for the worldwide elec- clear power plants growing most rapidly?
tric generating capacity of nuclear power plants?

Year 1960 1965 1970 1975 1980 1985 1990 1995 2000
Electric Capacity 1 5 16 71 135 250 328 340 347
Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That Are Shaping Our Future.

28. The table shows the population of the world, in bil- growing more slowly than it was only a few years
lions, over time. The population appears to be ago, so it may have passed its point of inflection.

Year 1950 1955 1960 1965 1970 1975

Population 2.556 2.780 3.039 3.345 3.707 4.086

Year 1980 1985 1990 1995 1999

Population 4.454 4.851 5.277 5.682 6

Source: Lester R. Brown et al., Vital Signs 2000: The Environmental Trends That
Are Shaping Our Future.

a. Estimate the parameters for the logistic model f. Use the difference equation to predict when the
that fits these data. world’s population will reach 7 billion.
b. What is the limiting value for the world’s popu- 29. Recall that the inflection point for the logistic
lation, based on this model? model
c. From the table, estimate when the world’s popu- Pn  aPn  bPn2
lation was growing most rapidly.
d. From your model in part (a), when was the occurs at a height of 21 L. At this point, the function
is growing most rapidly. Show algebraically that the
world’s population growing most rapidly?
maximum value of Pn is 14 bL2.
e. Use the difference equation to predict the
world’s population in 2010.

5.4 Newton’s Laws of Cooling and Heating

In a familiar scene in TV crime shows, the medical examiner studies the homicide
victim’s body and knowledgeably announces that “Mr. Jones died at approximately
1:30 in the morning.” We now develop the mathematics behind this type of con-
clusion. More generally, we investigate the rate at which any object cools or heats.

Newton’s Law of Cooling

Suppose that you heat a pizza in an oven set at 450°F and then remove it to cool
on a kitchen counter where the temperature is a constant 70°F. How fast does the
temperature of the pizza drop until it reaches room temperature? Clearly, the
gord.3896.05.pgs 4/24/03 10:02 AM Page 408

408 CHAPTER 5 Modeling with Difference Equations

temperature of the pizza drops most rapidly at first because of the large differ-
ence between the pizza temperature and the room temperature. As the pizza
cools, the rate at which the temperature decreases slows. That is, the hotter the
pizza, the faster the temperature drops; it drops most slowly when the pizza’s
temperature is close to room temperature. Geometrically, we expect the graph of
the temperature as a function of time to be decreasing and concave up, as shown
in Figure 5.38. According to a physical principle developed by Isaac Newton, the
change in temperature Tn after n time periods is proportional to the difference
between the temperature of the pizza 1Tn 2 and the temperature of the surround-
ing air 170°F2 , or Tn  70. Thus
Tn

450
T0
Temperature (°F )

R = 70

Time

FIGURE 5.38

Tn  a . 1Tn  70 2,
where the constant of proportionality a  0. Because the temperature change
Tn is negative and both Tn  70 and a are positive, there must be a minus sign in
front of a. Equivalently, if we add Tn to both sides of the equation, we have the dif-
ference equation model
Tn1  Tn  a . 1Tn  70 2
 11  a 2Tn  70a.
More generally, if the room temperature is any constant R, the analogous difference
equation is
Tn1  11  a 2Tn  aR. (1)
This equation and the principle behind it are both known as Newton’s law of
cooling.
To solve this difference equation, we use the results summarized at the end of
Section 5.1. Note that Difference Equation (1) has the same form as
xn1  axn  B (2)
with
a  1  a and B  aR.
The solution to Difference Equation (2) is
xn  L  1x0  L2an,
where L  B> 11  a2 is the limiting value.
gord.3896.05.pgs 4/24/03 10:02 AM Page 409

5.4 Newton’s Laws of Cooling and Heating 409

For Newton’s law of cooling, a is a positive fraction, so a  1  a is between

0 and 1. Also, we have
B aR aR
L    R.
1a 1  11  a2 a
Therefore, if the initial temperature of the pizza (or any other cooling object) is T0 ,
the solution to Difference Equation (1) is
Tn  R  1T0  R2 11  a2 n.
The constant a usually is determined from an additional temperature measure-
ment, as we illustrate in Examples 1–3 below.
First, let’s examine the behavior of this solution function. The term
1T0  R2 11  a2 n is an exponential decay function because a  1  a lies between
0 and 1. This term is added to the constant R, so the solution function is decreasing
in a concave up manner as it decays to a level of R as n increases. This behavior pat-
tern is precisely what we predicted in Figure 5.38.
We can also see this behavior directly from the formula
Tn  R  1T0  R2 11  a2 n.
As n increases, the exponential decay term 11  a2 n approaches 0 because 1  a
lies between 0 and 1. So again Tn S R as n S .
In Example 1 of Section 4.8, we constructed a function to fit a set of data ob-
tained from an experiment on cooling. A temperature probe was warmed to
42.3°C and plunged into cold water at about 8.6°C to cool. The shifted exponential
function that best fit the data was T1t2  8.6  35.439410.84802 t. Note that this
function has the identical form as the solution sequence for the difference equation
model.
The mathematical model for Newton’s law of cooling may be summarized as follows.

Newton’s Law of Cooling

Assumptions
◆ The temperature R of the medium remains constant.
◆ The change in temperature is proportional to the difference between
the temperature of the object and the temperature of the medium.

Mathematical Model
◆ Difference equation:
Tn  a . 1Tn  R2 or Tn1  11  a 2 Tn  aR
◆ Solution: Tn  R  1T0  R2 11  a 2 n

EXAMPLE 1
Suppose that a cake is baking in an oven at 350°F. It is removed when its temperature is 180°F
and is left to cool in a kitchen at 70°F. After 10 minutes, the temperature of the cake is 125°F.
a. Find the solution to the difference equation based on Newton’s law of cooling.
b. Find the temperature of the cake after 15 minutes.
c. How long does it take the cake to cool to 75°F?
gord.3896.05.pgs 4/24/03 10:02 AM Page 410

410 CHAPTER 5 Modeling with Difference Equations

Solution

a. For T0  180 and R  70, the difference equation for the temperature at any time n is

Tn1  11  a2 Tn  70a.

The corresponding solution is

Tn  1180  70 2 11  a2 n  70
 11011  a2 n  70.

Further, when n  10, the temperature T10  125, so

T10  11011  a2 10  70  125.
Subtracting 70 from both sides gives

11011  a 2 10  125  70  55, so that 11  a2 10 

55
 0.5.
110
Taking the tenth root of both sides of this equation yields

1  a  0.51>10  0.933, and so a  1  0.933  0.067,

which is between 0 and 1, as we expected. The solution to the difference equation is

Tn  11011  a2 n  70
 70  11010.9332 n.

b. After n  15 minutes, the temperature of the cake is

T15  70  11010.9332 15
 70  38.87  108.9°.
c. To find the time needed for the cake to cool to 75°F, we need the value of n when
Tn  75. We begin with

Tn  11010.9332 n  70  75 and so 11010.9332 n  5.

We divide both sides by 110 to get

10.9332 n 
5
 0.045.
110
We take logarithms of both sides of this equation to obtain
n log10.933 2  log10.0452

and then solve for n:

log10.0452
n  44.7 minutes.
log10.9332

Thus it takes about three-quarters of an hour for the cake to cool to 75°F.
◆
gord.3896.05.pgs 4/24/03 10:02 AM Page 411

5.4 Newton’s Laws of Cooling and Heating 411

EXAMPLE 2
Mr. Jones’s body was found in his kitchen at 9 A.M. by the police who noted that the body
temperature was 77.3°F and that the room temperature was 70°F. According to the med-
ical examiner, the body temperature an hour later was 76.1°F. Assuming that Mr. Jones’s
body temperature was the normal 98.6°F at the time of death, at what time did he die?
Solution Using the solution to the difference equation for cooling, the body tempera-
ture after n hours is given by
Tn  R  1T0  R2 11  a2 n,
where R  70° and T0  98.6°. Therefore
Tn  70  198.6  70 2 11  a2 n
 70  28.611  a 2 n.
Because we don’t know the time of death (which corresponds to n  0 2, we don’t know
the value of n at 9 A.M. However, at 9 A.M., which is n hours after death, the body tem-
perature was 77.3°, so
Tn  70  28.611  a2 n  77.3,
or, after subtracting 70 from both sides,
28.611  a 2 n  7.3. (3)
An hour later at 10 A.M., which is n  1 hours after death, the body temperature was
Tn1  70  28.611  a2 n1  76.1,
or, again after subtracting 70 from both sides,
28.611  a2 n1  6.1. (4)
Dividing Equation (4) by Equation (3) yields

28.611  a2 n1 6.1

n  1  a   0.8356.
28.611  a2 7.3

Substituting this value into Equation (3) gives

28.610.83562 n  7.3,
or, equivalently, when we divide both sides by 28.6,
10.83562 n  0.2552.
To solve for n, we take logs of both sides of this equation and get
n log10.8356 2  log10.25522
so that
log10.25522
n  7.6.
log10.83562
The body was found at 9 A.M., which is n  7.6 hours after death. Thus the murder oc-
curred 7.6 hours (or 7 hours and 36 minutes) before 9 A.M., so we conclude that Mr. Jones
was murdered at approximately 1:24 A.M.
◆
gord.3896.05.pgs 4/24/03 10:02 AM Page 412

412 CHAPTER 5 Modeling with Difference Equations

EXAMPLE 3
A cup of hot coffee is left standing on a table in a room where the temperature is 70°F.
The temperature T, in °F, of the coffee is measured every minute t, and the results are
shown in the following table. Find a model that best fits these temperature readings as a
function of time.

t 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

T 186 182 178 175 171 168 165 162 159 156 153 152 148 145 143 141

Solution Based on our knowledge of cooling curves, we would expect the best fit to be
an exponential function that decays to 70°F. We can’t use our data-fitting techniques
from Chapter 3 to fit an exponential curve to the data because the function would decay
to 0°F rather than 70°F. As in Example 1 of Section 4.8, we must first subtract 70 from
each temperature reading, which is equivalent to introducing a vertical shift for the tem-
perature function, to get the following table.

t 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
T  70 116 112 108 105 101 98 95 92 89 86 83 82 78 75 73 71

The exponential function that best fits the shifted data is

T  70  115.610.967832 t,
where T is the temperature. The corresponding correlation coefficient is r  0.99945;
it is negative because the temperature readings are decreasing. This value for r is ex-
tremely close to 1, indicating an extremely good fit. Finally, we undo the vertical shift
by adding 70 to both sides of the preceding equation to obtain
T  115.610.967832 t  70
as the best model for the temperature of the coffee, as illustrated in Figure 5.39. Note
that this function has the same form as the solution we obtained for the difference equa-
tion for Newton’s law of cooling.

190

180
Temperature (°F)

170
T = 115.6(0.96781) t + 70
160

150

140

t
0 3 6 9 12 15
FIGURE 5.39 Time (minutes)

◆
gord.3896.05.pgs 4/24/03 10:02 AM Page 413

5.4 Newton’s Laws of Cooling and Heating 413

Newton’s Law of Heating

Similar mathematical methods apply if an object is being warmed rather than
cooled. The corresponding principle is known as Newton’s law of heating, which
is based on the assumption that the change in temperature Tn (now an increase)
is proportional to the difference between the temperature R of the medium (the
room, the freezer, the oven, etc.) and the temperature of the object Tn . This as-
sumption leads to the difference equation
Tn  a . 1Tn  R2 ,
which is identical to the difference equation for Newton’s law of cooling. Let’s see
why it is the same. Because the object is being heated, the change in temperature
Tn of the object must be positive. Also, the temperature Tn of the object is always
less than R, so Tn  R is negative. Finally, because the constant of proportionality
a is positive, the product of the terms on the right-hand side will be positive.
We solve this difference equation the same way we solved the one for cooling to get
Tn  R  1T0  R2 11  a2 n.

EXAMPLE 4
A chicken is removed from the refrigerator at a temperature of 40°F and placed into an
oven kept at a constant temperature of 350°F. After 10 minutes, the temperature of the
chicken is 70°F. The chicken is considered cooked when its temperature reaches 180°F.
How long must it remain in the oven until it is cooked?
Solution The solution of the difference equation Tn  a . 1Tn  350 2, with R  350
and T0  40, is
Tn  350  140  350 2 11  a2 n  350  31011  a 2 n .
After 10 minutes,
T10  350  31011  a 2 10  70.
Therefore
11  a 2 10 
280
31011  a2 10  350  70  280 so that  0.903.
310
Taking the tenth root of both sides, we obtain
1  a  0.9898.
Consequently, the solution to the difference equation for the temperature of the chicken
is
Tn  350  31010.98982 n.
We must now find how long it takes the temperature to reach 180°F. Doing so re-
quires finding the value of n for which
Tn  350  31010.98982 n  180 so that 31010.98982 n  350  180  170.
Dividing by 310 gives

10.98982 n 
170
 0.5484.
310
Taking logs of both sides of this equation yields
n log10.98982  log10.54842 ,
gord.3896.05.pgs 4/24/03 10:02 AM Page 414

414 CHAPTER 5 Modeling with Difference Equations

and so
log10.54842
n  58.6 minutes.
log10.98982
Hence the chicken will be ready in just under an hour.
◆
Problems
1. In Example 2 we assumed that the initial body tem- 8. A cup of boiling water 1100°C2 was placed in a re-
perature was a normal 98.6°F. Suppose that Mr. frigerator kept at 7°C at 8 A.M., and the following
Jones had a slight fever of 100°F when he was mur- readings were obtained.
dered. How much difference does that make to the
estimated time of death?
Time (minutes) 1 7 21 45
2. In Example 2 we also assumed that the room temper-
ature was kept constant at 70°F. Actually, the home Temperature (C) 89 71 53 36
heating system probably cycled on and off, so the ac- Time (minutes) 73 90 123 152
tual temperature might have oscillated between 67°F
and 72°F, say. How might you take this variation into Temperature (C) 25 20 14 11
account in predicting the time of death? How much
of a difference would it make? Determine the best exponential fit to these data and use
3. In Example 4 we presumed that the temperature of it to predict when the water temperature will be 9°C.
the oven is 350°F for cooking a chicken. Suppose in- 9. The following data are printed on a carton of milk
stead that the temperature is set at 325°F. How to indicate how many days the milk will last with-
much difference does that make in the time needed out spoiling at different temperatures.
to cook the chicken to 180°F?
4. A pot of bubbling pudding 1212°F2 is removed
from the stove and put immediately into a refriger- Temperature (F) 32 40 45 50 60 70
ator at 40°. After 10 minutes, the temperature of the Time (days) 24 11 5.5 2 1 0.5
pudding is 160°. Find the temperature after 1 hour.
How long does it take for the temperature to drop
a. Determine the linear, exponential, and power
to 75°?
functions that best fit these data.
5. Sam takes a can of soda at room temperature 170°F2
and puts it in a freezer 10°F2 to chill quickly. After 10
b. Which of the three functions seems to be the best
fit to the data?
minutes, the temperature of the soda is 60°. How
c. Using the model that you think is the best fit,
long does it take the temperature to drop to 40°?
how long should milk last in a refrigerator kept
6. A bowl of cold soup is taken out of the refrigerator at 35°F?
136°F2 and placed in a heated oven 1375°F2 to warm.
10. A cool potato at temperature 60°F is placed in an
After 10 minutes, the temperature of the soup is 120°.
oven kept at a constant 350°.
How long does it take for the soup to reach 200°?
a. Sketch the graph of the temperature of the potato
7. Professor Smith’s body was found in a large walk-in
as a function of time.
refrigerator in the laboratory at 9 A.M. by the police
who noted that the body temperature was 67.3°F Use the concavity of your graph from part (a) to
and that the refrigerator temperature was 40°. An answer the questions in parts (b)–(d).
hour later, with the body still in the refrigerator, the b. Suppose that you measure the temperature of the
medical examiner found the body temperature to be potato after 5 minutes and find that it is 109° and
63.1°. Assuming that the body temperature at death that, after 7 minutes, it is 127°. Use this informa-
was 98.6°, at what time did Professor Smith die? tion to estimate the temperature after 10 minutes.
gord.3896.05.pgs 4/24/03 10:02 AM Page 415

5.4 Newton’s Laws of Cooling and Heating 415

Is the actual temperature higher or lower than your 12. A cup of hot chocolate (temperature ) is placed
estimate? How do you know? on a table where the air temperature is 70°. Suppose
c. Suppose that you are told that the temperature that it takes 12 minutes for the drink to cool to 100°.
of the potato after 12 minutes is 150°. Use this Let r1 represent the average rate of decrease in temper-
information and the temperature after 5 minutes ature per minute over the full 12-minute period, let r2
to estimate the temperature after 10 minutes. Is be the average rate of decrease over the first 6 minutes,
the actual temperature higher or lower than your and let r3 be the average rate of decrease over the last
estimate? How do you know? 6 minutes. List these three rates in increasing order.
d. How might you use the results from parts 13. Suppose that it takes t1 minutes for a raw potato, start-
(b) and (c) to come up with a better estimate ing at 70°F, to reach 200° in an oven. At that time, it is
of the temperature after 10 minutes? removed from the oven and put on the table where it
cools. Suppose that the potato takes t2 minutes to
11. Your Thanksgiving turkey is taken from a refrigera-
reach 70°. Is t1  t2 , t1  t2 , or t1  t2 ? Explain.
14. A potato at room temperature 1T0 2 is placed in an
tor at 40°F and is cooked in an oven kept at a con-
stant temperature of 350°. The temperature T of the
bird is 70° after 30 minutes and is 96° after 60 min- oven at temperature R. Construct the actual solu-
utes. From your knowledge of the pattern of temper- tion to the corresponding difference equation for
ature rise, decide which of the following temperature heating in terms of the various parameters. What
readings are possible and which are impossible. is the formula for the temperature of the potato if
b. T145 2  85°
T  70°F and R  350°F?
a. T1452  80°
c. T1752  105° d. T175 2  115° 15. In Section 4.8, we presented a set of data from a
cooling experiment. The temperature readings, in
180°F degrees Celsius, were as follows.

Time Temp Time Temp Time Temp Time Temp

1 42.30 10 14.77 19 10.17 28 9.04

2 36.03 11 13.82 20 9.92 29 8.91

3 30.85 12 13.11 21 9.80 30 8.83

4 26.77 13 12.51 22 9.67 31 8.78

5 23.58 14 11.91 23 9.54 32 8.78

6 20.93 15 11.54 24 9.42 33 8.78

7 18.79 16 11.17 25 9.29 34 8.78

8 17.08 17 10.67 26 9.16 35 8.66

9 15.82 18 10.42 27 9.16 36 8.66

According to Newton’s law of cooling, the change in b. Plot the points 1T  8.6, T2 that you calculat-
temperature T is proportional to the difference be- ed. In what type of pattern do they fall?
tween the temperature of the object and the tempera- c. Find the equation of the line that best fits these
ture of the medium (in this case the temperature of points and write a difference equation for T for
the cool water is 8.6°C). the temperature data.
a. Extend the table, with additional columns for d. Solve the difference equation for the tempera-
T  8.6 and also for the differences T between ture T as a function of time.
successive temperature readings.
gord.3896.05.pgs 4/24/03 10:02 AM Page 416

416 CHAPTER 5 Modeling with Difference Equations

5.5 Geometric Sequences and Their Sums

Consider the difference equation for exponential growth
xn1  rxn
whose solution is given by
xn  x0r n,
where x0 is any initial value. Using the notation for sequences, we can write this so-
lution as
5x0 , x0r, x0r 2, x0r 3, x0r 4, . . . , x0r n, . . . 6.
Recall that such a sequence is called a geometric sequence or exponential sequence.
For simplicity, suppose that x0  1, so that the sequence reduces to
51, r, r 2, r 3, r 4, . . . , r n, . . . 6.
For instance, if r  5, we have the geometric sequence
51, 5, 52, 53, 54, . . . , 5n, . . . 6  51, 5, 25, 125, 625, . . . , 5n, . . . 6.
The difference equation xn1  rxn reinforces the fact that each term in any
geometric sequence is a constant multiple of the preceding term, or, equivalent-
ly, there is a common ratio r between successive terms. Alternatively, xn is an ex-
ponential function of n, and we know that the ratio of successive values is a
constant—namely, the growth or decay factor r. In the preceding sequence, each
term is five times the preceding term, so the common ratio r of each pair of suc-
cessive terms is 5.
We can deduce a considerable amount of information about the behavior of
the terms in a geometric sequence 51, r, r 2, r 3, . . . , r n, . . . 6 from the value of the
common ratio r. We summarize this information as follows.
1. If r  1, the terms are successively larger and approach infinity. We say
that they increase monotonically.
2. If r  1, all the terms are equal to 1.
3. If 0  r  1, the terms are successively smaller and approach zero. We say
that they decrease monotonically.
4. If r  0, all terms after the initial term, 1, are 0.
5. If 1  r  0, the terms oscillate between positive and negative values,
each is numerically smaller than the preceding term, and they approach 0.
6. If r  1, the terms oscillate between 1 and 1.
7. If r  1, the terms oscillate between positive and negative values, and
each term is numerically larger than the preceding term.
If the terms of a sequence approach a single value as n approaches , written
xn S , we say that the sequence converges to that value. Otherwise, we say that
the sequence diverges.
The following seven sequences illustrate these properties.
1. r  5: 51, 5, 52, 53, 54, . . . , 5n, . . . 6 S , so this sequence diverges as
n S . (Note that the terms in this sequence increase monotonically.)
2. r  1: 51, 1, 1, 1, 1, . . . 6 S 1, so this sequence converges to 1 as n S .
gord.3896.05.pgs 4/24/03 10:02 AM Page 417

5.5 Geometric Sequences and Their Sums 417

3. r  12 : 51, 12 , 1 12 2 2, 1 12 2 3, 1 12 2 4, . . . , 1 12 2 n, . . . 6  51, 0.5, 0.25, 0.125, 0.0625,

. . . 6 S 0, so this sequence converges to 0 as n S . (Note that the terms
in this sequence decrease monotonically.)
4. r  0: 51, 0, 0, 0, 0, . . . 6 S 0, so this sequence converges to 0 as n S 
5. r   14 : 51,  14 , 1 14 2 2, 1 14 2 3, 1 14 2 4, . . . , 1 14 2 n, . . . 6  51, 0.25,
0.0625, 0.015625, 0.00390625, . . . 6 S 0, so this sequence converges to
0 as n S . (Note that the terms in this sequence oscillate between posi-
tive and negative values, but that each term is numerically smaller than the
previous term; the terms are not monotonic.)
6. r  1: 51, 1, 112 2, 11 2 3, 112 4, . . . , 11 2 n, . . . 6  51, 1, 1, 1,
1, 1, . . . 6, so this sequence does not converge to a single value but di-
verges as n S .
7. r  2: 51, 2, 12 2 2, 122 3, 12 2 4, . . . , 12 2 n, . . . 6  51, 2, 4, 8,
16, 32, 64, 128, . . . 6 S  , so this sequence diverges as n S .
The terms oscillate between positive and negative values, but each term is
numerically larger than the previous term; the terms are not monotonic.

The Sum of the Terms in a Geometric Sequence

Geometric sequences arise in a great variety of applications, and usually they are
accompanied by the related question: What is the sum of the terms? That is, for any
geometric sequence with constant ratio r, what is
1  r  r2  r3  r4  . . .  rn
for any given value of n?
For instance, there is a very old puzzle in which gold coins are placed on a
chessboard according to the pattern: one coin on the first square, two on the sec-
ond, four on the third, and so on to 263 on the 64th square. The problem then is:
Find the total number of coins. That is, find the sum
1  2  22  23  24  . . .  263.
Before attempting to solve this specific problem, let’s look at the more general
problem of finding a formula for the sum of the terms in any geometric sequence.
We introduce the following notation. Let
S0  1, S1  1  r, S2  1  r  r 2,
and, in general, for the sum of the n terms 1, r, r 2, r 3, r 4, . . . , r n,
Sn  1  r  r 2  . . .  r n.
We want to find a formula for the sum Sn for any n. We multiply the expression
for Sn by the common ratio r to obtain
rSn  r11  r  r 2  . . .  r n 2  r  r 2  r 3  . . .  r n  r n1.
When we subtract the expression for rSn from the expression for Sn , all the inter-
mediate terms r, r 2, r 3, . . . , r n cancel out and we are left with
Sn  rSn  1  r n1.
We factor out the term Sn on the left-hand side and get
11  r2Sn  1  r n1.
gord.3896.05.pgs 4/24/03 10:02 AM Page 418

418 CHAPTER 5 Modeling with Difference Equations

If r
1, we can divide both sides by 1  r to get
1  r n1
Sn  .
1r

The sum of the terms 1, r, r 2, . . . , r n of a finite geometric sequence is

1  r n1
1  r  r2  r3  . . .  rn  ,
1r

provided that r
1.

EXAMPLE 1
What is the total number of gold coins that would be needed on a chessboard according
to the pattern previously described?
Solution The number of coins is
1  2  22  23  24  . . .  263.
The common ratio in this sum of terms of a geometric sequence is r  2. So, for n  63,
we get
S63  1  2  22  . . .  263
1  264

12
 264  1  1.844674  1019,
which is more than 18 quintillion.
◆

In two of the seven possible cases for a geometric sequence—namely,

0  r  1 and 1  r  0—the terms of the sequence approach 0. Also, when
r  0, all terms after the initial term 1 are 0. In the other 4 cases listed, the terms do
not approach 0. Let’s see what this means for the sum of the terms of the entire geo-
metric sequence, not just the sum of the first n terms.
Using the formula for the sum of the terms from 1 to r n,
1  r n1
Sn  ,
1r
we see that, whenever 1  r  1, the expression r n1 becomes ever smaller
and approaches 0 as n increases. As a result, even though we are adding more
and more terms, eventually the additional terms are all so small that together
they contribute virtually nothing to the sum. This fact enables us to give
meaning to the sum of all the terms of an infinite geometric sequence, provid-
ed that the terms decrease rapidly enough. In other words,
10 1
1  r  r2  r3  . . .  rn  . . .   ,
1r 1r
provided that 1  r  1, or, equivalently, 0 r 0  1.
gord.3896.05.pgs 4/24/03 10:02 AM Page 419

5.5 Geometric Sequences and Their Sums 419

However, if r  1 or if r 1, the values r n1 do not approach 0 as n ap-

proaches , so no finite value for the sum of the infinite sequence is possible. Thus
the sum of an infinite geometric sequence makes sense only when the common
ratio r is strictly between 1 and 1.

The sum of the terms of an infinite geometric sequence is

1
1  r  r2  r3  . . .  rn  . . .  ,
1r
provided that 1  r  1.

For instance, the sum of the terms in the infinite geometric sequence

1 2 1 3
 a b  a b  ... 
1 1 1
1   2.
2 2 2 1  11>22 11>2 2

Think About This Add enough of the terms from the preceding sequence to convince yourself that
this result is reasonable. ❐

The facts and results regarding geometric sequences just described occur fre-
quently in applications of mathematics. We encounter such sequences repeatedly
later in this book. For now, let’s consider several additional situations in which they
arise.
In our discussion of the elimination of a drug from the body in Section 5.1, we
saw that the kidneys remove about 25% of any Prozac in the bloodstream every
24 hours and that a person takes the same dose D0  80 mg every day. We modeled
this situation with the difference equation
Dn1  0.75Dn  80
and created the formula
Dn  320  24010.752 n
for the solution sequence. Let’s now find this formula by using our knowledge of
the sum of a geometric sequence.

EXAMPLE 2
a. Find a formula for the level of Prozac in the body after any number of days, using the
sum of a geometric sequence.
b. Use the result from part (a) to find the level of Prozac in the bloodstream after 5 days and
after 10 days.
c. What is the limiting value L for Prozac in the body?

Solution

a. From the difference equation and the initial value D0  80, we have

D1  0.75D0  80  0.75180 2  80  11  0.75 2 180 2 .

gord.3896.05.pgs 4/24/03 10:02 AM Page 420

420 CHAPTER 5 Modeling with Difference Equations

Similarly,
D2  0.75D1  80
 0.7511  0.75280  80  11  0.75  0.752 2 80.
Furthermore,
D3  0.75D2  80
 0.7511  0.75  0.752 2 80  80  11  0.75  0.752  0.753 280.
In general, after n days,
Dn  8011  0.75  0.752  0.753  . . .  0.75n 2 .
The expression inside the parentheses is the sum of the terms from 1 to 0.75n in a
geometric sequence with common ratio r  0.75. Therefore the sum of these terms
is given by
1  10.75 2 n1 1  10.752 n1
Dn  80 c d  80 c d
1  0.75 0.25
 3203 1  10.752 n1 4  320  32010.752 10.75 2 n
 320  24010.752 n,
which is identical to the formula that we created in Section 5.1 using difference equa-
tions.
b. After n  5 days, we have

D5  320  24010.752 5  263.0469 mg,

which agrees with the value we obtained in Section 5.1. Similarly,
D10  320  24010.752 10  306.485 mg.

c. Finally, because r  0.75  1, the limiting value for the sum of this geometric se-
quence is
L  D0 . 11  r  r 2  r 3  . . . 2

 80 a b 
1 80
1r 1  0.75
80
  320 mg,
0.25

which is the same value we found in Section 5.1 for the drug maintenance level.
◆
When Will Our Oil Run Out?
An ongoing political debate having major economic and social implications has to
do with energy policy both at home and abroad. One aspect of this debate centers
on the use of petroleum, both as a source of energy in vehicles and power plants
and as a component of oil-based products such as plastics. Estimated worldwide
oil reserves in 2000 were about 2250 billion barrels, and worldwide oil consump-
tion in 2000 was about 26 billion barrels.* How long will the oil last?
*Source: Energy Information Administration, www.eia.doe.gov/ieu/.
gord.3896.05.pgs 4/24/03 10:03 AM Page 421

5.5 Geometric Sequences and Their Sums 421

If the current rate of consumption remains constant at 26 billion barrels per

year, known oil reserves will run out in

2250
 86.5 years.
26
But oil consumption hasn’t been constant; in fact, some estimates indicate that
worldwide oil consumption has been growing at an annual rate of about 2%. How
do we then calculate how long the known oil reserves will last?

EXAMPLE 3
Calculate how long the estimated 2250 billion barrel oil reserves worldwide will last if oil
consumption was 26 billion barrels in 2000 and continues to grow at a 2% annual rate.
Solution Annual oil consumption C can be modeled by the exponential growth function
C1t2  2611.022 t,
where t is the number of years since 2000. The total oil consumed from 2000 on is then
approximately
C10 2  C112  C12 2  C13 2  . . . .
Let n be the first year when the total passes the 2250 billion barrel level. We need to solve
for the value of n that gives
C10 2  C11 2  C12 2  . . .  C1n2  2250.
This expression is equivalent to
2611.022 0  2611.022 1  2611.02 2 2  . . .  2611.022 n  2250
or, when we factor out the common factor of 26,
263 1  11.022 1  11.02 2 2  . . .  11.022 n 4  2250.
The expression in the brackets is the sum of the first n terms of a geometric sequence
with common ratio r  1.02, so
1  1.02n1 1  1.02n1
26 a b  26 a b
1  1.02 0.02
1.02n1  1
 26 a b  130011.02n1  1 2 .
0.02
We therefore need to solve the equation
130011.02n1  1 2  2250
for n. We divide through by 1300 to get
1.02n1  1  1.7308
or, by adding 1 to both sides, we obtain
1.02n1  2.7308.
To extract n  1 from the exponent, we use logarithms to get
log 11.02n1 2  1n  1 2log 1.02  log 2.7308,
gord.3896.05.pgs 4/24/03 10:03 AM Page 422

422 CHAPTER 5 Modeling with Difference Equations

so that

log 2.7308
n1  50.73
log 1.02

and therefore n  49.73. We therefore conclude that all the world’s known oil reserves
will be depleted in just under 50 years from 2000 if consumption continues to grow at an
annual rate of 2%.
◆
The results in Example 3 are startling because all the industrialized nations de-
pend so heavily on oil. Many people therefore advocate major efforts to discover
new oil supplies. Let’s see what doing so would gain.
EXAMPLE 4
Suppose that, as a result of major efforts to find new sources of oil, the worldwide oil re-
serves are doubled to 4500 billion barrels. How long will it take to use it all if oil con-
sumption continues to grow at the current 2% annual rate?
Solution We now have to find the value of n for which
C10 2  C11 2  C12 2  . . .  C1n2  2122502  4500,
which is equivalent to the sum of terms in the geometric sequence
263 1  11.022 1  11.02 2 2  . . .  11.022 n 4  4500.
Following the same algebraic development as in Example 3, we have to solve
130011.02n1  1 2  4500
for n. Dividing both sides of the equation by 1300, we get
1.02n1  1  3.4615 or 1.02n1  4.4615.
Taking logarithms gives
1n  1 2log 1.02  log 4.4615,
so that
log 4.4615
n1  75.52
log 1.02

and hence n  74.52. Note that doubling the total oil reserve didn’t double the 50 years
we found in Example 3; it merely added less than 25 more years.
◆
Other people advocate reducing the rate of oil consumption by using alternative
energy sources (especially renewable sources), by developing more energy efficient ve-
hicles, by encouraging conservation, and by taking other actions. Let’s see what such
measures can accomplish.
EXAMPLE 5
Suppose that, as a result of conservation efforts, the annual rate of growth of oil con-
sumption drops to 1.5% from 2%. How long will the current worldwide oil reserve last?
Solution Our new exponential model for oil consumption is
C1t2  2611.0152 t,
gord.3896.05.pgs 4/24/03 10:03 AM Page 423

5.5 Geometric Sequences and Their Sums 423

where t is still the number of years since 2000. We now have to solve
26 31  11.0152 1  11.015 2 2  . . .  11.0152 n 4  2250
for n. When we find the sum of the terms in this geometric sequence and simplify, as we
did before, we get
1733.3311.015n1  12  2250.
Dividing by 1733.33, we get
1.015n1  1  1.2981 or 1.015n1  2.2981.
Taking logarithms yields
1n  1 2 log 1.015  log 2.2981,
so that
log 2.2981
n1  55.9,
log 1.015
and n  54.9 years. Hence decreasing the annual rate at which oil is consumed by one-
half percentage point gives the world an additional 5 years of oil reserves.
◆
Think About This Is it more likely that the rate at which oil consumption is increasing can be reduced
through conservation measures or that an amount of undiscovered oil equal to the
total known oil reserves can be found? ❐

A Bouncing Ball
We now consider another application involving the sum of a geometric sequence.
EXAMPLE 6
Suppose that a properly inflated basketball is designed to bounce back to three-quarters
of the height from which it is dropped.
a. If such a ball is initially dropped from a height of 10 feet, find the total vertical dis-
tance it travels on the first 10 bounces; the first 20 bounces; the first 30 bounces.
b. What total vertical distance does the ball cover if, theoretically, it keeps bouncing in-
definitely?
Solution

a. Figure 5.40 shows that the vertical distance the ball travels is 10 feet until the first
bounce plus 2 times the distance (up and then down) between the first and second
bounces, or

10
3
4
(10) = 7.5

3 3
( ( 10 = 5.625
4 4
3
Height

( 34 ( (10)

FIGURE 5.40 Time

gord.3896.05.pgs 4/24/03 10:03 AM Page 424

424 CHAPTER 5 Modeling with Difference Equations

2 a . 10b ,
3
4
plus 2 times the distance between the second and third bounces, or
3 2
2 c a . 10b d  2 a b . 10,
3 3
4 4 4
and so on.
The total vertical distance traveled on the first n bounces is therefore
3 2 3 3 3 n
Dn  10  2 a b . 10  2 a b . 10  2 a b . 10  . . .  2 a b . 10
3
4 4 4 4
3 2 3 3 ... 3 n
 10  20 a b  20 a b  20 a b   20 a b
3
4 4 4 4
2 3 n
 10  20 c a b  a b  a b  . . .  a b d
3 3 3 3
4 4 4 4
3 2 ... 3 n1
 10  20 a b c 1  a b  a b   a b d
3 3
4 4 4 4
2 n1
 10  15 c 1  a b  a b  . . .  a b d .
3 3 3
4 4 4
Note that the expression in the brackets on the preceding line is the sum of a finite
number of terms of a geometric sequence with common ratio r  34 . Therefore the
total vertical distance traveled by the ball during the first n bounces is
1  13>4 2 n 1  13>4 2 n
Dn  10  15 a b  10  15 a b
1  13>4 2 1>4
3 n
 10  60 c 1  a b d .
4
Consequently, the vertical distance traveled by the ball during the first 10 bounces is
3 10
D10  10  60 c 1  a b d
4
 10  56.621  66.621,
or about 66.6 feet. During the first 20 bounces, the ball travels
3 20
D20  10  60 c 1  a b d  69.810,
4
or about 69.8 feet. And during the first 30 bounces, it travels
3 30
D30  10  60 c 1  a b d  69.989,
4
or about 69.99 feet.
b. Little extra distance is contributed by the 20th through the 30th bounce, and all sub-
sequent bounces will contribute even less. In fact, because the common ratio r  34 is
between 1 and 1, we know that 1 34 2 n S 0 as n increases and so we can sum all of the
terms of the infinite geometric sequence to obtain
10  15 31  34  1 34 2 2  1 34 2 3  . . .  1 34 2 n  . . . 4
 10  15 3 1  113>4 2 4  10  151 1>4
1
2
gord.3896.05.pgs 4/24/03 10:03 AM Page 425

5.5 Geometric Sequences and Their Sums 425

 10  1514 2  70 feet.
That is, theoretically, if the ball were to continue bouncing forever, the total vertical
distance it would travel would be 70 feet.
◆

Problems
1. Find the sum 1  r  r 2  . . .  r n, with r  12, grow at an annual rate of 2% and 26 billion bar-
for n  10, n  20, and n  30. rels were consumed in 2000, determine how long
2. Repeat Problem 1 with r  0.2. the oil reserves will last.
3. Repeat Problem 1 with r  0.8. 15. Suppose that, as undeveloped countries become
more industrialized, the annual growth rate in oil
4. Repeat Problem 1 with r  0.8. consumption increases to 2.5%. Based on the cur-
5. Repeat Problem 1 with r  1.5. rent 2250 billion barrel estimate for worldwide oil
6. Repeat Problem 1 with r  2.5. reserves and the estimated worldwide oil consump-
7. Suppose that 6000 new cases of a certain disease oc- tion of 26 billion barrels in 2000, how long will it be
curred in 1960. If the number of new cases dimin- until all the known oil is used?
ished 20% per year since then, what is the total 16. Repeat Example 6 if the initial height of the ball is 6
number of people who contracted this disease from feet.
1960 through 2000? from 2000 through 2010? 17. Repeat Example 6 if the initial height of the ball is
8. Repeat Problem 7 if the number of new cases of the 12 feet.
disease increased 20% per year since 1960. 18. Repeat Example 6 if the ball bounces back to 80%
9. In 1980, the United States used approximately 2.5 bil- of its height. By how much does the total distance
lion kilowatt-hours of electricity. If electric usage traveled by the ball change compared to a 75%
grew by 2% per year, find the total amount of elec- bounce?
tricity used between 1980 and 2000. 19. Repeat Example 6 if the ball bounces back 23 of its
10. The United States produced 195,000 metric tons of height.
wheat in 1984. If production increased by 10% per 20. The repeating decimal 0.222222 . . . can be thought
year find the total amount of wheat produced be- of as
tween 1984 and 2002.
2 2 2
11. The United States produced 70,600 metric tons of    . . ..
rice in 1984. If rice production fell by 9% per year, 10 100 1000
find the total amount of rice produced between What is the sum of this geometric sequence?
1984 and 2002.
21. The repeating decimal 0.252525 . . . can be thought
12. At age 22, Ken gets his first job paying $35,000 per of as
year. If he stays with the same employer and gets an
25 25 25
average annual increase of 4% each year, what will    . . ..
be his total earnings over his entire career by the 100 10,000 1,000,000
time he retires at age 65? What is the sum, as a simple fraction, of this geo-
13. In 1986, a total of 70,000 pages of new mathemat- metric sequence?
ical research was published. If the amount of re- 22. A geometric sequence is based on the fact that the ratio
search grew at the rate of 8% per year, find the of successive terms is constant: xn1>xn  r. Suppose
total amount of new mathematics research pub- instead that the ratio of successive terms in a sequence
lished between 1986 and 2000. is a linear function—say, xn1>xn  rn, for n  1.
14. Suppose that several immense new oil fields are a. How does the growth rate of this sequence com-
discovered that increase the worldwide oil re- pare to that for a geometric sequence?
serves tenfold. If oil consumption continues to b. What is the solution of xn1>xn  n, for n  1?
gord.3896.05.pgs 4/24/03 10:03 AM Page 426

426 CHAPTER 5 Modeling with Difference Equations

c. What is the solution of xn1>xn  5n, for n  1? c. How does the estimate in part (b) compare to the
d. What is the solution of xn1>xn  rn for n  1, actual total gross obtained by adding the entries in
for any r? the table?
e. What is the solution of xn1>xn  rn  b, for 26. The estimated worldwide reserve of natural gas at
n  1? the end of 1999 was 5200 billion cubic feet. The
23. How would the solution of xn1>xn  n2 compare worldwide consumption of natural gas in 1999 was
to the solution of xn1>xn  n, for n  1? Con- about 84 billion cubic feet and was growing at an
struct a formula for this solution. annual rate of about 2.9%. Source: Energy Information Ad-
24. Consider xn1>xn  n1 . How does its solution be- ministration, www.eia.doe. gov/ieu/.
have? How does it compare to the solution of
xn1>xn  1>r , for any r  1? Construct a formula
a. If this growth rate continues, how long will the
known reserves of natural gas last?
for this solution.
b. If new discoveries of natural gas triple the
25. The table below shows the box-office gross, in mil- known reserves, how long will it take to deplete
lions of dollars, of the movie Star Wars: The Phan- the natural gas reserves at the current growth
tom Menace during its first 10 weeks in the theaters. rate in consumption?
a. Find the exponential function that models the c. How long will the current 5200 billion cubic feet
box office gross for the movie as a function of of natural gas last if the annual growth rate of
the number of weeks in release. consumption rises to 3.5%?
b. Use the exponential function from (a) to esti- d. How long will the current 5200 billion cubic feet
mate the total box office gross for The Phantom of natural gas last if the annual growth rate of
Menace during its first ten weekends in theaters. consumption falls to 2%?

Weekend 1 2 3 4 5 6 7 8 9 10

Gross 101.4 48.7 41.2 31.1 23.6 21.5 12.0 8.0 9.6 5.8
Source: Internet Movie Database: http://us.imdb.com.

Chapter Summary

In this chapter we introduced difference equations and their solutions. We also dis-
cussed the use of difference equations as mathematical models of population
growth and other phenomena. More specifically we showed the following.
◆ How to present the solution sequence to a difference equation either as a
closed-form expression for the nth term or as a sequence of numbers gener-
ated term-by-term.
◆ The fact that the solution to a difference equation depends on the initial
condition with each different initial condition giving rise to a different solu-
tion sequence.
◆ How to model the level of a drug in the bloodstream with difference equa-
tions.
◆ How to find the maintenance level associated with a drug and what it
means.
◆ How to model exponential growth and decay processes with difference
equations.
◆ How to model population growth with the Fibonacci difference equation.
gord.3896.05.pgs 4/24/03 10:03 AM Page 427

Review Problems 427

◆ How to model inhibited population growth with the logistic model, includ-
ing finding the maximum sustainable population.
◆ How to interpret the behavior of a logistic curve.
◆ How to estimate the logistic coefficients from a set of data.
◆ How to model temperature decrease with a difference equation based on
Newton’s law of cooling.
◆ How to model temperature increase with a difference equation based on
Newton’s law of heating.
◆ How to interpret the behavior of the solutions for the heating and cooling
models.
◆ How to sum the first n terms of a finite geometric sequence.
◆ How to sum all the terms of an infinite geometric sequence, provided that
the common ratio is between 1 and 1.
◆ How to apply the formulas for the sum of the terms in a geometric se-
quence.

Review Problems
1. Write the first five terms of each sequence. is lost each day through evaporation or filters. What is
a. an  6n  1 the maintenance level of chlorine in the reservoir?
3n 7. The difference equation vn1  1.30vn  0.00002vn2
b. tn  , n0 models the number of people in a town who have
n
c. rn  1  10.32 n VCRs as a function of the year n. Initially, 40 house-
holds had VCRs. Estimate how many people will
2. Determine the first five terms in the solution se- eventually have VCRs. Determine the year in which
quence of each difference equation. half the population has VCRs.
a. xn1  xn  8, x0  2 8. A population grows according to the logistic model
b. xn1  xn  8, x0  12 from an initial size of 1000 to a final size of 12,000.
1 The annual growth rate is 20%. What is the inhibiting
c. xn1  xn , x0  5
3 constant? Write the difference equation that describes
d. xn1  xn  132 n, x0  10 this population for any year n.
3. Determine the first five values in each solution se- 9. The size of the fish population in a stream grows in
quence. accordance with the logistic model at an annual
a. yn2  yn1  yn , y0  2, y1  7 rate of 30% with an inhibiting constant of 0.04%.
b. yn2  yn1  yn , y0  3, y1  7 Write the difference equation for each situation.
4. A drug is administered every 6 hours. The kidneys a. The fish population grows according to the
eliminate 60% of the drug over that period. If the logistic model.
original dose is 100 mg, how much of the drug re- b. The fish population grows according to the
mains in the body after 8 days? logistic model, but the state game warden allows
2000 fish to be caught and removed from the
5. A drug is administered every 4 hours. The kidneys stream each year.
eliminate about 70% of the drug over the 4-hour c. The fish population grows according to the
period. The initial dose of the medicine is 100 mg. logistic model, and the state game warden stocks
How much should the repeated dosage be to ensure the stream with 400 new fish per year.
a maintenance level of 30 mg? d. The fish population grows according to the logistic
6. Chlorine is added to a town’s water supply reservoir at model, but about 10% of the population is caught
the rate of 30 lb>day. An estimated 20% of this amount every year.
gord.3896.05.pgs 4/24/03 10:03 AM Page 428

428 CHAPTER 5 Modeling with Difference Equations

10. Write the difference equation for each scenario, so he adds a quart of oil weekly. The capacity of the
draw a graph of the behavior of the solution, and engine is 6 quarts of oil.
answer the question. a. Write a difference equation for the amount of
a. Pancake syrup used to be 100% maple syrup, but oil in the engine as a function of time, meas-
over the years the amount of maple syrup has ured in weeks.
been reduced. Suppose that in 1970 a company b. Find the solution to the difference equation.
began reducing the amount of maple syrup in its c. Use the solution to predict when the level of oil
product by 15% per year. What percentage of in the engine just after the weekly quart is added
maple syrup is in its product in 2005? will be down to 5 quarts.
b. John’s investment in the stock market has been For each difference equation in Problems 12–19, decide
growing by 10% per year. He adds $2000 a year whether the behavior of the solution is an increasing
to his investment. If he had $50,000 invested in concave up pattern, a decreasing concave up pattern, an
1995, how much does he have invested in 2005? increasing concave down pattern, a decreasing concave
c. Advertising in the print media has been less im- down pattern, or none of these patterns. In each case,
portant recently than it was in the past. In 1998, x0  50.
a company decided to decrease its budget for ad-
12. xn  511.042 n
vertising in the print media by $20,000 per year.
Also, the company increased its budget for tele- 13. xn  5n2.5
vision by 10% per year. If the budgets were each 14. xn  810.852 n
$2 million in 1998, how much is the company 15. xn  4n0.3
budgeting for print and how much for television 16. xn  12n1.2
at the end of 5 years? How is the total advertising 17. xn  5n  3
budget changing?
18. xn  25  2n
11. Chris’s old car has a major oil leak. He estimates that
it loses about 25% of the oil in the engine every week, 19. xn  12  310.902 n
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 429

6
Introduction to
Trigonometry
6.1 The Tangent of an Angle
Historically, trigonometry was developed to solve problems involving right trian-
gles. Thus, for the right triangle shown in Figure 6.1, if we know any two of the three
sides a, b, and c, we can easily find the third side with the Pythagorean theorem
c 2  a 2  b 2,
where c is the length of the hypotenuse. But, there is no simple way to find the two
unknown angles. To find them, we need to use trigonometry.
c
b Similarly, if only one of the three sides and one angle are known, we can easily
find the other angle (the two nonright angles must sum to 90° because they are
complementary angles). However, without trigonometry, there is no simple way to
find the lengths of the other two sides.
a The basic idea behind trigonometry is a fundamental geometric fact about
right triangles. The two right triangles shown in Figure 6.2 share the angle u (lower-
FIGURE 6.1 case Greek letter theta). Therefore the remaining angle in both triangles must be the
same. We denote it f (the lowercase Greek letter phi). Because all three angles in
both triangles are the same, the triangles are similar (see Appendix A4). As a conse-
quence, once an angle u (other than the right angle) has been specified in a right tri-
φ angle, that triangle is similar to every other right triangle having the same angle u.
In the smaller triangle shown in Figure 6.2, from the point of view of the
c2
b2
angle u, there is an adjacent side, denoted by a1 ; an opposite side, denoted by b1 ;
φ and the hypotenuse, denoted by c1 . In the larger triangle, also from the point of
c1 b1 view of the angle u, the adjacent side is a2 , the opposite side is b2 , and the hypotenuse
θ a1
is c2 . The triangles are similar, so that their corresponding sides are proportional, and
a2 a1 b1 c1
  .
a2 b2 c2
FIGURE 6.2
Equivalently, once an angle u has been specified, the ratio of corresponding sides of
these right triangles will be the same. In particular, among several other compara-
ble ratios,
a1 a2 a1 a2 b1 b2
 ,  , and  .
b1 b2 c1 c2 c1 c2

429
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 430

430 CHAPTER 6 Introduction to Trigonometry

That is, each of these ratios depends solely on the angle u, not on the dimensions
of the triangle. It is these ratios, and their dependence on the angle u, that form
the basis of trigonometry. In this section, we begin by examining one of these
three ratios.

The Tangent of an Angle

Suppose that your math instructor has assigned you the task of calculating the
height of a tall flagpole in the middle of campus. The direct approach would be to
climb to the top, release a string until the bottom reaches the ground, and then
measure the length of string. Obviously, this method presents some practical diffi-
culties, and you would likely try to come up with some less physical approach.
Assume that, when you go out to the flagpole, you notice that the pole is cast-
ing a 66-foot-long shadow. How can you use this piece of extra information to de-
termine the height of the pole? Suppose that you enlist the aid of a friend Ron, who
is exactly six feet tall. Have him stand in the shadow cast by the pole so that the tip
of his shadow falls exactly on the same spot A as the tip of the shadow of the flag-
pole, as illustrated in Figure 6.3. Also, suppose that the length of his shadow is 814 ,
or 8.25, feet. The two triangles ABC and ADE are similar because the angles are the
same, and so the corresponding sides are proportional. Therefore
Ron’s height height of pole

length of his shadow length of pole’s shadow
6 height of pole
 .
8.25 66

1 6
8 –4
A D B
FIGURE 6.3 66

Multiplying both sides by 66 yields

6166 2
Height of the pole   48 feet.
8.25
Is this result correct? You can check it with the help of another friend, Sue, who
is five feet tall. Have her stand so that the tip of her shadow matches the end of the
pole’s shadow. Suppose that the length of her shadow is 6 78 , or 6.875, feet, which
leads to right triangle AFG that is similar to the previous two, as illustrated in Fig-
ure 6.4. Because the corresponding sides are proportional, we get
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 431

6.1 The Tangent of an Angle 431

E
G
6
5
A 7 B
6 –8 F D
FIGURE 6.4 66

Sue’s height height of pole


length of her shadow length of pole’s shadow
5 height of pole
 .
6.875 66
Again, we find that
5166 2
Height of the pole   48 feet.
6.875
Let’s look at this situation from a slightly more sophisticated point of view. In
each of the three right triangles shown in Figure 6.5, the various lengths are differ-
ent but the angles in corresponding positions are all the same, so all three triangles
are similar. The angle u (which is the same as angle CAB, angle EAD, and angle
GAF) is called the angle of inclination. Using a protractor, we measure this angle
and find that u is about 36°. In fact, in any right triangle where the angle of incli-
nation is 36°, the ratio of the vertical height (the opposite side) to the horizontal
distance or width (the adjacent side) will always be the same; in this case,
Height 6 5
  7  0.727.
Width 8.25 68

FIGURE 6.5 θ
A F D B

Of course, if the angle u has a different value—say, u  40°—the configura-

tion of height and width is different and their ratio therefore is different. The ratio
of height to width, or opposite side to adjacent side, in a right triangle depends
only on the size of the angle u, so this ratio is a function of the angle. We call this
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 432

432 CHAPTER 6 Introduction to Trigonometry

function the tangent of the angle, the tangent ratio, or the tangent function and
write it as
opposite height
tan u   .
adjacent width
Use your calculator, in Degree mode, to verify that tan 36°  0.7265. (Note that
the values of the tangent function, as well as the other trigonometric functions that
we discuss in Section 6.2, typically are irrational numbers, but we usually give the
values to three or four decimal places.)
Because we are concerned exclusively with right triangles here, the angle u
must be between 0° and 90°, and so for now the domain of the tangent function
consists of all angles 0°  u  90°. (Later we show how we can extend it to a larg-
er domain.) Also, we can have a right triangle in any possible orientation, as shown
in Figure 6.6, so the words height and width may not be appropriate. Instead, we
typically think of the tangent ratio for an angle u as follows.

opposite
tan u 
adjacent

Adjacent

Opposite
θ

Hypotenuse
FIGURE 6.6

From the point of view of the other angle f in the triangle, the opposite and
adjacent sides are reversed, as depicted in Figure 6.7. Note also that the angles u and
f are complementary angles.

The Tangent of Some “Special” Angles

Recall from geometry that in any 45°– 45°– 90° right triangle the two sides flanking
the hypotenuse are equal and, by the Pythagorean theorem,
c2  a2  a2  2a2,
so c  12 a. That is, the hypotenuse must be 12 times the length of either side, as il-
lustrated in Figure 6.8. In this triangle with angle u  45° and sides a, a, and 12 a,

45°
φ

Hypotenuse √2a a
Adjacent

θ 45°
FIGURE 6.7 Opposite FIGURE 6.8
a
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 433

6.1 The Tangent of an Angle 433

opposite a
tan 45°    1.
adjacent a

You can easily verify that tan 45°  1 on your calculator. (Be sure that your calcu-
lator is set in Degree mode.)
Similarly, recall from geometry that in any 30°–60°–90° right triangle, the side
opposite the 30° angle is one-half the hypotenuse, or, equivalently, the hypotenuse
is twice the side opposite the 30° angle. In such a triangle, suppose that the side op-
posite the 30° angle has length a so that the hypotenuse has length 2a, as shown in
Figure 6.9. We find the length of the third side from the Pythagorean theorem. Be-
cause a2  b2  c2, we have b2  c2  a2, so
b2  12a2 2  a2  4a2  a2  3a2 so that b  23 a.

60°

2a
a

30°

b = √3a

FIGURE 6.9

Consequently, for an angle of 30°, the ratio of the opposite side to the adjacent side is
a 1
tan 30°    0.577.
13 a 13
Alternatively, using a calculator, we find tan 30°  0.577.
Similarly, to find the tangent of 60°, we see from Figure 6.9 that the side oppo-
site the 60° angle is 13 a and the side adjacent to it is a, so that
13 a
tan 60°   13  1.732,
a
which you can also check on your calculator.
For any angle u between 0° and 90°, you can use a calculator to obtain the cor-
responding value for tan u. For instance, to three decimal place accuracy,
tan 10°  0.176,
tan 20°  0.364,
tan 50°  1.192,
tan 80°  5.671.
Note that as u increases toward 90°, the value of tan u also increases; that is, the
tangent is an increasing function of u, at least between 0° and 90°. Does that make
sense? Imagine walking toward the 556-foot-high Washington Monument while
keeping your eye fixed on the top of the monument, as illustrated in Figure 6.10.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 434

434 CHAPTER 6 Introduction to Trigonometry

The opposite side (the vertical height) remains the same, 556 feet, while the adja-
cent side (the horizontal distance) gets smaller and smaller. The closer you get to
the monument, the larger the angle of inclination and the larger the ratio of the
fixed vertical height to the diminishing horizontal distance. By the time your eye is
practically touching the side of the monument, and the angle is virtually 90°, the
value for the tangent function has gotten very large indeed. The tangent function is
not defined for u  90° because the length of the adjacent side would be zero.
What about the tangent of 0°? Suppose that you’re standing across the street
from a glass elevator that is descending along the outside of a tall building, as illus-
trated in Figure 6.11. Now the adjacent side (the horizontal distance) is fixed, the
opposite side (the vertical height) is decreasing, and the angle u is decreasing to-
ward 0°. Therefore the value of the tangent function is likewise diminishing be-
cause it is the ratio of the decreasing vertical height and the fixed horizontal
distance. Clearly, tan 0° is 0. We therefore conclude that the domain of the tangent
function can be extended at least to 0°  u  90°.

h
θ1
θ1 θ2 θ3 θ2
θ3

FIGURE 6.10 FIGURE 6.11

Behavior of the Tangent Function

Let’s consider the values for the tangent function and investigate their growth pat-
tern. Using a calculator, we obtain the following values.

U 0° 10° 20° 30° 40° 50° 60° 70° 80° 90°

tan U 0 0.176 0.364 0.577 0.839 1.192 1.732 2.747 5.671 UNDEF

Note that, as the angle u increases from 0° to 10° to 20°, and so on, the tangent
function is growing ever more quickly, so the function is concave up. The graph of
the tangent function y  tan u for angles between 0° and 90° is shown in Figure
6.12. It passes through the origin and grows in a concave up pattern, approaching a
vertical asymptote as u approaches 90°.
How does the growth pattern compare to that of an exponential function? If
you examine the successive ratios of the values of tan u, you will find that they are
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 435

6.1 The Tangent of an Angle 435

20

y = tan θ
10

θ
0 15° 30° 45° 60° 75° 90°

FIGURE 6.12

not constant, but rather are increasing considerably. In fact, the tangent function
grows extremely rapidly near u  90° because u  90° is a vertical asymptote for
the function. You might want to look at its graph on your function grapher for an-
gles between 0° and somewhat less than 90°. We examine the properties of the tan-
gent function in considerably more detail in Section 7.4.

Think About This Construct a table of values for the tangent function y  tan u for u  80°, 81°,
82°, . . . , 89° and plot the points. Repeat for u  89°, 89.1°, 89.2°, . . . , 89.9°. ❐

Using the Tangent Ratio

Suppose that we have a right triangle in which we can measure one of the angles
other than the 90° angle with a protractor. We can find the tangent of that angle
with a calculator. Then, if we know the length of either the adjacent side or the op-
posite side, we can easily find the length of the other side without involving Ron,
Sue, or anyone else to solve the type of problem we used to begin this discussion.

E XAMPLE 1
A flagpole casts a shadow of length 66 feet. If the angle of inclination from the tip of the
shadow to the top of the flagpole is 36°, find the height of the flagpole.
Solution Figure 6.13 shows that
opposite H
tan 36°   ,
adjacent 66

H=?

θ = 36°
66 feet

FIGURE 6.13
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 436

436 CHAPTER 6 Introduction to Trigonometry

so
H  66 tan 36°  47.952,
or about 48 feet high.
◆
Note the approach used in Example 1. The first, and key, step was to draw a
sketch of the situation, in which we identified all known parts of the right triangle,
and marked the unknown parts. We then set up the tangent ratio and used it to
find the unknown quantity.

E XAMPLE 2
While hiking through the mountains, you come to the edge of a deep gorge and wonder
how far it is to the other side. A vertical tree is rooted on your side at the edge of the
gorge. From a point 15 feet up in the tree, you find that the angle of depression (meas-
ured down from the horizontal at eye level) to the opposite edge of the gorge is 22°. How
far is it across the gorge?
Solution The height (15 feet) to the point in the tree and the unknown distance D
across the gorge form two sides of a right triangle, as depicted in Figure 6.14. Note that
the 22° angle of depression is not an angle of the triangle. However, it does determine the
measures of the triangle’s angles u and f, based on some simple geometry. First, the
angle u  68° because it is the complement of 22°. Second, the angle f  22° because f
is the complement of u  68°. We therefore have
tan u  tan 68°  2.475.

22°
15 θ = 68°
φ = 22°
D=?

FIGURE 6.14

From the triangle shown in Figure 6.14,

D
tan u   2.475
15
so that
D  1512.4752  37.125.
Therefore the gorge is about 37 feet across.
◆
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 437

6.1 The Tangent of an Angle 437

Note that if we worked with the angle f instead, we would obtain the same result:
15
tan f  tan 22° 
D
so that
15
D  37.13.
tan 22°

Finding an Angle in a Triangle

We often face the problem of determining an angle u in a right triangle when we
know two of the sides. For example, let the two sides of the right triangle shown in
Figure 6.15 be a  20 and b  13, so that
b 13
tan u    0.65.
a 20

b = 13

θ
a = 20

FIGURE 6.15

Suppose that we want to find u. We know from the table of values we constructed
previously for the tangent function that tan 30°  0.577 and tan 40°  0.839. Be-
cause the values for the tangent are strictly increasing, we expect u to be between
30° and 40°. We can improve on these rough estimates by trial and error. For in-
stance, using a calculator, we might find that tan 35°  0.7002 (too high),
tan 32°  0.625 (too low), tan 34°  0.6745 (slightly too high), and so on.
A far more effective method is to use the inverse of the tangent function, which
gives the angle whose tangent has a particular value. (We discuss this inverse func-
tion in detail in Section 7.4.) For now, on your calculator simply press either 2nd
or INV followed by TAN and then the known tangent value. For this example, INV
TAN 0.65 returns 33.024. That is, 33.024° is the angle whose tangent value is 0.65.
You can check on your calculator that tan 33.024°  0.65.
The inverse tangent of a number x is usually written as either arctan x or
Tan1x. We will use the first notation, arctan x.
E XAMPLE 3
A ski slope drops 1500 feet vertically in the process of covering 4300 feet horizontally.
a. What is the angle of inclination of the ski slope?
b. What is the actual distance that a skier will ski down the slope?
Solution

a. We start with a sketch of the ski slope, as shown in Figure 6.16. From geometry, the
angle u equals the angle inside the triangle at the end of the ski run (they are alternate
angles between parallel lines). Therefore
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 438

438 CHAPTER 6 Introduction to Trigonometry

θ
D=?
1500 feet
θ
4300 feet

FIGURE 6.16

opposite 1500
tan u    0.3488,
adjacent 4300
so that
u  arctan10.3488 2  19.2288,
or about 19.2°.
b. The actual distance skied D is simply the length of the hypotenuse. Therefore, from
the Pythagorean theorem,
D2  15002  43002,
so that
215002  43002  220,740,000  4554.12,
or about 4554 feet.
◆
In general, problems in right angle trigonometry typically involve knowing a
small amount of information about a right triangle and using that information in-
telligently to determine values for the other parts (either the sides or the angles) of
the triangle. In fact, there are only a limited number of possibilities. We list these
cases (based on the right triangle shown in Figure 6.17) in the following table. We
leave the last column for you to complete. Decide on an appropriate strategy for
finding each of the missing pieces, based on the information given or previously
determined.

Given Objective Strategy

a and b Find c.

Find u.

a and c Find b.

Find u.

b and c Find a.

Find u.

a and u Find b.

Find c.

b and u Find a.
(continued )
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 439

6.1 The Tangent of an Angle 439

Given Objective Strategy

Find c.

c and u Find a. Cannot be done simply by using tan u.

Find b. Cannot be done simply by using tan u.

c
b

θ
a

FIGURE 6.17

We examine the last case when c and u are known—which cannot be solved by
using the tangent of an angle—in Section 6.2.
Whenever you face any problem involving a right triangle, your first step should
always be to draw a simple picture of the situation to identify the different parts of
the triangle and see how they are related. Your drawing will help you determine
which strategy, if any, to use to solve for the remaining parts of the triangle.

Problems
1. a. Use a ruler to measure, as accurately as possible, 2. u  52° and b  12 3. u  16° and a  12
the lengths AB, AC, AD, AE, AF, AG, BE, CF, and 4. c  15 and a  6 5. c  30 and b  18
DG.
6. a  72 and b  47

c
F b

E
θ
a

7. The shadow of a flagpole is 50 feet long. A line of

A B C D sight from the tip of the shadow to the tip of the
pole makes an angle of 28° with the ground. How
high is the pole?
b. Using your values from part (a), calculate the ra- 8. You want to find the distance across a straight, fast-
tios AB AC AD BE BE CF DG CF DG
AE , AF , AG , AB , AE , AF , AG , AC , and AD . flowing river. You find two vertical trees that are di-
c. Group the ratios that you calculated in part (b) rectly across the river from one another at points A
that appear to be equal. Identify any patterns and B so that the angle at B is a right angle, as
that help you explain why certain ratios have the shown in the accompanying diagram. You then
same values. measure a distance of 32 feet to another tree at
For Problems 2–6, refer to the accompanying figure. point C on the edge of the river on your side. From
Use the information given to find all other parts of the the tree at C, you find that the angle ACB is 56°.
triangle. Find the distance across the river.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 440

440 CHAPTER 6 Introduction to Trigonometry

a. What is the shortest length for a ramp that meets

this requirement?
b. How far is the start of the ramp from the base of
the platform?
A B
12. Jill is standing at the top of a vertical cliff and Jack is
standing 25 feet away from the foot of the cliff and es-
32
timates that the angle of elevation u from his position
to Jill’s is 40°. Approximately how high is the cliff?
56° 13. Suppose that Jack’s measurement in Problem 12 of
the distance to the cliff is off by 1 foot. How much
C difference does this error make in the calculated
height of the cliff? (Hint: Recalculate your answer to
9. The line of sight from the top of a lighthouse to a Jet- Problem 12 two ways, once for a distance of 24 feet
Ski out on the water makes an angle of depression of and then for a distance of 26 feet.)
5° with the horizontal. The lighthouse is 50 feet high.
14. Suppose that Jack’s estimate of the angle of elevation
5°
in Problem 12 is off by 1°. How much difference
does this error make in the calculated height of the
50 ft
cliff?
a. How far is the Jet-Ski from the base of the 15. Suppose that Jill, at the top of the cliff, wants to find
lighthouse? the horizontal distance from the foot of the cliff to
b. What is the straight-line distance from the top of where Jack is standing without climbing down and
the lighthouse to the Jet-Ski? measuring it directly. She drops a rock at the end of
a long measuring tape down the cliff and finds that
10. A helicopter is hovering over a particular spot with
the height of the cliff is about 75 feet. Next, she
its searchlight trained on an injured hiker on the
measures the angle of depression from her position
ground. Because of tricky wind currents, the pilot
to Jack’s to be approximately 70°. How far is Jack
can’t get the copter any closer to the hiker. The
from the foot of the cliff?
angle that the searchlight makes with the ground is
20°. If the copter pilot estimates that she is at a 16. The installation instructions for a TV satellite re-
height of 300 feet above the ground, how far away, ceiver at a particular location call for it to be aimed
horizontally, is the injured hiker? at an angle of 68° from the horizontal. Unfortunate-
ly, your protractor is broken. Devise a strategy that
will help you aim the dish in the proper direction.
RESCUE

17. a. Find the missing entries in the table.

300 ft
20° tan U 0 0.5 1 1.5 2 2.5 3
U
b. Plot the points (tan u, u) and connect them with
11. A wheelchair ramp is to be built from ground level
a smooth curve. (This is part of the graph of the
to a platform 7 feet above the ground. The angle of
function y  arctan u.)
inclination with the ground is required to be no
greater than 15°.

6.2 The Sine and Cosine of an Angle

Suppose that you’re flying a kite at the end of 400 feet of string and are curious
about how high the kite is. How can you find its height? Figure 6.18 shows that the
length of string is simply the hypotenuse of the right triangle. You can measure the
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 441

6.2 The Sine and Cosine of an Angle 441

angle of inclination u that the kite string makes with the horizontal—say, u  37°.
Recall that this case is the last one presented in the strategy table in Section 6.1,
where we pointed out that the tangent function is of no help. We must devise a dif-
ferent strategy to determine the height y of the kite.

400 y=?

FIGURE 6.18

Using a yardstick, you could measure 5 feet along the kite string and then
measure the height from the horizontal to that point on the string; say that you get
3 feet. As shown in Figure 6.19, you have a pair of similar right triangles ABC and
ADE, so you know that their corresponding sides are proportional. Consequently,
height of kite height to point on string

length of hypotenuse length of string to that point
y 3
 .
400 5

400
y
E
5
3
37°
A D B

FIGURE 6.19

Therefore the height of the kite is

y  400 a b  240,
3
5
or 240 feet above the “horizontal.” (If, in fact, you hold the kite string chest-high,
say, 4 feet above the ground, the kite is 240 feet above your hand. Hence the kite is
actually 244 feet above the ground.)

The Sine of an Angle

The key to solving this problem is to construct the ratio of the height and the hy-
potenuse of the right triangle. In our discussion of the tangent of an angle, we in-
dicated that, for any angle u, we can construct infinitely many right triangles that
are all similar. Thus the ratio of the height and the hypotenuse will be the same for
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 442

442 CHAPTER 6 Introduction to Trigonometry

all these similar triangles, as illustrated in Figure 6.20. Because the ratio changes as
the angle changes, this ratio is a function of the angle u. We define this ratio to be
the sine of the angle, or the sine function, and write it as follows.

opposite
sin u 
hypotenuse

se
te nu
po Opposite
Hy
e
us
oten opposite
p
hy
θ adjacent
Adjacent

FIGURE 6.20

As with the tangent function, opposite refers to the side opposite the angle u re-
gardless of the orientation of the right triangle. You must think of the opposite side
in terms of the angle u, not as the side of a triangle that is in some particular loca-
tion, such as the vertical position.

The Behavior of the Sine Function

For now we’re concerned only with right triangles, so the domain of the sine
function consists of angles between 0° and 90°. The following comments apply
only to this situation. In Section 6.3, we consider cases for which this restriction
is lifted, leading to more interesting and useful behavior patterns for the sine
function.
As with the tangent ratio, you can get the values for the sine of any angle in a
right triangle using your calculator in Degree mode. For instance,
sin 10°  0.174,
sin 20°  0.342,
sin 30°  0.5,
sin 40°  0.643,
sin 75°  0.966.
Note that the values for the sine function y  sin u are increasing as the angle u in-
creases from 0° to 90°. Further, the sine function grows more rapidly for small an-
gles and less rapidly as angles get closer to 90°, so that these values follow a concave
down pattern. Figure 6.21 shows a graph of the sine function for u between 0° and
90°. (We discuss the sine’s behavior for angles outside this interval in Section 6.3.)
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 443

6.2 The Sine and Cosine of an Angle 443

y = sin θ

θ
0 90°

FIGURE 6.21

The Sine of Some “Special” Angles

As we did with the tangent function in Section 6.1, let’s consider some of the spe-
cial angles, notably u  0°, 30°, 45°, 60°, and 90°, from the point of view of the sine
function. To begin, think about what happens in a right triangle as the angle
shrinks to 0° for a fixed hypotenuse c. (Imagine the kite nosediving toward the
ground at the end of the taut string as shown in Figure 6.22.) The length of the op-
posite side also decreases to 0, so
opposite 0
sin 0°    0.
hypotenuse c
What about sin 90°? Again, for a fixed hypotenuse c, think about what happens in a
right triangle as the angle increases to 90°. (Although improbable, imagine the kite
moving directly overhead so that the height of the kite becomes equal to the length
of the string, as illustrated in Figure 6.23.) The length of the opposite side in the
triangle grows until it approaches the length of the hypotenuse, so
c
sin 90°   1.
c
Verify these two facts on your calculator.
Now let’s look at the other special angles. As shown in Figure 6.24, when
u  45°, we have a right triangle with two angles of 45° and the two corresponding
sides of equal length, say a. Recall that, by the Pythagorean theorem, the length of

45°

√2a
c a

c θ →90° 45°
θ →0° a

FIGURE 6.22 FIGURE 6.23 FIGURE 6.24

Gordon.3896.06.pgs 4/28/03 1:33 PM Page 444

444 CHAPTER 6 Introduction to Trigonometry

the hypotenuse is 12 a. Hence

opposite a 1
sin 45°     0.707.
hypotenuse 12 a 12
Similarly, as shown in Figure 6.25, when u  30°, the remaining angle is 60°.
Recall that, in any 30°–60°–90° right triangle, the length of the side opposite the
30° angle is half the length of the hypotenuse. If the hypotenuse has length 2a, the
opposite side has length a. Consequently,
opposite a
sin 30°    0.5.
hypotenuse 2a

60°

2a
a

30°

b = √3a

FIGURE 6.25

By the Pythagorean theorem, if the remaining side has length b, then

b2  12a2 2  a2  4a2  a2  3a2,
so b  13 a. We therefore have
13 a 13
sin 60°    0.866,
2a 2
as we found previously by using a calculator.
We summarize these findings as follows

U 0° 30° 45° 60° 90°

1 13
sin U 0 0.5  0.707  0.866 1
12 2

Note that the values for the sine of any angle in a right triangle must always lie be-
tween 0 and 1. The reason is that the sine is the ratio of the opposite side and the
hypotenuse, and in any right triangle the hypotenuse is always the longest side.

Applications of the Sine Function

We apply the sine function in Examples 1–3 to illustrate its use in solving different
types of everyday problems.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 445

6.2 The Sine and Cosine of an Angle 445

E XAMPLE 1
A highway through the mountains has a stretch that drops at a grade of 5°. If you drive a
distance of 12 miles along this road, how far do you descend vertically?
Solution To help visualize the situation, we “straighten out” all curves in the road and
sketch the situation, as shown in Figure 6.26, which is not to scale. Note that a 5° grade
also can be thought of as a 5° angle of descent or a 5° angle of declination. We know that
the length of the hypotenuse is 12 miles. We let y be the vertical drop, and get
y
sin 5°  so that y  121sin 5°2  1.05.
12

5°
12
y

θ = 5°

FIGURE 6.26

Consequently, along this stretch of highway, the road drops about 1.05 miles, or about
5544 feet.
◆

E XAMPLE 2
A tall tree has been uprooted during a storm. It is tilted over and supported near its top
by a vertical wall, as shown in Figure 6.27. The actual horizontal distance from the tree’s
roots to the wall is 42 feet and the angle of elevation of the tree is estimated to be 35°.

φ
L=? H=?

θ = 35°
42

FIGURE 6.27

a. Estimate the length of the tree.

b. Estimate how high on the wall the top of the tree is lodged.
Solution

a. Note that
H
sin u  sin 35°  ,
L
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 446

446 CHAPTER 6 Introduction to Trigonometry

which involves two unknowns; thus we cannot solve the equation. Instead, we must
work with the remaining angle f in the triangle, which is f  90°  35°  55°.
Therefore we have
42
sin 55°  ,
L
so that
42
L  51.27 feet.
sin 55°
(Note that we could also have used the tangent of 35° to determine the height H of the
triangle and then used the Pythagorean theorem to find the length of the hypotenuse.)
b. We now use the Pythagorean theorem to find the height of the triangle:
H 2  51.272  422  864.613,
so that

H  29.4 feet.
◆
Think About This Suppose that the estimate of the angle in Example 2 is off by 5°, either high or
low. How much difference would this error make in the answers to parts (a) and
(b) of Example 2. ❐

Often, we face the problem of determining an angle when we know the

value of the sine of that angle. For instance, if the hypotenuse of a right triangle is
20 and the side opposite the angle u is 15, as shown in Figure 6.28, then
sin u  0.75. What is the angle u? We could find it by trial and error (we know that
sin 45°  0.707 and sin 60°  0.866, so we might try 50°, and so on). A far more
effective approach is to use the inverse sine function, which gives the angle whose
sine has a particular value. We write this inverse function as arcsin x, for any given
value x (although sin1x is also used). We discuss the inverse sine function in detail
in Section 7.3. For now, with your calculator, simply press either 2nd or INV, fol-
lowed by SIN, and then the known value of the sine function—say, 0.75: the cal-
culator returns 48.590. To verify that it is the correct angle, we check that
sin 48.59  0.749996  0.75.

20 15

FIGURE 6.28
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 447

6.2 The Sine and Cosine of an Angle 447

E XAMPLE 3
A 12-foot-long ladder is leaning against a wall. If the foot of the ladder is 4 feet from the
wall, what is the angle of inclination of the ladder?
Solution We start with a sketch of the situation, as shown in Figure 6.29. To use the
φ sine function, we have to consider the point of view of the angle f, so
4 1
sin f  
12 12 3
and therefore
1
f  arcsin  19.47°.
3
θ =?
Hence the angle of inclination of the ladder is
4
u  90°  19.47°  70.53°.
FIGURE 6.29
◆

In Section 6.1, we asked you to complete a table outlining strategies for solving
for all the parts of a right triangle given various combinations of sides and angles.
In all but one of those cases, you could determine all the other parts of the triangle
by using the tangent. Now we ask you to complete the table again by deciding on
appropriate strategies to determine the parts of a right triangle by using the sine
instead of the tangent. Refer to Figure 6.30.

Given Objective Strategy

a and b Find c.

Find u.

a and c Find b.
Find u.

b and c Find a.

Find u.
a and u Find b.

Find c.

b and u Find a.

Find c.

c and u Find a.

Find b.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 448

448 CHAPTER 6 Introduction to Trigonometry

c
b

θ
a

FIGURE 6.30

In practice, which function you apply doesn’t matter so long as you use a cor-
rect strategy. Thus for most of the cases, a variety of different approaches will give
the correct answers. Incidentally, together the tangent function and the sine func-
tion allow you to solve for all the parts of any right triangle in all six cases.

The Cosine of an Angle

So far, we’ve considered two of the six possible ratios among the sides of a right
triangle:
opposite opposite
tan u  and sin u 
adjacent hypotenuse
One other ratio is very useful: the ratio of the adjacent side and the hy-
potenuse, which also depends only on the angle u. We now define a third trigono-
metric function, the cosine of an angle, or the cosine function, as follows.

adjacent
cos u 
hypotenuse

The Behavior of the Cosine Function

As with the sine and the tangent, you can get the values for the cosine of any angle
in a right triangle using your calculator. For instance,
cos 10°  0.985,
cos 20°  0.940,
cos 30°  0.866,
cos 40°  0.766,
cos 50°  0.643.
These values are decreasing more slowly for small angles and more rapidly as an-
gles get closer to 90°. Therefore the values for the cosine function decrease in a con-
cave down pattern for angles between 0° and 90°. Figure 6.31 shows a graph of the
cosine function y  cos u for u between 0° and 90°. Note that it starts at a height of
1 and decreases toward 0 in a concave down pattern as u approaches 90°.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 449

6.2 The Sine and Cosine of an Angle 449

y = cos θ

θ
0 90°

FIGURE 6.31

The Cosine of Some “Special” Angles

Again, let’s consider the special angles u  0°, 30°, 45°, 60°, and 90°. To begin, what
is cos 0°? Think about a right triangle in which the hypotenuse remains constant
and the angle u shrinks to 0°. (Imagine again the kite as it nosedives toward the
ground on a windy day so that the string remains taut, as shown in Figure 6.32.)
The hypotenuse gets closer and closer to the adjacent side, so
adjacent
cos 0°   1.
hypotenuse

400
θ →0°

FIGURE 6.32

Similarly, think about a right triangle in which the hypotenuse remains fixed and
the angle approaches 90°. (Imagine the kite moving directly overhead, as shown in
Figure 6.33.) The adjacent side gets closer to 0, so
cos 90°  0.

400

θ →90°

FIGURE 6.33
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 450

450 CHAPTER 6 Introduction to Trigonometry

Next, let’s look at the other special angles. As shown in Figure 6.34, when u  45°,
adjacent a 1
cos 45°     0.707.
hypotenuse 12 a 12

Think About This This result is the same value we found for the sine of 45°. Explain why they are
the same. ❐

Also, when u  30°, Figure 6.35 shows that

13 a 13
cos 30°    0.866,
2a 2
which is the same as sin 60°. Similarly,
a 1
cos 60°   ,
2a 2
which is the same as sin 30°.

60°
45°

2a
a
√2a a

30°
45°
a √3a

FIGURE 6.34 FIGURE 6.35

We summarize these key values for the cosine function as follows.

U 0° 30° 45° 60° 90°

13 1
cos U 1  0.866  0.707 0.5 1
2 12

Think About This Explain why cos 30°  sin 60° and cos 60°  sin 30°. ❐

Applications of the Cosine Function

We use the cosine function in Examples 4 and 5 to illustrate its value in solving a
couple of rather simple problems.

E XAMPLE 4
To get onto a straight water slide at an amusement park requires climbing a flight of
steps 60 feet high. The slide itself is inclined downward at a 42° angle. How long is the ac-
tual slide?
Solution Figure 6.36 indicates that the angle in the right triangle is 48° and that the ad-
jacent side is 60 feet long. Therefore, to find the length L of the slide, we use
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 451

6.2 The Sine and Cosine of an Angle 451

42°

θ = 48°

60 L

FIGURE 6.36

adjacent 60
cos 48°   .
hypotenuse L
Thus
60
L  89.67.
cos 48°
So the slide is almost 90 feet long.
◆
Think About This Can you solve Example 4 by using the sine function instead? the tangent function? ❐

E XAMPLE 5
A 30 foot ramp extends 24 feet horizontally.
a. What is the angle of elevation of the ramp?
b. How high does the ramp extend?
Solution

a. We start with a sketch of the situation, as shown in Figure 6.37. The hypotenuse has
length 30 feet and the base (which is the adjacent side from the point of view of the
unknown angle u) is 24 feet. Therefore
adjacent 24
cos u   .
hypotenuse 30

30 H=?

θ =?
24

FIGURE 6.37
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 452

452 CHAPTER 6 Introduction to Trigonometry

To find the angle u, we undo the cosine using the inverse cosine (see Section 7.3), so that

u  arccos a b  36.87°,
24
30
or about 37°.
b. We can solve for the height H in a variety of ways. Probably the simplest is to use the
Pythagorean theorem, which gives
H 2  302  242  324,
so that
H  18 feet.
◆

Applications from Physics

The trigonometric functions arise frequently in applications of the physical sci-
ences. Many physical quantities, such as force and velocity, involve both a direction
and a size. Such quantities are known as vectors, and we look at them more formally
in Section 10.1. For now, we consider some physical applications informally to illus-
trate the use of trigonometry.
Imagine pushing against a window that is stuck in order to open it. You exert a
certain force, but the effect of that force depends on the angle u at which you exert
it. If the angle is primarily vertical, most of the effect of your effort is applied to
push the window upward, as shown in Figure 6.38(a). If the angle is more horizon-
tal, as shown in Figure 6.38(b), only a small portion of your effort is applied to
moving the window upward while most of your effort is wasted in the horizontal
direction, effectively pushing the window outward. The total force exerted can be
broken into two parts, one horizontal and the other vertical, based on the angle at
which the force is applied. We illustrate this principle in Examples 6–8.

Effective
e
rc

force
Fo

rce Effective
Fo
force
θ θ
(a) (b)

FIGURE 6.38

E XAMPLE 6
A 30 pound force is exerted at an angle of 20° with the vertical to push a stuck window up-
ward. Find the effective value of the force actually exerted to move the window vertically.
Solution We begin with a sketch of the situation, as shown in Figure 6.39, where the force
being exerted is represented by the hypotenuse of the right triangle. We let the lengths of
the sides equal the sizes of the forces. Thus the hypotenuse has length 30. The portion F
of the force effectively applied to move the window vertically upward is the vertical side
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 453

6.2 The Sine and Cosine of an Angle 453

of this triangle. From the point of view of the 20° angle, the effective force is exerted
along the adjacent side of the triangle, which suggests using the cosine function. In
particular,
F
cos 20°   0.9397
30
so that
F  30 cos 20°  28.19,
or slightly more than 28 pounds of the 30 pounds of the force is applied to moving
the window.

20°

30 F

FIGURE 6.39

◆
E XAMPLE 7
A sailboat is out on a still lake where the wind is blowing at a speed of 16 mph from the
northeast, as shown in Figure 6.40. How fast is the sailboat moving toward the west? to-
ward the south?
Solution Because the wind is blowing from the northeast, the angle it makes with the
horizontal is 45°. The wind is actually pushing the sailboat toward the southwest at
16 mph, as represented by the hypotenuse of the right triangle shown in Figure 6.40. We
want to find the speed w in the westward direction, as indicated by the horizontal side of
the triangle, and the speed s in the southward direction, as indicated by the vertical side
of the triangle.

w=?

45°

s=?
16 mph

45°

FIGURE 6.40
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 454

454 CHAPTER 6 Introduction to Trigonometry

Let’s first determine the sailboat’s speed toward the west. Because the side opposite
that angle is the unknown and we have the hypotenuse, we use the sine function. Thus
w
sin 45° 
16
so that
w  16 sin 45°  11.314.
That is, the sailboat is moving at slightly more than 11 mph toward the west.
To find the speed toward the south, we simply observe that, because the angle in this
right triangle is 45°, the two sides are equal and so s  11.314 also. Thus the sailboat is
also moving at slightly more than 11 mph toward the south.
◆
Suppose that the wind is not quite blowing from the northeast but from some
angle other than 45°. Example 8 demonstrates how the solution changes accordingly.

E XAMPLE 8
A sailboat is out on a still lake where the wind is blowing at a speed of 16 mph from the
northeasterly direction of 32° east of north, as shown in Figure 6.41. How fast is the sail-
boat moving toward the west? toward the south?

w=?

s=?
16 mph

32°

FIGURE 6.41

Solution As in Example 7, the wind is actually pushing the sailboat in a southwesterly

direction at 16 mph, as represented by the hypotenuse in the right triangle in Figure
6.41. But now the angle is 32° instead of 45°. We again indicate the westward speed w
along the horizontal side of the triangle and the southward speed s along the vertical side
of the triangle.
We first determine the speed toward the west. Using the sine function, we have
w
sin 32° 
16
so that
w  16 sin 32°  8.479.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 455

6.2 The Sine and Cosine of an Angle 455

That is, the sailboat is moving at about 812 mph toward the west.
To find the speed toward the south, we need to determine the remaining side of the
right triangle. We do so by using the Pythagorean theorem:
s2  162  w2  162  8.4792  184.107
so that
s  13.569.
Thus the sailboat is moving at about 1312 mph toward the south.
◆
Problems

1. Use a ruler to measure the three sides of the triangle the end of the pendulum reaches compared to its
shown. Based on the measurements, what are your lowest height when it passes the vertical?
best estimates for sin u, cos u, and tan u? What is 11. From takeoff, an airplane reaches a height of 2 miles
your estimate for the angle u? (10,560 feet) in the process of covering 20 miles
horizontally.
a. Find the average angle of ascent of the airplane
as it climbs.
b. Is the actual path upward of the airplane a
straight line, or is the path curved in a concave up
θ =? pattern or in a concave down pattern? Explain
your reasoning.
Problems 2–7 refer to the accompanying figure. Use the c. If the airplane were to climb along a straight-line
information given to find all other parts of the triangle. path, find the distance it would travel as it goes
from the ground to the 2-mile height. Is the dis-
tance that the airplane actually travels greater than
or less than the distance you calculated? Why?
c
b 12. When the space shuttle comes in for a landing at
Cape Canaveral, its descent to the ground for the
final 10,000 feet of height is at an angle of 19° with
θ
the horizontal.
a
a. What actual distance does the shuttle traverse
2. u  52° and b  15 3. u  16° and c  12 along this final glide path?
4. c  22 and a  16 5. c  30 and b  8 b. How far from touchdown, horizontally, should the
6. a  12 and b  9 7. a  42 and u  72° shuttle be when it passes the 10,000-foot altitude?
8. A road up a hill is inclined at 11° to the horizontal. 13. Jack and Jill are about to climb a 400-foot-high hill.
A driver starts driving up this hill and, by checking If the angle of ascent is 52° from the horizontal,
the odometer, discovers that the steep portion of what is the actual distance they will cover to reach
the road extends for three-quarters of a mile. How the summit on a straight track?
much has the car gained in altitude?
9. With its radar, an aircraft spots another aircraft 10,000
feet away at an angle of depression of 15°. Find the x
horizontal distance from one aircraft to the other. 400 ft
10. As a pendulum of length 21 inches swings back
52°
and forth, the maximum angle it makes from the
vertical is u  18°. What is the greatest height that
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 456

456 CHAPTER 6 Introduction to Trigonometry

14. When Jack and Jill came tumbling down from the most of the slices will be ellipses with a minor axis of 3
top of the hill in Problem 13, their angle of descent inches. Suppose that you slice the roll at an angle of
was 61° from the horizontal. What is the actual dis- 27° to the vertical. Find the length of the major axis of
tance they covered while tumbling down? each elliptical slice. (See Appendix A7; ellipses and
15. A javelin is 1 meter long. When it lands after being their properties are covered in detail in Section 9.3.)
thrown, its base is 0.6 meters 1  60 cm2 vertically 20. An escalator rises at a 26° angle with the horizontal.
above the ground. What angle does the javelin make? If it rises 28 feet vertically, what is its length?
21. A safety regulation limits the maximum angle of in-
clination for the ladder on a fire truck to 72°. If a
hook-and-ladder fire truck has a ladder that can ex-
tend to a length of 90 feet, what is the maximum
1m height that it can reach?
0.6 m
22. A balloonist is trying to cross the Atlantic Ocean. If
the wind is blowing at 40 mph from the northwest,
θ what is the actual airspeed at which the balloon is
traveling eastward toward Europe?
16. Problem 15 is unrealistic because the point of the 23. The wind in Problem 22 now shifts slightly and in-
javelin is going to be embedded in the ground. Sup- creases in speed so that it is now blowing at 50 mph
pose that 92 cm of the javelin is visible above the from 40° north of west. What is the actual airspeed
ground, and that its base is still 60 cm vertically at which the balloon is moving eastward?
above ground level. What angle does the javelin
24. a. Find the missing entries in the table.
make with the ground?
17. You must hammer a 3-inch-long nail into a piece
of wood 2 inches thick. Find the steepest angle at sin U 0 0.2 0.4 0.6 0.8 1
which you can hammer the nail all the way into the
U
wood without it coming out the opposite side.
b. Plot the points (sin u, u) and connect them with
a smooth curve.
c. This curve is part of the graph of what function?
25. a. Find the missing entries in the table using only
θ your answers to Problem 24.
n. 2 in.
3i

18. When an airplane takes off, it climbs at an angle of cos U 1 0.8 0.6 0.4 0.2 0
16° at a speed of 180 feet per second. How high is U
the plane after 1 second? after 2 seconds?
19. The cranberry sauce to go with your holiday turkey b. Plot the points (cos u, u) and connect them with
comes out of a can and has a diameter of 3 inches. a smooth curve.
When you slice the roll of cranberry sauce at an angle, c. This curve is part of the graph of what function?

6.3 The Sine, Cosine, and Tangent in General

So far, we’ve considered the trigonometric functions only for angles in a right tri-
angle—that is, angles between 0° and 90°. However, we often encounter situations
in which we need to consider angles larger than 90°. A natural question is: How do
we adapt the ideas previously discussed to such cases? Before we address that ques-
tion, let’s choose such an angle—say, u  125°—and find out what happens when
we use a calculator. We get
sin 125°  0.819, cos 125°  0.574 and tan 125°  1.428.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 457

6.3 The Sine, Cosine, and Tangent in General 457

Let’s see how these values are defined and why they have the indicated signs.

Angles Between 90° and 180°

Consider any angle u between 90° and 180°; imagine it with one side, its initial
side, along the positive x-axis and the other side, its terminal side, in the second
quadrant, as depicted in Figure 6.42. By convention, we measure such an angle
starting along the positive x-axis and rotating counterclockwise. The terminal
side forms the hypotenuse of a right triangle in the second quadrant when we
drop a vertical line from the terminal side to the x-axis. Suppose that the hy-
potenuse of this right triangle is h, the length of the vertical side is y, and the
length of the horizontal side is x. By convention, because y extends up from the
horizontal axis, we think of it as positive. Because x extends to the left of the ver-
tical axis, we think of it as negative. By convention, the hypotenuse is always con-
sidered positive. The angle f in this right triangle is the supplement of the angle
u because u  f  180°; thus u  180°  f, or f  180°  u. The angle f is
sometimes called the reference angle.
y

terminal
side

h
y

θ
φ
x
x<0 initial side

FIGURE 6.42

We define the trigonometric functions when the angle u is between 90° and
180° in terms of the comparable values for the angle f. Therefore
opposite y
sin u   ,
hypotenuse h
adjacent x
cos u   , x0
hypotenuse h
opposite y
tan u   , x  0.
adjacent x
As previously mentioned, for any angle u between 90° and 180° with its terminal
side in the second quadrant, x is negative and y and h are positive. Thus the cosine
and tangent of that angle are negative, whereas the sine is positive, as we saw with
sin 125°  0.819, cos 125°  0.574, and tan 125°  1.428.

If 90°  u  180°,
sin u  0,
cos u  0,
tan u  0.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 458

458 CHAPTER 6 Introduction to Trigonometry

Angles Between 180° and 270°

What about an angle u between 180° and 270° whose terminal side is in the third
quadrant, as depicted in Figure 6.43. We construct a right triangle in the third
quadrant by drawing a vertical line from the terminal side to the x-axis that deter-
mines a reference angle f. Now both the x- and y-values are negative, and
u  180°  f. As before, we define the trigonometric functions for u in terms of
this reference angle f by using the appropriate lengths in the right triangle so that
opposite y
sin u   , y0
hypotenuse h
adjacent x
cos u   , x0
hypotenuse h
opposite y
tan u   , x  0 and y  0.
adjacent x
y

θ
x<0
x
φ
y<0
h

FIGURE 6.43

For any angle between 180° and 270°, the values of x and y are negative, so the sine
and cosine are negative. However, for these angles, the tangent is positive because it
is the quotient of two negative quantities.

If 180°  u  270°,
sin u  0,
cos u  0,
tan u  0.

Think About This Suppose that u  211° so that f  31°. Use your calculator to find the values for
sin 211°, cos 211°, and tan 211°. How do they compare with sin 31°, cos 31°, and
tan 31°? ❐

Angles Between 270° and 360°

Next, consider an angle u between 270° and 360° whose terminal side is in the
fourth quadrant, as shown in Figure 6.44. Once more, we construct a right tri-
angle by drawing a vertical line from the terminal side to the x-axis. We define
each of the trigonometric functions in terms of the reference angle f in that tri-
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 459

6.3 The Sine, Cosine, and Tangent in General 459

angle. Now the y-value is negative, the x-value is positive, and u  f  360°, so
u  360°  f. Also,
opposite y
sin u   , y0
hypotenuse h
adjacent x
cos u   ,
hypotenuse h
opposite y
tan u   , y  0.
adjacent x
Therefore, because y is negative, for any angle between 270° and 360°, the cosine is
positive and the sine and the tangent are both negative.
y

θ
x
x
φ
y<0
h

FIGURE 6.44

If 270°  u  360°,
sin u  0,
cos u  0,
tan u  0.

Angles Greater than 360°

What happens if u is greater than 360°—say, 410°? As shown in Figure 6.45, we can
construct such an angle by looping around a full 360° and then an additional 50°;
essentially, this angle is equivalent to an angle of f  50° in the first quadrant.
Using a calculator, we find that
y

h y

θ = 410° φ = 50°
x
x

FIGURE 6.45
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 460

460 CHAPTER 6 Introduction to Trigonometry

sin 410°  0.766  sin 50°,

cos 410°  0.643  cos 50°,
tan 410°  1.192  tan 50°.
Similarly, if u  775°, we make two full rotations (accounting for 2  360° 
720°), which leaves an angle f  55°, and so
sin 775°  0.819  sin 55°,
cos 775°  0.574  cos 55°,
tan 775°  1.428  tan 55°.
Note that the values for the three trigonometric functions repeat every 360°.
Therefore they are periodic functions because their behavior repeats. The smallest
interval over which the pattern repeats is called the period. The periods of the sine
and cosine functions are both 360°. In general, for any angle u, we have the following.

sin1u  360°2  sin u

cos1u  360°2  cos u

However, the period of the tangent function is 180° because its values repeat
every 180°. Thus, for any angle u, we have the following.

tan1u  180°2  tan u

Check these identities on your calculator either numerically with a variety of dif-
ferent values for u or graphically by comparing the graphs of y  tan x and
y  tan1x  180 2.

Angles Less Than 0°

Finally consider a negative angle—say, u  30°—drawn clockwise, as shown
in Figure 6.46. This angle is equivalent to a positive angle of 330° because both
angles have the same terminal side. Note that y is negative and that x is positive.
We therefore have
y

330°
x
x
θ = −30°
y<0
h

FIGURE 6.46
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 461

6.3 The Sine, Cosine, and Tangent in General 461

y
sin u  , y0
h
x
cos u  ,
h
y
tan u  , y  0,
x
as we have already discussed for angles in the fourth quadrant.
Figure 6.47 summarizes the information about the signs of the three trigono-
metric functions, based on the quadrant containing the terminal side.

II I

sine +
sine + cosine +
tangent +

tangent + cosine +

III IV

FIGURE 6.47

The Graph of the Sine Function

Let’s summarize what we already know about the sine function to determine its
overall behavior pattern. For u between 0° and 90°, y  sin u increases from 0 to
1. For u between 90° and 180°, y  sin u decreases from 1 to 0. For u between 180°
and 270°, y  sin u continues to decrease from 0 to 1. For u between 270° and
360°, y  sin u increases from 1 to 0. Thus the sine function has a maximum
value of 1 and a minimum value of 1. This oscillatory pattern continues indefi-
nitely in both directions (for u  360° and for u  0°). Use your function graph-
er with u between 500° and 500°, say, to observe this pattern.

Think About This Give a similar summary for the behavior of the cosine function. ❐

You can visualize the behavior of the trigonometric functions by looking at

their graphs. Figure 6.48(a) shows the graph of the function y  sin u for u be-
tween 0° and 360° and how the graph relates to the signs of sin u in the four quad-
rants shown in Figure 6.47. In Figure 6.48(b) we expand the graph of y  sin u to
show its behavior between 360° and 720°. This portion of the curve consists of
three full cycles, or repetitions, of the basic sine curve that occurs between 0° and
360°, which is one full period of the function. Also, the curve oscillates between a
minimum height of y  1 and a maximum height of y  1. In particular, the
sine curve reaches its maximum when
u  . . . , 270°, 90°, 450°, . . . ,
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 462

462 CHAPTER 6 Introduction to Trigonometry

y y

I II III IV I II III IV
1 1

θ θ
0 90 180 270 360 –360 –180 180 360 540 720

–1 –1

FIGURE 6.48 (a) (b)

and it reaches its minimum when

u  . . . , 90°, 270°, 630°, . . .
In addition, the sine curve is concave down between u  0° and u  180° and is
concave up from u  180° to u  360° and again in every other cycle.
You should carefully distinguish between the information shown in Figure 6.47
regarding the signs of the sine function in different quadrants and what happens in the
graph of the sine function shown in Figure 6.48(a). The quadrants referred to in Fig-
ure 6.47 are based on a coordinate system with y versus x, and values for the angle u
are measured by rotating the terminal side of the angle. These are not the same quad-
rants shown in Figure 6.48(a) because that graph shows y as a function of u, with u
measured horizontally as it takes on values in the different quadrants. In particular,
angles in the first quadrant in Figure 6.47 correspond to the portion of the u-axis in
Figure 6.48(a) between u  0° and u  90°; the second quadrant in Figure 6.47 cor-
responds to the portion of the u-axis between u  90° and u  180° in Figure
6.48(a); and so on. We have marked these differences in Figures 6.48(a) and (b) with
Roman numerals and corresponding shadings for the different quadrants to help
make the point. Be sure that you understand these subtle differences before going on.

The Graph of the Cosine Function

Figure 6.49 shows the graph of the cosine function y  cos u from u  360° to
u  720°. Use the graph to answer the following questions: Where is the cosine
function increasing? Where is it decreasing? What are its maximum and minimum
values? Where do they occur? Where is the cosine function concave up? Where is it
concave down? Also, be sure that you understand how the information shown in
Figure 6.47 on the sign of the cosine function in the different quadrants relates to
the behavior depicted in the graph of the cosine function.
y

I II III IV
1

θ
–360 –180 180 360 540 720

–1
FIGURE 6.49
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 463

6.4 Relationships among Trigonometric Functions 463

Problems
In Problems 1–12, find the value of each quantity by 31. tan 1500°2 32. sin 11000°2
using only the information in the table. Do not use the 33. Consider the function f 1x2  sin x cos x.
trigonometric function keys on your calculator. (Hint: a. Determine the sign of f 1x2 for x between 0° and
Start by drawing a picture of each angle.) 90°. between 90° and 180°. between 450° and
540°.
b. For what values of x between 0° and 540° is
f 1x 2  0?
U 0 30 45 60
sin U 0 0.5 0.707 0.866 c. Use the results of part (a) to sketch a rough
cos U 1 0.866 0.707 0.5
graph of f 1x 2.
d. Does the function appear to be periodic? If so,
tan U 0 0.577 1 1.732 what is its period?
34. a. Find the missing entries in the table below.
1. sin 225° 2. cos 210° b. Plot the points 1sin u, u2 and connect them with
3. tan 135° 4. sin 150° a smooth curve.
c. This curve is the graph of what function?
5. sin 330° 6. cos 390°
7. cos 315° 8. tan 240°
9. cos 840° 10. sin 1450°2 sin U 1 0.75 0.5 0.25 0 0.25 0.5 0.75 1

11. cos 1240°2 12. tan 1225°2 U

Decide whether each quantity is positive, negative, or
zero without calculating its value. Give a reason for 35. a. Find the missing entries in the table below using
your answer. only your answers to Problem 34.
13. sin 300° 14. cos 450° b. Plot the points 1cos u, u 2 and connect them with
a smooth curve.
15. cos 240° 16. sin 210° c. This curve is the graph of what function?
17. tan 300° 18. tan 450°
19. cos 270° 20. sin 225°
cos U 1 0.75 0.5 0.25 0 0.25 0.5 0.75 1
21. sin 215° 22. cos 320°
23. cos 520° 24. sin 885° U
25. tan 925° 26. sin 1000°
27. cos 1000° 28. tan 1000°
29. sin 1480°2 30. cos 1500°2

6.4 Relationships among Trigonometric Functions

Consider the following values for the sine and cosine of the special angles between
0° and 90°.

U 0 30 45 60 90

1 13
sin U 0 0.5  0.707  0.866 0
12 2
13 1
cos U 1  0.866  0.707 0.5 1
2 12
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 464

464 CHAPTER 6 Introduction to Trigonometry

Comparing the values of the sine and cosine, you will notice that the values asso-
ciated with 0° and 90° are reversed, as are the values for 30° and 60°, so that
sin 30°  cos 60° and sin 60°  cos 30°. This is no coincidence. In general, for
any angle u,
cos u  sin 190°  u 2 and sin u  cos 190°  u2.
Why are these two relationships true? Consider the right triangle shown in Fig-
ure 6.50. Side a, opposite angle u, is the side adjacent to the angle 90°  u. Similar-
ly, side b, adjacent to angle u, is the side opposite angle 90°  u. That is, their roles
are reversed, depending on which angle, u or 90°  u, you consider.

c
b

90° − θ
FIGURE 6.50 a

One reason why the trigonometric functions are so useful is that there are
many interrelationships among them, as the two preceding formulas demon-
strate. They are called identities because they hold for every possible value of the
variable u. But other identities involving relationships between the trigonometric
functions are far more important. Let’s consider again the right triangle shown in
Figure 6.50 and the definitions of the sine and cosine:
a b
sin u  and cos u  .
c c
If we multiply both sides of these equations by c, we obtain
a  c sin u and b  c cos u.
We substitute these expressions into the formula for the tangent function to get
a c sin u
tan u   ,
b c cos u
provided that cos u 0. Therefore, for any such angle u, we have the following
identity.

sin u
tan u  , if cos u 0
cos u

For instance, if u  74°, sin 74°  0.9613, cos 74°  0.2756, and
sin 74°
tan 74°  3.4874   3.488.
cos 74°
If we use more than four digits for sin 74° and cos 74°, the result would be more
accurate.
Next let’s apply the Pythagorean theorem to the right triangle shown in Figure 6.50:
a 2  b 2  c 2.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 465

6.4 Relationships among Trigonometric Functions 465

When we substitute a  c sin u and b  c cos u, we have

1c sin u 2 2  1c cos u 2 2  c2
or
c2 1sin u 2 2  c2 1cos u 2 2  c2.
Dividing both sides by c2 (which is not 0) gives
1sin u 2 2  1cos u 2 2  1,
which holds for any angle u. For convenience, it is customary to write
1sin u2 2  sin2u and 1cos u 2 2  cos2u
and we have the following identity.

The Pythagorean Identity

sin2u  cos2u  1

Check this result on your calculator by using different values for u. Be careful
to enter the expressions as (SIN X)^2 and (COS X)^2. For instance, if u  74°
again, we have
sin2 174°2  cos2 174°2  10.96132 2  10.27562 2  1.000053.
The discrepancy is due to rounding errors. If we use more digits in sin 74° and
cos 74°, the result would be even closer to 1.
Let’s now start with the Pythagorean identity and divide both sides by cos2u.
We then get
sin2u cos2u 1
2  2  ,
cos u cos u cos2u
or, equivalently, because sin u>cos u  tan u,
1
tan2u  1  .
cos2u
This relationship is more commonly written in the following form.

1
1  tan2u 
cos2u

We summarize the definitions and special relationships among the three

trigonometric functions as follows.

opposite
sin u 
hypotenuse
adjacent
cos u 
hypotenuse
opposite
tan u 
adjacent
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 466

466 CHAPTER 6 Introduction to Trigonometry

sin u
tan u 
cos u
sin2u  cos2u  1
1
1  tan2u 
cos2u

We investigate many other relationships between these three functions in Section 8.1.

E XAMPLE 1
Suppose that the SIN and TAN keys on your calculator are broken. You can use the COS
key to find that cos 20°  0.940. Determine the values for sin 20° and tan 20°.
Solution We illustrate three different approaches to solving this problem.
Method 1
Using the Pythagorean relationship,
sin2u  cos2u  1,
we find that
sin2u  1  cos2u
for any angle u. Therefore, when u  20°,
sin2 120°2  1  cos2 120°2  1  10.9402 2  0.1170.
We take the square root of both sides to find
sin 20°  20.1170  0.342.
Further, we have
sin 20° 0.342
tan 20°    0.364.
cos 20° 0.940
Method 2
Figure 6.51 shows that sin 20°  b>c. However, cos 70°  b>c also, and we can use the
“broken” calculator to find cos 70°  0.342. Therefore sin 20°  0.342 also. Knowing
sin 20° and cos 20°, we now can find tan 20°  0.364, as we did in Method 1.

c 70°
b

20°
FIGURE 6.51 a

Method 3
We know that
adjacent
cos 20°  0.940  ,
hypotenuse
so from the triangle shown in Figure 6.52, the ratio a>c must be 0.940. Thus we can as-
sume, for instance, that a  94 and c  100. (There are infinitely many other possibilities;
another is a  0.940 and c  1.) Consequently, using the Pythagorean theorem, we can
find the third side b:
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 467

6.4 Relationships among Trigonometric Functions 467

60°

2a
a

30°

√3a

FIGURE 6.52

b2  c2  a2  1002  942  1164;

b  21164  34.117.
As a result, we have
b 34.117
sin 20°    0.341;
c 100
b 34.117
tan 20°    0.363.
a 94
Both of these values differ slightly from the results in Methods 1 and 2 because of rounding.
◆

E XAMPLE 2
Suppose that the SIN and COS keys on your calculator are broken. Using the TAN key,
you find that tan 25°  0.466. Find the sine and cosine of this angle (a) by using
trigonometric identities and (b) by constructing an appropriate triangle.
Solution

a. We have the relationship

1  tan2 125°2 
1
cos 125°2
2

so that

1  10.4662 2  1.2172 
1
cos 125°2
2 .

Consequently,

cos2 125°2 
1
 0.8216.
1.2172
Taking the positive square root, we have
cos 25°  0.906.
Furthermore, because cos2 125°2  0.8216, we use the Pythagorean identity to find that
sin2 125°2  1  cos2 125°2  0.178.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 468

468 CHAPTER 6 Introduction to Trigonometry

Taking the positive square root, we have

sin 25°  0.422.

b. Because
opposite
tan 25°   0.466
adjacent
we can construct a right triangle in which the length of the side opposite the 25°
angle is 466, say, and the adjacent side is 1000, as shown in Figure 6.53. To find the
hypotenuse H of this triangle, we have
H 2  4662  10002  1,217,156
so that, when we take the positive square root,
H  1103.25.

H=?
466

θ = 25°
1000

FIGURE 6.53

For this triangle, we now have the desired values

opposite 466
sin 25°    0.422
hypotenuse 1103.25
and
adjacent 1000
cos 25°    0.906.
hypotenuse 1103.25

E XAMPLE 3
Simplify the expression sin3x  sin x cos2x by using one of the trigonometric identities.
Solution We first factor out the common factor of sin x to get
sin3x  sin x cos2x  sin x 1sin2x  cos2x 2 .
Using the Pythagorean identity yields
sin3x  sin x cos2x  sin x # 11 2  sin x.
◆
Think About This Verify graphically that the given expression sin3x  sin x cos2x and the final expres-
sion sin x in Example 3 are identically equal for all values of x by graphing both. ❐
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 469

6.5 The Law of Sines and the Law of Cosines 469

Problems
1. Suppose that the COS and TAN keys on your calcu- 8. Suppose that, for a certain angle in the third quad-
lator are broken. You can use your SIN key to find rant, tan u  0.75. Find the cosine and sine of u
that, for some angle u in the first quadrant, algebraically.
sin u  0.3. Determine the values for cos u and 9. Suppose that, for a certain angle in the fourth quad-
tan u. What is the angle u? rant, sin u  0.7. Find the cosine and tangent of
2. Suppose that the SIN and TAN keys on your cal- u algebraically.
culator are broken. You can use your COS key to 10. Simplify the expression sin2x cos x  cos3x by using
find that, for some angle u in the first quadrant, one of the trigonometric identities.
cos u  0.4. Determine the values for sin u and tan u 11. Consider the two equations:
algebraically. What is the angle u? tan x
3. Suppose that, for a certain angle u in the first quad- i.  sin x
cos x
rant, sin u  0.6. Using paper and pencil only, find sin x
the cosine and tangent of u. ii.  cos x
tan x
4. Suppose that, for a certain angle u in the first quad- a. Determine graphically which of these equations
rant, cos u  0.6. Using paper and pencil only, find represents an identity that is true for every
the sine and tangent of u. value of x, except for those points where the de-
5. Suppose that, for a certain angle u in the first quad- nominator is 0, and which is not an identity.
rant, tan u  34 . Using paper and pencil only, find b. Prove algebraically, using trigonometric identi-
the sine and cosine of u. ties, that the identity is indeed true.
6. Suppose that, for a certain angle u in the first c. For the equation that is not an identity, find two
quadrant, tan u  1.2. Find the cosine and sine different values of x that satisfy the equation.
of u algebraically.
7. Suppose that, for a certain angle in the second
quadrant, sin u  0.52. Find the cosine and tangent
of u algebraically.

Exercising Your Algebra Skills

Use appropriate trigonometric identities to simplify 6. cos3x  sin2x cos x
each expression. 7. cos x  tan2x cos x
1. cos x tan x 1
8. tan2u 
2. 11  sin x2 11  sin x2 cos2u
3. 11  cos x2 11  cos x2
9. a1  b a1  b
1 1
4. 1sin u  cos u2 2 cos x cos x
5. 1sin u  cos u2 2

6.5 The Law of Sines and the Law of Cosines

When we first introduced trigonometry, our original development was present-
ed in terms of angles in right triangles. We subsequently extended the defini-
tions of the sine, cosine, and tangent to angles larger than 90°. We now consider
some additional properties of the sine and cosine in any triangle, not just in a
right triangle.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 470

470 CHAPTER 6 Introduction to Trigonometry

The Law of Sines

Consider the triangle ABC shown in Figure 6.54 where the sides opposite the angles
A, B, and C are denoted by a, b, and c, respectively. All three angles are acute; that is,
each is less than 90°. Later we consider the case where one angle is greater than 90°.

c b
y

B D C
FIGURE 6.54 a

Suppose that we drop a perpendicular from the vertex at angle A to point D on

base a. That perpendicular line AD, whose length we call y, produces two right tri-
angles. In triangle ABD we have
y
sin B  so that y  c sin B.
c
Similarly, in triangle ACD we have
y
sin C  so that y  b sin C.
b
These two expressions for y must be equal, so
y  c sin B  b sin C,
and therefore
sin B sin C
 .
b c
However, we could just as easily have drawn a perpendicular from the vertex at
angle B to the opposite side b. In that case, using the same reasoning, we get
sin A sin C
 .
a c
Together, these results yield
sin A sin B sin C
 
a b c

for any triangle with three acute angles.

What about a triangle with an angle greater than 90°? Consider the one shown
in Figure 6.55. We can still drop a perpendicular of length y from the vertex at
angle A to point D on an extension of side a, as shown. This line forms two right
triangles. Clearly, in the large right triangle ACD,
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 471

6.5 The Law of Sines and the Law of Cosines 471

b
y c

φ
FIGURE 6.55 D B a C

y
sin C  so that y  b sin C.
b
To determine the sine of angle B, we use the angle f in the smaller right triangle
ABD. We thus find that
y
sin B  so that y  c sin B.
c
Consequently,
y  c sin B  b sin C,
so that again we have
sin B sin C
 .
b c
We can similarly drop a perpendicular either from the vertex at angle B onto side
b or from the vertex at angle C onto an extension of side c and obtain a similar
relationship involving
sin A
.
a
What we have just proved is called the law of sines.

The Law of Sines

In any triangle,
sin A sin B sin C
 
a b c

The law of sines can be used to find all the remaining sides and angles in any trian-
gle if two sides and one angle are known or one side and two angles are known,
provided that the known combination of sides and angles includes one angle and
the side opposite it.

Think About This Draw a triangle in which two sides and one angle are known and the law of sines
will not apply. Then draw a triangle in which one side and two angles are known
and the law of sines will not apply. ❐
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 472

472 CHAPTER 6 Introduction to Trigonometry

We illustrate use of the law of sines in Example 1.

E XAMPLE 1
The Federal Communications Commission (FCC) is attempting to locate a pirate radio
station by a method called triangulation. The FCC set up two monitoring stations
30 miles apart on an east–west line and took simultaneous readings on the direction of
the radio signal. The westernmost monitor measured the signal as coming from a direc-
tion 42° north of east; the other monitor measured the signal as coming from a point 56°
north of west. Where is the pirate station located?
Solution The information recorded determines the triangle shown in Figure 6.56. The
two monitoring stations are located 30 miles apart at the points A and B. The signal di-
rections determine the angles of 42° and 56°. The pirate is located at point C. Hence the
angle at C must be 180°  42°  56°  82°.
C

82°

b a

42° 56°
west east
A c = 30 B

FIGURE 6.56

We now apply the law of sines to find the lengths of sides a and b. Using angles
A  42° and C  82°, we find
sin A sin C sin 42° sin 82°
 or 
a c a 30
so that
30 sin 42°
a  20.27.
sin 82°
Similarly, to find b we apply the law of sines, using the angles B and C, to get
sin B sin C sin 56° sin 82°
 or 
b c b 30
so that
30 sin 56°
b  25.12.
sin 82°
Therefore the pirate station is located 25.12 miles from station A in a direction of 42° to-
ward the northeast and 20.27 miles from station B in a direction of 56° toward the
northwest. The point C is determined precisely by these two facts.
◆
In Example 1 we used the law of sines when two angles and one side of a trian-
gle are known. The law of sines can also be used when two sides (say, a and b) and
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 473

6.5 The Law of Sines and the Law of Cosines 473

the angle opposite one of these sides (either A or B) are known. However, depend-
ing on the sizes of the two known sides, it is possible to obtain either a unique an-
swer or two distinct configurations for the triangle. This ambiguous case occurs
when we try to find the angle from its sine. Recall that there will be two angles—
one less than 90° and the other greater than 90°—that both have the same sine
value. We ask you to explore possible ambiguous cases in the Problems at the end
of this section.
Another complication may arise when we’re using the law of sines if we know
two sides and the angle opposite one of them. If in the midst of such a set of calcu-
lations, we obtain a sine value greater than 1, it indicates that the values we’re
working with could not have come from a real triangle. Again, you will encounter
such a case in the Problems at the end of this section.

The Law of Cosines

There is one useful relationship involving the cosine function that relates the three
sides of any triangle and any one of its three angles. Consider the triangle shown in
Figure 6.57 with sides a, b, and c, where the angle opposite side c is C. The sides and
angle are related by the law of cosines.

a
c
Law of Cosines
In any triangle,
C
b c2  a2  b2  2ab cos C

FIGURE 6.57

Note that the triangle need not be a right triangle; the law of cosines applies to any
triangle. The law of cosines allows us to determine (1) the length of the side oppo-
site a known angle if the other two sides are known, or (2) any angle if the three
sides of the triangle are known.
We prove this formula for the case where the triangle has three acute angles; a
similar argument applies if one of the angles is greater than 90°. Also, to make
things easier we assume that one of the vertices is at the origin and that one of the
sides of the triangle lies on the x-axis, as shown in Figure 6.58. Note that the coor-
dinates of the point P are 1a, 0 2 and that the coordinates of the point Q are at
x  b cos C and y  b sin C. As a result, the length of side c is just the distance
from P to Q and we can find it by using the usual formula for the distance between
two points (see Appendix A5):

Q (b cosC, b sinC)

b c
y = b sinC

C B (a, 0)
FIGURE 6.58 0 x = b cosC a P
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 474

474 CHAPTER 6 Introduction to Trigonometry

c  distance from P to Q  21b cos C  a2 2  1b sin C2 2.

We square both sides to eliminate the square root and obtain
c2  1b cos C  a2 2  1b sin C2 2
 1b2cos2 C  2ab cos C  a2 2  b2sin2 C
 1b2cos2 C  b2sin2 C2  a2  2ab cos C
 b2 1cos2 C  sin2 C2  a2  2ab cos C
 b2  a2  2ab cos C Pythagorean identity
 a2  b2  2ab cos C
If we have a right triangle in which angle C  90°, then cos C  0 and the law of
cosines reduces to the Pythagorean theorem c 2  a2  b2.
We use the law of cosines in Examples 2–4.

E XAMPLE 2
Let ABC be a triangle with sides a  5 and b  7, and let the angle C between the two
sides a and b be 60°.
a. Find the third side c. A
b. Find the other two angles A and B.
c=? b=7
Solution

a. We begin with Figure 6.59. Using the law of cosines, we have

B C = 60°
c2  a2  b2  2ab cos C
 52  72  215 2 17 2 cos 60°
a=5

 25  49  701 12 2  39. FIGURE 6.59

Thus the third side is c  139  6.245.

b. To find the angle A, we use the law of sines:
sin A sin C sin A sin 60°
 or  .
a c 5 6.245
Therefore
5 sin 60°
sin A   0.6934,
6.245
so that
A  arcsin10.6934 2  43.9°.

Consequently,
B  180°  60°  43.9°  76.1°

◆
E XAMPLE 3
In a standard baseball infield, the four bases are at the corners of a square whose sides are
90 feet in length. The pitcher’s mound is 60 feet, 6 inches, or 60.5 feet, from home plate
on a line through second base, as illustrated in Figure 6.60. The distance from the pitch-
er’s mound to second base is about 67 feet.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 475

6.5 The Law of Sines and the Law of Cosines 475

ft

90
90

ft
d=?
θ
60.5 ft
ft

90
45° 90

ft
Home
plate

FIGURE 6.60

a. How far is the pitcher’s mound from first base?

b. What is the angle at the pitcher’s mound between home plate and first base?

Solution

a. Note that the angle at home plate between the mound and first base is 45°, so we know
the angle opposite the unknown length d. Therefore, using the law of cosines, we have

d 2  60.52  902  2160.52 190 2cos 45°

 3660.25  8100  2160.5 2 190 2 cos 45°  4059.86.

When we take the square root of both sides, we find that

d  63.72 feet.
b. Note that the angle u at the pitcher’s mound between home plate and first base must
be more than 90°. We use the law of sines:

sin u sin 45°


90 d
so that
90 sin 45°
sin u   0.9987.
63.72

Therefore, using the inverse sine function, we get u  arcsin10.99872  87.08°. Be-
cause this is less than 90°, it must be the angle at the pitcher’s mound between first base
and second base. The desired angle is the supplement, 180°  87.08°  92. 92°.
◆

In Section 6.2, we considered some physical examples in which a force or a ve-

locity could be broken into two parts, one horizontal and the other vertical. At the
time, we were limited to particularly simple examples where, for instance, a boat
was out on a still lake with the wind blowing, but there was no mention of a cur-
rent. We now look at a more complicated situation.
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 476

476 CHAPTER 6 Introduction to Trigonometry

E XAMPLE 4
The pilot of a small plane is flying due east at its top speed of 200 mph. The wind is blowing
out of the northwest at a speed of 40 mph. The wind pushes the plane in a direction south
of east and increases its airspeed to more than the 200 mph, as illustrated in Figure 6.61.

a. What is the actual airspeed of the plane due to its own engines and the wind?
b. What is the actual direction that the plane flies?
P = 200 mph P = 200 mph
θ =? θ =? 45°
W = 40 mph 135° W = 40 mph
Actual = ? Actual = ?

FIGURE 6.61 FIGURE 6.62

Solution

a. We first consider the speeds. The plane itself contributes a horizontal airspeed of
P  200 mph. The wind comes from the northwest at W  40 mph, so the associat-
ed angle inside the triangle in Figure 6.62 is 180°  45°  135°. The actual airspeed
s of the plane is the length of the remaining side in the triangle, and we find it by
using the law of cosines:
s2  P 2  W 2  2PW cos 135°
 2002  402  21200 2 140 2 10.7072  52,912.
Taking the positive square root yields
s  230.026,
or about 230 mph.
b. To find the angle u in the triangle, we use the law of sines:
sin u sin 135°
 ,
40 230
so that
40 sin 135°
sin u   0.1230.
230
Therefore
u  arcsin10.12302  7.07°,
or the plane is actually flying about 7° south of east.
◆

Problems
For Problems 1–6, refer to the notation for the sides 1. A  26°, B  63°, b  12
and angles in the accompanying figure. Use the infor- 2. A  47°, C  72°, c  60
mation given to find all other parts of the triangle. 3. A  35°, B  65°, c  24
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 477

6.5 The Law of Sines and the Law of Cosines 477

A 11. Find the angle opposite the side of a triangle whose

length is 6 if the lengths of the other two sides are
11 and 8, as shown in the accompanying figure.

c b

11
6
B a C

4. A  40°, a  10, b  6 (Hint: Is it possible to have

8
two different values for B?)
5. A  40°, a  10, b  12 (Hint: Is it possible to
have two different values for B?) 12. A TV camera is positioned 100 feet behind home
plate. The center fielder is standing 300 feet from
6. A  40°, a  10, b  18
home plate and directly on the line from home
7. Two ships at sea are 50 miles apart on a plate through second base. The batter hits a long fly
north–south line when they both receive an SOS ball toward right-center field that the center fielder
signal from a third ship in trouble. One ship re- catches against the wall, 380 feet from home plate.
ceives the SOS from a direction of 41° north of If the TV camera had to pan through an angle of 8°
east. The other ship receives the signal from a di- in following the center fielder from the point where
rection of 54° south of east. Where is the third he was standing to the point where he made the
ship? spectacular catch, how far did he have to run?
8. You want to find the distance across a fast-flowing 13. Using a map of the United States (which ignores
river. You pick two large trees, at points A and B, the fact that the earth is round), New York City is
that are 35 feet apart along the edge of the river on 1851 miles from Denver and about 10° north of
your side. You then spot another tree on the oppo- east from Denver. New Orleans is 1282 miles
site side of the river at point C. The angle CAB at from Denver and about 36° south of east from
point A is 43°; the angle CBA at point B is 52°. Find Denver. Estimate the distance from New York
the distance across the river. City to New Orleans.
9. Problems 4–6 involved three cases in which the law
of sines works very differently because of the rela-
tive sizes of sides a and b. Based on those results, ex-
plain the following statements.
1851 miles
a. Given a value for angle A, there will always be 10° New York
Denver City
one triangle whenever b  a. 36°
b. There will be two different possible triangles
whenever b is somewhat larger than a. 1282 miles

c. There will be no triangle whenever b is much New Orleans

larger than a.
10. Find the angle C if a  7, b  11, and c  5, as
shown in the accompanying figure. 14. Chicago is 695 miles from Atlanta, and Seattle is
2756 miles from Atlanta. The same map of the Unit-
ed States as in Problem 13 shows that Chicago lies at
an angle of about 65° north of west from Atlanta
a=7 c=5
and that Seattle lies at an angle of about 29° north of
west from Atlanta. Estimate the distance from
C
b = 11 Chicago to Seattle.
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 478

478 CHAPTER 6 Introduction to Trigonometry

15. A 40 meter vertical tower is to be built and supported carries a protractor.) From Al’s position, the angle
by several guy wires anchored to the ground 25 me- of elevation to the balloon is 28°; from Bob’s posi-
ters from the base of the tower on flat land. Find the tion, the angle of elevation is 35°.
length of the guy wires and the angle they make with a. How high is the balloon?
the ground. b. How far is each of them from where the balloon
is moored?
19. A straight tunnel passes through a mountain. An
observer has a clear view of the two ends of the tun-
nel. The distance from her position to the tunnel
40 m entrance toward her left is 520 meters, and the dis-
tance to the entrance toward her right is 440 me-
ters. If the angle subtended by the two tunnel
θ entrances is 39°, how long is the tunnel?
25 m

16. The 40 meter tower in Problem 15 is to be built on

the side of a hill that slopes upward at an angle of
10° from the horizontal. One guy wire will be posi-
tioned directly uphill from the tower and another
will be positioned directly downhill from the tower. 520 ft 440 ft
Each guy wire will be anchored 25 meters from the
base of the tower. Find the lengths of the two guy 39°
wires and the angles they make with the sloping
ground.

20. Meryl and Bernice are walking along a straight

beach when they observe a small island in the dis-
40 m tance and wonder if they can swim out to it. To esti-
mate the distance, they separate and walk 400
meters apart. From Meryl’s perspective, the angle to
the island is 27° and from Bernice’s perspective the
25 m 25 m
10°
angle is 39°. How far is the island from the shore?

17. A communications satellite passes directly over

Phoenix and Los Angeles, which are 340 miles
apart. At some point in its orbit when the satellite is
27° 39°
between Phoenix and Los Angeles, its angle of ele-
400 m
vation from Phoenix is 52° and its angle of eleva-
tion from Los Angles is 72°.
a. How far is the satellite from Phoenix at that moment? Problems 21 and 22 refer to Figure 6.57 in the text.
b. How high is the satellite above the Earth? 21. Find c if a  5, b  3, and C  20°.
18. Al and Bob are driving toward a moored hot air 22. Find c if a  7, b  4, and C  25°.
balloon from opposite sides of the balloon and are
in contact via cell phones. When they are 4 miles
apart, they both take sightings on the balloon. (Of
course, everyone who is about to take a balloon ride
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 479

Review Problems 479

Chapter Summary

In this chapter we introduced some of the fundamental ideas and applications of

trigonometry as they apply to right triangles. In particular, we discussed:
◆ The definition of the tangent ratio in terms of the sides of a right triangle.
◆ How to use the tangent ratio to solve problems involving right triangles.
◆ The graph of the tangent function between 0° and 90°.
◆ The definition of the sine and cosine ratios in terms of the sides of a right
triangle.
◆ How to use the sine and cosine to solve problems involving right triangles.
◆ The graphs of the sine and cosine functions between 0° and 90°.
◆ How to extend the sine, cosine, and tangent functions to angles beyond 0° to
90°.
◆ The fundamental identities that relate the sine, the cosine, and the tangent
functions.
◆ How to use the law of sines and the law of cosines to solve various types of
problems.

Review Problems
1. A TV cameraman is standing on a platform 75 feet b. If u  25°, how far has the runner gone since
from a straight portion of a race track and is focus- she passed the point closest to the cameraman?
ing on the runner in the lead as she runs from left c. When the runner has gone 150 feet past the point
to right. closest to the cameraman, through what interval
of angles has the cameraman panned while fo-
cusing on her?
3. The next assignment for the TV cameraman in
75 feet Problems 1 and 2 is to videotape the liftoff of the
θ d
space shuttle. The cameraman is positioned at
ground level 500 meters from the launch pad.
a. Write a formula for the height y of the shuttle as
a function of the angle of inclination a.
a. Write a formula for the distance d from the cam- b. Find the height of the shuttle when a  20°.
era to the runner as a function of the angle u, as c. Find the height of the shuttle when a  40°.
shown in the accompanying diagram. d. Find the angle of inclination when the shuttle is
b. Suppose that the maximum distance for which at a height of 2000 meters.
the camera lens can get a good image is 240 feet. 4. A swimming pool is 60 feet long and 25 feet wide. It
Through what interval of angles can the camera- is 3 feet deep at the shallow end and 12 feet deep at
man pan while focusing on the runner? the deep end.
c. What might be appropriate values for the do-
main of this function? 3 ft
25 ft
2. The camera in Problem 1 is again focused on the 12 ft
lead runner in the race. 60 ft.
a. Write a formula for the distance that the runner
covers from the instant that she passes closest to
the cameraman as a function of the angle u.
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 480

480 CHAPTER 6 Introduction to Trigonometry

a. Find the angle of depression of the bottom of a. How tall is the building?
the pool. b. How high is the window in the hotel?
b. Find the equation of the line along the bottom of 9. The Earth is 93 million miles from the sun, and the
the pool extending from the shallow end to the moon is 240,000 miles from Earth. When the moon
deep end along one of the long sides of the pool. is exactly half full, the Earth, the moon, and the sun
c. Find the equation of the line along the bottom of form a right triangle with the right angle at the
the pool that extends from one corner to the op- moon. Calculate, correct to two decimal places, the
posite corner of the pool. angle at the Earth in this triangle.
5. A salami is 4 inches in diameter. However, the man
in the deli department slices it at an angle of 28° so Sun Moon
that each slice comes out oval for a fancier presen-
tation. What is the longest length of each slice of the
salami? 240,000 mi.
93,000,000 mi.
6. A piece of metal 96 inches long by 36 inches wide is θ
to be made into a watering trough by bending up
12 inches of the metal along each long side, as shown Earth
in the accompanying figure.
10. To calculate the height of a mountain above a level
12 in. plain, two measurements are necessary. Suppose
12 in.
that, from a certain point on the plain, the angle of
12 in. 96 in. elevation to the top of a mountain is a  34°. The
observer then moves 1000 meters closer to the
a. If the two metal sides are bent up at angles of mountain, takes a second reading, and gets an angle
35°, how deep is the trough? of elevation of b  37°. How tall is the mountain?
b. If the two metal sides are bent up at angles of
55°, how deep is the trough? 11. As you sit in class waiting for the end of the period,
you notice that the length of the minute hand on
7. The shape of the trough in Problem 6 has a trape- the wall clock is 10 inches.
zoidal cross section because the top and bottom are
parallel. The volume of water that the trough can a. How far vertically does the point of the minute
hold is then its length, 96 inches, times its cross- hand rise from 45 minutes after the hour until
sectional area, 21 1b1  b2 2h, where h is the height of 50 minutes after the hour?
the trough and b1 and b2 are the horizontal lengths b. Explain why the point of the minute hand can-
of the top and bottom of each cross section. not rise by the same amount from 50 minutes
after the hour until 55 minutes after the hour.
a. What volume of water can the trough hold if the
c. During what other 5-minute time intervals over
two edges are bent up at angles of 35°?
the course of an hour does the minute hand ei-
b. What volume of water can the trough hold if the
ther rise or fall vertically by that same amount,
two edges are bent up at angles of 55°?
as in part (a)?
c. Write a formula for the total volume of water
that the trough can hold as a function of the 12. The tangent of some angle in the first quadrant is
angle u at which the two sides are bent up. 1.20.
d. Use your function grapher to estimate the angle a. Find the sine and cosine of that angle, using only
that produces a trough that will hold the maxi- appropriate trigonometric identities.
mum amount of water. b. Find the sine and cosine of that angle by con-
8. A tall building stands across the street from a structing an appropriate triangle.
hotel—a distance of 220 feet. From one of the hotel 13. Because of a storm, a tree is inclined at an angle of
windows, a guest in the hotel observes that the 10° from the vertical. From a point 70 feet from the
angle of inclination to the roof of the building is 36° base of the tree, the angle of elevation to the top of
and that the angle of depression to the base of the the tree is 26°, with the tree leaning toward the ob-
building is 23°. server, as shown in the figure on the next page.
Gordon.3896.06.pgs 11/21/03 10:47 AM Page 481

Review Problems 481

b. At the moment the quarterback throws the ball,

what is the angle between the line of scrimmage
and the line from the quarterback to the receiv-
10°
er’s original position on the line of scrimmage?
26° c. Find the distance that the ball travels from the
70 ft. quarterback to the receiver.
d. To catch the ball straight on, the receiver has to
a. Find the height of the tree. turn toward the quarterback. Through what
b. Find how high the top of the tree is above angle u must the receiver turn in order to face di-
ground level. rectly toward the quarterback?
c. How do your answers to parts (a) and (b) 16. A motorboat leaves its dock and motors 6 miles due
change if the tree is leaning away from instead of north. It then changes course, heading northeaster-
toward the observer? ly at an angle of 58° east of north for 14 miles. At
14. Two forest rangers are in observation towers that point, the pilot decides to turn back and head
16 miles apart on an east–west line, and each spots a directly to the dock.
fire. From the point of view of the ranger in tower a. How far is it from the turnaround point back to
A, the direction of the fire is 41° north of east. From the dock?
the point of view of the ranger in tower B, the fire is b. Through what angle does the motorboat have
35° north of west. Find the distance to the fire from to turn in order to be pointed directly back to
each tower. the dock?
15. In a football passing play, the wide receiver lines up c. If the motorboat moves at a roughly constant
12 yards to the right of the quarterback on the line speed of 18 mph, how long will the return trip
of scrimmage. The quarterback drops straight back take? How long does the entire outing take?
5 yards before throwing the football. The wide re- d. The motorboat gets 6 mpg. If the refueling station
ceiver runs straight down the field 30 yards before at the dock charges $2.85 per gallon, how much
turning to catch the pass. did the entire outing cost?
17. A steep, snow-covered mountain rises 2700 feet above
the surrounding plain and rises at an angle of 68° to
the horizontal. A ski lift is to be built from a point 750
θ feet from the base of the mountain to the summit.
a. What will be the angle of inclination of the cable
for the ski lift?
b. What will be the length of the cable?
18. The sides of a triangle have lengths 8 cm, 14 cm,
30 yards
and 19 cm.

14 8

12 yards θ =?
5 yards 19

a. What is the angle u?

b. What is the height of the triangle?
a. At the moment the quarterback throws the ball, c. What is the area of the triangle?
what is the distance from the quarterback to the d. Write a formula for the area of any triangle given
receiver’s original position on the line of scrim- the lengths a, b, and c of the three sides and one
mage just before the play started? of the angles u.
Gordon.3896.06.pgs 4/28/03 1:33 PM Page 482
Gord.3896.07.pgs 4/28/03 11:55 AM Page 483

7
Modeling Periodic
Behavior
7.1 Introduction to the Sine and Cosine Functions
One of the most common behavior patterns in nature is a periodic oscillatory effect—a
pattern that repeats over and over. For instance, think about how the ocean level
varies at a beach between low tide and high tide approximately every 12 hours. If low
tide occurs at midnight, high tide will occur at about 6 A.M., low tide will occur again
at about noon, and so on indefinitely. This periodic oscillatory behavior is shown in
Figure 7.1. Recall that the word periodic refers to the fact that this phenomenon re-
peats indefinitely and that the period is the time needed to complete one full cycle. If
it takes 12 hours to complete a full cycle, the period is 12 hours.
Similarly, consider the number of hours of daylight each day in a particular lo-
cation. The minimum number of hours of daylight occurs on the winter solstice,
December 21, the “shortest” day of the year. The number of hours of daylight in-
creases slowly until the maximum daylight occurs on the summer solstice, June 21,
the “longest” day, and then decreases to the same minimum the following Decem-
ber 21. This oscillatory behavior repeats year after year. For instance, suppose that,
at some location, there are 10 hours of daylight on the shortest day of the year and
14 hours of daylight on the longest day. The number of hours of daylight over the
course of several years can be represented by the graph shown in Figure 7.2, which
has the same shape as that in Figure 7.1.

H
14
12
Hours of daylight

10
8
6
4
2

t 0 150 300 450 600 750

6 A.M. noon 6 P.M. Days

FIGURE 7.1 FIGURE 7.2

483
Gord.3896.07.pgs 4/28/03 11:55 AM Page 484

484 CHAPTER 7 Modeling Periodic Behavior

In this chapter, we consider how to model such periodic phenomena by intro-

ducing some new functions that have this type of periodic, oscillatory behavior. We
can then use the ideas on stretching and shifting functions from Section 4.7 to create
two families of functions that are used to model any such periodic phenomenon.
To begin, imagine a clock on a wall that is running backward, so that the hands
move counterclockwise, as shown in Figure 7.3. (Unfortunately, the people who orig-
inally developed the mathematics relating to periodic, oscillatory behavior chose
counterclockwise as their convention and we’re stuck with it). In particular, picture
the motion of the arrowhead on the minute hand. Let the horizontal axis be the line
through the 3 and 9 positions on the clock. Suppose that the minute hand is 5 inches
long and that we start the process at the instant the minute hand is pointing straight
up. Every 60 minutes, the point of the minute hand moves from a maximum height of
5 inches above the center (when it is pointing straight up) to a minimum height of
5 inches below the center (when it is pointing straight down) and then back up to-
ward to a maximum height of 5 inches again—and it repeats this cycle indefinitely.
Thus the height of the arrowhead on the minute hand, as a function of time t, re-
peatedly traces a path that is periodic and oscillatory. The arrowhead traces the
path shown in Figure 7.4 over the first 3 hours, or 180 minutes—it oscillates be-
tween 5 and 5 every 60 minutes. This type of function is just what we need to
model periodic phenomena.

5
Height of arrowhead

t
0 60 120 180

–5

FIGURE 7.3 FIGURE 7.4 Minutes

To develop this process more formally, we use the unit circle—a circle with ra-
dius 1 centered at the origin—as shown on the left in Figure 7.5. A point P with co-
ordinates 1x, y2 lies on this circle if
x 2  y2  1.

y y

D D′
C 1
E C′
E′
B
B′
1
y
F θ A A′ F′
x θ
x
(1, 0)
G G′
I I′
–1
FIGURE 7.5 H H′
Gord.3896.07.pgs 4/28/03 11:55 AM Page 485

7.1 Introduction to the Sine and Cosine Functions 485

This point P corresponds to an angle u measured counterclockwise from the posi-

tive x-axis. For instance, when u  0°, P is on the positive x-axis; when u  90°, P
is on the positive y-axis. We consider separately the horizontal distance x to the left
or right of the y-axis and the vertical height y above or below the x-axis. Each of
these quantities is actually a function of u, the angle at the center of the circle.
Let’s start with the height y. In the triangle at the left in Figure 7.5,
opposite y
sin u    y,
hypotenuse 1
so the sine of any angle u precisely equals the y-coordinate of the corresponding
point on the unit circle. For instance, consider the point A at 11, 0 2 at the extreme
right of the circle as the starting point; it has height y  0 and an angle of inclina-
tion u  0, so y  sin 0  0. We set up a second coordinate system where y is a
function of u, as shown in the graph on the right in Figure 7.5. When we think of y
as a function of u, the pair 1u, y2  10, 02 corresponds to the point A at the origin
in the graph on the right.
Be sure that you distinguish carefully between the coordinate systems in the
two diagrams—the circle on the left, where we think of the height y as a function of
the horizontal distance x, and the associated graph being created on the right
showing the height y as a function of the angle u. In the circle, x is measured hori-
zontally, whereas in the graph u is measured horizontally. The position of any
point—say, B—on the circle can be specified either (1) in terms of its x- and
y-coordinates or (2) in terms of its angle of inclination u and its vertical distance y
above or below the horizontal axis of the circle. It is these 1u, y2 pairs that are
graphed on the right.

The Sine Function

We know that there are 360° in a circle, so we can trace the complete circle as u
runs from 0° to 360°. Also, we can continue tracing the circle over and over as u
increases beyond 360°; in fact, every 360° we complete another full revolution
around the circle.
Let’s consider all the points P on the circle. We labeled several specific points as
A, B, C, D, E, . . . , I in Figure 7.5. Each point corresponds to a particular angle u,
and we plot the heights y corresponding to these angles u in the graph at the right.
For instance, the angle u corresponding to point B is 30° and its height is 12 , or 0.5.
As point P traces the circle, starting at point A and passing through the points B, C,
D, E, . . . , I, . . . , the height y on the circle rises from 0 (at A) to a maximum of 1 (at
D when u  90°), then decreases past 0 (at F when u  180°) to a minimum
height of 1 (at H when u  270°), and then back up to 0 as P finally returns to
the starting point A, having gone through a full 360°. This motion of P continues
as P traces the circle again and again, and the identical pattern of heights recurs re-
peatedly, every 360°. The oscillatory pattern shown in the right-hand graph is peri-
odic; it repeats forever with a period of 360° and is part of the graph of the sine
function, f 1u 2  sin u, for any angle u. It is this periodic, oscillatory effect that we
need to model periodic phenomena.

Think About This You can observe the development of the sine function dynamically by using your
graphing calculator. Set the mode for radians (we discuss this topic shortly), for
parametric graphing (we discuss this topic in Chapter 9), and for simultaneous
plotting. The independent variable used is now typically t instead of x. See
Gord.3896.07.pgs 4/28/03 11:55 AM Page 486

486 CHAPTER 7 Modeling Periodic Behavior

the instructions for your particular calculator if necessary. Go to the Y= menu

and enter
X1  cos t
Y1  sin t
X2  t
Y2  sin t

For the viewing window, set t between 0 and 7, t  0.1, set x between 1.5 and 7,
and set y between 2 and 2. When the graphs are drawn, the circle (somewhat flat-
tened because of the screen dimensions) and the sine curve are produced simulta-
neously. Watch how the heights of points on the circle precisely match the heights
on the sine curve for any angle t. You may also want to trace this behavior with
your fingers on the graphs in Figure 7.5. ❐

We summarize these ideas about the sine function as follows.

The sine function y

y  sin u 1

represents the height that a point

on the unit circle is above or below
the horizontal axis as a function of
the angle u. The graph of the sine θ
90° 180° 270° 360°
function is shown at the right.

–1

Again, note that the graph of the sine function oscillates between a maximum
height of 1 and a minimum height of 1. Also, the basic shape repeats every 360°,
so the behavior pattern you see from 0° to 360° occurs again from u  360° to
u  720°, again from 720° to 1080°, and so on. Similarly, the same pattern occurs
between u  360° and 0°, between u  720° and 360°, and so on. Thus the
sine function is a periodic function and its period is 360°, as shown in Figure 7.6.

The graph of the sine function passes through the origin and oscillates be-
tween 1 and 1 every 360°.

In addition, the sine curve reaches its maximum height of 1 at u  90° and
again at u  450° 1 90°  360°2 , 810° 1 90°  2  360°2, . . . , as well as at
u  270° 1 90°  360°2 , 630° 1 90°  2  360°2, . . . Similarly, the sine
curve reaches its minimum height of 1 at u  270°, 630°, . . . , and at u 90°,
450°, 810°, . . . Note also that the sine curve crosses the horizontal axis at the
origin where u  0°, again at u  180° (corresponding to the extreme left-hand
point on the unit circle), yet again at u  360°, and so on indefinitely. In fact, the
sine function has zeros at every integer multiple of 180°.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 487

7.1 Introduction to the Sine and Cosine Functions 487

x
–720° –360° 0° 360° 720° 1080°

–1
FIGURE 7.6

The Cosine Function

The sine function reflects the vertical height y above or below the horizontal axis in the
unit circle. Next, we consider the horizontal distance x from the vertical axis of the unit
circle to points P at 1x, y2 on the circle, as shown on the left in Figure 7.7. We now treat
x as a function of u. Thus, in the graph of 1u, x 2 on the right in Figure 7.7, the angle u
is measured along the horizontal axis and the distance x is measured vertically.

y x

D 1 A′
C B′
E I′
B C′
1
y
F θ A D′ H′
x θ
x
(1, 0)
G
I E′
G′
–1
FIGURE 7.7 H F′

The initial point A at 11, 02 on the circle lies at a distance of 1 to the right of
the vertical axis, so the corresponding point A on the graph at the right has a
height of x  1. The succeeding points B and C are closer to the vertical axis of
the circle and so, because the x-values are smaller than 1, the corresponding
heights on the graph are smaller. The point D is on the vertical axis, so its distance
from the vertical axis is 0. The points E and F are to the left of the vertical axis of
the circle, so the corresponding heights on the graph are negative. In fact, F is lo-
cated where u  180° at the extreme left point of the circle at a distance of 1
from the vertical axis, so the corresponding point F on the graph is a minimum.
As the angle u continues to increase, the points on the circle approach the ver-
tical axis from the left, and the corresponding points on the graph now rise to-
ward 0. Eventually, the tracing point P on the circle passes the vertical axis at H
and approaches the initial point A where u  360°. The horizontal distance that P
is from the axis changes from negative, to 0, to positive and approaches the dis-
tance 1 to the right of the vertical axis, which is where we started. Simultaneous-
ly, the graph on the right crosses the u-axis and rises to its initial starting height
Gord.3896.07.pgs 4/28/03 11:55 AM Page 488

488 CHAPTER 7 Modeling Periodic Behavior

of 1. Allowing u to continue beyond u  360° we see that the previous pattern re-
peats exactly and indefinitely.
The graph shown on the right in Figure 7.7 is also a periodic, oscillatory curve.
This curve, which corresponds to the horizontal distances from the vertical axis of
the unit circle to points on the circle is the graph of the cosine function,
g1u2  cos u. The cosine function, like the sine function, is periodic and repeats
every 360°, so its period is also 360°. The maximum value of the cosine function is
1, which occurs at u  0°, 360°, 720°, . . . , as well as at u  360°, 720°, . . . The
minimum value of the cosine function is 1, which occurs at u  180°,
540°, . . . The cosine function has zeros when u  90°, 270°, 450°, . . .

The cosine curves passes through the point 10, 1 2 and oscillates between 1
and 1 every 360°.

We summarize these ideas about the cosine function as follows.

The cosine function, y

y  cos u 1

represents the horizontal distance

that a point on the unit circle is to
the right or left of the vertical axis
θ
as a function of the angle u. The 0° 90° 180° 270° 360°
graph of the cosine function is
shown at the right.

–1

You may find the preceding construction of the cosine curve somewhat easier
to visualize by using the following trick. Rotate the circle shown in Figure 7.7
through an angle of 90° counterclockwise, as shown in Figure 7.8. Each horizontal

A′

x
A
y

FIGURE 7.8
Gord.3896.07.pgs 4/28/03 11:55 AM Page 489

7.1 Introduction to the Sine and Cosine Functions 489

distance x is then transformed into an equivalent “height” above or below the new
horizontal axis. These “heights” produce the heights of the points on the cosine
curve shown in the graph on the right in Figure 7.7.
Because of the way that the sine and cosine functions can be defined in terms
of the unit circle, they are sometimes called circular functions.
Figure 7.9 shows both the sine and cosine graphs from u  0° to u  540°.
Clearly, these two functions are closely related. Both have the same shape and each
can be thought of as arising from the other by an appropriate horizontal shift. If we
shift the sine curve to the left by 90°, we get the cosine curve, so
cos u  sin1u  90°2.
Alternatively, if we shift the cosine curve to the right by 90°, we get the sine curve, so
sin u  cos1u  90°2.

y = sin θ y = cos θ
1

θ
0° 90° 180° 270° 360° 450° 540°

–1
FIGURE 7.9

Moreover, the unit-circle definition suggests what is perhaps the most impor-
tant relationship between the two functions. Figure 7.10 shows that the vertical
height y to a point 1x, y2 on the unit circle equals sin u. Similarly, the horizontal
distance x from the vertical axis to the same point equals cos u. Because x  cos u
and y  sin u must satisfy the equation of the unit circle,
x 2  y2  1,
it follows that
1cos u 2 2  1sin u 2 2  1.

(x, y)
y = sin θ

x
x = cos θ

FIGURE 7.10
Gord.3896.07.pgs 4/28/03 11:55 AM Page 490

490 CHAPTER 7 Modeling Periodic Behavior

Recall that we write

1cos u2 2 as cos2 u and 1sin u 2 2 as sin2u.
Note that this is another way to prove the Pythagorean identity that we presented in
Section 6.4.

The Pythagorean identity is

sin2u  cos2 u  1
for any angle u.

Think About This Use your function grapher to graph the function f 1x 2  sin2x  cos2x for any in-
terval of x values. What does it look like? [You will likely have to enter the function
as (sin x)^2+(cos x)^2.] ❐

Radian Measure
Because we want to use the sine and cosine functions to model phenomena that are
periodic over time, such as the heights of tides or the number of hours of daylight,
we need a function of time t rather than a function of an angle u. Therefore we
need a way to avoid angles measured in degrees in our definitions of these func-
tions. To do so, we introduce an alternative unit, called the radian, for measuring
an angle. In the circle of radius 1 shown in Figure 7.11, we begin on the horizontal
axis at the point A at 11, 0 2. We move counterclockwise around this circle and
measure off a distance equal to the radius, or 1. This distance produces an angle a
whose size is defined as one radian. In degrees, this angle is approximately 57°, as
we show shortly.

1 1
α = 1 rad A
x
1 (1, 0)

FIGURE 7.11

We next develop a way to convert between radians and degrees. The length of
the arc that defines one radian equals the radius of the unit circle. Because r  1,
the total circumference of the circle is 2pr  2p. Moreover, the angle a represents
a fraction of the full 360° in the circle. As a result, we can set up the proportion
Fraction of the total angle Fraction of the total circumference

Total angle Total circumference
Gord.3896.07.pgs 4/28/03 11:55 AM Page 491

7.1 Introduction to the Sine and Cosine Functions 491

or
1 radian 1
 .
360° 2p
Cross-multiplying, we get
2p radians  360°
or
p radians  180°
Alternatively, dividing both sides by p gives
180 °
1 radian  a b  57.29578°,
p
or about 57.3°. Furthermore, because p radians  180°, we can divide both sides
by 180 to get
p
1°  radians.
180
If we perform the same construction in any circle with radius r—that is, if we
measure an arc whose length equals the radius r—the corresponding angle would
be the same 1 radian, or about 57.3°. Thus an angle measured in radians is the
same no matter what the size of the circle. More important, radians are not tied di-
rectly to angles the way degrees are. Using radians, we can consider any variable
and apply the sine and cosine functions to it. Thus we can use a variable represent-
ing time, height, or any other desired quantity as the independent variable with ei-
ther the sine or the cosine function.

E XAMPLE
Use the fact that 180°  p radians to obtain the radian measure for the common angles
90°, 60°, 45°, and 30°.
Solution If we divide 180° by 2, we get
180° p p
90°   radians  .
2 2 2
Similarly,
180° p p
60°   radians  ;
3 3 3
180° p p
45°   radians  ;
4 4 4
180° p p
30°   radians  .
6 6 6
◆
To summarize, we have the following relationships.

180°  p radians
In particular,
p p p p
30°  , 45°  , 60°  , and 90°  .
6 4 3 2
Gord.3896.07.pgs 4/28/03 11:55 AM Page 492

492 CHAPTER 7 Modeling Periodic Behavior

These results occur often in applications of the trigonometric functions, and you
need to know them.
For these standard, or special, angles, we have the following values for the sine
and cosine functions (which we derived in Chapter 6).

23
sin 30°  sin a b  0.5  cos 30°  cos a b 
p 1 p
 0.866
6 2 6 2
22 22
sin 45°  sin a b  cos 45°  cos a b 
p p
 0.707  0.707
4 2 4 2
23
sin 60°  sin a b  cos 60°  cos a b   0.5
p p 1
 0.866
3 2 3 2

sin 90°  sin a b  1 cos 90°  cos a b  0

p p
2 2

Be sure that you know how to use your calculator to obtain the value for the
sine or cosine of any argument, both in degrees and radians. We strongly recom-
mend that you permanently set your calculator mode to Radians; we work with
radians almost exclusively from this point on.
Note that radians are not always given in terms of p. For instance, we might
have x  0.5 radians or x  2 radians or x  4.27 radians.

The Behavior of the Sine and Cosine Functions

Finally, let’s consider the important aspects of the behavior of the sine and cosine
functions. In general, we want to answer several questions for any function.
1. Where is it increasing?
2. Where is it decreasing?
3. Where is it concave up?
4. Where is it concave down?
5. Where are its points of inflection?
6. Where are its zeros?
7. Where does it achieve its maximum value, and what is that maximum value?
8. Where does it achieve its minimum value, and what is that minimum value?
9. Is it periodic? If so, what is its period?
We can answer all these questions about the sine and cosine functions by examin-
ing their graphs and applying ideas developed earlier in this book. However, for
other functions that may not be as well known, answering some of these questions
requires the use of calculus.
Let’s consider the behavior of the sine function y  sin x. It is evident from its
graph that the sine curve increases for x between 0 and p>2, then decreases from
p>2 to 3p>2, and then increases from 3p>2 to 2p—and repeats this cycle there-
after. This behavior is also clear from the unit circle definition of the sine.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 493

7.1 Introduction to the Sine and Cosine Functions 493

Further, the sine curve is concave down for x between 0 and p, concave up for x
between p and 2p, and then repeats this cycle thereafter. Consequently, the sine
curve has points of inflection at x  0, p, 2p, . . . , where its concavity changes.
In addition, the sine function has zeros when x  0, p, 2p, . . . A special
characteristic of the function f 1x 2  sin x is that its zeros and its points of inflec-
tion are identical, which is not the case for most other common functions.
Finally, the sine function achieves its maximum value of 1 at x  p>2, at
x  5p>2, at x  9p>2, . . . and at x  3p>2, at x  7p>2, . . . The sine
function achieves its minimum value of 1 at x  3p>2, at x  7p>2, at
x  11p>2, . . . and at x  p>2, at x  5p>2, . . .
We ask you to describe the behavior of the cosine function in the Problems at
the end of this section.
In summary, the key points about the sine and cosine functions are:
The sine function passes through the origin and oscillates between 1 and
1 every 2p.
The cosine function passes through the point 10, 12 and oscillates between 1
and 1 every 2p.

Problems
1. Janis trims her fingernails every Saturday morning. 4. Convert each angle from degrees to radians.
Sketch the graph of the length of her nails as a func- a. 15° b. 75° c. 120°
tion of time. Can this process be modeled by a peri- d. 150° e. 225° f. 315°
odic function? If it is periodic, what is the period? g. 270° h. 240° i. 135°
2. Harry gets a haircut on the first of every month. j. 210°
Sketch the graph of the length of his hair as a func- 5. Convert each angle from radians to degrees.
tion of time. Can this process be modeled by a peri- 3p 4p 2p
odic function? If it is periodic, what is the period? a. b. c.
4 5 3
3. In the accompanying figure, the circle on the left d. 1.5 e. 2.5 f. 3
has been subdivided every 15° from u  0° to p 5p 3p
g. h. i. 
u  360°. Use the heights from the horizontal axis 8 3 2
to the associated points on the circle to construct 5p
j. 
the graph of the sine function for u  0° to 3
u  360° on the axes at the right. 6. For f 1u 2  5 sin u, evaluate each function.
a. f 130°2 b. f 145°2 c. f 160°2
y
y

1
0.8
0.6
0.4
0.2
x 0 θ
60° 120° 180° 240° 300° 360°
–0.2
–0.4
–0.6
–0.8
–1
Gord.3896.07.pgs 4/28/03 11:55 AM Page 494

494 CHAPTER 7 Modeling Periodic Behavior

d. f 1120°2 e. f 115°2 f. f 1873°2 lynxes are the predators and hares are the prey. The
figure below is based on records kept by the Hud-
g. f a b h. f a b i. f a b
p p p
4 3 12 son’s Bay Trading Company on the number of each
species caught by fur trappers from 1845 through
j. f a b k. f 15.272 l. f 125.6142
p
1935. The graphs indicate that both populations
6
For f 1u2  sin 2u, evaluate each function.
change in roughly periodic cycles.
7.
a. f 130°2 b. f 145°2 c. f 1120°2
a. Estimate the period of the cycle for the lynxes.
d. f 1225°2 e. f 1p>32 f. f 1p>12 2
b. Estimate the period of the cycle for the hares.
g. f 13p>82 h. f 12p>72
c. Estimate the years in which the lynx population
reached its maximum and minimum values.
8. At the end of this section we posed nine questions d. Estimate the years in which the hare population
about the behavior of any function. Answer these reached its maximum and minimum values.
questions for the cosine function. e. Can you find any relationship between the
9. With your calculator set in radians, graph the two lengths of the periods in parts (a) and (b) and
functions y  sin x and y  cos1x  p>2 2 . What the times in parts (c) and (d)? If so, what is it?
do you observe? Explain what you observed. f. Estimate the years when the hare population
10. Plot the functions y  cos x and y  sin1x  p>22. passed its points of inflection. How do they
Explain why you see only one graph. (If you see two compare to any of the times you found in parts
graphs, check that your calculator MODE is set for (c) and (d)?
Radians.) 12. Consider the functions y  cos x and y 
11. The population growth patterns of two species are sin1#$%#$ 2 . What could #$%#$ represent so that
interrelated when one species preys on the other. the two graphs are identical? Is there only one cor-
This situation occurs in northern Canada where rect answer to this question? Explain.

Hare
160
Lynx
140

120
Number in thousands

100

80

60

40

20

1845 1855 1865 1875 1885 1895 1905 1915 1925 1935
Year

7.2 Modeling Periodic Behavior with the Sine and Cosine

Because of their behavior patterns, the sine and cosine functions are used as the
mathematical models to represent most periodic phenomena. For example, the
number of hours of daylight H any day of the year in San Diego can be modeled by
the function
H1t2  12  2.4 sin c 1t  802 d ,
2p
365
Gord.3896.07.pgs 4/28/03 11:55 AM Page 495

7.2 Modeling Periodic Behavior with Sine and Cosine 495

where t is the number of days from the first of the year (t  1 on January 1). We
begin by using the formula for some predictions and then see where the formula
comes from.

E XAMPLE 1
Based on the model, how many hours of daylight are there in San Diego on (a) February 15?
(b) March 21? (c) June 21?
Solution
a. February 15 is the 46th day of the year (31 days in January plus 15 more in February),
so t  46. Using a calculator set to radian mode, we find that

H146 2  12  2.4 sin c 146  80 2 d

2p
365

 12  2.4 sin c 134 2 d  10.67 hours.

2p
365

b. March 21 is the t  31  28  21  80th day of the year, so

H180 2  12  2.4 sin c 180  80 2 d  12 hours.

2p
365
Incidentally, March 21 is the spring equinox, which means that there are 12 hours of
daylight and 12 hours of darkness, so the model gives the right prediction.
c. June 21 is the t  31  28  31  30  31  21  172nd day of the year, so the
number of hours of daylight on June 21(the first day of summer and so the “longest”
day of the year) is

H1172 2  12  2.4 sin c 1172  80 2 d  14.40 hours.

2p
365

Think About This Without using a calculator, find the number of hours of daylight in San Diego on
December 21, the 355th (and “shortest”) day. ❐

Later in this section we show how a similar formula can be developed for any
city. For now, let’s see what the different numbers 12, 2.4, 365, and 80 in the formula

H1t2  12  2.4 sin c 1t  80 2 d

2p
365
for the number of hours of daylight in San Diego actually represent. Obviously,
365 represents the number of days in a year; it tells us how long it takes for a full
cycle to be completed. Over a full year, the average number of hours of daylight is
12 hours per day—the days when there are more than 12 hours of daylight exactly
counterbalance the days when there are fewer than 12 hours of daylight. Alterna-
tively, averaging the number of hours of daylight for all 365 days gives 12 hours. So
the 12 represents the average, or middle, value for the sine function.
Next, as we found in Example 1, the longest day in San Diego has 14.4 hours of
daylight and the shortest day has 9.6 hours of daylight. Note that 14.4 is 2.4 hours
more than the average level of 12 and that 9.6 is 2.4 hours less than 12. Of course,
Gord.3896.07.pgs 4/28/03 11:55 AM Page 496

496 CHAPTER 7 Modeling Periodic Behavior

the maximum and minimum number of hours of daylight at any particular loca-
tion, as well as the number on any specific date, depend on the location itself and
are therefore modeled by a function slightly different from H; think about how
long a “day” is during the winter or the summer in the far north, the so-called “land
of the midnight sun”.
The graph of H, the number of hours of daylight in San Diego over a 3-year in-
terval, is shown in Figure 7.12. It has the same shape as the graph of the basic sine or
cosine function. However, it does not oscillate about the horizontal axis; rather it os-
cillates about the horizontal line H  12, which represents the average number of
hours of daylight over a full year, so it is shifted up by 12 hours. Also, its maximum
and minimum “heights” above the horizontal line H  12 are no longer 1 and
1 as with the basic sine and cosine functions. Instead, the graph varies from a
minimum of 9.6 hours to a maximum of 14.4 hours, which is 2.4 hours either side
of the average 12, so the sine function has been stretched by a factor of 2.4.

14.4

12

9.6

t
FIGURE 7.12 365

Note some additional differences: The graph is shifted to the right compared
to the sine curve (the curve does not “start” at the vertical axis where t  0 and
H  12). Also, the period is 365 days, rather than the usual 2p radians, or 360°.
(Incidentally, the ancient Babylonians believed that the length of a year was 360 days.
That’s why we divide a circle into 360 degrees.)
This particular function H1t2 differs from the standard, or base, function
y  sin x that we discussed in Section 7.1 in four ways:
1. a vertical shift,
2. an oscillation other than from 1 to 1 (a stretch),
3. the length of a cycle, and
4. the “starting” point of the cycle (a horizontal shift).
Understanding how to incorporate these variations is crucial for applying the sine
and cosine functions to describe periodic phenomena. We therefore focus on each
in detail.
The equation for the number of hours of daylight in San Diego is

H  12  2.4 sin c 1t  80 2 d .
2p
365
Consider the more general sinusoidal function
S1x2  D  A sin3B1x  C2 4,
Gord.3896.07.pgs 4/28/03 11:55 AM Page 497

7.2 Modeling Periodic Behavior with Sine and Cosine 497

where A, B, C, and D are all constants and x is the independent variable. In the San
Diego situation, D  12, A  2.4, B  2p>365, and C  80. Let’s investigate how
each of these four parameters affects the graph of the basic sine curve. To do so, we
consider each parameter separately.

The Vertical Shift or Midline

To show the significance of the D term, we consider the simpler function with
A  1, B  1, and C  0:
S1x 2  D  sin x.
We know that y  sin x oscillates repeatedly between 1 and 1. What is the ef-
fect of adding a constant D? From our discussion in Sections 4.6 and 4.7, we know
that D raises or lowers the basic sine curve by the amount D. The graph of
S1x 2  2  sin x has the same shape as the basic sine function but is shifted up 2
units; it oscillates about the horizontal line y  2, between 1 and 3 units above the
x-axis as shown in Figure 7.13. Similarly, the graph of S1x 2  5  sin x oscillates
about the horizontal line y  5, between 6 and 4.

3
y = 2 + sin x
2

1
y = sin x

x
0 π 2π

FIGURE 7.13 –1

Thus the effect of the constant D in

S1x 2  D  sin x
is to produce a sinusoidal curve that oscillates about the horizontal line y  D, be-
tween D  1 and D  1. If D is positive, the sine curve is shifted upward D units;
if D is negative, the curve is shifted downward D units. The number D is the
vertical shift or the midline. In the formula for the number of hours of daylight in
San Diego, the vertical shift, or midline, is 12.

The Amplitude
We next investigate the effect of the multiplicative constant A in the general equa-
tion of a sinusoidal function y  D  A sin3B1x  C2 4. We set D  0, B  1,
and C  0 to consider the simpler function
S1x 2  A sin x.
For example, if A  2, we get S1x 2  2 sin x, whose graph is shown in Figure 7.14,
where it is compared to the basic curve for the sine function, y  sin x (for which
Gord.3896.07.pgs 4/28/03 11:55 AM Page 498

498 CHAPTER 7 Modeling Periodic Behavior

A  1). For comparison, we also show the graph of T1x 2  12 sin x. Although the
basic sine function oscillates between 1 and 1, the transformed function
S1x2  2 sin x oscillates between 2 and 2 and the transformed function
T1x2  12 sin x oscillates between  12 and  12 . In general, the effect of multiplying
the sine function by a constant A is to increase its vertical height above and below
the midline by the factor 0 A 0 .

2 A=2

A=1
1
1
A= 2

x
0 π 2π

–1

FIGURE 7.14 –2

To show why the absolute value is necessary, we consider the graph of

S1x2  4 sin x,
which has the same shape as the basic sine curve, but oscillates between 4 and
4. The main difference is that the graph of this function is flipped over the x-axis
compared to the graph of y  sin x. We naturally think of it as being four times as
high as the base curve, not 4 times as high. (Draw simultaneously the graphs of
y  sin x and y  4 sin x using your function grapher.) This is the same effect
of the negative multiple that we encountered in Section 4.6. However, this curve
has the same period 12p  360°2 and the same zeros (x  0, p, 2p, . . . ) as
the basic sine curve.
The quantity 0 A 0 is called the amplitude of the sine function. In the expres-
sion for H1t2 modeling the number of hours of daylight in San Diego, the ampli-
tude is 2.4.
In Example 2 we show what happens when we combine the two transforma-
tions to construct a new function.

E XAMPLE 2
Analyze the graph of
S1x 2  2  3 sin x.

Solution In this formula, 2 is the vertical shift, or midline, and 3 is the amplitude. The
effect of multiplying the sine function by 3 is to stretch it vertically by a factor of 3, so
that 3 sin x oscillates between 3 and 3. Adding the constant 2 to the function 3 sin x
simply raises the entire curve 2 units vertically. Consequently, the combined effect is to
produce a sinusoidal function that oscillates from 3 units below the horizontal line
y  2 to 3 units above the line; that is, from 1 to 5, as shown in Figure 7.15.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 499

7.2 Modeling Periodic Behavior with Sine and Cosine 499

4 y = 2 + 3 sin x

2
y = sin x
1

x
0 π 2π
–1

FIGURE 7.15 –2

◆
Incidentally, the function y  2  3 sin x is not the same as y  5 sin x; the
coefficients cannot be combined because 2 and 3 sin x are not like terms. Graph
both functions to see that they produce very different results. Also, y  2  3 sin x
is not the same as y  3  2 sin x —each parameter has its own role to play.
Use your function grapher to examine the graphs of several functions of the
form y  D  A sin x for different values of A and D. Predict and then observe how
the different constant values are reflected in the corresponding sinusoidal curve.

The Frequency and the Period

We next consider the effect of the parameter B, which multiplies the term 1x  C2 in
S1x 2  D  A sin3 B1x  C2 4.
To concentrate on B only, we take C  0, D  0, and A  1. We also assume that
B 0. Consider how the function
S1x 2  sin12x 2
compares to the basic curve y  sin x, as shown in Figure 7.16. The resulting sinu-
soidal curve y  sin12x 2 completes two full cycles between x  0 and x  2p,
compared to one full cycle for the basic sine curve.

y
y = sin x y = sin 2x
1

π
x
0 2π

–1

FIGURE 7.16
Gord.3896.07.pgs 4/28/03 11:55 AM Page 500

500 CHAPTER 7 Modeling Periodic Behavior

Similarly, the graph of

S1x 2  sin13x 2
shown in Figure 7.17 completes three full cycles across the interval from 0 to 2p.
Based on these two results, we expect that the graph of
S1x 2  sin1nx 2,
for any positive integer n, will complete n full cycles between x  0 and x  2p.
Try this with your function grapher for values of n such as 5 or 8.

y
y = sin x y = sin 3x
1

x
0 π 2π

FIGURE 7.17 –1

What happens if the positive multiple B is not an integer? The graph of

y  sin1 12 x 2 is shown in Figure 7.18. The function y  sin1 12 x 2 completes half a
complete cycle between 0 and 2p; it actually requires an interval of values for x
from 0 to 4p to complete a full cycle.
Similarly, Figure 7.19 shows that the function y  sin12.5x 2 completes 2.5 full
cycles between 0 and 2p. (Trace the graph with your finger and count the cycles.)
It therefore completes one full cycle in 1>2.5  0.4  2>5 of this interval.

y y
y = sin x y = sin x
1 y = sin 1
x 1 y = sin (2.5x)
2

x x
0 π 2π 0 π 2π

–1 –1

FIGURE 7.18 FIGURE 7.19

Mathematicians call the parameter B in y  sin1Bx 2 the frequency of the sinu-

soidal function. It tells us the number of complete cycles that occur between x  0
and x  2p  360°. For instance, the function y  sin16x2 completes six full cycles
across this interval, whereas the function y  sin1 38x 2 completes 3>8 of one cycle.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 501

7.2 Modeling Periodic Behavior with Sine and Cosine 501

Note that standard usage is to write y  sin 2x or y  sin 5x, say, rather than
y  sin12x 2 or y  sin15x 2, even though it is less precise a notation. But parentheses
are essential on calculators and computers.
As with any periodic function, the period of a sinusoidal function y  sin Bx is
the length of the interval needed to complete one full cycle. For y  sin 2x, a full
cycle is completed in any interval of x-values of length p (see Figure 7.16), so the
period is p radians  180°. For y  sin 3x, the period is

a b 2p  a b 360°  120°,
1 2p 1
or
3 3 3
(see Figure 7.17). For y  sin 1 12x 2 , the period is

12p 2  4p radians,
1
1
2

or 720° (see Figure 7.18).

In general, the period of y  sin Bx is
2p 2p
Period   .
B frequency

Thus, for instance, for y  sin12.5x 2, the period is

 12p 2  p
2p 2p 2 4
Period  
2.5 5>2 5 5
because it is the length of the interval needed for this sinusoidal function to complete
one full cycle (see Figure 7.19). This result agrees with our earlier statement that the
function y  sin12.5x 2 completes one full cycle in 2>5 of the interval from 0 to 2p.
We have shown that the period of any periodic function is the length of the in-
terval needed to complete one full cycle. Alternatively, if we start with the period B,
2p
Frequency  .
period
In the formula for the number of hours of daylight in San Diego

H  12  2.4 sin c 1t  80 2 d ,
2p
365
the frequency of the sinusoidal curve is
2p
Frequency   0.0172.
365
The period of the sinusoidal curve is
2p 2p
Period    365 days.
frequency 12p>365 2
Thus, as we would expect, the period is 1 year.
In summary we have the following.

2p 2p
period  frequency 
frequency period
Gord.3896.07.pgs 4/28/03 11:55 AM Page 502

502 CHAPTER 7 Modeling Periodic Behavior

Note that in engineering, the frequency is defined somewhat differently. In-

stead of meaning the number of cycles in 2p radians, engineers consider the num-
ber of cycles in a given length of time—say, cycles per second. They then write
1
Frequency 
period
and write the sinusoidal function in the form y  sin12pBt2 , where B, not 2pB, is
the frequency. Unfortunately, this slight difference in terminology is so deeply embed-
ded in the two fields that it is not possible for either field to change to match the other.

The Phase Shift

Finally, we consider the role of the parameter C in
S1x2  D  A sin3 B1x  C2 4.
We simplify the discussion by taking A  1, B  1, and D  0 so that we con-
sider only
S1x 2  sin1x  C2.
From Section 4.7, we know that the term 1x  C2 should have the effect of a
horizontal shift to the right when C is positive and to the left when C is negative.
Figure 7.20 compares the graph of S1x 2  sin1x  p>4 2 to the basic curve
y  sin x. The two curves appear similar, but S1x 2 is shifted to the left (backward)
by p>4, or 18 of 2p (which is one-eighth of a full cycle). Similarly, Figure 7.21 shows
the graph of T1x 2  sin1x  p>3 2. It has been shifted to the right (forward) by
p>3, or 16 of 2p (which is one-sixth of a full cycle).

y y
− π4 π
3
1 1

π
y = sin x y = sin (x − 3 )

x x
0 3π π 7π 2π 0 π π 4π 2π
4 4 3 3

π
y = sin (x + 4 ) y = sin x

–1 –1

FIGURE 7.20 FIGURE 7.21

In general, the parameter C shifts a sinusoidal curve to the left or the right by
the amount C. If C is positive in the term 1x  C2, as in y  sin1x  p>3 2 , the
curve is shifted to the right by C; if C is negative in 1x  C2 , as in
y  sin1x  p>4 2, the curve is shifted to the left by C. This parameter is called the
phase shift, instead of the horizontal shift, in the context of sinusoidal functions.
In the expression for the daylight function for San Diego

H  12  2.4 sin c 1t  80 2 d ,
2p
365
the phase shift is 80 days.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 503

7.2 Modeling Periodic Behavior with Sine and Cosine 503

E XAMPLE 3
What is the significance of the phase shift in the formula for H?
Solution The phase shift shifts the curve to the right by 80 days. Recall that the 80th
day of the year is March 21, which is the spring equinox (the day when there are equal
numbers of hours of daylight and darkness). On this day, the graph for the sinusoidal
function crosses the midline, or average level, of D  12 hours.
◆

In general, the phase shift for a sine function corresponds to the first point to
the right of the origin where the curve crosses the midline while increasing. Equiv-
alently, it occurs midway, horizontally, from a minimum point to a maximum.
We summarize all the results for the San Diego daylight function in Figure 7.22.

14.4
2.4 = amplitude
vertical shift = 12

9.6

t
FIGURE 7.22 0 80 = phase shift 365 = period

E XAMPLE 4
The water at a boat dock is 7 feet deep at low tide and 11 feet deep at high tide. On a cer-
tain day, low tide occurs at 4 A.M. and high tide at 10 A.M. Find an equation for the height
of the tide y as a function of time t.

Solution We use the given information to sketch the graph of a sinusoidal curve in Fig-
ure 7.23.

Height of tide

11

t
FIGURE 7.23 0 4 10 16 22
Gord.3896.07.pgs 4/28/03 11:55 AM Page 504

504 CHAPTER 7 Modeling Periodic Behavior

Because the tide ranges from a minimum height of 7 feet to a maximum height of
11 feet above sea bottom, the curve oscillates about the middle value of 9 feet, which is
the vertical shift, or midline. Also, the amplitude of this sinusoidal curve is 2. Further, the
time interval between the minimum and maximum heights of the water level is 6 hours;
consequently, a complete tide cycle takes 12 hours, so the period is 12 hours. As a result,
2p 2p p
Frequency    .
period 12 6
Finally, because the tide level increases from 4 A.M. to 10 A.M., the curve passes across the
middle height of 9 feet halfway between 4 A.M. and 10 A.M. (or at 7 A.M.), which gives the
phase shift. (The graph shows that, even though the tide function also crosses the 9-foot
level at 1 A.M., the function is decreasing there and so this does not give the phase shift.)
Therefore the height y of the water at any time t is modeled by

y  9  2 sin c 1t  7 2 d .
p
6
◆
E XAMPLE 5
The air conditioning in a home is set to go on when the temperature reaches 74°F and to
go off when the temperature drops to 68°. This cycle repeats every 20 minutes. If the
temperature in the house at noon is 71° and rising, write a sinusoidal function to model
the temperature as a function of the number of minutes t since noon.
Solution A sinusoidal function is of the form T  D  A sin3B1t  C2 4, where A, B,
C, and D must be determined. We know that the temperature oscillates between 68°
and 74°, so it is centered about 71°, which is the vertical shift, or midline, D. Further,
because the size of the oscillation above and below this midline is 3, we know that the
amplitude A  3. We also know that the length of the cycle is 20 minutes, so the period
is 20 and therefore the frequency B  2p>20  p>10. Finally, because the house tem-
perature reaches 71°—the level of the vertical shift—at noon when t  0, the phase
shift is 0, so C  0. Therefore our model for the temperature of the house as a function
of time is

T  71  3 sin a
p
tb .
10
Figure 7.24 shows the graph of this sinusoidal function for the first 60 minutes.

T
π

74
T = 71 + 3 sin ( t)
10
Temperature (°F)

72

70

68

t
0 10 20 30 40 50 60
FIGURE 7.24 Time (minutes)

◆
Gord.3896.07.pgs 4/28/03 11:55 AM Page 505

7.2 Modeling Periodic Behavior with Sine and Cosine 505

E XAMPLE 6
For part of the year, the temperature T in the Colorado Rockies can be modeled by the
function
T1t2  60  10 sin a tb ,
p
12
where t is measured in hours and t  0 is at 9 A.M. In Example 5 of Section 2.2, we used
the linear function
C1T2  4T  160
to model the chirp rate C (in chirps per minute) of the snow tree cricket as a function of
the air temperature T (in ° F).
a. Express the chirp rate as a function of time.
b. How fast is the cricket chirping at 5 P.M.?
c. What are the domain and range of this function?

Solution
a. The chirp rate in C  4T  160 is measured in chirps per minute and the time t in
the formula for the temperature as a function of time is measured in hours. To
make things consistent, we convert the chirp rate to chirps per hour by multiplying
by 60 (minutes per hour) to get
C  6014T  160 2  240T  9600.
We now have C as a function of T, where T is a function of t, so C is a composite
function,
C  f 1t2  240T  9600

 240 c 60  10 sin a tb d  9600

p
12
 14,400  2400 sin a tb  9600
p
12
 4800  2400 sin a tb ,
p
12
where t is measured in hours since 9 A.M. and C is chirps per hour, as illustrated in
Figure 7.25. (Note that we could have gone the other way and converted everything to
minutes, but then the numbers get quite large.)
b. At 5 P.M., when t  8 hours after 9 A.M., the chirp rate is

C  f 182  4800  2400 sin a # 8b  6878 chirps per hour.

p
12
C
π
C = 4800 + 2400 sin ( t)
12
7200

6000
Chirps per hour

4800

3600

2400

t
4 8 12 16 20 24
FIGURE 7.25 Time (minutes)
Gord.3896.07.pgs 4/28/03 11:55 AM Page 506

506 CHAPTER 7 Modeling Periodic Behavior

c. The independent variable is the time t in hours from 9 A.M. on a particular day. The
practical domain of the function f depends on how long the functions that are the
component models make sense. The model for the temperature is good for only
part of the year; let’s assume that it applies only for a 30-day period. If the initial
time is at the beginning of that time period, the domain would last for the follow-
ing 30 days; if the initial time is at the middle of that time period, the domain
would extend from 15 days before to 15 days after. As for the range, look at the
function f. It oscillates above and below 4800 chirps per hour, from a minimum of
4800  2400  2400 to a maximum of 4800  2400  7200 chirps per hour. So
the range is 2400 to 7200.

◆
Identical ideas about vertical shift, amplitude, period, frequency, and phase
shift apply to cosine functions of the form
y  D  A cos 3 B1x  C2 4,
whose behavior also is described as sinusoidal. The midline D serves to raise or
lower the “center” of the cosine curve; the amplitude A stretches or shrinks the co-
sine curve vertically about the midline; the frequency B represents the number of
cycles over an interval of 2p; and the parameter C is the phase shift, which shifts
the cosine curve to the left or the right, depending on the sign of C. The only dif-
ference between working with sines and cosines lies in finding the phase shift. For
a cosine function, the phase shift corresponds to the first point to the right of the
origin where the curve reaches its maximum.

E XAMPLE 7

Describe the graph of the sinusoidal function y  5  3 cos c 2 ax  b d.

p
4
Solution The basic cosine curve is multiplied by 3, so its amplitude is 3. Because the
midline is 5, the function extends from a minimum of 5  3  2 to a maximum of
5  3  8. Because the frequency is 2, there are two complete cycles between 0 and 2p,
so the period is p. Finally, the phase shift is p>4, so the curve is shifted to the right by
p>4. The graph of this function from x  0 to x  2p, shown in Figure 7.26, illustrates
all these effects.

y
π
[(
y = 5 + 3 cos 2 x − 4 )]
8

x
o
3

FIGURE 7.26

◆
Gord.3896.07.pgs 4/28/03 11:55 AM Page 507

7.2 Modeling Periodic Behavior with Sine and Cosine 507

Combining Sinusoidal and Exponential Functions

Consider a spring hanging vertically from the ceiling with a weight attached at the
bottom, as shown in Figure 7.27. The weight is pulled down and then released, so
the weight bobs up and down with smaller and smaller oscillations until it settles
to a stop in the original rest position, called its equilibrium. Figure 7.28 displays a
graph of this vertical displacement y as a function of time t.

rest
W t

FIGURE 7.27 FIGURE 7.28

E XAMPLE 8
Figure 7.29 shows the results of recording the vertical oscillations of an object attached
to a particular spring as a function of time from t  0 to t  2p. Construct a function
that models this behavior pattern.

5
Height

1
t
0 2π ≈ 6.28
FIGURE 7.29 Time

Solution What type of function could have this kind of behavior pattern? The oscilla-
tory effect certainly suggests a sinusoidal function, either a sine or a cosine, but the am-
plitude is not constant. In fact, the oscillations eventually die out. The overall effect of
the decreasing amplitude might suggest either a decaying power function or an expo-
nential decay function. Because there is a finite starting value, not a vertical asymptote,
at time t  0, a power function is not appropriate. An exponential decay term makes
more sense. Moreover, based on our discussion in Section 4.6, we might be tempted to
consider the product of such an exponential decay function and a sinusoidal function.
Two possible formulas for functions that combine these two behavior patterns are
y  Abtsin ct or y  Abtcos ct,
with b
1. You can think of the decaying exponential function as a variable amplitude
that decays to 0 over time. In addition, there is a vertical shift, so the possible functions
Gord.3896.07.pgs 4/28/03 11:55 AM Page 508

508 CHAPTER 7 Modeling Periodic Behavior

are y  D  Abtsin ct or y  D  Abtcos ct. Our task is to determine values for the
four parameters A, b, c, and D.
First, we note that the initial height of the object is approximately y  7. Starting
from there, the object drops at first. Its height decreases from a maximum, which sug-
gests a cosine function rather than a sine function.
Second, we note that the final, or equilibrium, height for the object is about 4, so the
object seems to be oscillating about a height of y  4. The maximum height is 7, or 3 units
above this equilibrium level, so the form for the function seems to be
y  f 1t2  4  3btcos ct,
where f 102  4  3b0cos 0  4  3112 11 2  7.
Further, we estimate from the graph that, between t  0 and t  2p  6.28, there
are about five complete diminishing cycles. Thus the frequency for the cosine function is
approximately 5, giving the equation
y  f 1t2  4  3btcos 5t.
Finally, consider the exponential decay curve g1t2  3bt that is superimposed over
the successive peaks of the decaying sinusoidal function in Figure 7.30. It starts with an
initial height of 7 and decays to a final level of 4. Using a ruler, we can estimate that it has
dropped halfway (to a height of 5.5) at about t  2, Therefore we use t  2 as an ap-
proximation for the half-life of the pure exponential decay function g1t2  3bt. Thus we
must solve the equation
g12 2  3b2  12 13 2 ,
which gives
b2  0.5 so that b  20.5  0.707.
y

6
5.5
5
Height

1
t
0 t≈2 2π ≈ 6.28
FIGURE 7.30 Time

We can now estimate that the desired function is given by

y  f 1t2  4  310.7072 tcos 5t.
Verify that this function matches the required pattern for the oscillation shown in
Figure 7.29 by using your function grapher to graph f between 0 and 2p.
◆
Incidentally, we have implicitly assumed throughout this section that all peri-
odic processes follow a sinusoidal pattern (either a sine or a cosine curve) precise-
ly. In practice, this assumption may be expecting a lot. Think about the length of
Gord.3896.07.pgs 4/28/03 11:55 AM Page 509

7.2 Modeling Periodic Behavior with Sine and Cosine 509

Janis’s fingernails in Problem 1 from Section 7.1. The nail length is a periodic func-
tion, but it is not sinusoidal. Even if you observe that the overall pattern for some
periodic process is smooth and appears to be that of a sine curve, you have no
guarantee that the behavior is exactly sinusoidal. Nevertheless, sinusoidal func-
tions are your best models for such types of periodic phenomena and consequent-
ly are the models that you should use when faced with such behavior.
Finally, just as you can construct linear, exponential, power, and other functions
to fit a set of data, you often will be faced with the problem of having a set of data
that exhibits a periodic pattern and wanting to find the periodic function that best
fits the data. We ask you to explore several such cases in the following Problems.

Problems
1. Decide which of the following functions are peri- 2. Find the number of hours of daylight in San Diego
odic. For those that are periodic, what is the period? on March 1, on May 12, on July 4.
(Assume that each graph continues in the same pat-
tern indefinitely to the left and right.)
y y y

x x
–3 –2 –1 1 2 3 –3 –1 1 3
x
–3 –2 –1 1 2 3
(a) (b) (c)

y y y

x x x
–3 –2 –1 0 1 2 –3 –2 –1 0 1 2 –3 –2 –1 0 1 2

(d) (e) (f)

y y y

x
x 1 2 3
x –1 0 1 2
0
(g) (h) (i)

x
10 40

(j)
Gord.3896.07.pgs 4/28/03 11:55 AM Page 510

510 CHAPTER 7 Modeling Periodic Behavior

3. The number of hours of daylight in Montreal is crest, say) on the open seas is about 20 times the
given by height of the wave (trough to crest). (This rule
H1t2  12  3.6 sin c 1t  802 d ,
2p doesn’t apply near coastlines where waves are much
365 choppier and their intervals shorter.)
where t is the number of days from the 1st of the year.
20y
a. What is the amplitude of this function? y
b. What is the period of this function?
c. What is the number of hours of daylight on the
a. Write a formula for ocean waves that are 4 feet
shortest day of the year?
high in moderately calm seas.
d. What is the number of hours of daylight on the
b. Write a formula for ocean waves that are 15 feet
longest day of the year?
high in rough seas.
4. The shortest day of the year in Fairbanks, Alaska,
10. Meryl is a normal individual with a pulse rate of
has 3.70 hours of daylight. Find a formula for the
72 beats per minute and a blood pressure of 120
number of hours of daylight there on any day of
over 80. Thus her heart is beating 72 times each
the year.
minute and her blood pressure is oscillating be-
5. Write a formula giving the number of hours of tween a low (diastolic) reading of 80 and a high
darkness in San Diego as a function of the day of (systolic) reading of 120. Assume that the oscilla-
the year. tion in Meryl’s blood pressure can be modeled by a
6. Consider Example 1 regarding the height of the tide sinusoidal function.
at a dock. Suppose that low tide still occurs at 4 A.M. a. What is the period of this sinusoid?
but that high tide actually occurs at 10:30 A.M. Find b. What is the frequency of this sinusoid?
an equation for the height of the tide as a function c. What is the equation of this sinusoid?
of time t.
11. Your Thanksgiving turkey is taken from a refrigerator
7. The Bay of Fundy in eastern Canada is known for at 40°F and placed in an oven set at 350°. Suppose
the highest tides in the world. The tides there rise that the temperature of the bird is 130° after 60 min-
and fall by as much as 50 feet. If the tidal cycle takes utes. You know that an oven cycles on and off as some
11 hours, find a sinusoidal function that models the of the heat escapes. Suppose that the cycle occurs
tides in the bay. For convenience, assume that low every 10 minutes and that the actual temperatures
tide corresponds to a height of 0. inside the oven oscillate between 340° and 360°.
8. The thermostat in Sylvia’s home in Baltimore is set a. Use this information to construct a sinusoidal
at 66°F. Whenever the temperature drops to 66° function to model the temperature of the oven
(roughly every 30 minutes), the furnace comes on as a function of time t.
and stays on until the temperature reaches 70°. b. Use Newton’s law of heating from Section 5.4
a. Write a sinusoidal function that models this to estimate how large a variation is possible in
situation. the temperature of the turkey after 60 minutes,
b. Gary’s thermostat in upper New York State is set and after 100 minutes. (Hint: Solve the prob-
the same way. How would the model you created lem with the minimum and maximum oven
in part (a) change to reflect Gary’s climate? temperatures.)
c. Jodi, who lives in central Florida, likewise has 12. A standard radio has two bands—the AM (amplitude
her thermostat set to come on at 66°F. How modulation band) and the FM (frequency modula-
would you change the models you created for tion band). In one case, the amplitude of a sinu-
parts (a) and (b) to reflect her climate? soidal wave is modulated (varied) to produce the
d. Is a sinusoidal function necessarily a good desired output sounds; in the other, the frequency
model? Explain. (Hint: Think about the rates at of a sinusoidal wave is modulated. Which of the fol-
which the temperature increases and decreases.) lowing represents an AM sound and which repre-
9. Ocean waves move in a roughly sinusoidal pattern. sents an FM sound?
As a rule of thumb, the length of a wave (crest to
Gord.3896.07.pgs 4/28/03 11:55 AM Page 511

7.2 Modeling Periodic Behavior with Sine and Cosine 511

0 2 4 6 8 10 0 2 4 6 8 10
(a) (b)

13. Two successive turning points of a sinusoidal func- la for a sinusoidal function that can be used to
tion are at 18, 722 and 120, 302. model the average daytime high temperature in
a. Write a possible formula for this function, using Fairbanks as a function of the day of the year.
a sine function. 16. Sketch by hand the graph of each function. Draw
b. Write a possible formula for this function, using the basic curve y  sin x or y  cos x on the same
a cosine function. set of axes for comparison. (Do not use your func-
14. Two successive inflection points of a sinusoidal tion grapher.)
function are at 16, 202 and 118, 202 ; the maximum b. y  3 sin a xb
1
a. y  3 sin 4x
attained by the function is 43. 2
c. y  2 sin 3x d. y  4 cos 2x
a. Write a possible formula for this function, using e. y  3 cos 2x f. y  4  2 sin x
a sine function.
g. y  sin ax  b h. y  3 sin a2x  b
p p
b. Write a possible formula for this function, using
a cosine function. 4 6

i. y  4  2 cos ax  b
15. Suppose that the historical average daytime high p
temperature in Fairbanks ranges from a low of 3
20°F to a high of 64°F and that the coldest day of 17. Write a possible formula for each sinusoidal func-
the year, historically, is the 40th day. Write a formu- tion 1a21l2 from its graph.

y y y y

2 5 4 3

6π 180°
x x
x x
90° 10π

–2 –5 –4 –3

(a) (b) (c) (d)

y y y
y
3 16
1.5 5

x 1 x
8π π
4
x
π
–1.5 –5 x
π
4
(e) (f) (g)
(h)
Gord.3896.07.pgs 4/28/03 11:55 AM Page 512

512 CHAPTER 7 Modeling Periodic Behavior

y y y
y

x 6 4
π 1
2

–3 4 x
x
2

–6 x –4
π –1
2

(i) (j) (k) (l)

18. Identify a possible formula for each of the six sinu-

soidal functions f 1 , f 2 , . . . , f 6 whose values are
given in the table. Note that there are many possible
correct answers.

x f1 f2 f3 f4 f5 f6
6 0.279 2.279 0.559 0.537 0.141 0.721
5 0.959 2.959 1.918 0.544 0.598 0.041

4 0.757 2.757 1.514 0.989 0.909 0.243

3 0.141 1.859 0.282 0.279 0.997 1.141

2 0.909 1.091 1.819 0.757 0.841 1.909

1 0.841 1.159 1.683 0.909 0.479 1.841

0 0.000 2.000 0.000 0.000 0.000 1.000

1 0.841 2.841 1.683 0.909 0.479 0.159

2 0.909 2.909 1.819 0.757 0.841 0.091

3 0.141 2.141 0.282 0.279 0.997 0.859

4 0.757 1.243 1.514 0.989 0.909 1.757

5 0.959 1.041 1.918 0.544 0.598 1.959

6 0.279 1.721 0.559 0.537 0.141 1.279

19. The table gives the outdoor temperatures in °F in

Chicago during one 24-hour period.

Midnight 2 A.M. 4 A.M. 6 A.M. 8 A.M. 10 A.M.

53 48 47 49 53 59
Noon 2 P.M. 4 P.M. 6 P.M. 8 P.M. 10 P.M. Midnight

66 71 68 65 58 54 53
If you were to fit a sinusoidal function to this set of
data, what is the vertical shift? the amplitude? the
period? the frequency? What is the equation of the
resulting sinusoidal function?
Gord.3896.07.pgs 4/28/03 11:55 AM Page 513

7.2 Modeling Periodic Behavior with Sine and Cosine 513

Month Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec

Avg. daily high temp. (F) 65.2 64.4 65.9 67.8 68.6 71.3 75.6 77.6 76.8 74.6 69.9 66.1

20. The table above shows the average daytime high a. Assuming that the temperature behavior in Dallas
temperature each month in San Diego. is periodic from year to year, determine a sinu-
a. Construct a sinusoidal function that best fits soidal function that models the average daytime
these data. high temperature in Dallas.
b. How does the phase shift for this function com- b. The values shown in the table are temperatures
pare to the phase shift used in the text for the roughly every 2 weeks, but two entries are miss-
number of hours of daylight in San Diego? In ing. Use your model from part (a) to predict the
particular, explain in practical terms why the si- average daytime high temperature in Dallas on
nusoidal function for air temperature lags be- the missing dates.
hind the function for hours of daylight.
21. The table gives the average daytime high tempera-
ture in Dallas on different days of the year (roughly
every 2 weeks), based on historical weather records.

Day 1 15 32 46 60 74 91 105 121 135 152

Avg. daily high temp. (F) 55 53 56 59 63 67 72 77 81 84 89

Day 196 213 227 244 258 274 288 305 319 335 349

Avg. daily high temp. (F) 98 99 98 94 90 85 80 72 66 61 58

22. The table shows the average number of tornados re-

ported in the United States per month based on his-
torical records.

Month Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec

Tornados 16 24 60 111 191 179 96 66 41 26 31 22

Determine a sinusoidal function that models the would be the effect of a planet in orbit about the
monthly number of tornados as a function of time. pulsar.
23. For a normal adult at rest, the rate R, in liters per sec- a. From the figure, estimate the length of the year
ond, at which air flows in and out of the lungs can be for the planet.
modeled by the function R1t2  0.85 sin3 12p>52t4, b. Use Kepler’s law from Example 4 in Section 3.6
where t is measured in seconds. The person is inhal- (assuming that the same coefficient applies) to
ing when R 0 and exhaling when R
0. How calculate the distance from this planet to its star.
many times does the person breathe per minute? c. Assuming that the orbit of this new planet is
24. Astronomers recently reported the discovery of the circular, how fast is it moving in its orbit about
first known planets outside the solar system. They the pulsar?
found three worlds orbiting around a pulsar, a ro- d. For comparison, Earth takes 365 days to com-
tating star that emits radiation with constant fre- plete one revolution about the sun at a distance
quency. For this pulsar, the astronomers detected of about 93 million miles. How fast is Earth
slight variations in the intensity of the radiation, as moving in its orbit about the sun?
shown in the accompanying figure. This variation
Gord.3896.07.pgs 4/28/03 11:55 AM Page 514

514 CHAPTER 7 Modeling Periodic Behavior

Variations (milliseconds) 26. A Ferris wheel is 12 meters in diameter and com-

0.005 pletes one full revolution every 20 seconds. If the
bottom of the Ferris wheel is 2 meters above the
0.003 ground, write a formula for the height above
ground of a person on the Ferris wheel as a func-
0.001
0 Days
tion of time.
5 10 15 20 25 30 35 40
–0.001 27. Certain stars are called variable stars because their
brightness increases and decreases in a periodic
–0.003
manner. The brightest variable star that can be ob-
–0.005 served from Earth is Delta Cephei, whose bright-
ness varies between a minimum brightness (or
magnitude) level of 3.65 and a maximum bright-
25. You are taking a ride on a Ferris wheel that is 200 ness of 4.35, with a cycle of 5.4 days. Write an equa-
feet in diameter and has a bottom point 10 feet tion that represents the brightness of Delta Cephei
above the ground. Suppose that the Ferris wheel ro- as a function of time t based on t  0 at an instant
tates twice every minute and, from your friend’s when the star has minimal brightness.
viewpoint on the ground, is rotating clockwise.
28. Many people believe that a person’s life is deter-
200 ft mined by three independent cycles, called bio-
rhythms. One cycle, with a period of 23 days,
represents the physical or health dimension of a
person, H1t2  sin12pt>23 2, where time t is meas-
ured in days starting at birth. A second cycle, with
a period of 28 days, represents the emotional or
sensitivity aspects of a person, E1t2  sin12pt>28 2.
A third cycle, with a period of 33 days, represents
the mental or intellectual aspects of an individual,
M1t2  sin12pt>33 2 .
a. Suppose that Tony was born on January 1.
Consider the 60-day period immediately fol-
10 ft
lowing his 20th birthday. What set of values for
t are appropriate?
a. Sketch your height y above the ground as a func- b. Which days would you recommend as being
tion of time t. suitable for Tony to compete in a track-and-
b. Find a formula for your height y as a function field meet?
of t. Does it agree with your rough sketch in c. Which days would you recommend as being
part (a)? good days for Tony to ask his girlfriend to
c. Find a formula for the horizontal distance x marry him?
from the center of the wheel as a function of t. d. Which days could you suggest as days on which
d. Find all intervals of t values for which you are Tony could hope to have a major exam at school?
moving forward. Indicate these intervals on the e. Are there any days when you would recommend
graph of the function in part (c). What do you that Tony simply not get out of bed?
observe? f. Are there any days when all the signs are highly
e. Suppose that the Ferris wheel rotates in the op- positive?
posite direction, so it is now moving counter- 29. As part of a study on the possibility of global
clockwise. How do your answers to parts (c) and warming at a National Science Foundation math
(d) change? modeling workshop at Pellissippi State College,
f. Find a formula relating your height y above the the accompanying scatterplot was produced. It
ground and the horizontal distance x from the suggests that the average global temperature val-
vertical axis through the center of the wheel. ues appear to oscillate about the regression line,
Gord.3896.07.pgs 4/28/03 11:55 AM Page 515

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 515

T  0.0042t  14.67, where t represents years T

since 1880.
15.2
a. Use the scatterplot to estimate the parameters 15.1
for a sinusoidal function that oscillates above

Temperature (°C)
15.0
and below the indicated line. What is the equa-
tion of the resulting function? 14.9
b. Use your function grapher to draw the graph of 14.8
that function. Does it have the correct shape? 14.7
c. What is your prediction for the average global
14.6
temperature in 2005, based on the combination
t
of the given linear function and the sinusoidal 0 10 20 30 40 50 60 70 80 90 100 110
function you created? Time (years since 1880)

7.3 Solving Equations with Sine and Cosine:

The Inverse Functions
We have shown that the number of hours H of daylight in San Diego as a function
of the day of the year t is given by

H1t2  12  2.4 sin c 1t  802 d .

2p
365
Suppose that we now ask: When will there be 13 hours of daylight? That is, on
which day t of the year will H  13? To find this date, as illustrated in Figure 7.31,
we must solve for the independent variable t in the equation

H1t2  12  2.4 sin c 1t  802 d  13.

2p
365

13
12

t=?
t
80
FIGURE 7.31 (March 21)

We can solve such an equation in a variety of ways, as we illustrate in Examples

1 and 2.

E XAMPLE 1
Determine graphically all days on which there will be 13 hours of daylight in San Diego.
Solution The function H1t2 oscillates between 9.6 and 14.4, so there are 13 hours of
daylight on two different days each year, as indicated in Figure 7.31. One of these days
Gord.3896.07.pgs 4/28/03 11:55 AM Page 516

516 CHAPTER 7 Modeling Periodic Behavior

occurs in the spring when the days are lengthening; the other occurs during the fall when
the days are shortening. When we trace the graph of this function, we find that the two
solutions are t  105 and t  238. The 105th day of the year is April 15 (31 days in
January  28 in February  31 in March  15 in April  105). The 238th day of the
year is August 26. Moreover, these same values will occur every year because the function
is periodic. Check this result on your calculator by evaluating H1105 2 and H1238 2.
◆

E XAMPLE 2
Determine algebraically when there will be 13 hours of daylight in San Diego.
Solution We start with the equation

H1t2  12  2.4 sin c 1t  802 d  13.

2p
365
To solve algebraically for t, we first subtract 12 from both sides:

2.4 sin c 1t  80 2 d  1.
2p
365
We next divide both sides by 2.4:

sin c 1t  80 2 d 
2p 1
 0.417.
365 2.4
Our task now is to extract the variable t from the argument of the sine function.
Compare this problem to the situation we repeatedly faced of extracting the variable
from an exponential function, such as 10x. We solved that problem by using logarithms
to undo the exponential function. The reason that this method works is because the ex-
ponential and logarithmic functions are inverse functions of one another.
Similarly, we can undo the sine function by using the inverse sine function. You can
do so on your calculator by pressing either INV or 2nd followed by SIN to get the arcsine
function. When you do this in radian mode, you will find that
arcsin 0.417  0.430;
that is, the value whose sine is 0.417 is 0.430 radians. Therefore

1t  80 2  arcsin 0.417  0.430.

2p
365
To solve for t, we now multiply both sides by 365 and get
2p1t  80 2  10.4302 1365 2  156.95.
Dividing both sides by 2p yields
156.95
t  80   24.98  25.
2p
Hence
t  25  80  105.
That is, there will be 13 hours of daylight in San Diego on approximately the 105th day
of the year (April 15), which is the same answer we obtained graphically in Example 1.
◆
Gord.3896.07.pgs 4/28/03 11:55 AM Page 517

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 517

Actually, the result in Example 2 is not complete because it is only one of the
two possible days each year on which the sinusoidal function H passes across the
13-hour level. However, this is the only solution that you can get directly from a
calculator or computer when you use the inverse sine function. You can determine
the other day when 13 hours of daylight occurs by using the following line of rea-
soning, based on some key facts about the sine function. We found that the solu-
tion t  105 corresponds to April 15, which is 25 days after the spring equinox on
March 21. Using the symmetry of the sine curve, we should expect that there will
also be 13 hours of daylight 25 days before the fall equinox on September 21. But
25 days before September 21 is August 27, which is roughly the other solution we
found in Example 1. Finally, because of the periodicity of the sine function, there
will be 13 hours of daylight in San Diego on April 15 and August 27 every year.

The Inverse Sine Function

Let’s now examine more carefully what the inverse sine function is all about and
the reason for the limitation in Example 2. Recall that a continuous function f has
an inverse f 1 when it is either strictly increasing or strictly decreasing or, equiva-
lently, if it satisfies the horizontal line test. Obviously, the sine function does not
fulfill either of these conditions because of its shape. The only way to obtain an in-
verse for the sine function is to restrict its domain, as we did for the parabola in
Section 2.9 when we considered only the right side of the parabola y  x 2. We thus
use only a small portion of the sine curve y  sin u—where the function is strictly
increasing. By convention, the restricted domain for the sine function is from
u  p>2 to u  p>2, as depicted in Figure 7.32.

y = sin θ

θ
– π2 π
2

FIGURE 7.32 –1

Let’s work temporarily in degrees. Suppose that

sin u  0.825
and u is some unknown value. From the graph of the sine function, we expect that
u will be closer to 90° than to 0°. The inverse sine function, arcsin y or Sin1 y, al-
lows us to solve for the correct value of u. That is, if
sin u  0.825,
then
u  arcsin 10.8252  55.59°.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 518

518 CHAPTER 7 Modeling Periodic Behavior

Equivalently, if we use radians instead of degrees, then

u  arcsin 10.825 2  0.9702 radians.
Thus the inverse sine function undoes the sine function to extract any value u be-
tween 90° and 90° or, equivalently, between p>2 and p>2 radians. Check this
result by taking the sine of 55.59° or 0.9702 radians. It may be helpful to think of
arcsin y as the angle whose sine is y.
In general, for any function f, if y  f 1x 2 and x  g1y2  f 1 1y2 are inverse
functions, we know that
f 1 1 f 1x 2 2  x, for all x in the domain of f ;
f 1 f 1 1y2 2  y, for all y in the domain of f 1.
In the case of the sine function and its inverse, we have
arcsin1sin u 2  u,
p p
 u ;
2 2
sin1arcsin y2  y, 1  y  1.
As we found previously, when we wanted to find the day on which there would
be 13 hours of daylight in San Diego, the inverse sine function returned only
t  105. This is the only answer we get because we must restrict the domain in
order to have an inverse function.
Let’s try a few values for y to see the effects of the inverse sine function.
If y  0.2, arcsin 0.2  11.5°.
If y  0.6, arcsin 0.6  36.9°.
If y  0.95, arcsin 0.95  71.8°.
If y  0.4, arcsin10.4 2  23.6°.
If y  0.88, arcsin10.882  61.6°.
What if you try to find arcsin12.5 2? Most calculators will return an error mes-
sage, usually indicating a problem with the domain. The reason is that you are try-
ing to find a number whose sine is 2.5. But the only permissible values for the sine
function are between 1 and 1. Thus, if you try to use any value outside this inter-
val, the inverse sine function is not defined and you will get an error message on
most calculators. (Some models return a complex number instead of an error mes-
sage, but that is a topic for considerably more advanced mathematics courses.)
You have seen that when you use any value for y between 1 and 1, the inverse
sine function returns an answer between 90° and 90° in degree mode or between
p>2 and p>2 in radian mode. These values are called the principal values of the
inverse sine. The inverse sine function does not give any values larger than 90° (or
p>2) or smaller than 90° (or p>2). It is up to you to realize that, for any real
number k between 1 and 1, there are infinitely many values for u whose sine is k.
You can find these values by visualizing the graphs of the sine function and the
horizontal line y  k and determining all points where they intersect, that is,
where sin u  k. You can also use what you know about the symmetry of each arch
of the sine curve. Finally, you can always estimate these values graphically, an ap-
proach that is usually straightforward. In Examples 3–5 we demonstrate the alge-
braic approach and the associated reasoning, because it is more difficult than the
graphical approach.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 519

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 519

E XAMPLE 3
Find all values of u, in degrees, for which sin u  0.6.
Solution If y  sin u  0.6,
arcsin 0.6  36.9°  37°,
which occurs while the sine curve is increasing. We have to find the second solution
which occurs while the sine curve is decreasing. We know that the sine curve reaches its
maximum height of 1 at 90° and is symmetric about 90°. Therefore, because 37° is 53°
before 90°, a second value at which the sine function reaches the same height of 0.6 oc-
curs 53° after 90°, or at u  143°, as shown in Figure 7.33. (Verify that sin 143°  0.6
with your calculator.)
y

1
y = sin θ

0.6

53° 53°
θ
FIGURE 7.33 0 37° 90° 143° 180°

Further, because of the periodicity of the sine curve, we know that the same be-
havior pattern will repeat every 360°. Therefore, other values of u whose sine is 0.6
are at 37°  360°  397°, at 143°  360°  503°, at 37°  21360°2  757°, at 143° 
21360°2  863°, and so on. Verify some of these values with your calculator as well.
◆
In Example 5 of Section 7.2, we created the sinusoidal function

T  71  3 sin a tb
p
10
to model the temperature in a house where the air conditioning control is set to
turn on the air conditioner when the temperature rises to 74°F and to turn it off
when the temperature drops to 68°F, a cycle that repeats every 20 minutes. Also, we
were told that, at noon, the temperature was 71°F and rising. We now consider
some inverse predictions based on this model.

E XAMPLE 4
Use the sinusoidal model to determine all times between noon and 1 P.M. when the tem-
perature in the house is 70°F.
Solution We need to solve the equation

T  71  3 sin a
p
tb  70.
10
We show an algebraic solution. We first subtract 71 from both sides and get

3 sin a tb  1 so that sin a tb   .

p p 1
10 10 3
Gord.3896.07.pgs 4/28/03 11:55 AM Page 520

520 CHAPTER 7 Modeling Periodic Behavior

Consequently, in radians

t  arcsin a b  0.3398,
p 1
10 3
so that
10.33982  1.08,
10
t
p
or about 1 minute before noon. The air conditioning cycle takes 20 minutes, so we con-
clude that the temperature will be 70° again about one minute before 12:20, or at about
12:19, again at about 12:39, and once more at about 12:59. All these values occur while
the sine curve is increasing, as shown in Figure 7.34.
To find the times that the temperature is 70° while the sine curve is decreasing, we
reason as follows. A complete cycle takes 20 minutes, so a half cycle takes 10 minutes.
The first time the temperature reaches 70° while the curve is increasing is at about
t  19 minutes, or 1 minute before the end of the cycle. Therefore, from the symmetry of
the sine curve over the first 20 minutes, as illustrated in Figure 7.35, the first time the
sine curve passes the 70° level while the curve is decreasing must occur about 1 minute
after the middle (t  10 minutes) of the cycle. That is, the other solution is at about
t  11 minutes, or at about 12:11. Because of the periodic nature of the sine function,
the 70° temperature will also occur at about 12:31 and at about 12:51.

T
π T
75
T = 71 + 3 sin ( t)
10

π
T = 71 + 3 sin ( t)
Temperature (°F)

10
73

71
71°
70°
69

1 1
t
0 10 20 30 40 50 60 t
10 11 19 20
Minutes since noon

FIGURE 7.34 FIGURE 7.35

◆
E XAMPLE 5
In Example 6 of Section 7.2, we created the sinusoidal function

C  4800  2400 sin a

p
tb
12
to model a cricket’s chirp rate C in chirps per hour, where t is measured in hours since 9
A.M. At what times, if any, does the cricket chirp at a rate of 6000 times per hour?

Solution We first solve this problem graphically. The graph of C over a 24-hour period
starting at 9 A.M. (when t  0) is shown in Figure 7.36. Note the horizontal line at a
height of 6000. The times when the cricket chirps at a rate of 6000 times per hour are the
points at which the curve intersects the line. If we zoom in on the graph about these two
points, we find that t is approximately t  2 hours after 9 A.M. (or about 11 A.M.) and
t  10 hours after 9 A.M. (or about 7 P.M.).
Gord.3896.07.pgs 4/28/03 11:55 AM Page 521

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 521

C
π
7200
C = 4800 + 2400 sin ( t)
12

6000

Chrips per hour

4800

3600

2400

t
0 4 8 12 16 20 24
FIGURE 7.36 Hours

Alternatively, let’s solve for t algebraically from

C  4800  2400 sin a

p
tb  6000.
12
We subtract 4800 from both sides to get

2400 sin a
p
tb  1200
12
so that

sin a
p 1200
tb   0.5.
12 2400
Using the inverse sine function in radian mode, we get

t  arcsin10.5 2  0.5236
p
12
so that

10.52362  2.0000,
12
t
p
or 2 hours after 9 A.M., which is 11 A.M., as we found graphically.
To find the other time of day when the cricket is chirping 6000 times per hour, we
note that the period of the sinusoidal function is 24 hours, starting at 9 A.M., so a half
cycle takes 12 hours. The symmetry of the sinusoidal function, as depicted in Figure 7.37,
shows that the other time that the curve passes a height of 6000 must be 2 hours before
the 12-hour mark at 9 P.M. (or 10 hours after 9 A.M.), which is at 7 P.M.

C
π
C = 4800 + 2400 sin ( t)
12

6000

4800
2 2
t
FIGURE 7.37 9 A.M. Time 9 P.M.

◆
Gord.3896.07.pgs 4/28/03 11:55 AM Page 522

522 CHAPTER 7 Modeling Periodic Behavior

We summarize the important properties of the inverse sine function as follows.

Properties of the Inverse Sine Function y  arcsin x

1. arcsin x is defined only for values of x (the domain) between 1 and 1.
2. The principal values for y (the range) lie between p>2 and
p>2 radians.
3. arcsin 1sin y2  y,
p p
for   y  .
2 2
4. sin 1arcsin x 2  x, for 1  x  1.

In general, suppose that f 1 is the inverse of a function f. If we write the two

functions in terms of the same independent variable x so that y  f 1x 2 and
y  f 1 1x 2, their graphs are mirror images of each other about the line y  x, as
we showed in Section 2.9. (Visualize the graphs of the exponential and logarith-
mic functions.) Using this fact, we can easily construct the graph of the inverse
sine function from our knowledge of the graph of the sine function. We begin
with the sine curve shown in Figure 7.38(a), in which the portion corresponding
to the domain from p>2 to p>2 is highlighted. When we reflect the highlighted
portion of the sine curve about the line y  x, we get the graph of the inverse sine
function; both graphs are shown in Figure 7.38(b). Finally, the graph of the in-
verse sine function alone is shown in Figure 7.38(c). Note that y  arcsin x exists
only for values of x between 1 and 1 and that the corresponding heights range
from p>2 to p>2.

y y y
y=x
y = arcsin x
π y = arcsin x
1 1 2

y = sin x y = sin x
x x x
– π2 π
2
– π2 –1 1 π
2
–1 1

–1 –1 – π2

FIGURE 7.38 (a) (b) (c)

The Inverse Cosine Function

Suppose that
cos u  0.3
and we want to find a number u whose cosine has this value. To solve this equation,
we introduce the inverse cosine function, y = arccos x, using ideas that are similar to
those that led to the inverse sine function. We use the inverse cosine (press 2nd
COS or INV COS on your calculator, in Degree mode) to get
u  arccos 0.3  72.5°.
Verify that cos 72.5°  0.3.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 523

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 523

We know that the only possible values for y  cos u lie between 1 and 1.
For the inverse cosine function u  arccos y, we must restrict the domain to a suit-
able portion of the cosine curve where the cosine function is either strictly increas-
ing or strictly decreasing. By convention, we consider only values of u between 0°
and 180° (or equivalently between 0 and p radians) where the cosine function is
strictly decreasing, as shown in Figure 7.39. These are the principal values for the
inverse cosine and they are the only values that your calculator will return when
you use the inverse cosine function. As with the inverse sine function, you will have
to use what you know about the behavior of the cosine graph, including the sym-
metry of the arches on the graph, if you want to determine all other numbers hav-
ing the specified cosine value.

1
y = cos x

θ
0 π 2π

FIGURE 7.39 –1

E XAMPLE 6
Find all values of u in both degrees and radians for which cos u  0.92.
Solution Although we can solve this problem graphically, we illustrate the details of
the algebraic solution. Because cos u  0.92,
u  arccos10.92 2  23°
in degree mode. Because of the periodicity of the cosine function, we also know that the
value of 0.92 will repeat every 360°, so the solutions include
u  23°, 23°  360°, 23°  21360°2 , 23°  31360°2, . . . ,
or, in general,
u  23°  n # 1360°2, for any integer n  0.
Moreover the graph of the cosine function shown in Figure 7.40 indicates that there
must be another value of u just before 360° whose cosine is also 0.92. In particular, be-
cause our first solution is u  23°, the other value must be 23° before 360°, or
360°  23°  337°. Thus the solutions also include
u  337°, 337°  360°, 337°  21360°2, 337°  31360°2 , . . . ,
or, in general,
u  337°  n # 1360°2, for any integer n  0.
Gord.3896.07.pgs 4/28/03 11:55 AM Page 524

524 CHAPTER 7 Modeling Periodic Behavior

y
y = cos θ
0.92

23°

θ
0 23° 337° 360°

FIGURE 7.40

Using radian measure gives the equivalent solutions

u  0.401  2np radians and u  5.882  2np radians, for any integer n  0.
Moreover, the cosine function is symmetric about the vertical axis u  0 (it is an
even function; see Appendix D). Therefore we know that the same patterns repeat with
negative values for u, so all the other solutions are
u  23°  n # 1360°2 and u  337°  n # 1360°2 , for any integer n  0.
Equivalently, with radian measure, u  0.401  2np radians and u  5.882  2np,
for any integer n  0.
◆
E XAMPLE 7
A clock is mounted on the wall with its center 7 feet (or 84 inches) above the floor. Sup-
pose that the minute hand is 5 inches long.
a. Write a formula for the vertical height y of the arrowhead on the minute hand above or
below the horizontal line through the center of the clock as a function of the time t in
minutes from the instant that the minute hand is pointing vertically upward to the 12.
b. Determine all times during the first hour when the arrowhead on the minute hand is
2 inches above that horizontal line.
Solution
a. The midline, or vertical shift, is 84 inches above floor level. The minute hand is 5 inch-
es long, so the height of the arrowhead on the hand oscillates between 84  5  79
and 84  5  89 inches above the floor, giving an amplitude of 5. This cycle repeats
every 60 minutes, so the period is 60 and the frequency is 2p>60  p>30. Because
t  0 corresponds to the instant when the hand is pointing vertically upward, the ini-
tial height for the arrowhead is y  84  5  89 inches. This suggests that we use a
cosine function with a phase shift of 0 as our model. The resulting function is

y  84  5 cos a tb .
p
30
b. To find all times when the arrowhead is 2 inches above the midline (equivalently,
when the arrowhead is 84  2  86 inches above the floor), we need to solve the
equation

84  5 cos a
p
tb  86.
30
Gord.3896.07.pgs 4/28/03 11:55 AM Page 525

7.3 Solving Equations with Sine and Cosine: The Inverse Functions 525

Proceeding algebraically, we have

5 cos a tb  2 so that cos a tb   0.4.

p p 2
30 30 5
Therefore, in radians,
p
t  arccos 0.4  1.1593
30
so that

11.15932  11.07,
30
t
p
or about 11 minutes after the hour, as illustrated in Figure 7.41.

y
π
Height (inches) 89
y = 84 + 5 cos ( t)
30

86
84

79

t
0 11 49 60
FIGURE 7.41 Time (minutes)

Because the first solution occurs at about 11 minutes after the hour, we know from
the symmetry of the cosine curve that the same height must occur at about 11 minutes
before the next hour, or at t  49.
◆
We summarize the important properties of the inverse cosine function as follows.

Properties of the Inverse Cosine Function y  arccos x

1. arccos x is defined only for values of x (the domain) between 1 and 1.
2. The principal values for y (the range) lie between 0 and p radians.
3. arccos 1cos y2  y, for 0  y  p.
4. cos 1arccos x 2  x, for 1  x  1.

Finally, as with the graph of the inverse sine function, the graph of the inverse
cosine function y  arccos x is the mirror image of the cosine graph about the line
y  x, as shown in Figure 7.42. Note that the inverse cosine is defined only for x
between 1 and 1 and that the inverse cosine values lie between 0 and p.
As a final note, you may find the names arcsine and arccosine to be rather
strange. To see where they come from, think about how we defined radians. In a
unit circle, we measured a length of 1 and defined the corresponding angle to be
1 radian. The same is true for any angle—its measure in radians equals the length
of arc along the unit circle. So, to solve sin u  a, say, we find the angle u that
equals the length of an arc on the circle corresponding to the value of a. Thus we
have arcsin a.
Gord.3896.07.pgs 4/28/03 11:56 AM Page 526

526 CHAPTER 7 Modeling Periodic Behavior

y=x
π
y = arccos x

y = cos x
x
–1 1

FIGURE 7.42 –π

Problems
1. On what days of the year will San Diego have d. Suppose that the temperature at noon is 68°. Re-
11 hours of daylight? 10 hours of daylight? 9 hours peat parts (a) and (b).
of daylight? 5. One of the dangers at places that have very high
2. The height of water at a dock is given by the formu- tides, such as Canada’s Bay of Fundy, is the rate at
la h1t2  10  4 sin1pt>62, where t is measured in which the tide can come in and potentially trap un-
hours since midnight. wary visitors. Use the formula you devised for a si-
a. When does high tide occur? nusoidal function that models the heights of the
b. When does low tide occur? tides at the Bay of Fundy in Problem 7 of Section 7.2
c. When does the water level reach 8 feet? 10 feet? to determine how long it takes for the water level to
11 feet? 12 feet? rise 5 feet
3. An air conditioner is being used to cool a room. a. from a point of low tide.
The temperature T oscillates according to the for- b. from a point at the average tide level.
mula T1t2  69  3 sin1pt>102, where t is meas- 6. A Ferris wheel is 12 meters in diameter and com-
ured in minutes after 9 A.M. pletes one full revolution every 20 seconds. The
a. At about what temperature is the thermostat set? bottom of the Ferris wheel is 2 meters above the
(i.e., when does the air conditioner kick in?) ground. In Problem 26 of Section 7.2, you were
b. At about what temperature does the air condi- asked to write a formula for the height above
tioner kick out? ground of a person on the Ferris wheel as a func-
c. When does the room temperature reach 70°F? tion of time. Use that model to determine the times
67°F? at which a person is 10 meters above the ground.
4. The thermostat in an apartment is set to turn the 7. The historical average daytime high temperature in
heat on when the temperature falls to 64°F and to Fairbanks ranges from a low of 20°F to a high of
turn it off when the temperature rises to 70°F. This 64°F, and the coldest day of the year, historically, is
cycle takes 15 minutes. the 40th day. In Problem 15 of Section 7.2, you were
asked to write a formula for a sinusoidal function
a. Write a formula for the temperature T as a func-
that can be used to model the average daytime high
tion of time t, where t is the number of minutes
temperature in Fairbanks as a function of the day of
after noon. Assume that the temperature at
the year. Use this model to determine the days on
noon is 70°.
which the high temperature in Fairbanks will be 0°.
b. Determine all times between noon and 1 P.M.
when the temperature is 66°. 8. The table below gives the outdoor temperatures in
c. Suppose that the temperature at noon is 67°. Re- Chicago during one 24-hour period:
peat parts (a) and (b).

Time Midnight 2 A.M. 4 A.M. 6 A.M. 8 A.M. 10 A.M. Noon 2 P.M. 4 P.M. 6 P.M. 8 P.M. 10 P.M. Midnight

Temp.(F) 53 48 47 49 53 59 66 71 68 65 58 54 53
Gord.3896.07.pgs 4/28/03 11:56 AM Page 527

7.4 The Tangent Function 527

In Problem 19 of Section 7.2, you were asked to cre- 20 of Section 7.2, you were asked to construct a si-
ate the equation of the sinusoidal function that fits nusoidal function that fits these data. Use that
these data. Use that model to determine the times at model to determine the months on which the aver-
which the temperature in Chicago will be (a) 50° age daytime high temperature in San Diego will be
and (b) 60°. (a) 65°, (b) 70°, and (c) 80°.
9. The table below shows the average daytime high
temperature each month in San Diego. In Problem

Month Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec

Avg. daily high temp. (F) 65.2 64.4 65.9 67.8 68.6 71.3 75.6 77.6 76.8 74.6 69.9 66.1

10. A 25-foot ladder is leaning against the side of a to the building. Use your function grapher to draw
building and begins to slip. Write a formula for the the graph of this function. What are appropriate val-
angle u that the ladder makes with the ground as a ues for the domain of the function? Is the graph con-
function of the distance x from the foot of the ladder cave up or concave down? When is it maximum?

Exercising Your Algebra Skills

Several of the following equations do not have solu- 5. 4 sin u  3 6. 5 sin 2x  3
tions. By inspection, decide which ones do not (give 7. 5 cos 2x  3 8. 3 cos x  2
reasons) and then find the solutions to the remaining
9. 5 sin 2x  3 cos x
equations.
1. sin u  4 2. sin u  0.4
3. 3 sin u  4 4. 4 sin u  3

7.4 The Tangent Function

We have considered many situations that can be modeled with either the sine or
cosine function. In this section we return to the third trigonometric function, the
tangent, and consider its properties and some applications. Recall from Section 6.1
that the tangent is defined by the ratio
opposite
tan u 
adjacent
of sides in a right triangle in terms of an angle u, as shown in Figure 7.43.

The Graph of y  tan x

As with any function, our first concern is to determine the graph of the tangent
function to help us understand its behavior. The graph of the tangent function is
shown in Figure 7.44, from which we observe the following characteristics.
1. The tangent function is periodic with period p; the tangent graph com-
pletes one full cycle between x  3p>2 and p>2, another full cycle be-
tween p>2 and p>2, a third cycle between p>2 and 3p>2, and so on.
Each segment is called a branch of the graph.
2. The tangent function has zeros at x  0, p, 2p, . . .
Gord.3896.07.pgs 4/28/03 11:56 AM Page 528

528 CHAPTER 7 Modeling Periodic Behavior

opposite
tan θ =
adjacent y = tan x

opposite x
– 32π – π2 π
2
3π
2

θ
adjacent

FIGURE 7.43 FIGURE 7.44

3. The tangent graph has vertical asymptotes at x  p>2, 3p>2,

5p>2, . . . , so the tangent function is not defined at these points (which
separate the branches).
4. The tangent is an increasing function for all intervals between successive
vertical asymptotes.
5. The tangent curve is first concave down and then concave up on each
branch. The point of inflection on each branch occurs at the point where
the tangent curve crosses the x-axis: at x  0, p, 2p, . . .
To understand the behavior of the tangent function, we consider the funda-
mental relationship.

sin x
tan x  , for any value of x for which cos x 0.
cos x

Because the tangent is the quotient of two functions, we can analyze this rela-
tionship in the same way that we analyzed the behavior of rational functions in
Section 4.6. First, the tangent function must have a zero wherever the numerator,
sin x, is 0. This corresponds to x  0, p, 2p, . . . , which clearly agrees with
what the graph of the tangent function shows. Second, the tangent function is un-
defined and therefore has a vertical asymptote wherever the denominator, cos x, is
0. This occurs at x  p>2, 3p>2, 5p>2, . . . , which again agrees with what
the graph of the tangent function shows.
Now let’s see how these ideas help in understanding the graph of the tangent
function. Consider what happens between x  0 and p>2. The sine function
(the numerator for y  tan x ) is positive and increasing toward 1, whereas the
cosine function (the denominator for y  tan x ) is positive and decreasing to-
ward 0. Because both are positive, tan x must be positive between 0 and p>2.
Also, the ratio involves a numerator that is getting larger and a denominator
that is getting smaller and approaching 0, so there is a vertical asymptote at
x  p>2. The tangent is a positive function that increases toward  as x ap-
proaches p>2 from the left.
Similarly, between x  p>2 and x  0, the sine function is negative and in-
creasing toward 0, whereas the cosine function is positive and increasing toward 1.
Gord.3896.07.pgs 4/28/03 11:56 AM Page 529

7.4 The Tangent Function 529

Thus their ratio is negative and increases toward 0 as x increases toward 0. More-
over, as x approaches p>2 from the right, the tangent ratio becomes ever more
negative and eventually approaches .
Next, why is the period of the tangent function p when the periods for the
sine and the cosine are both 2p? Visualize the sine curve from 0 to p and then
from p to 2p, as shown in Figure 7.45. If you flip the second half of the curve
over the x-axis, you get a curve identical to the first half. So the values for sin x
between p and 2p are the same as those between 0 and p, but with the signs re-
versed. The same is true for the cosine between p and 2p—its values repeat
those for the cosine between 0 and p, but with the signs reversed, as shown in
Figure 7.46.

y y
y = sin x y = −sin x y = cos x y = −cos x
1 1

x x
0 π 2π 0 π 2π

–1 –1

FIGURE 7.45 FIGURE 7.46

So, when we take the ratio sin x>cos x for x between p and 2p 1x 3p>2 2, the
numerator and the denominator have the same numerical values as the ratio of
sin x>cos x for x between 0 and p 1x p>2 2 , but the signs of both the numerator
and the denominator are reversed. In the quotient, these reversed signs cancel, so
that the values for tan x from p to 2p match the corresponding values for tan x
from 0 to p. But, if the same values are repeated, the function is periodic and there-
fore its period is p.
Finally, because the tangent function is periodic with period p, it repeats
the behavior that we’ve outlined here, leading to the graph previously shown in
Figure 7.44. For reference purposes, you should know that

sin 0 0 p
tan 0    0 and tan  tan 45°  1.
cos 0 1 4

E XAMPLE 1
A video cameraman is taping a 100 meter dash down a straight track. He is positioned
halfway along the track 40 meters from the inside lane where the race’s favorite is run-
ning. He plans to focus his camera on the favorite throughout the race.
a. Write a formula for a function that models the distance d from the runner to the
point A on the track as a function of the angle u, as illustrated in Figure 7.47.
b. What is the runner’s distance from the line extending from the cameraman to the
middle of the track when the angle u  30°?
c. What is the runner’s distance from that line when u  0.6 radian?
Gord.3896.07.pgs 4/28/03 11:56 AM Page 530

530 CHAPTER 7 Modeling Periodic Behavior

A
d
40 feet
θ

FIGURE 7.47

Solution
a. From Figure 7.47,

so that d  40 tan u  f 1u2 .

d
tan u 
40

b. When u  30°,

d  f 130°2  40 tan 30°  23.09 meters.

c. When u  0.6 radian,

d  f 10.6 2  40 tan 0.6  27.37 meters.

◆

We can write more general tangent functions of the form

y  D  A tan3 B1x  C2 4,

where D is the vertical shift or midline, A is the amplitude, B is the frequency, and
C is the phase shift. These ideas are the same as those that we encountered with
general sinusoidal functions in Section 7.2. We explore some of these ideas for the
tangent function in the Problems at the end of this section.

The Inverse Tangent

Suppose that we have an equation such as tan u  1.5. To find a value of u that sat-
isfies this equation, we use the inverse tangent function, arctan x, that gives the
number whose tangent value is x. Using a calculator, we find

u  arctan 1.5  0.9828 radian  56.31°.

As with the inverse sine and inverse cosine functions, we have to restrict the
domain of the tangent function in order to define the inverse tangent function.
By convention, the principal values for the tangent function are from p>2 to
p>2 where the tangent function is strictly increasing. Accordingly, a calculator
returns a value only between p>2 and p>2 (or between 90° and 90°) for the
inverse tangent.
We summarize the important properties of the inverse tangent function as
follows.
Gord.3896.07.pgs 4/28/03 11:56 AM Page 531

7.4 The Tangent Function 531

Properties of the Inverse Tangent Function y  arctan x

1. arctan x is defined for values of x (the domain) between  and .
2. The principal values for y (the range) lie between p>2 and p>2
radians.
3. arctan 1tan y2  y, for p>2
y
p>2.
4. tan 1arctan x 2  x, for 
x
.

Finally, as with the graphs of the other inverse functions, the graph of the in-
verse tangent function y  arctan x is the mirror image of the tangent graph about
the line y  x, as shown in Figure 7.48, where x goes from 15 radians to 15 radi-
ans. Note how the curve levels off to the right at a height of p>2  1.57 and to the
left at a height of about p>2  1.57. These are a pair of horizontal asymptotes.

2
π
2 y = arctan x

x
−15 0 15

– π2

−2
FIGURE 7.48

If you need other values of u outside the interval from p>2 to p>2, you will
have to determine them by using what you know about the symmetry of the graph
of the tangent function.
E XAMPLE 2
You enter a movie theater that has a screen 20 feet high positioned 5 feet above your eye
level. If you sit too far back in the theater, the screen appears too small because your
viewing angle is too small. If you sit too close to the screen, the picture will seem distort-
ed because your viewing angle is again too small.
a. Find a formula giving the viewing angle u as a function of your distance d from the
screen, as illustrated in Figure 7.49.
b. What is your viewing angle u if you sit 40 feet back from the screen?

20

θ
α 5
FIGURE 7.49 d
Gord.3896.07.pgs 4/28/03 11:56 AM Page 532

532 CHAPTER 7 Modeling Periodic Behavior

c. Does the viewing angle increase or decrease if you move farther back? In particular, esti-
mate the distance you should sit from the screen to get the largest possible viewing angle.
Solution
a. Your viewing angle u is the angle subtended by the screen. There is no direct way to
get an expression for u, because the associated triangle shown in Figure 7.49 is not a
right triangle. Thus you have to get an expression for u in a somewhat indirect man-
ner. To do so, introduce the angle a shown in Figure 7.49 representing the angle from
your eye level vertically upward to the bottom of the screen. This gives you two right
triangles. In the smaller triangle,
5 5
tan a  so that a  arctan .
d d
In the larger triangle,

tan 1u  a2 
25 25
so that u  a  arctan .
d d
Consequently, the desired expression for u is

u  1u  a 2  a  arctan  arctan  f 1d2 .

25 5
d d
(Note that these two terms cannot be combined algebraically.)
b. For d  40 feet,

u  f 1402  arctan
25 5
 arctan  0.4342 radians,
40 40
or about 24.9°.
c. To determine what happens to this viewing angle u as you move farther back, just re-
place the 40-foot distance with somewhat larger values—say, 41 or 45 feet. Alterna-
tively, graph the function f that gives the angle u as a function of the distance d. If you
graph this function on the interval from 0 to 50, say, as shown in Figure 7.50, you can
determine the behavior of this function for u more thoroughly. The function increas-
es rapidly, starting at d  0, and rises to a maximum viewing angle when d is ap-
proximately 11 feet. You can verify this result on your calculator. Then the function
slowly decreases as d increases thereafter. Therefore if you move farther back from the
screen, the viewing angle will decrease.

d
0 50
FIGURE 7.50

◆
Gord.3896.07.pgs 4/28/03 11:56 AM Page 533

7.4 The Tangent Function 533

Finally, we note that it is possible to consider three other trigonometric

functions—the cotangent, the secant, and the cosecant—which are just reciprocals
of the tangent, the cosine, and the sine functions, respectively. These functions are
defined as
1 1 1
cot u  , sec u  , and csc u  .
tan u cos u sin u
Be sure that you understand that these are reciprocals of the basic trigonometric
functions. They have nothing to do with the associated inverse functions.
The cosecant, secant, and cotangent functions have been useful in the past be-
cause they simplified the hand calculations required in working with certain
trigonometric problems. However, with technology, working with the actual recip-
rocals is just as easy as using csc u, sec u, and cot u. Consequently, the cotangent,
secant, and cosecant functions gradually are being laid to rest, and we don’t con-
sider them further.

Problems
1. Each of the figures (a)–(f) shows the graph of y y
y  tan x, where x is in degrees.
40 40
a. Write an equation for a tangent function with 30 30
20 20
frequency 2 and sketch its graph superimposed 10 10
over the graph of y  tan x in Figure (a). x x
−90 −30 30 60 90 −90 −30 30 60 90
b. Write an equation for a tangent function with −20 −20
frequency 12 and sketch its graph superimposed −30 −30
− 40 − 40
over the graph of y  tan x in Figure (b).
c. Write an equation for a tangent function with (c) (d)
amplitude 3 and sketch its graph superimposed y y
over the graph of y  tan x in Figure (c).
40 40
d. Write an equation for a tangent function with 30 30
amplitude 2 and sketch its graph superim- 20 20
posed over the graph of y  tan x in Figure (d). 10
x
10
x
e. Write an equation for a tangent function with −90 −30 30 60 90 −90 −30 30 60 90
phase shift of 30° and sketch its graph superim- −20 −20
−30 −30
posed over the graph of y  tan x in Figure (e). − 40 − 40
f. Write an equation for a tangent function with ver- (e) (f)
tical shift of 10 and sketch its graph superim-
posed over the graph of y  tan x in Figure (f).
2. Write a possible formula involving tangent func-
tions for each function (a)–(c) shown.
y y y
40 40
30 30
20 20
10 10
x x x
−90 −30 30 60 90 −90 −30 30 60 90 1
−20 −20
−30 −30
− 40 − 40
(a) (b) (a)
Gord.3896.07.pgs 11/21/03 10:50 AM Page 534

534 CHAPTER 7 Modeling Periodic Behavior

y y

25
x
− π4 π
4
π x 54°
2 32°
60 m
(b) (c)

3. Write a possible formula involving a tangent function 6. The lines y  x, y  2x, y  3x, and y  4x all
for the function whose values are given in the table. pass through the origin. Find the angle each line
makes with the x-axis.
7. a. For the general equation of a line through the
U p>3 p>4 p>6 0 p>6 p>4 p>3
origin y  mx, interpret the meaning of the
f (U) UNDEF 2.414 1 0 1 2.414 UNDEF slope m in terms of trig functions.
b. What is the significance of the slope m in
y  mx  b from this point of view?
4. The Statue of Liberty is 46 meters tall and stands on
a base that is also 46 meters tall. Find an expression 8. Use the graph of y  sin x to sketch the graph of its
for the angle subtended by the statue from ground reciprocal function y  1>sin x. (This is the cose-
level as a function of distance from the base of the cant function.)
statue. Use this function to estimate graphically the 9. Use the graph of y  cos x to help you sketch the
distance when the angle is maximum. Approxi- graph of its reciprocal function y  1>cos x. (This
mately what is this maximum angle? is the secant function.)
10. Use the graph of y  tan x to sketch the graph of
y  1>tan x. (This is the cotangent function.)
11. A 5-foot high painting is hanging on the wall of an
art museum when a photographer takes a picture of
46 it. The lens of his camera is 1 foot below the bottom
of the painting when he snaps the picture.

5 ft
46
θ
β
1 ft
α x ft

5. A tall smokestack extends from the roof of a large

industrial plant. At a point 60 meters from the base
of the building, the angle of elevation to the roof a. Find a formula for the angle b subtended by the
(the bottom of the smokestack) is 32°, and the painting at the camera’s lens at a distance of x
angle of elevation to the top of the smokestack is feet from the wall.
54°. Find both the height of the building and the b. Using your function grapher, estimate the dis-
height of the smokestack. tance x from the wall at which the photographer
Gord.3896.07.pgs 4/28/03 11:56 AM Page 535

Chapter Summary 535

should position his camera to subtend the great- where M0 is the mass of the object at rest 1v  0 2
est possible angle with the painting. and c is the speed of light (about 186,282 miles
12. A TV cameraman is videotaping the liftoff of the per second). Suppose that an object has a rest
space shuttle. The cameraman is positioned at mass of M0  1 unit.
ground level 500 meters from the launch pad and is a. Construct a table of values for the mass of the
tracking the shuttle as it rises. object for each of the following speeds expressed
a. Write a formula for the angle of inclination a to as a fraction of the speed of light: v  0, 0.5c,
the shuttle as a function of the height y of the 0.9c, 0.95c, 0.99c, and 0.999c.
shuttle above the ground. b. Sketch a graph showing the behavior of the mass
b. Find the angle of inclination a when the shuttle M of an object as its speed approaches the speed
is 1000 meters high. of light.
c. Find the angle of inclination a when the shuttle c. The speed of light is the physical equivalent of a
is 2000 meters high. vertical asymptote. Write the formula for a func-
tion involving the tangent that can be used to
13. According to Einstein’s theory of relativity, the mass
model the mass of an object as a function of its
M of an object increases as its speed v increases ac-
speed expressed as a fraction of the speed of light.
cording to the formula
v2 1>2
M  f 1v2   M0 # a1  b
M0
,
v2 c2
1
B c2

Exercising Your Algebra Skills

Solve each trigonometric equation. 5. 5 sin u  4 cos u 6. 4 sin u  5 cos u
1. 4 tan u  5 2. 5 tan u  4 7. sin u  cos u  0 8. 4 sin u  3 cos u  0
3. cos u  sin u 4. 2 cos u  sin u

Chapter Summary

In this chapter, we introduced the use of the sine and cosine functions for model-
ing periodic phenomena and the tangent function. In particular, we discussed the
following:
◆ The behavior of the sine and cosine functions.
◆ How to convert between radian measure and degree measure.
◆ What the vertical shift or midline means for the sine and cosine functions.
◆ What amplitude means for the sine and cosine functions.
◆ What frequency means for the sine and cosine functions.
◆ What period means for the sine and cosine functions.
◆ What phase shift means for the sine and cosine functions.
◆ How to use the sine and cosine functions to model periodic behavior.
◆ How to fit sine and cosine functions to data.
◆ The behavior of the inverse sine and inverse cosine functions.
Gord.3896.07.pgs 4/28/03 11:56 AM Page 536

536 CHAPTER 7 Modeling Periodic Behavior

◆ How to solve trigonometric equations, using the inverse sine and inverse co-
sine functions.
◆ The behavior of the tangent function.
◆ The behavior of the inverse tangent function.
◆ How to solve trigonometric equations, using the inverse tangent function.

Review Problems
1. The student with whom you are working finishes a b. What do the variables represent?
problem and announces her answer is cos 15.70 2. c. What are the units?
You get an answer in the form sin 17.27082. Under d. What are possible values for the domain and
what circumstances are these answers the same? range?
2. Suppose that u is 60°. 20. Bernice is swinging on a playground swing whose
a. Find two positive angles and two negative angles supporting crossbar is 11 feet above the ground and
that have the same sine as u. the length of the chain to her seat is 8 feet. At the
b. Write the angles from part (a) in radian form. end of each swing, she makes an angle of 60° with
the vertical and it takes her 3 seconds to complete
3. Let u  45°.
each full cycle.
a. Find two positive angles and two negative angles
with the same cosine as u. 8 ft
b. Write the radian form of the angles from part (a). 60°

For each sinusoidal function in Problems 4–11, identify

the vertical shift, the amplitude, the frequency, the peri- 11 ft

od, and the phase shift.

3 seconds

4. y  325  10 sin a tb
2p
9 a. Write a sinusoidal function that can be used to
5. y  63  3 sin a tb
2p model the height of the seat above the ground as
25 a function of time t.
b. Write a sinusoidal function that can be used to
6. y  71  2 cos a
2p
tb model the horizontal displacement from directly
15
under the crossbar as a function of time t.
7. y  80  13 cos c 1t  152 d
2p 21. A bungee jumper dives off a bridge that spans across
24 a deep gorge. The bungee cord initially stretches to a
8. y  38  8 sin c 1t  52 d
2p maximum length of 200 feet before the jumper be-
24 gins her first rebound. Over the course of the next

9. y  100  25 sin a
2p
tb
72

10. y  100  25 sin a

2p
tb
97

11. y  145  40 sin a

2p
tb
83
12–19. Each of the functions in Problems 4–11 can be a
model for a common periodic phenomenon.
For each function,
a. describe a phenomenon that each function could
model.
Gord.3896.07.pgs 4/28/03 11:56 AM Page 537

Review Problems 537

60 seconds, she bounces up and down with ever- 24. The child in Problem 23 is also moving forward
diminishing oscillations, each lasting about 6 sec- 20 inches with each bounce of the pogo stick.
onds, until she comes to rest about 160 feet below a. Write a sinusoidal function to model the path of
the bridge. Write the equation of a decaying oscilla- the child’s feet—that is, the height y above the
tory function that models the height of the bungee ground as a function of the horizontal distance x
jumper as a function of time as measured from the covered.
instant the cord is extended to its maximum stretch. b. By comparing the graph of the function you cre-
22. If a car’s engine is operating at 2000 rpm, its pistons ated in part (a) to your image of what is actually
are moving up and down 2000 times per minute. happening, explain why the sinusoidal model
Thus, in a four-cylinder engine, each piston moves may not make sense.
up and down 500 times per minute. Suppose that the c. Look at the graph of the absolute value of the
total vertical distance that a piston moves is 3 inches. function you created in part (a). Is it a better or
worse model for the behavior you envision?
3 in. 3 in. To solve Problems 25–27, use the fact that, if an arc of
length s on a circle of radius r subtends an angle of u ra-
dians, then s  r u.
25. The distance between two points P and Q on the
Earth is measured as the distance along the arc of
the circle through P and Q and centered at the cen-
ter of the Earth O. The radius of the Earth is about
a. Write a sinusoidal function that models the 4000 miles. Find the distance from P to Q if the
height of the piston as a function of time in min- angle POQ has the following measurements.
utes, based on the midline for the height. p p
b. Write a sinusoidal function that models the a. b.
4 3
height of the piston as a function of time in min- 5p
utes, based on the lowest height of the piston. c. d. 15°
6
23. A pogo stick consists of a spring in a vertical tube
26. A wheel of radius 2 feet rotates at a constant rate of
with two fixed pedals on which a person stands and
180 revolutions per minute.
jumps up and down. Suppose that a child on a pogo
stick hops up and down every 3 seconds and that a. How many radians per minute are swept by
the height of the pedals varies from 4 inches above the wheel?
the ground to 14 inches above the ground. Write a b. How far does a point on the rim of the wheel
sinusoidal function to model the height of the ped- travel in 1 minute?
als above ground level as a function of time. (Hint: 27. Find the diameter of the tires on your car. Assume
Assume that the pedals are at the midline level at that the car is traveling at 60 mph and determine the
the start.) number of revolutions the tire makes every minute.
28. On the same set of axes, graph the functions
S1x 2  2 sin x, R1x 2  2 sin 3x
and T1x 2  2 sin 0.5x.
Clearly mark the zeros of each function.
29. For each function give the frequency, period, ampli-
tude, and phase shift.

a. y  5  2 cos a xb
3
4

b. y  5  2 cos a x  pb
14 in. 3
4 in. 4
Gord.3896.07.pgs 11/21/03 10:50 AM Page 538

538 CHAPTER 7 Modeling Periodic Behavior

b. Repeat part (a) with the function y 

c. y  5  2 cos apx  b
3
4 sin 3 arcsin1x 2 4 for x between 1 and 1.

d. y  5  2 cos c 1px  12 d
3 32. Solve for u.
4 a. 4 sin u  6 cos u
30. Determine the values of x for which each function b. 2 cos u  sin u
in Problem 29 equals 6. c. 3 tan u  21  0
31. a. Graph the function y  arcsin 3 sin1x 2 4 for x be- 33. Solve for x.
tween 10 and 10 radians. Explain why you get a. arctan x  1.35
the pattern you do. b. arcsin x  0.5
gord.3896.08.pgs 4/24/03 10:11 AM Page 539

8
More About the
Trigonometric Functions
8.1 Relationships Among Trigonometric Functions
In many applications of trigonometry, particularly in calculus, it often is necessary
to transform one trigonometric function into another using an appropriate
trigonometric identity. Recall that an identity is a relationship that is true for all val-
ues of the variable. For instance, the Pythagorean identity
sin2x  cos2x  1 (1)
that we discussed in Section 6.4 holds for every value of x.
However, suppose that we ask whether sin x  cos x equals 1. Figure 8.1 shows
a portion (one complete cycle) of the graph of y  sin x  cos x. Note that the
function is not identically equal to 1 because its graph is not a horizontal line of
height 1. Although there are several specific values of x for which sin x  cos x
equals 1 (such as x  0, x  p>2 and x  2p), the relationship does not hold for
every value of x. So sin x  cos x  1 is not an identity, but simply an equation
that holds for some specific values of the variable.

2
y = sin x + cos x
1.5
y=1
1
0.5
x
π π 3π 2π
2 2
–0.5
–1
–1.5
FIGURE 8.1 –2

539
gord.3896.08.pgs 4/24/03 10:11 AM Page 540

540 CHAPTER 8 More About the Trigonometric Functions

Identities Involving the Sine and Cosine

Consider again the Pythagorean identity (1). We can use it to transform sines to
cosines with
cos2x  1  sin2x
so that
cos x  21  sin2x .
Similarly, we can transform cosines to sines by using
sin2x  1  cos2x
so that
sin x  21  cos2x .
Each of these equations holds for all values of the variable x, so each is an identity.

The Reflection Identities

We explore several other useful relationships among the trigonometric functions
here. Two properties of the sine and cosine functions are
sin 1x 2  sin x (2)
cos 1x 2  cos x, (3)
for any x. These two relationships, known as the reflection identities, are easy to
see graphically. The graph of the cosine function is symmetric about the vertical y-
axis, as illustrated in Figure 8.2. That is, for any positive value of x, the height of the
cosine function is the same to the left of the y-axis (at x) as it is at the same dis-
tance to the right of the y-axis (at x). Thus
cos1x 2  cos x
for any value of x. We discussed this same type of behavior in Section 2.7 for power
functions with even powers such as f 1x 2  x2 and g1x 2  x4. For this reason, the
cosine function is called an even function.

y = cos x

y y

x
– π2 –x x
π
2

FIGURE 8.2

However, the sine curve is not symmetric about the y-axis. Rather, if you move
a distance of x to the left of the y-axis and consider the height to the sine curve, it is
gord.3896.08.pgs 4/24/03 10:11 AM Page 541

8.1 Relationships Among Trigonometric Functions 541

equivalent, but opposite in sign, to the height you get if you move the same dis-
tance x to the right of the y-axis, as shown in Figure 8.3. Thus
sin1x 2  sin x,
for any value of x. We encountered this type of behavior with power functions such
as g1x 2  x3 when the power is odd. As a result, the sine function is called an odd
function.

y
y = sin x

–x y
x
–π π
–y x

FIGURE 8.3

We discuss even and odd functions again in Section 8.2 when we describe con-
nections between polynomial functions and trigonometric functions.

Think About This Write a reflection identity for the tangent function. ❐

The Double-Angle Identities

We next consider some additional relationships involving the sine and cosine.
The Pythagorean identity says that sin2x  cos2x  1, which is equivalent to
cos2x  sin2x  1. What happens if we take the difference instead of the sum? Fig-
ure 8.4 shows the graph of y  cos2x  sin2x, for x between 0 and 2p. It is a sinu-
soidal curve that oscillates between 1 and 1 and completes two full cycles
between 0 and 2p, so it has a period of p and a frequency of 2. But these features
exactly describe the function y  cos 2x. Therefore it seems that
cos2x  sin2x  cos 2x,
or, equivalently,
cos 2x  cos2x  sin2x. (4)

y = cos 2 x − sin2 x
1

x
0 π π 3π 2π
2 2

FIGURE 8.4 –1
gord.3896.08.pgs 4/24/03 10:11 AM Page 542

542 CHAPTER 8 More About the Trigonometric Functions

You can verify this relationship numerically by substituting any value of x into
Equation (4). Alternatively, you can verify this relationship graphically by examin-
ing the graphs of the functions y  cos 2x and y  cos2x  sin2x.
We can rewrite Equation (4) in several alternative, but equivalent, forms by
making use of the Pythagorean identity (1). Thus
cos 2x  cos2x  sin2x  11  sin2x 2  sin2x
so that
cos 2x  1  2 sin2x. (4a)
Similarly, we can rewrite Equation (4) as
cos 2x  cos2x  sin2x  cos2x  11  cos2x 2
so that
cos 2x  2 cos2x  1. (4b)
Verify the identities in Equations (4a) and (4b) visually by using your function
grapher and numerically by substituting several different values for x into the
equations.
We next consider sin 2x. Suppose that we want to express sin 2x in an equiva-
lent form that does not show the frequency 2 explicitly. Is it possible that sin 2x and
2 sin x are equivalent? Graph the two functions and you’ll see that they cannot be
the same. The first, y  sin 2x, is a sinusoidal curve with an amplitude of 1 and a
frequency of 2, so its values oscillate between 1 and 1 and it completes two full
cycles between x  0 and x  2p. The second function, y  2 sin x, is a sinu-
soidal curve with an amplitude of 2 and a frequency of 1, so its values oscillate be-
tween 2 and 2 and it completes one full cycle between 0 and 2p.
The actual relationship for sin 2x is
sin 2x  2 sin x cos x. (5)
You can verify Equation (5) graphically on your function grapher. When you graph
the two functions y  sin 2x and y  2 sin x cos x simultaneously, you will see only
one graph—the second traces precisely over the first. You can also verify this result nu-
merically: Pick any value for x and evaluate sin 2x and 2 sin x cos x. The results will be
identical for every value of x, thus supporting the fact that Equation (5) is an identity.
The identities in Equations (4), (4a), (4b), and (5) are known as the double-
angle identities for the sine and cosine.

The Sum and Difference Identities

The double-angle identities in Equations (4) and (5) actually are special cases of
more general identities known as the sum and difference identities for sine and co-
sine that are formally derived in any trigonometry text. The sum identities are
sin1x  y2  sin x cos y  cos x sin y (6)
cos1x  y2  cos x cos y  sin x sin y. (7)
To show how the double-angle identities are derived from these formulas, we set
y  x in Equations (6) and (7). For instance, in Equation (6),
sin1x  x 2  sin x cos x  cos x sin x  2 sin x cos x,
gord.3896.08.pgs 4/24/03 10:11 AM Page 543

8.1 Relationships Among Trigonometric Functions 543

giving
sin12x 2  2 sin x cos x.
The same process in Equation (7) produces the double-angle formula for the cosine.
Similarly, we can replace y with y in the two sum identities, Equations (6)
and (7), and then use the reflection identities to derive the difference identities for
the sine and cosine:
sin1x  y2  sin x cos 1y2  cos x sin 1y2
 sin x cos y  cos x sin y (8)
cos1x  y2  cos x cos 1y2  sin x sin 1y2
 cos x cos y  sin x sin y. (9)

E XAMPLE 1
Show that cos1x  p>2 2  sin x for all x by using the difference identity for the cosine.
Solution Using the difference identity in Equation (9), we have

cos ax  b  cos x cos  sin x sin

p p p
2 2 2
 cos x 10 2  sin x 11 2  sin x.
# #
◆
E XAMPLE 2
Reduce sin 3x to an equivalent expression involving only sines and not including any
multiple angles.
Solution We write
sin 3x  sin12x  x 2  sin 2x cos x  cos 2x sin x Sum identity

 12 sin x cos x 2cos x  11  2 sin2x 2sin x Double angle identity

 2 sin x cos2x  sin x  2 sin3x

 2 sin x # 11  sin2x 2  sin x  2 sin3x Pythagorean identity

 2 sin x  2 sin3x  sin x  2 sin3x

 3 sin x  4 sin3x.
◆
E XAMPLE 3
Reduce sin 4x to an equivalent expression involving sines and cosines that has no multi-
ple angles.
Solution Following the approach in Example 2, we write
sin 4x  sin13x  x2

 sin 3x cos x  cos 3x sin x. Sum identity

gord.3896.08.pgs 4/24/03 10:11 AM Page 544

544 CHAPTER 8 More About the Trigonometric Functions

But, this expression involves expanding cos 3x, and we haven’t worked that out yet.
(You are asked to do so in a problem at the end of this section.) Alternatively, we could
start with
sin 4x  sin12x  2x 2  sin3212x 2 4

 2 sin 2x cos 2x Double angle identity

 212 sin x cos x 2 1cos2x  sin2x2 Double angle identities

 4 sin x cos3x  4 sin3 x cos x.

◆

The Half-Angle Identities

Occasionally we face the reverse problem of starting with powers of the sine or
cosine—say, sin3x or cos4x––and having to eliminate all powers by rewriting the
expression in terms of sines and cosines with multiple angles. To eliminate the
powers, we make use of two additional identities. Starting with the double-angle
identity in Equation (4a),
cos 2x  1  2 sin2x,
we have
2 sin2x  1  cos 2x
so that

sin2x  11  cos 2x 2.
1
(10)
2
Similarly, if we start with the double-angle identity in Equation (4b),
cos 2x  2 cos2x  1,
we get
2 cos2x  1  cos 2x
or

cos2x  11  cos 2x 2 .
1
(11)
2
The identities in Equations (10) and (11) are the half-angle identities. Verify them
graphically on your function grapher. We illustrate their use in Example 4.

E XAMPLE 4
Rewrite cos4x in terms of cosines of multiple angles by eliminating all exponents.
Solution Using Equation (11), we have
2
cos4x  1cos2x 2 2  c 11  cos 2x 2 d
1
2
gord.3896.08.pgs 4/24/03 10:11 AM Page 545

8.1 Relationships Among Trigonometric Functions 545

 3 1  2 cos 2x  cos2 12x 2 4.

1
4
This expression involves cos2 12x2 , so we apply Equation (11) again to get

e 1  2 cos 2x  31  cos 212x 2 4 f

1 1
cos4x 
4 2

a1  2 cos 2x   cos 4xb

1 1 1

4 2 2

a  2 cos 2x  cos 4xb

1 3 1

4 2 2
3 1 1
  cos 2x  cos 4x.
8 2 8

For easy reference, we list all the fundamental trigonometric identities involv-
ing the sine and cosine functions. These identities reappear both in this course and
in later mathematics and associated courses.

Trigonometric Identities
Pythagorean identity: sin2x  cos2x  1 (1)
Reflection identities: sin1x 2  sin x (2)
cos1x 2  cos x (3)
Double-angle identities: cos 2x  cos2x  sin2x (4)
 1  2 sin2x  2 cos2x  1 (4a,b)
sin 2x  2 sin x cos x (5)
Sum identities: sin1x  y2  sin x cos y  cos x sin y (6)
cos1x  y2  cos x cos y  sin x sin y (7)
Difference identities: sin1x  y2  sin x cos y  cos x sin y (8)
cos1x  y2  cos x cos y  sin x sin y (9)

sin2x  11  cos 2x 2
1
Half-angle identities: (10)
2

cos2x  11  cos 2x 2
1
(11)
2

Using the Trigonometric Identities

Suppose that a projectile, such as a cannonball or a high-pressure stream of water, is
shot off with an initial velocity v0 at an angle u with the horizontal, as shown in Fig-
ure 8.5. The distance R that the cannonball or the water travels—its range—depends
on the angle u. For very small angles, the range is minimal because gravity pulls the
gord.3896.08.pgs 4/24/03 10:11 AM Page 546

546 CHAPTER 8 More About the Trigonometric Functions

object to the ground quickly. For very large angles (close to 90°), the object is shot
almost vertically upward and comes back to the ground fairly near the point at
which it was released. For moderately sized angles, the range is considerably larger.

α
x

FIGURE 8.5

It is shown in physics that the range R is given by

2v02sin u cos u
R
ˇ

,
g
where g is the acceleration due to gravity.

E XAMPLE 5
Use a trigonometric identity to simplify the formula for the range of a projectile and use
the result to determine the angle that leads to the maximum range for any initial velocity.
Solution The range is
2v02sin u cos u
R
ˇ

.
g
Because sin 2u  2 sin u cos u, this expression for the range reduces to
v02sin 2u
R
ˇ

.
g
Because g and v0 are fixed, the range is maximal when sin 2u is maximal and the
largest value of the sine function is 1, which occurs when 2u  90° or u  45°. There-
fore a projectile subject only to the force of gravity has a maximum range when the ini-
tial angle u  45°.
◆
In most derivations in physics and engineering involving wave phenomena
such as electromagnetic waves (e.g., radio signals or electric currents in a circuit),
sound waves, or water waves, the height y of the wave as a function of time t is usu-
ally given in the form
y  A sin kt  B cos kt,
where A and B are constants and k is the frequency. In Example 6, we show how this
type of expression can be simplified by using a trigonometric identity to give far more
insight into the behavior of the wave than this fairly complicated expression provides.
gord.3896.08.pgs 4/24/03 10:11 AM Page 547

8.1 Relationships Among Trigonometric Functions 547

E XAMPLE 6
The equation of a wave is y  4 sin t  3 cos t. Use a trigonometric identity to explain
the behavior of this wave.
Solution The graph of this function between t  0 and t  2p is shown in Figure 8.6.
It looks like a sine wave shifted horizontally to the right by about p>6, or 30°. Also, the
amplitude of this wave seems to be about 5, compared to the amplitudes of 4 and 3 in
the two terms of the function. Finally, the period of this wave seems to be about 2p. As a
result, the equation of the wave y  4 sin t  3 cos t appears to be equivalent to
y  C sin1t  D2, where the amplitude C  5 and the phase shift D is about p>6. Let’s
see why.

y = 4 sin t − 3 cos t
5

t
0 π π 3π 2π
2 2

FIGURE 8.6 –5

The seemingly equivalent form for the wave y  C sin1t  D2 suggests using the
difference formula for the sine, or
C sin1t  D2  C # 1sin t cos D  cos t sin D 2. (12)
Also, the fact that the individual amplitudes in the original formula are 4 and 3 and the ap-
parent amplitude we observe for the wave is about C  5 suggests the Pythagorean theorem
242  32  5.
Factoring 5 out of the original formula for the wave yields

y  4 sin t  3 cos t  5 a sin t  cos tb .

4 3
5 5
Comparing this expression to Equation (12) suggests that we make the association
4 3
C  5, cos D  , and sin D  .
5 5
If cos D  4>5,
4
D  arccos  36.87°,
5
which is close to what we predicted for the phase shift, based on the graph shown in Fig-
ure 8.6. To be sure that this result is consistent with the third condition, we see that if
sin D  3>5,
3
D  arcsin  36.87°.
5
gord.3896.08.pgs 4/24/03 10:11 AM Page 548

548 CHAPTER 8 More About the Trigonometric Functions

Consequently, the wave formula

y  4 sin t  3 cos t  51cos D sin t  sin D cos t2
 5 sin1t  36.87°2
and the original wave is the same as a pure sinusoidal function centered about y  0
with amplitude 5, period 2p  360°, and a phase shift of about 36.87°, or 0.6435 radian.
◆
Identities Involving the Tangent
Just as there are trigonometric identities relating the sine and cosine, there are iden-
tities involving the tangent function. We encountered two of them in Section 6.4.
The first is the key identity relating the tangent function to the sine and the cosine:
sin x
tan x  .
cos x
We also derived an analog to the Pythagorean identity in Section 6.4:
1
tan2x  1  ,
cos2x
provided that cos x  0.
Likewise, there are double-angle, sum, and difference formulas, and so on, for
the tangent function. We investigate a double-angle identity and a sum identity in
the Problems. If you’re interested, you can find more details about them in any
trigonometry textbook.

Problems
1. Using ideas on amplitude and frequency, explain 10. cos 3b  3 cos3b  1
why cos 3x cannot be identically equal to 3 cos x. 11. sin23x  cos 6x  cos23x
2. Using ideas on amplitude, explain why cos 2x  12. cos22x  311  sin 2x 2
2 cos2x  1 is reasonable. (Recognize that such an 13. sin1cos x 2  cos1sin x 2
sin1cos x 2  sin x cos x
argument is not a proof.)
14.
Examine each equation in Problems 3–14 graphically to
see if the relationship may be an identity. If it is not an 15. Express cos 3x in terms of powers of sin x and cos x,
identity, attempt to locate graphically or numerically at but with no multiple angles.
least one point that lies on both curves. If it seems to be 16. Express cos 4x in terms of powers of sin x and cos x,
an identity, prove it algebraically. but with no multiple angles.
3. sin3x  cos3x  1 17. Express cos 5x in terms of powers of sin x and cos x,
4. cos 3x  cos3x  sin3x but with no multiple angles.
sin 2x 18. Examine the results of Problems 15–17 and the for-
5.  2 cos x mula for cos 2x. Are there any patterns in the terms?
sin x If so, what are they?
6. 11  cos u2 11  cos u2  sin2u 19. By setting y  x in the sum identity in Equation
7. sin 3x  3 sin x (7), show that you get the double-angle identity in
cos2u Equation (4).
8.  1  sin u
1  sin u 20. Rewrite sin4x in terms of multiple angles by elimi-
1  cos a sin a nating all exponents.
9. 
sin a 1  cos a
gord.3896.08.pgs 4/24/03 10:11 AM Page 549

8.1 Relationships Among Trigonometric Functions 549

21. Rewrite sin2x cos2x in terms of multiple angles by ground. Use the formula for the range of a projec-
eliminating all exponents. tile to determine the angle of inclination of the ball
22. a. Sketch the graph of sin1x  p>2 2. What familiar as it comes off the bat.
function do you get from this phase shift?
b. Use the sum identity for the sine function to show
that sin1x  p>22 actually equals that function.
23. a. Repeat Problem 22 for sin1x  p2. 5 ft

b. Repeat Problem 22 for cos1x  p>2 2. 320 ft

24. In the half-angle identities in Equations (10) and

(11), let y  2x, so that x  12 y. Rewrite each iden- 29. The accompanying figure shows the graph of the
tity in terms of y to see why they are called half- function y  sin4x  cos4x from x  0 to 2p. The
angle identities. curve suggests that the function is equivalent to
some sinusoidal function.
25. William Tell is about to shoot the most important
arrow of his life. His son is standing 250 feet away.
y
Tell releases the arrow at a height of 5 feet above the
ground with an initial speed of 180 feet per second. y = sin 4 x + cos 4 x
1
The height of the center of the apple on his son’s
head is also 5 feet above the ground. Find alge-
braically two different angles a at which Tell should
release the arrow in order to have it pass through 0.5
the apple without hitting the boy.

5 ft x
0 π π 3π 2π
2 2

250 ft
a. By examining the graph carefully on your func-
26. Suppose that William Tell’s son is actually a foot tion grapher, estimate values for each parameter
shorter than in Problem 26 so that the center of the to find a sinusoidal function that seems to have
apple is now 4 feet above the ground and that the the matching behavior pattern. (Hint: The pa-
arrow comes off the bow string at a height of 5 feet. rameters should be simple fractions or whole
Estimate, graphically, two different angles a at numbers.)
which Tell should release the arrow in order for it to b. Superimpose the graph of your function over
pass through the apple without hitting the boy. the graph of y  sin4x  cos4x to verify that
27. In Example 6, we converted the wave y  4 sin t  they do appear to be the same.
3 cos t to the equivalent pure sinusoidal expression c. Use the half-angle identities for sine and cosine
y  5 sin1t  36.87°2 . repeatedly to prove that y  sin4x  cos4x does
a. Convert this formula to a pure cosine curve by reduce to the expression you conjectured.
an appropriate horizontal shift. 30. Repeat Problem 29 with the function y 
b. Repeat the derivation in Example 6 by using the sin6x  cos6x.
sum or difference identity for cosines to derive 31. Refer to the functions shown in Problem 1 of Sec-
the equivalent formula as a cosine wave. tion 7.2 and decide which are odd, even, or neither.
c. How does the result in part (b) compare to the 32. a. Use some ideas from Section 5.5 on the sum of
result in part (a)? the terms in an exponential sequence to explain
28. A baseball player hits a ball with an initial velocity why you can calculate the value of
of 120 feet per second at a height of 5 feet above the 1  sin x  sin2x  sin3x  sin4x  . . .
ground. The ball is caught 320 feet from home plate
by an outfielder whose glove is also 5 feet above the 1
as .
1  sin x
gord.3896.08.pgs 11/21/03 10:52 AM Page 550

550 CHAPTER 8 More About the Trigonometric Functions

Are there any values of x for which this approach by using the double-angle identities for sine and
does not work? cosine. (Hint: Divide both the numerator and
b. What formula would you get for the sum of the denominator by cos2x. 2
the terms b. Derive the addition identity for the tangent,
1  sin x  sin2x  . . . sinnx tan x  tan y
tan1x  y2 
for any given positive integer n? 1  tan x tan y
33. Use the result of Problem 32(b) with different val- by using the addition formulas for sine and co-
ues of n to calculate the value of sine. (Hint: Divide both the numerator and the
1  sin x  sin2x  sin3x  sin4 x  . . . , denominator by cos x cos y.)
for x  p>6 correct to three decimal places. Now Examine each equation in Problems 38–46 graphically
suppose that you want to do so for x  p>3 in- to see whether the relationship may be an identity. If it
stead. Will you need approximately the same num- is not an identity, attempt to locate graphically or nu-
ber of terms, more terms, or fewer terms to get the merically at least one point that lies on both curves. If it
same three decimal place accuracy? Explain. seems to be an identity, prove it algebraically.
34. a. Verify graphically that 1 1
38. 1  2 
1 1 tan u sin2u
tan u   , 39. tan 2u  2 tan u
40. tan2x  sin2x  1tan x sin x 2 2
tan u sin u cos u
for all u for which the denominators are nonzero.
b. Show algebraically that the expression in part (a) 1
41. 1  tan2x 
is an identity. (Hint: Transform tan u to equiva- cos2x
lent expressions in sin u and cos u.) a 1  cos a a
42. tan  (Hint: Let  u.)
35. Use appropriate trigonometric identities to show that 2 sin a 2
1 2 43. tan x  1  2 tan x
2
 tan x  . 44. 1  cos 2x  tan x sin 2x
tan x tan 2x
36. Use the identity in Problem 35 to derive a double- 45. tan1sin x 2  tan x sin x
angle formula for the tangent function. 46. cos1tan x 2  tan1cos x 2
37. a. Derive the double-angle identity 47. What is wrong with the following “proof ”?

cos1tan x 2  cos a b  sin x.

2 tan x sin x
tan 2x 
1  tan2x cos x

8.2 Approximating Sine and Cosine

with Polynomials
Have you ever wondered what happens when you press either the SIN or COS key
on your calculator and the value for the function appears? How does the calculator
actually find the values of these functions?

Approximating the Sine Function

In this section, we consider one approach that has been used to compute function
values. We begin by examining the graph of the sine function, with x measured in
radians, as shown in Figure 8.7(a). We zoom in on the portion of the curve close to
the origin, as marked by the box; the corresponding curve is shown in Figure 8.7(b).
If we zoom in still further about the origin, as marked by the box in Figure 8.7(b),
we get the portion of the sine curve shown in Figure 8.7(c). This final graph looks
gord.3896.08.pgs 4/24/03 10:11 AM Page 551

8.2 Approximating Sine and Cosine with Polynomials 551

like a straight line rather than a portion of a curve. (In fact, if you zoom in suffi-
ciently on any smooth curve, it will eventually look like a straight line.)

y y y
y = sin x y = sin x y = sin x

x x x
–π π –1 1 –0.25 0.25

FIGURE 8.7 (a) (b) (c)

Linear Approximation to the Sine When x is very close to the origin, the sine
curve looks like a line. Let’s find the equation of this “line.” Because it passes
through the origin, the vertical intercept must be 0. To find the slope, we need a
second point. If we trace along the sine curve very close to the origin, we find that
x  0.001 and y  sin 0.001  0.0009999998 is a point on the sine curve. The
slope of the line through this point and the origin is
0.0009999998  0
m  0.9999998  1.
0.001  0
Therefore the equation of a line that very closely hugs the sine curve near the origin
is y  x. We show the graph of this line, along with the sine curve, in Figure 8.8.

y
g(x) = x
1
f (x) = sin x

x
–3 –1 1 3

–1
FIGURE 8.8

Observe that, when x is very close to 0, the graphs of f 1x 2  sin x and

g1x 2  x are very close to one another. In fact, when x is very close to 0, the two
graphs are virtually indistinguishable. That is,

sin x  x, if x is very close to 0

Of course, as the value of x gets farther from 0, the sine curve eventually bends
away from the line y  x.
gord.3896.08.pgs 4/24/03 10:11 AM Page 552

552 CHAPTER 8 More About the Trigonometric Functions

To show this result numerically, we look at some values of x to see how close
x sin x x sin x the values along the line match the values of the sine function.
0 0 From the table at the left, we see that, when x is extremely close to 0, the value
of sin x is almost identical to x itself, but the farther that x is from the origin, the
0.1 0.100 0.1 0.100 less accurate the approximation. Thus, whenever x is very close to 0, we can re-
0.2 0.199 0.2 0.199 place sin x with x for the purposes of approximating the value of sin x. For in-
stance, to approximate sin10.002432, we could say that
0.3 0.296 0.3 0.296
sin10.002432  0.00243.
0.4 0.389 0.4 0.389
Using a calculator gives sin10.002432  0.0024299976, so the approximation is ac-
0.5 0.479 0.5 0.479 curate to five decimal places.
We can approximate the sine function with a linear function in a different way
0.6 0.565 0.6 0.565
by using methods of linear regression. We use a set of points that lie on the sine
0.7 0.644 0.7 0.644 curve y  sin x very close to the origin. They are shown rounded to 6 decimal
places in Table 8.1. The line that best fits these “data” is y  0.9999258x with a
correlation coefficient r  1.000000000, which tells us that a line with slope of
about 1 is virtually perfect. Thus we again see that when x is very close to 0,
sin x  x. However, if we move too far away from x  0, the accuracy of the ap-
proximation breaks down. For instance, we would not want to approximate
sin10.752 with the value x  0.75 because sin10.752  0.6816; the value x  0.75
is too far from x  0 for the approximation to be good.

TABLE 8.1

x 0.025 0.02 0.015 0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
y  sin x 0.024997 0.019999 0.014999 0.01 0.005 0 0.005 0.01 0.014999 0.019999 0.024997

This idea of approximating a function such as y  sin x with a simpler func-

tion (often a linear function) is an essential principle in mathematics. We use this
principle to approximate the values of trigonometric functions because it is impos-
sible to calculate them directly with algebraic methods.

Improving on the Linear Approximation to the Sine

Unfortunately, as we have noted, the linear approximation to the sine curve is only
accurate if x is very close to the origin. As we take values of x farther and farther
from the origin, the sine curve bends ever more sharply and eventually bends away
from the line. Let’s see how we can improve on the linear approximation sin x  x
when x is somewhat farther from 0. To do so, we need a simple curve (at least one
that is simpler to work with than the sine function) that bends in a similar manner.
For computational purposes, the simplest curves are usually polynomials.
In Figure 8.7, we zoomed in on the sine curve very close to the origin so that
the curve looked like a line. Now we zoom out a bit to see what happens for values
of x from 3 to 3, as previously shown in Figure 8.8. Although the line is indistin-
guishable from the sine curve near the origin (roughly from x  0.6 to x  0.6),
the sine curve bends away from the line as the first pair of turning points in the sine
curve come into view. In fact, the overall shape of this portion of the sine curve is
quite suggestive of a cubic polynomial with a negative leading coefficient. (Recog-
nize that, if you zoom out a bit farther, more turning points appear and the cubic-
gord.3896.08.pgs 4/24/03 10:11 AM Page 553

8.2 Approximating Sine and Cosine with Polynomials 553

like appearance disappears.) This result suggests that we try to approximate this
portion of the sine curve with a cubic curve. We use the data in Table 8.1 that we
used for the linear fit, but now fit a cubic polynomial instead. We then get the cubic
y  0.1666601x3  0x2  0.999999999x  0.
Note that (1) the constant term is 0, which assures us that the cubic passes through
the origin; (2) the coefficient of the linear term is essentially 1; (3) the coefficient of
the quadratic term is 0; (4) the leading coefficient is negative, which is what we ex-
pected; and (5) the value of the leading coefficient, 0.1666601, is quite close to
1>6  0.16666667. Thus a cubic polynomial that approximates the sine func-
tion is
x3
T3 1x 2    x,
6
or, equivalently,
x3
sin x  x 
6
when x is fairly close to 0.
Figure 8.9 shows both the sine curve and the cubic polynomial for x from 3.5
to 3.5. The two curves are indistinguishable from about x  1.2 to x  1.2,
which extends over a considerably larger interval than the linear approximation,
which is accurate only from about x  0.6 to x  0.6.

3
y= x– x 1
6

y = sin x

x
–3.5 –1.5 –0.5 0.5 1.5 3.5

–1
FIGURE 8.9

To illustrate the accuracy of the approximation for sin x using the cubic poly-
nomial for values of x near 0, we try x  0.125, say, and find that
10.1252 3
sin10.125 2  10.1252   0.1246744,
6
which agrees with the true value of sin10.125 2  0.1246747 to six decimal places.
If we move farther from 0 and try x  0.7, we find that
10.7 2 3
sin10.7 2  10.7 2   0.643,
6
gord.3896.08.pgs 4/24/03 10:11 AM Page 554

554 CHAPTER 8 More About the Trigonometric Functions

compared to the actual value of sin10.7 2  0.644, which is correct to the nearest
hundredth, so the approximation is still fairly accurate. However, the graphs in Fig-
ure 8.9 show that the two curves eventually diverge. Thus if we take x too far from
0, the accuracy of the approximation diminishes. Moreover, the farther from x  0
we go, the worse the approximation is. For instance, if x  1 radian  57°,
13
sin 1  1   0.83333,
6
compared to the correct value of sin 1  0.84147. If x  1.5 radians,
11.52 3
sin11.5 2  11.5 2   0.93750,
6
compared to the correct value of sin11.5 2  0.99749. If x  2 radians,
23
sin 2  2   0.66667,
6
compared to the correct value of sin 2  0.90930. If x  p radians,
p3
sin p  p   2.02612,
6
compared to the correct value sin p  0. In fact, this last approximation is so bad
that it gives us a value, 2.02612, outside the range of the sine function.
What if we wanted to improve on the approximation still further so that we
could use it to estimate values for sin x when x is still farther from the origin? Con-
sider the graph of the sine curve from x  6 to x  6 shown in Figure 8.10. It
has four turning points and three inflection points, which suggests that the sine
curve looks like a polynomial of degree 5. Although graphing calculators don’t fit a
fifth degree polynomial to a set of data, that task can be accomplished by many
software packages. Using a spreadsheet, we find that the fifth degree polynomial
that fits the data in Table 8.1 is
T5 1x 2  0.0083x5  0x4  0.1667x3  0x2  0.999999999x  0,
or essentially

x3
T5 1x 2  0.0083x5   x.
6

1 y = sin x

x
–6 0 6

FIGURE 8.10 –1
gord.3896.08.pgs 4/24/03 10:11 AM Page 555

8.2 Approximating Sine and Cosine with Polynomials 555

Note that this polynomial has a positive leading coefficient, so it increases toward
the right, as we want. It also has a 0 constant coefficient, so it passes through the
origin. Note also that the only change from the cubic polynomial to this fifth de-
gree polynomial is the fifth degree term—all other terms remained the same.
Because 0.0083  1>120, we can write this polynomial as
x5 x3 x3 x5
T5 1x2    x or T5 1x 2  x   .
120 6 6 120
A simple and interesting pattern is developing here with the coefficients: 6 
3  2  1  3! and 120  5  4  3  2  1  5!. (See Appendix A2 for a
discussion of factorial notation.) So we can rewrite the approximation formula for
the sine function as
x3 x5
sin x  x   .
3! 5!
Figure 8.11 shows the graphs of the sine function and the fifth degree polynomial for
x between 4 and 4. The two curves are indistinguishable for x between roughly 2
and 2, so we have achieved a considerable improvement over the cubic approxima-
tion, which was a good match for x between roughly 1.2 and 1.2.

y = T5 (x)
1

x
–4 –2 –1 1 2 3 4

y = sin x
–1
FIGURE 8.11

To verify the accuracy of this approximation, let’s see how much improvement
we get compared to the previous values. The results are shown in the following table.

x sin x T3 1x 2 T5 1x 2
0.7 0.64422 0.643 0.64423

1 0.84147 0.83333 0.84167

1.5 0.99749 0.93750 1.00078

2 0.90930 0.66667 0.93333

p 0 2.02612 0.52404
gord.3896.08.pgs 4/24/03 10:11 AM Page 556

556 CHAPTER 8 More About the Trigonometric Functions

The fifth degree approximation is better still because we get more accurate esti-
mates for the values of sin x over larger intervals of x-values centered at 0.
We can continue this process, using higher degree polynomials, and get even
better approximations. However, before doing so, let’s examine the sequence of
polynomial approximations we have so far. They are
sin x  x,
x3
sin x  x  ,
3!
x3 x5
sin x  x   .
3! 5!
First, each successive polynomial involves just one additional term, compared to
the preceding polynomial. Second, each polynomial involves only odd powers—
and we know that the sine function is an odd function. This means that both the
sine function and the approximating polynomials are symmetric about the origin.
Third, the signs of successive coefficients alternate. Fourth, there is a definite pat-
tern involving factorials in the coefficients. These polynomials are known as Taylor
polynomial approximations after English mathematician Brook Taylor, who in-
vestigated them in the early 1700s.

Think About This Predict the next higher degree polynomial approximation to sin x. How accurate
is this approximation for the values x  0.7, 1, 1.5, 2, and p? ❐

Improving the Approximation Using the Behavior of sin x We could continue

this process and construct Taylor polynomial approximations of higher and higher
degree. However, that isn’t necessary if we cleverly use some of the basic behavioral
properties of the sine function. First, recall the reflection identity
sin1x 2  sin x.
It allows us to approximate sin x when x is negative simply by using the correspon-
ding positive value for x and reversing the sign of the estimate.
Second, we know that the sine function is periodic with period 2p. Therefore, if
x is any number greater than 2p  6.28, the value of sin x is the same as the value
of sin x0 , where x0 is the corresponding number between 0 and 2p radians. Conse-
quently, we need only obtain an approximation that is accurate as far out as 2p. We
can handle anything beyond that by reducing the value of x to an appropriate value
x0 between 0 and 2p by “removing” all multiples of 2p, as illustrated in Figure 8.12.

y = sin x

x
x0 π x
2

sin x = sin x 0
FIGURE 8.12
gord.3896.08.pgs 4/24/03 10:11 AM Page 557

8.2 Approximating Sine and Cosine with Polynomials 557

Now visualize the portion of the sine curve between x  p and x  2p, as
shown in Figure 8.13. It has the same shape as the portion from x  0 to x  p,
only “flipped” across the x-axis. Thus, if we have a value of x between p and 2p
(where sin x is negative), there is a point between 0 and p, namely at x  p, where
the sine function has the same value, but with a positive sign. That is,
sin x  sin1x  p 2.
So, all we need is an approximation that is sufficiently accurate for x between 0 and p.

y
y = sin x
1
y = –sin x

x
0 x− π π x 2π

FIGURE 8.13 –1

Think About This Use an appropriate trigonometric identity to show that sin x  sin 1x  p2. ❐

Now visualize the sine curve from x  0 to x  p. The two halves are sym-
metric, as shown in Figure 8.14. Therefore, for any point x between p>2 and p,
the value of sin x is the same as that at a corresponding point between 0 and p>2.
So, all we need is an approximation to sin x that is sufficiently accurate for x be-
tween 0 and p>2. The previous fifth degree polynomial T5 1x 2 gives two-decimal
accuracy for any value of x in this interval. If we want more than two-decimal ac-
curacy, we have to use a higher degree polynomial—say, the seventh degree Tay-
lor polynomial that we asked you to produce in a previous Think About This
exercise.

1 y = sin x

π π
x− 2
x− 2
x
0 π−x π x π
FIGURE 8.14 2
gord.3896.08.pgs 4/24/03 10:12 AM Page 558

558 CHAPTER 8 More About the Trigonometric Functions

Think About This Try the seventh degree polynomial

x3 x5 x7
sin x  x   
3! 5! 7!

for various values of x in the interval from 0 to p>2. Does it provide four-decimal
accuracy? Is that adequate? If not, what would you do? ❐

Approximating the Cosine Function

We now consider the comparable problem of approximating the cosine function by
using a polynomial. If you zoom in on the cosine curve very close to x  0, it ap-
pears indistinguishable from a horizontal line. In fact, because cos 0  1, that line
must be y  1. So, for x very close to 0,
cos x  1.
However, once you move away from x  0, the cosine curve bends away from the
line y  1.
Let’s now look at the cosine curve in a somewhat wider interval about x  0—
say, from x  2 to x  2, as shown in Figure 8.15. Its overall shape suggests a
parabola opening downward. Therefore we try to approximate the cosine function
with a quadratic function so long as x remains fairly close to 0.

y
1 y = cos x

x
− π2 0 π
2
FIGURE 8.15 –0.2

Approximating y  cos x Using Data Analysis There are several ways to find
an equation for such a quadratic. One way is to fit a quadratic function to some set
of values for cos x when x is relatively close to 0. Consider the values in Table 8.2.

TABLE 8.2

x 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08

y  cos x 0.9968 0.9982 0.9992 0.9998 1 0.9998 0.9992 0.9982 0.9968

Using a calculator, we find that the quadratic function that fits these data is
y  0.49973x2  0x  0.99999.
The constant term is essentially 1 and the leading coefficient is approximately
0.5. Hence we have the following approximation to the cosine function near
x  0.
gord.3896.08.pgs 4/24/03 10:12 AM Page 559

8.2 Approximating Sine and Cosine with Polynomials 559

x2
cos x  1 
2

Figure 8.16 shows the graphs of the cosine function and this quadratic Taylor
approximating polynomial for x between 2 and 2. The two are virtually indistin-
guishable for x between about 0.8 and 0.8. For instance, cos10.5 2  0.87758
compared to the value of the approximating quadratic,
10.5 2 2
cos10.5 2  1   0.875,
2
so we have two decimal place accuracy.

y
1 y = cos x

x
–2 0 2

2
y = 1– x
2

FIGURE 8.16 –1

If we zoom out somewhat farther on the graph of the cosine curve—say, for x
between 5 and 5, as shown in Figure 8.17—the cosine function no longer sug-
gests a quadratic function. This portion of the cosine curve has four real roots,
three turning points, and two inflection points, which suggest a polynomial of de-
gree 4. Using a calculator, we find a quartic function that fits the data values in
Table 8.2 is
y  0.041653x4  0x3  0.499999x2  0x  0.999999.

1
y = cos x

x
–5 0 5

–1
FIGURE 8.17
gord.3896.08.pgs 4/24/03 10:12 AM Page 560

560 CHAPTER 8 More About the Trigonometric Functions

Note that the coefficients of both odd-powered terms are 0, the constant coeffi-
cient is essentially 1, and the quadratic coefficient is essentially 12 . As for the lead-
ing coefficient, 0.041653,
1
 24.00787,
0.041653
so the leading coefficient is essentially 1>24  1>4!. Therefore we have the fourth
degree Taylor polynomial approximation
x2 x4
cos x  1   .
2 4!
Figure 8.18 shows the graph of this polynomial T4 1x 2  1  x2>2  x4>4! and
the cosine function for x between 5 and 5. Observe that T4 1x 2 tries hard to cap-
ture the pattern in the cosine curve. In fact, the polynomial is an excellent match
to the cosine for x between roughly 1.5 and 1.5. For instance, if x  0.5, then
T4 10.5 2  0.87760, compared to cos10.5 2  0.87758; similarly, T4 112  0.541666,
compared to cos11 2  0.54030. If we choose a value of x too far from 0, the approx-
imation breaks down. Thus T4 11.5 2  0.08594, compared to cos11.5 2  0.07074.

y = T4 (x)
1

y = cos x

x
–5 0 5

–1
FIGURE 8.18

If we zoom out still farther on the cosine curve—say, from x  8 to x  8—

two more turning points come into view, which suggests that we could get a better
approximation to the cosine with a sixth degree polynomial. Using the values in
Table 8.2 and a spreadsheet, we find, after rounding the coefficients, that
x2 x4 x6
cos x  T6 1x 2  1    .
2 4! 6!
Summarizing these results, the successive Taylor polynomial approximations
to the cosine function are:
x2
cos x  1  ,
2
x2 x4
cos x  1  
2 4!
x2 x4 x6
cos x  1    .
2 4! 6!
gord.3896.08.pgs 4/24/03 10:12 AM Page 561

8.2 Approximating Sine and Cosine with Polynomials 561

As with the sine approximations, (1) each successive polynomial involves just one
additional term; (2) each polynomial involves only even powers, and we know that
the cosine function is an even function; (3) the signs of successive coefficients al-
ternate; and (4) there is a clear pattern in the coefficients involving factorials.
The graphs of these polynomials, as well as that of the cosine curve, are shown
in Figure 8.19. Note that each successive polynomial fits the cosine curve more ac-
curately over a larger and larger interval centered at x  0. You should examine
these successive approximations using your function grapher.

1
T4

x
–3.5 –2.5 –0.5 0.5 2.5 3.5

y = cos x

–1
FIGURE 8.19
y = T2 T6

Think About This How could you improve on the sixth degree polynomial approximation to the co-
sine? By eye, over what interval does it appear to be a good fit to the cosine curve? ❐

Think About This Devise a scheme to reduce any value of x to an equivalent value that allows you to
use the smallest possible interval of x-values. How accurate is the fourth degree
Taylor polynomial on this interval (i.e., what is the largest error between the cosine
and the polynomial)? How accurate is the sixth degree polynomial? ❐

Approximating sin x and cos x Using Trigonometric Identities

We now approach the problem of approximating the sine and cosine from a different
viewpoint using several trigonometric identities. Recall the double-angle identity
cos 2u  1  2 sin2u
from Equation (4a) of Section 8.1. If we let x  2u so that x>2  u, the expression
for cos 2u  cos x becomes
x 2
cos x  1  2 sin2 a b  1  2 c sin a b d .
x
2 2
When u is close to 0, we have sin u  u, so that

sin a b  .
x x
2 2
Consequently we can approximate cos x by
x 2 x2
cos x  1  2 a b  1 
2 2
gord.3896.08.pgs 4/24/03 10:12 AM Page 562

562 CHAPTER 8 More About the Trigonometric Functions

when x is close to 0. Note that this approximation is identical to the quadratic Tay-
lor polynomial that we obtained previously by using data analysis techniques.
We now use the double-angle formula for the sine,
sin 2u  2 sin u cos u
with x  2u so that x>2  u, which gives

sin x  2 sin a b cos a b .

x x
(13)
2 2
We use the linear approximation for sin x  x and the quadratic approximation
cos x  1  x2>2 to get
1x>2 2 2 x2
sin a b  and cos a b  1 
x x x
1 .
2 2 2 2 8
Substituting these expressions into Equation (13), we obtain an approximation for
sin x that improves on sin x  x:
x2 x2
sin x  2 a b a1  b  x a1  b
x
2 8 8
3
x
x .
8
This is a cubic function that approximates the sine function near x  0, but it is
slightly different from the third degree Taylor approximation, sin x  x  x3>6. We
compare these two cubic approximations in the Problems at the end of this section.
Figure 8.20 shows the graph of this cubic along with the sine function on the
interval from 3 to 3. The two curves seem almost identical for x between 1 and
1; they are reasonably close between 2.5 and 1 and again between 1 and 2.5; but
they apparently begin to diverge farther from 0. Note how much better this cubic
function seems to approximate sin x than our linear approximation sin x  x.

1
y = sin x

3
y=x– x
8
x
–3 –2 –1 1 2 3

–1
FIGURE 8.20

Think About This Check numerically on your calculator how close the cubic is to the sine function at
x  0.5, at x  1, and at x  1.5. ❐

We can continue this process to produce still better approximations to both

the cosine and the sine functions by using the same trigonometric identities.
gord.3896.08.pgs 4/24/03 10:12 AM Page 563

8.2 Approximating Sine and Cosine with Polynomials 563

For instance, using the double-angle formula cos 2u  1  2 sin2u and our new
approximation
x3
sin x  x  ,
8
we get
x 2
cos x  1  2 sin2 a b  1  2 c sin a b d
x
2 2
1x>22 3 2
 1  2 c 1x>2 2  d .
8
After some algebraic simplification, we eventually get
x2 x4 x6
cos x  1    .
2 32 2048
The two graphs in Figure 8.21 illustrate that this sixth degree polynomial P6 1x 2 is
an almost perfect match to the cosine function from about x  1.5 to about
x  1.5. It is quite accurate from about x  2 to about x  1.5 and from
about x  1.5 to about x  2; thereafter its accuracy diminishes. For comparison,
Figure 8.22 shows three graphs: the basic cosine curve, the initial quadratic ap-
proximation P2 1x 2  1  x2>2, and this sixth degree polynomial approximation
P6 1x 2 . The higher degree polynomial is clearly a much better fit. It follows the
bends of the cosine curve and stays close to it over a wider interval of x-values.

x
–4 –3 –2 –1 1 2 3 4

y = cos x

–1

FIGURE 8.21 y = P6 (x)

x
–4 –3 –2 –1 1 2 3 4

y = cos x

–1
y = P2 (x)
FIGURE 8.22 y = P6 (x)
gord.3896.08.pgs 4/24/03 10:12 AM Page 564

564 CHAPTER 8 More About the Trigonometric Functions

Approximating sin x and cos x Using Taylor Polynomials

We could continue this process to get better polynomial approximations to both sin
x and cos x by using the trigonometric identities to construct polynomials of still
higher degree. Unfortunately, each successive improvement is based on a series of ap-
proximations—we used sin x  x to generate the approximation
x2
cos x  1  ,
2
and so on. The approximation errors in this process mount up and give less than the
best possible approximation at each successive stage. For instance, we first found

sin x  x  T1 1x 2
and then used it to find

x3
sin x  x   Q3 1x 2.
8
Actually, as you will learn in calculus, the best possible cubic curve to approxi-
mate the sine curve near x  0 is the Taylor polynomial of degree 3.

x3
sin x  x   T3 1x 2 .
6
It is identical to the cubic polynomial we obtained earlier based on fitting a
cubic function to a set of values of the sine function near 0. Figure 8.23 shows
the graph of this cubic and the underlying sine curve. Figure 8.24 shows the sine
curve and the two different cubic approximations: the Taylor approximation of
degree three, T3 1x 2  x  x3>6, and the polynomial of degree three based on
the trig identities, Q3 1x 2  x  x3>8. Note that T3 1x 2 remains closer to the sine
curve over a wider interval than Q3 1x 2 does. Note also that T3 1x 2 bends in such a
way that it remains very close to the sine curve over a relatively large portion of
its first arch.

3
y= x– x 1
6

y = sin x

x
–3.5 –1.5 –0.5 0.5 1.5 3.5

–1

FIGURE 8.23
gord.3896.08.pgs 4/24/03 10:12 AM Page 565

8.2 Approximating Sine and Cosine with Polynomials 565

3
y= x– x 1
6

y = sin x

x
–3.5 –1.5 –0.5 0.5 1.5 3.5

3
y= x– x
–1 8
FIGURE 8.24

Use your function grapher to view what happens in a more dynamic way. In
particular, examine the three curves near x  0 to see that T3 actually is closer to
the sine curve than Q3 is.
As a final note, let’s look at the ideas we have developed in this section from a
somewhat different perspective. Until now, we have interpreted Taylor polynomials
as a means of approximating one function by a polynomial. An alternative inter-
pretation is that we have been constructing a function (or a portion of a function)
from simpler functions. That is, we have been constructing the trigonometric
functions by using polynomials as the fundamental building blocks. More specifi-
cally, we have used linear combinations of power terms (i.e., sums of constant mul-
tiples of power terms) as these fundamental building blocks. This idea of using
linear combinations of basic mathematical elements to construct more complicat-
ed mathematical structures is a continuing theme throughout mathematics.
Problems
1. Use the Taylor polynomial approximation to b. Add 2 columns to the table, one for sin x  x
f 1x2  cos x of degree 2 to estimate the value of the and another for sin x  1x  x3>6 2, to compare
cosine function for x  0, 0.1, 0.2, 0.3, 0.4, 0.5, and the linear and cubic approximations to the cor-
0.6. Compare each estimate to the correct value. rect value for each x.
c. Use your function grapher to graph y 
2. Use the values you calculated in Problem 1 to con- sin x  x. What type of function does it appear
struct a table that has a column containing the error to be?
in the approximation (the difference between the d. Use polynomial regression to find an appropri-
estimate and the correct value). Analyze the column ate polynomial to fit the data values of sin x  x
of errors. Do they appear to grow approximately versus x.
linearly? exponentially? quadratically? cubically? e. Repeat part (c) with y  sin x  1x  x3>62 .
3. Use the Taylor polynomial approximation to 5. Construct a table of values of sin x for x  4p>25,
f 1x2  sin x of degree 3 to construct a table of esti- 3p>25, 2p>25, . . . , 4p>25. Use your calculator
mates for the values of the sine function when x  0, to find the cubic polynomial that fits this set of sine
0.1, 0.2, 0.3, 0.4, 0.5, and 0.6. Calculate the errors and values. How close does it come to the cubic Taylor
analyze them the same way you did in Problem 2. polynomial approximation sin x  x  x3>6?
4. a. Construct a table containing Taylor polynomial 6. Use the Taylor polynomial approximation of de-
approximations of degrees n  1 and n  3 to gree n  5 to f 1x 2  sin x to find a polynomial
f 1x2  sin x for x  p>3, p>4, p>6, 0, approximation of degree n  5 to g1x2  sin1x 2.
p>6, p>4, p>3. Is the result surprising? Explain.
gord.3896.08.pgs 4/24/03 10:12 AM Page 566

566 CHAPTER 8 More About the Trigonometric Functions

7. Use the Taylor polynomial approximation of degree function close to x  0. What limiting value
n  5 to f 1x2  sin x to find a polynomial approx- does this function appear to approach?
imation of degree n  10 to g1x2  sin1x2 2 . Graph b. Use the linear Taylor polynomial approximation
both g1x 2 and your approximation to it on the in- to sin x to explain why the limiting value you
terval from p to p. Based on this graph, over found in part (a) appears to make sense.
what interval does your polynomial seem to be a 13. In calculus, you will have to determine the value of
good approximation to g1x2? sin1x  x2  sin x
8. a. Use the Taylor polynomial approximation of de- ,
gree n  5 to f 1x2  sin x to find a polynomial
x
approximation of degree n  5 to h1x2  sin 2x. where x is a very small quantity.
b. What do you get if you multiply the polynomial a. Estimate the value of this quotient by using lin-
approximation of degree n  3 to f 1x 2  sin x ear approximations to both sine expressions.
by the polynomial approximation of degree b. Estimate the value of this quotient by using a cubic
n  4 to g1x2  cos x? approximation to both sin x and sin1x  x 2 .
c. Graph h1x 2 and twice the product of the two ap- c. With the cubic approximation, suppose that x
proximations found in part (b). What do you is actually 0. What does the resulting expression
observe? Explain why. suggest?
9. Write the Taylor polynomial approximation of degree 14. The exponential function f 1x 2  ex with base
3 to f 1x 2  sin x and the approximation of degree 4 e  2.71828 . . . is used extensively in mathemat-
to g1x2  cos x. Square each expression and add ics and the sciences. As with the trig functions, its
them. What do you get? What do you think will hap- values are calculated using Taylor polynomial
pen if you use higher degree approximations? Explain. approximations:
10. In the text, we used the double-angle identity ex  1  x,
cos 2u  1  2 sin2u with u  x>2 to construct
x2
the approximation ex  1  x  ,
2!
x2 x4 x6
cos x  1    . x2 x3
2 32 2048 ex  1  x   ,
2! 3!
Instead, use the alternative form of the identity
cos 2u  cos2u  sin2u and any lower degree poly- and so on. Use these and any further approxima-
nomials to find a different approximation to cos x. tions that you need to approximate the values of
11. Repeat Problem 10 with the third form of the double- a. e0.1
angle identity cos 2u  2 cos2u  1 to construct still b. e0.1
another approximation formula for cos x. c. Use the given polynomials and any additional ap-
12. The function f 1x2  1sin x2>x is not defined at proximations to ex that you need to estimate the
x  0. value of e reasonably accurately. What degree
polynomial will produce two-decimal accuracy?
a. Use values of x  0.1, 0.01, 0.001, 0.0001, three-decimal accuracy? four-decimal accuracy?
0.00001, . . . to investigate the behavior of this

8.3 Properties of Complex Numbers

One of the most amazing developments in the history of mathematics was the in-
troduction of complex numbers to solve quadratic equations. For example, if
x2  4  0, then x2  4, so that x  14  2i, and the two roots are
x  2i and x  2i, where i  11. Similarly, from the quadratic formula, the
roots of x2  2x  10  0 are

12 2  212 2 2  4110 2 2  236 2  16i2

x    1  3i,
2 2 2
gord.3896.08.pgs 4/24/03 10:12 AM Page 567

8.3 Properties of Complex Numbers 567

or x  1  3i and x  1  3i.
In our exploration of the nature of the roots of polynomials in Section 4.4, we
demonstrated that quadratic, cubic, and higher degree polynomials have a surpris-
ingly high proportion of complex zeros. We now develop a way to visualize com-
plex numbers that gives a deeper understanding of the processes that lead to such
polynomial equations.
Any complex number z  a  bi is composed of two parts, a real part, a, and
an imaginary part, b. For instance, in z  4  7i, the real part is 4 and the imagi-
nary part is 7. We occasionally write a  Re1z2 and b  Im1z2 , respectively. Note
that a and b are both real numbers; it is the combination a  bi that is a complex
number. In the special case when b  0, the complex number z  a  bi reduces
to a real number. In another special case where a  0, the complex number z reduces
to a pure imaginary number, bi.
The arithmetic of complex numbers, for the most part, is quite straightfor-
ward, and we review it briefly in Appendix E. Because i  11 , it follows that

i2  1 21 2 2  1
i3  1i 2 2 1i2  11 2 1i2  i
i4  1i 2 2 1i 2 2  11 2 11 2  1
i5  1i 4 2 1i2  11i2  i.
In fact, all higher powers of i simply cycle through the four “values” i, 1, i, and 1.
That is, i 6  i 2  1, i 7  i 3  i, i 8  i 4  1, i 9  i 5  i, and so on.
Visualizing complex numbers geometrically is extremely helpful. We do so by
using the complex plane, which is a two-dimensional coordinate system designed to
display a complex number z  a  bi. We measure the real part a horizontally and
the imaginary part b vertically. In Figure 8.25 we plot the complex number z  2  5i.
Note that it lies 2 units to the right and 5 units up from the origin. Similarly, the com-
plex numbers 1  3i, and 2  i are also plotted in Figure 8.25. Any purely real
number, such as 4 (which is 4  0i) or 6 (which is 6  0i), lies on the horizon-
tal axis. Any purely imaginary number, such as 4i (which is 0  4i) or 3i (which is
0  3i) lies on the vertical axis.

Im
2 + 5i
5
4i

–2 + i
1
Re
–2 0 1 2

FIGURE 8.25 –3 1 – 3i

Suppose that z  a  bi is an arbitrary complex number that we plot as a

point in the complex plane. We connect the point to the origin with a line segment,
which is the hypotenuse of a right triangle, as shown in Figure 8.26. The base of the
gord.3896.08.pgs 4/24/03 10:12 AM Page 568

568 CHAPTER 8 More About the Trigonometric Functions

triangle is a, the real part of z, and the height of the triangle is b, the size of the
imaginary part of z. The Pythagorean theorem gives the length of the hypotenuse as
1a2  b2, which we interpret as the size of the complex number z  a  bi. We
call it the modulus of the complex number and write it as

7 z7  2a2  b2 .

Im

z = a + bi

z
= √a 2 + b 2
b = Im(z)

θ
Re
FIGURE 8.26 0 a = Re(z)

For instance, if z  4  3i, then its modulus is

7 z7  242  32  225  5.
The complex numbers 4  3i, 4  3i, and 4  3i all have the same modulus
of 5. Sketch them to verify that this is indeed the case.

Think About This Are there any other points in the complex plane that also have a modulus of 5?
What can you say about all such complex numbers? ❐

We again consider the complex number z  a  bi and the associated right

triangle in the complex plane. We now focus on the angle u shown in Figure 8.26.
By convention, u is measured counterclockwise from the horizontal, or real, axis.
Thus
b
tan u  .
a
We also have the two further relations
a b
cos u  and sin u 
7 z7 7 z7
,

which lead to
a  7 z7 cos u and b  7 z7 sin u.
Consequently, we can write the original complex number z in the equivalent
trigonometric form
z  a  bi  7 z7 cos u  i7 z7 sin u
 7 z7 1cos u  i sin u 2.
gord.3896.08.pgs 4/24/03 10:12 AM Page 569

8.3 Properties of Complex Numbers 569

The trigonometric form for the complex number z  a  bi is

z  7 z7 1cos u  i sin u 2,

where

7 z 7  2a2  b2 and tan u  , a  0.

b
a

E XAMPLE 1
Find the trigonometric form for the complex number z  4  3i.
Solution For z  4  3i, we have 7 z7  5, so that
z  4  3i  51cos u  i sin u2 ,
where tan u  3>4, so that u  arctan 3>4  0.6435 radian or 36.87°.
◆
Think About This Use the value for the angle u in Example 1 to show that the trigonometric form for
the complex number z is identical to the original expression 4  3i. ❐

Powers of Complex Numbers

The trigonometric form for a complex number z  a  bi allows us to interpret z
as being located at a certain distance, the modulus, from the origin and rotated
through an angle u from the horizontal. This model gives us a way to gain some
special insights into powers of complex numbers.
E XAMPLE 2
For z  4  3i, (a) find z2 algebraically and (b) interpret z2 geometrically in the com-
plex plane.
Solution

a. If z  4  3i,
z2  14  3i2 2  42  214 2 13i2  13i2 2 1u  v2 2  u2  2uv  v2
 16  24i  91i2 2
 16  24i  9 i2  1
 7  24i.
This algebraic result provides no special insight into how z 2 is related to z.
b. We look at the trigonometric form for z  4  3i. The modulus is 7 z7  5, and the
associated angle is u  arctan13>4 2  0.6435 radians, or 36.87°, as in Example 1.
Now consider the trigonometric form for z 2. Its modulus is
7 z2 7  272  242  249  576  2625  25,
which is the square of the modulus of the original complex number z. Next, the angle
f associated with z2 is defined by
24 24
tan f  so that f  arctan  1.2870 radians or 73.74°,
7 7
gord.3896.08.pgs 4/24/03 10:12 AM Page 570

570 CHAPTER 8 More About the Trigonometric Functions

which is exactly twice the angle u (0.6435 radians or 36.87°) associated with z, as il-
lustrated in Figure 8.27.
Im
z2

25

z
73.74°
5

Re
FIGURE 8.27 0 36.87°

◆
For this particular complex number z  4  3i, z 2 is related to z by the
process of squaring the modulus and doubling the angle. Does this rule hold in
general? Let’s look at two other simple cases.
E XAMPLE 3
Find the modulus and angle associated with z2  12i2 2 and relate them to the modulus
and angle associated with z  2i.
Solution We know that z  2i  0  2i is located at a distance of 7 z7  2 from the
origin with an associated angle of u  p>2 measured in the usual positive direction
from the horizontal axis. We now consider z 2  12i2 2, which is
z 2  4i2  4  4  0i.
This complex number has modulus 4 and associated angle p because it is on the nega-
tive real axis. That is, the modulus of z 2  12i2 2 is the square of the modulus of z  2i,
and the associated angle p is twice the angle p>2 associated with z  2i.
◆
E XAMPLE 4
Find the modulus and angle associated with z 2  11  i2 2, where z  1  i, and relate
them to the corresponding modulus and angle for z.
Solution For z  1  i, the modulus is
7 z7  212  12  22 ,
and the associated angle is u  p>4. Further,
z 2  11  i2 2  1  2i  i2  1  2i  1  2i,
so 7 z2 7  24  2, the square of the modulus of z. The associated angle is p>2, or dou-
ble the angle associated with z. So, again, when we square a complex number, the modu-
lus is squared and the angle is doubled.
◆
gord.3896.08.pgs 4/24/03 10:12 AM Page 571

8.3 Properties of Complex Numbers 571

Let’s now consider any complex number z  a  bi in the equivalent trigono-

metric form
z  7 z7 1cos u  i sin u 2.
Squaring z gives
z 2  7 z7 2 1cos u  i sin u 2 2
 7 z7 2 1cos2u  2i cos u sin u  i 2 sin2u 2.
Using i2  1 and collecting the real and imaginary terms yields
z 2  7 z7 2 3 1cos2u  sin2u 2  2i cos u sin u 4 .
Now recall the double-angle identities:
cos 2u  cos2u  sin2u
sin 2u  2 sin u cos u.
Examining the real part of the previous expression for z2, we see that it equals cos 2u,
whereas the imaginary part equals sin 2u. Thus we have
z 2  7 z7 2 1cos 2u  i sin 2u 2 .
Geometrically, squaring any complex number always produces a new complex num-
ber whose modulus is the square of the original modulus and whose angle is double
the original angle. If the modulus of the original number is greater than 1, z 2 is a
“larger” complex number, as shown in Figure 8.28(a). If 7 z7 is smaller than 1, z 2
is a “smaller” complex number, as shown in Figure 8.28(b).

Im
z
> 1 Im
z
< 1

z2

z = a + bi

z
2
z2

z
2

2θ z = a + bi 2θ

θ θ
Re Re
0 0
FIGURE 8.28 (a) (b)

What about other powers of z  a  bi? Is there any pattern for zn when n 2?
E XAMPLE 5
Find the modulus and angle associated with z 3 when z  2i.
Solution The complex number z  2i is located at a distance of 2 from the origin and
at an angle of p>2. Now consider
z 3  12i2 3  8i 3  8i  0  8i.

It is located at a distance of 8 from the origin and is rotated through an angle of 3p>2,
which is triple p>2. Thus the modulus of 12i2 3 is the cube of the modulus of 2i, and the
associated angle 3p>2 is three times the angle p>2 associated with 2i.
◆
gord.3896.08.pgs 4/24/03 10:12 AM Page 572

572 CHAPTER 8 More About the Trigonometric Functions

Let’s find out whether the same pattern holds when we cube any complex
number z  a  bi. We do so by using the trigonometric form. Because
z 2  7 z7 2 1cos 2u  i sin 2u 2 ,
we have
z 3  z 2 1z2
 3 7 z7 2 1cos 2u  i sin 2u 2 4 3 7 z7 1cos u  i sin u 2 4
 7 z7 3 3 cos 2u cos u  i cos 2u sin u  i sin 2u cos u  i 2sin 2u sin u4.
Using i2  1 and collecting the real and imaginary terms, we get
z 3  7 z7 3 3 1cos 2u cos u  sin 2u sin u 2  i1cos 2u sin u  sin 2u cos u 2 4.
We now use the sum identities
cos1x  y2  cos x cos y  sin x sin y
sin1x  y2  sin x cos y  cos x sin y
with x  2u and y  u to get
z 3  7 z7 3 3cos 12u  u 2  i sin12u  u 2 4
 7 z7 3 1cos 3u  i sin 3u 2 .
Thus cubing any complex number always results in cubing the modulus and
tripling the rotation of the original complex number. It either “lengthens” the com-
plex number if the original modulus is greater than 1, as illustrated in Figure 8.29(a),
or “contracts” it if the modulus is less than 1, as illustrated in Figure 8.29(b). If the
modulus equals 1 and u  0, all that happens is a rotation.

Im
z
> 1 Im
z
< 1

z3

z
3
z3
z = a + bi

z
3
3θ z = a + bi 3θ

θ θ Re
Re
0 0
(a) (b)

FIGURE 8.29A FIGURE 8.29B

In the Problems at the end of this section, we ask you to show that
z4  z3 1z2  7 z7 4 3cos 4u  i sin 4u 4
and that, in general for any positive integer power n,
z n  z n1 1z2
 7 z7 n 1cos u  i sin u 2 n
 7 z7 n 1cos nu  i sin nu2 .
gord.3896.08.pgs 4/24/03 10:12 AM Page 573

8.3 Properties of Complex Numbers 573

This important and extremely useful result is known as DeMoivre’s theorem after
French mathematician Abraham DeMoivre who first discovered it.

DeMoivre’s Theorem
If
z  a  bi  7 z7 1cos u  i sin u 2

then
zn  7 z7 n 1cos nu  i sin nu 2

for any positive integer n.

Complex Conjugates
We know that complex numbers occur in complex conjugate pairs, such as
z  3  5i and z  3  5i when we use the quadratic formula. If z  a  bi is
any complex number, we write its conjugate as z  a  bi, which is shown geo-
metrically in Figure 8.30. Clearly, z and z have the same modulus, 1a2  b2 , so
7 z7  7 z7 . Also, if the angle associated with z is u, the angle associated with z is u.

Im
z = a + bi
=
z
(cos θ + i sin θ )

z

b

θ a
Re
–θ

–b

z

z = a − bi
FIGURE 8.30 =
z
(cos θ − i sin θ )

Using the reflection identities

cos 1u 2  cos u and sin1u 2  sin u,
we find that
z  a  bi  7 z7 3cos 1u 2  i sin 1u 2 4  7 z7 1cos u  i sin u 2.
Applying DeMoivre’s theorem to z gives
1 z 2 n  7 z7 n 1cos u  i sin u 2 n
 7 z7 n 1cos nu  i sin nu 2.
A simple extension of these ideas provides a way of visualizing both the prod-
uct and the quotient of any two complex numbers. We explore this approach in the
Problems at the end of this section.
gord.3896.08.pgs 4/24/03 10:12 AM Page 574

574 CHAPTER 8 More About the Trigonometric Functions

Problems
In Problems 1–9, find the modulus and the associated c. Hypothesize and prove an extension of DeMoivre’s
angle for each complex number. theorem that will allow you to multiply any two
1. z  4  3i complex numbers in trigonometric form. (Hint:
Your extension should reduce to DeMoivre’s theo-
2. z  5  12i rem for z 2 when w  z.)
3. z  12  5i d. Apply the rule that you discovered in part (b) to
4. z  15  20i find the product of
5. z  64  36i i. z  1  2i and w  1  2i
6. z  8  3i 1 23 23 1
ii. z   i and w   i.
7. z  5  7i 2 2 2 2
8. z  3  28 i 50. a. Hypothesize an extension of DeMoivre’s theo-
rem that will allow you to divide one complex
9. z  8  23 i number by another in trigonometric form.
10–18. Find the trigonometric form for each complex b. Apply the rule that you proposed in part (a) to
number in Problems 1–9. find the quotient of
19–22. For each complex number in Problems 1–4, i. z  1  2i and w  1  2i
find z 2 algebraically. 1 23 23 1
ii. z   i and w   i.
23–31 For each complex number in Problems 1–9, 2 2 2 2
find z 2 by using DeMoivre’s theorem. 51. a. Hypothesize an extension of DeMoivre’s theo-
32–35. For each complex number in Problems 1–4, rem that will allow you to determine the square
find z 3 algebraically. root of a complex number z.
b. Apply the rule that you proposed in part (a) to
36–44. For each complex number in Problems 1–9,
find the square root of
find z 3 by using DeMoivre’s theorem.
45. For z  1  2i, calculate and plot z n, for n  0, 1, 1 23
z  i.
2, 3, and 4. 2 2
46. Repeat Problem 45 for z  0.6  0.8i. What differ- c. Algebraically square the complex number that
ence do you observe about the behavior of the two you obtained in part (b) to verify that it actually is
sets of points? the square root of the original number in part (a).
47. Show that z4  z 3 . z  7 z7 4 1cos 4u  i sin 4u 2 for
d. Can you hypothesize a further extension of
DeMoivre’s theorem to extract any desired root
any complex number z. of a complex number? any desired rational
48. Prove DeMoivre’s theorem for any integer power n: power of a complex number? Explain.
z n  7 z7 n 1cos nu  i sin nu2 . 52. A negative real number can be thought of as being
Hint: Write z n  z n1 . z and assume that produced by rotating the corresponding positive
z n1  7 z7 n1 3cos11n  12u2  i sin1 1n  1 2 u 2 4. real number (which is located on the horizontal
axis) through an angle p in the complex plane. Use
49. Suppose that you have two complex numbers
this interpretation to explain why the product of
z  a  bi and w  c  di.
two negative numbers is positive.
a. What is the product of z and w algebraically?
53. Show that, for any pair of complex conjugates
z  a  bi and z  a  bi, z # z  7 z7 2.
b. What is the product of z and w using the
trigonometric forms of z and w?

8.4 The Road to Chaos

In this section we investigate some fascinating results that arise from iteration process-
es applied to complex numbers. Let’s begin with any complex number in trigonomet-
ric form—say, z0  7 z0 7 1cos u  i sin u)—and square it to produce z1  z02 . Using
ˇ
gord.3896.08.pgs 4/24/03 10:12 AM Page 575

8.4 The Road to Chaos 575

DeMoivre’s theorem, we know that the geometric result is a complex number whose
associated angle is 2u and whose modulus is 7 z0 7 2. Recall that, if 7 z0 7 1, we get a ro-
tation and an expansion to a “larger” complex number; if 7 z0 7
1, we get a rotation
and a contraction to a “smaller” complex number; if 7 z0 7  1, we get only a rotation.
Suppose that we next square z1 to produce z2  z12  z04 . If 7 z0 7 1, we get a
further rotation (to the angle 2  2u  4u) and a further expansion. If 7 z0 7
1,
ˇ ˇ

we get the same further rotation (to 4u) and a further contraction. If 7 z0 7  1, we
get only the rotation (to 4u).
What happens if we continue this process indefinitely to produce a sequence of
complex numbers z0 , z1  z02 , z2  z12 , z3  z22 , . . . ? The geometric behavior of
ˇ ˇ ˇ

the terms of this sequence can be predicted easily by extending the reasoning we
just used. If the modulus of the initial value z0 is greater than 1, each successive it-
erate is farther from the origin in the complex plane, at a larger angle, and the se-
quence clearly diverges in a counterclockwise spiral pattern for u 0, as shown in
Figure 8.31(a). If 7 z0 7
1, each successive term is closer to the origin; the succes-
sive iterates converge to 0 in a counterclockwise spiral pattern as each one is a fur-
ther rotation of the original angle u 0, as shown in Figure 8.31(b). Finally, if
7 z0 7  1, all successive iterates fall on the boundary of the unit circle centered at
the origin in the complex plane.

Im
z 0
> 1 Im
z 0
< 1
z3
z2
8θ
z1
4θ z2 z1
z0 z3 z0
2θ
Re Re
θ

(a) (b)

FIGURE 8.31A FIGURE 8.31B

The Julia Set

Let’s focus on the possible initial values for z0 . Any initial point inside the unit cir-
cle starts a sequence that spirals in to the origin; any initial point on the circle itself
starts a sequence that remains on the unit circle; and any initial point outside the
unit circle starts a sequence that spirals away toward infinity.
We can display this graphically in the following way. Visualize the unit circle
centered at the origin in the complex plane, as shown in Figure 8.32. The circle is
drawn in heavy black, the interior is shaded, and the region outside the circle is un-
shaded. Think of the unshaded region as indicating any point that begins a se-
quence that diverges, the shaded region as indicating those initial points for which
the sequence converges to 0, and the black as indicating those initial points for
which the sequence remains on the circle forever. The set of initial points for which
the resulting sequences do not diverge to infinity is known as the Julia set associat-
ed with the function f 1z2  z 2. (It is named after French mathematician Gaston
gord.3896.08.pgs 4/24/03 10:12 AM Page 576

576 CHAPTER 8 More About the Trigonometric Functions

Julia, who discovered the properties of these sets in the 1920s.) The Julia set associ-
ated with f 1z2  z 2 consists of the unit circle and all points inside it.

Im

Re
1

FIGURE 8.32

A relatively small change in what we have just done puts us on the road to
chaos. Instead of using f 1z2  z 2, let’s see what happens if we use f 1z2  z 2  C,
where C is any constant, either real or complex. (You may want to think of this as a
family of functions for different values of C.) We take z1  f 1z0 2  z02  C, so
that z2  f 1z1 2  z12  C, z3  f 1z2 2  z22  C, . . . We now consider a variety
ˇ

ˇ ˇ

of cases with different values for C and with different starting values z0 .
Let’s begin with C  2. If the initial value is z0  3,
z1  z02  C  9  2  11,
ˇ

z2  z12  C  121  2  123,

ˇ

z3  z22  C  1232  2  15,131, . . . ,

ˇ

and the sequence clearly diverges. Using the same C  2 with some other starting
values, we get
if z0  0.2, then z1  2.04, z2  6.1616, z3  39.9653, . . . ;
if z0  1  i, then z1  2  2i, z2  2  8i, z3  58  32i, . . . ;
if z0  0.5  0.2i, then z1  2.21  0.2i, z2  6.844  0.884i, z3  48.059  12.10i . . .
All three sequences seem to diverge to infinity. Of course, we can’t reach such a
conclusion based on just a few examples; they can, at best, suggest what may
happen.
Let’s use DeMoivre’s theorem to analyze the behavior of the successive iterates.
Suppose that z0 is any initial value inside the unit circle, so its modulus is less than
1. When we square it, the modulus for z02 is smaller still. However, when we add 2
ˇ

to it, the point is shifted 2 units to the right, so that z1 must be outside and to the
right of the unit circle.
Now suppose that z0 (or some subsequent iterate) is outside the unit circle. Its
modulus is greater than 1, so the modulus for z02 is larger still. When we add 2 to it,
ˇ

the point is again shifted 2 units to the right. For almost all possible values of z0 ,
the resulting z1 will be outside the unit circle.
There are some exceptions—say, z0  1.1i so that
z1  11.1i2 2  2  1.21  2  0.79.
gord.3896.08.pgs 4/24/03 10:12 AM Page 577

8.4 The Road to Chaos 577

However, it can be shown that, eventually all subsequent iterates will land outside
the circle and ultimately diverge to infinity. (Because each iteration involves a rota-
tion, at some stage one of the successive iterates will eventually land near the hori-
zontal axis to the right and the following iterate will be outside and to the right of
the unit circle.) Thus it turns out that, with f 1z2  z 2  2, for every initial point
in the complex plane, the resulting sequence diverges. The Julia set associated with
the function f 1z2  z 2  C, when C  2, will be completely empty because all
initial points give rise to sequences that eventually diverge. Our diagram of this
Julia set will be entirely unshaded because there are no initial points that start con-
vergent sequences.
Similarly, if C  2i, all sequences will diverge regardless of the initial value for
z0 . The additive constant 2i results in a shift upward of 2 units in the imaginary di-
rection. Pick several initial values for z0 (real, imaginary, or complex) and see what
happens when you calculate the successive iterates.
However, if C  0.2i and we start with z0  0.5  0.2i, we obtain
z1  10.5  0.2i2 2  0.2i  0.21  0.4i;
z2  0.1159  0.368i; z6  0.0394  0.1881i;
z3  0.1220  0.1147i; z7  0.0338  0.1852i;
z4  0.0017  0.1720i; z8  0.0332  0.1875i.
z5  0.0296  0.2006i; z9  0.0341  0.1876i.
The sequence apparently converges to some point in the complex plane.
Unfortunately, repeating this process for every possible starting value z0 is not
practical. Instead, we use a computer to perform such calculations for a large num-
ber of points in a grid to give a representative picture of what happens. As with the
previous cases, we leave any initial point that starts a sequence that diverges to in-
finity unmarked to become part of the unshaded region. We put a small dot at any
initial point that starts a sequence that converges to some point in the complex
plane so that it will be part of the shaded Julia set.
The resulting picture of the Julia set for the function f 1z2  z 2  0.2i is
shown in Figure 8.33.

FIGURE 8.33
gord.3896.08.pgs 4/24/03 10:12 AM Page 578

578 CHAPTER 8 More About the Trigonometric Functions

This picture does not indicate the limits of any of the sequences; it shows only
those points that start sequences that have limits. Typically, if you take points in the
interior of the shaded region, it turns out that nearby starting points tend to converge
to limits that are relatively close to one another. However, if you take points near the
boundary, very different results can occur. Initial points that are extremely close to-
gether can produce sequences that converge to radically different limits. The result is
an instance of mathematical chaos because the behavior has no predictable patterns.
Points that are very close together, provided that they are both near the boundary of
the Julia set, may well lead to sequences that behave very differently. If you were to
zoom in on the portion of the Julia set near these boundaries, you would see an ever
more intricate design illustrating how nearby points can start sequences that either
diverge or converge. They occur in a totally chaotic and unpredictable manner.
A striking illustration of this outcome is shown in Figure 8.34, which is the Julia
set corresponding to C  0.2  0.7i. Note how intricate the boundary appears.
Figure 8.35 shows the result of zooming in on the upper left corner of the Julia set
shown in Figure 8.34. Observe how roughly similar patterns repeat; this kind of rep-
etition is typical of what happens when you zoom in repeatedly on Julia sets associ-
ated with most values for C. Also, note how much more jagged the boundary looks
as more details appear in the magnified image in Figure 8.35, which also is typical.

FIGURE 8.34

FIGURE 8.35
gord.3896.08.pgs 4/24/03 10:12 AM Page 579

8.4 The Road to Chaos 579

The Julia set associated with a complex constant C can be far more intricate than
we have shown so far; it can, for instance, consist of a large variety of disconnected
pieces. It may even consist of nothing but a collection of isolated points like a set of dust
particles. You may want to experiment with some of these ideas yourself, using any of
the many computer programs available for displaying Julia sets for iterated functions.

The Mandelbrot Set

There is a completely different way of looking at these ideas. In the discussion of
Julia sets, we considered the function f 1z2  z 2  C, selected a particular value
for C, and then examined points in the complex plane as starting points z0 for iter-
ated sequences. Now let’s reverse this.
Suppose instead that we select a particular starting point z0 and examine the ef-
fects of using different values for the complex constant C in f 1z2  z 2  C. Thus
our view of the complex plane has shifted—it now represents all different constants
rather than all different starting points. In particular, suppose that we select z0  0 as
the starting point for all sequences. Then, for any constant C, z1  02  C  C,
z2  z12  C  C2  C, and so on. Clearly, if C is large (far from the origin in the
ˇ

complex plane), all successive iterates will be larger still and the successive points of
the sequence will diverge. However, if C is fairly small, the successive iterates may re-
main close to the origin and the sequence may converge to some finite complex value.
The Julia set associated with the function f 1z2  z 2  C consists of all initial
points z0 for which the sequences converge for a given constant C. Similarly, the
Mandelbrot set associated with the function f 1z2  z 2  C (named after French
mathematician Benoit Mandelbrot) consists of all constants C for which the se-
quences starting from z0  0 fail to diverge. For this initial point z0  0, the Man-
delbrot set illustrated in Figure 8.36 shows those constants C for which the
corresponding sequences remain close to the origin. As with a Julia set, the bound-
ary of the Mandelbrot set is an incredibly intricate structure. If you zoom in on it,
as shown in Figure 8.37, you will see remarkable shapes with no predictable pat-
terns; however, the original overall shape shown in Figure 8.36 appears to repeat at
all levels of magnification. The main heart-shaped portion of the Mandelbrot set is
called a cardioid, which we discuss in Chapter 9; the portion to the left of the car-
dioid is actually a circle.

FIGURE 8.36
gord.3896.08.pgs 4/24/03 10:12 AM Page 580

580 CHAPTER 8 More About the Trigonometric Functions

FIGURE 8.37A

FIGURE 8.37B

FIGURE 8.37C
gord.3896.08.pgs 11/21/03 10:52 AM Page 581

8.4 The Road to Chaos 581

These displays show the Mandelbrot set with different shadings to indicate
how quickly different sequences diverge from the starting value z0  0. When dif-
ferent colors are used, the results are even more dramatic. You may want to exam-
ine the Mandelbrot set, using one of the programs available for displaying it. All
such programs allow you to see the details at different levels of magnification as
you zoom in on the boundary. In theory, there is no limit to the degree of com-
plexity of the boundary. Such a shape is known as a fractal.
Many shareware programs are available (one of the most popular is called
FracInt) that will let you investigate both Julia and Mandelbrot sets. This subject is
one of the most exciting areas of current mathematical research, and many new
and important theorems have been proven in the last few years. These ideas have
also formed the basis for many of the computer graphics images that you have un-
doubtedly seen in today’s movies.

Problems
1. a. Use the quadratic formula to find a condition on 4. Explain graphically the significance of C in deter-
those values of C for which the sequence of iter- mining whether the iteration process based on
ates x  f 1x 2  x 2  C has a real limiting value. x  f 1x 2  x 2  C has a real limiting value by
b. Verify your condition in part (a) by using looking at the graphs of y  x 2  C and y  x.
C  0.1, starting with x0  0.5 and performing 5. Explain graphically why the iteration process based
enough iterations to see the eventual behavior. on the function x  f 1x 2  x 3  C must have at
c. Repeat part (b), using C  0.4. least one real limiting value.
2. a. What is the limiting value you expect if C  1>4 6. Consider iterations x  f 1x2 based on the function
for the sequence of function iterations based on f 1x 2  x  sin x.
x  f 1x 2  x 2  C ? a. Begin the iteration process at x0  2 and per-
b. Start the iteration process at x0  0.5  0.5i form enough iterations to allow you to recognize
and perform enough iterations to verify that the the limit of the resulting sequence.
process seems to be converging to your answer b. Repeat part (a), starting with x0  5.
for part (a). c. Repeat part (a), starting with x0  8. How does
c. Start the iteration process at x0  1  i and per- the limiting value compare to p?
form enough iterations to determine the eventu- d. Repeat part (c), starting with x0  15.
al behavior of the sequence of iterates. How e. Based on the function f, explain why all limits
could you have anticipated the result without will be some multiple of p.
7. Consider iterations x  f 1x2 based on the function
performing the actual calculations?
3. You can think of the iteration scheme for f 1x 2  x  cos x. Predict the possible values that
x  f 1x 2  x 2  C as the difference equation
xn1  f 1xn 2  xn2  C. What are the equilibrium
can arise for the limits based on the function f. Ver-
ˇ

ify whether your predictions are correct if you start

levels for the solutions to the difference equation? with initial values x0  1, 3, 7, and 12.
Under what conditions on C will the equilibrium
values be real?

Chapter Summary

In this chapter, we continued our discussion of trigometric functions. In

particular, we discussed the following:
◆ The fundamental identities that relate the sine and cosine functions.
◆ Some identities involving the tangent function.
◆ How to approximate the sine and cosine functions with polynomial functions.
gord.3896.08.pgs 4/24/03 10:12 AM Page 582

582 CHAPTER 8 More About the Trigonometric Functions

◆ How the accuracy of a polynomial approximation depends on the degree of

the polynomial.
◆ How to convert a complex number to its equivalent trigonometric form.
◆ How to construct powers of complex numbers with DeMoivre’s theorem.
◆ The Julia set that is associated with a function f 1z2 and the idea of chaos.
◆ The Mandelbrot set that is associated with a function f 1z2 .

Review Problems
Determine graphically which of the relationships in close to 1. Using your function grapher, estimate
Problems 1–9 might be identities and which clearly are how far the expression 3T3 1x 2 4 2  3T2 1x 2 4 2 is from
not identities. For those that appear to be identities, 1 for any value of x between 1 and 1 radian.
prove them algebraically. 14. a. Convert the complex numbers z  6  8i
1. sin x cos2x  sin3x  cos x and w  5  2i to trigonometric form.
2. sin x cos2x  sin3x  sin x b. Use the results from part (a) to find z . w, z>w,
and w>z.
3. sin x cos2x  sin3x  cos2x
15. A complex number z has modulus 3 and an associ-
4. sin x cos2x  sin3x  sin2x ated angle of 52°.
5. 1sin x  cos x2 2  1  sin 2x a. Write the complex number in trigonometric form.
sin u cos u 1 b. Write the complex number in the usual form
6.  
1  cos u sin u sin u z  a  bi.
1 1 2 c. Find the fifth power of this complex number z.
7.   d. Find the square root of this complex number z.
1  cos t 1  cos t sin2t
Use your function grapher to estimate the period for
8. cos4u  sin4u  cos 2u
each function. Express your answers as multiples of
9. cos6u  sin6u  cos 2u p.
10. Prove each identity. 16. f 1x 2  sin 3x  cos 2x
a. cos1x  y2  cos1x  y2  2 cos x cos y 17. f 1x 2  sin 3x  cos 4x
18. f 1x 2  sin 4x  cos 2x
b. sin1x  y2  sin1x  y2  2 sin x cos y

19. f 1x 2  sin 2x  cos 4x

11. Use the Taylor polynomial approximation of degree
n  3 for the sine function to estimate the value of
the function f 1x2  sin 3x at x  0.2. Sketch the 20. f 1x 2  sin a xb  cos 2x
1
graph of the function and the approximating poly- 2

21. f 1x 2  sin 3x  cos a xb

nomial on the same set of axes. 1
12. Repeat Problem 11 with degree n  5. Discuss any 2
differences you observe. 22. Based on your answers to Problems 16–21, conjec-
13. You know that sin x  T3 1x2  x  x 3>6 and ture a general rule for the period of the function
cos x  T2 1x2  1  x 2>2 when x is reasonably f 1x 2  sin mx  cos nx, for any m and n.
close to 0, so 3T3 1x2 4 2  3T2 1x2 4 2 should be fairly
gord.3896.09.pgs 4/24/03 10:15 AM Page 583

9
Geometric Models

9.1 Introduction to Coordinate Systems

What is a coordinate system? In simple terms, a coordinate system provides a way
to locate and identify points in the plane. In the usual rectangular or Cartesian coor-
dinate system, every point can be pictured in either of two ways. First, a point P
with coordinates 1x0 , y0 2 can be thought of as lying at the corner of a unique rec-
tangle whose opposite corner is at the origin and two of whose sides lie along the
two coordinate axes, as illustrated in Figure 9.1(a). The base of this rectangle is x0 ,
and its height is y0 . Second, the point P can be thought of as the intersection of two
perpendicular lines, one parallel to the y-axis at a distance of x0 from it and the
other parallel to the x-axis at a height of y0 from it, as illustrated in Figure 9.1(b).
y y

P(x 0 , y0) y = y0 P(x 0 , y0)

y0 x = x0

x x
x0
(a) (b)

FIGURE 9.1

Mathematicians have found that, in many situations, rectangular coordinates

are not the most natural or the most effective way to locate points and have devel-
oped alternative coordinate systems. One such approach involves the use of two
axes that are not perpendicular, but rather meet at the origin at some angle other
than a right angle. Points in such a slanted coordinate system can be located at the
opposing vertex of a parallelogram, as illustrated in Figure 9.2.

583
gord.3896.09.pgs 4/24/03 10:15 AM Page 584

584 CHAPTER 9 Geometric Models

P(u 0, v 0)
v0
u0
u
FIGURE 9.2

Another approach is to locate a point by using a circle centered at the origin O

instead of a rectangle. To do so requires specifying both the radius of the circle and
an angle u to indicate where on the circle the point is located. This approach leads
to the polar coordinate system, which is illustrated in Figure 9.3. We investigate
this coordinate system in Sections 9.6 and 9.7.

P(r, θ )
r

θ
O

FIGURE 9.3

Other approaches are used for particular applications that involve locating
points lying on some ellipse centered at the origin (an elliptic coordinate system),
on some parabola (a parabolic coordinate system), or on a hyperbola (a hyperbolic
coordinate system), as illustrated in Figures 9.4(a–c), respectively. In fact, the long
range navigation (LORAN) system used by navigators in ships and planes to locate
their positions is based on the fact that every point in a plane can be interpreted as
lying at the intersection of two hyperbolas in a hyperbolic coordinate system.
y y y

x x x

FIGURE 9.4

In this chapter, we first develop several special characteristics of the rectangu-

lar coordinate system and then show how to represent some extremely important
curves. Later we explore other ways to represent functions.
gord.3896.09.pgs 4/24/03 10:15 AM Page 585

9.2 Analytic Geometry 585

9.2 Analytic Geometry

One of the most useful and far-reaching developments in mathematics is Rene
Descartes’s idea of representing algebraic concepts geometrically. This approach,
known as analytic geometry, lets you visualize the mathematics graphically to com-
plement the algebraic approach that is based on symbols. Everything we have done in-
volving graphs of functions is an outgrowth of Descartes’s ideas. In this section we
examine some additional ideas involving points, lines, and circles in the plane.
We begin by considering the two points A at 1x 0 , y0 2 and B at 1x1 , y1 2 in the
plane. You already know how to find an equation of the line through them by using
either the point–slope form
y  y0  m1x  x0 2
or the slope–intercept form
y  mx  b,
where the slope of the line is
y1  y0
. m
x1  x 0
Alternatively, we have the implicit form for the equation of a line,
ax  by  c,
where c>a and c>b represent the x- and the y-intercepts of the line, respectively.

Distance Between Points

We now ask: What is the distance between the points A at 1x0 , y0 2 and B at 1x1 , y1 2?
We write this distance as 0 AB 0 . Figure 9.5 shows that the points A and B determine
a right triangle ABC; the coordinates of point C are 1x1 , y0 2 because C is at the same
horizontal distance as B (measured from the y-axis) and at the same vertical height
as A (measured from the x-axis). Moreover, the horizontal distance from A to C is
x1  x0 ; it is the change, or difference, in the x-coordinates. Similarly, the vertical
distance from C to B is y1  y0 ; it is the change in the y-coordinates. Consequent-
ly, the distance from A to B is the length of the hypotenuse of this right triangle.
The Pythagorean theorem therefore gives us the distance formula.

y
B(x 1, y 1)
y1

|AB| = √(x 1 – x 0 ) 2 + (y1 – y 0 ) 2

y1 – y 0

x1 – x 0
y0 C(x 1, y 0 )
A(x 0, y 0 )

x
FIGURE 9.5 x0 x1
gord.3896.09.pgs 4/24/03 10:15 AM Page 586

586 CHAPTER 9 Geometric Models

Distance Formula
The distance from point A at 1x0 , y0 2 to point B at 1x1 , y1 2 is
0 AB 0  21x1  x0 2 2  1y1  y0 2 2 .

E XAMPLE 1
Find the distance from the point A at 12, 52 to the point B at 16, 8 2.
Solution Applying the distance formula gives
0 AB 0  216  2 2 2  18  5 2 2
 216  19  225  5 units.

◆
Consider again the two points A at 1x0 , y0 2 and B at 1x1 , y1 2 in the plane. Suppose
that we want to determine the midpoint M of the line segment connecting A to B.
Figure 9.6 shows that the points A and B determine a right triangle ABC and that the
points A and M determine a smaller right triangle AMD. These two right triangles

y
B(x 1, y 1)

1 1
M(x 0 + 2
(x 1 – x 0 ), y 0 + 2
( y 1 – y 0 ))

1
2 AB 1
2 CB
1
2 AC
C(x 1, y 0 )
A(x 0, y 0 ) D (x 0 + 12 (x 1 – x 0 ), y 0 )

x
FIGURE 9.6

are similar, and hence their corresponding sides are proportional (see Appendix A4).
Because M is halfway from A to B, we see that D is halfway from A to C, and the
height DM is half the height CB. Thus the x-coordinate at D (and hence also at M) is
x  x0  1x1  x0 2 .
1
2
Similarly, because the height DM is half the height CB, the y-coordinate at M is
y  y0  1 y1  y0 2 .
1
2
We can rewrite these expressions as

x0  1x1  x0 2  x0  x1  x0  1x1  x0 2
1 1 1 1
2 2 2 2
and
y0  1y1  y0 2  y0  y1  y0  1 y1  y0 2 .
1 1 1 1
2 2 2 2
gord.3896.09.pgs 4/24/03 10:15 AM Page 587

9.2 Analytic Geometry 587

Thus the coordinates of the midpoint M of a line segment are simply the averages
of the x-coordinates and the y-coordinates of the endpoints, respectively.

Midpoint Formula
The midpoint M of the line segment from A at 1x0 , y0 2 to B at 1x1 , y1 2 is at

1x  x0 2, 1 y  y0 2
1 1
x  x0  y  y0 
2 1 2 1
or
x1  x0 y1  y0
x , y .
2 2

E XAMPLE 2
Find the midpoint of the line segment joining A at 11, 11 2 and B at 13, 7 2.
Solution The coordinates of the midpoint are
x  x0  1x1  x0 2
1
2
 1  13  1 2  1  1  2
1
2
and
1 y  y0 2
1
y  y0 
2 1
 11  17  11 2  11  14 2  9.
1 1
2 2
Alternatively,
x 1  x2 31 y1  y2 7  11
x   2 and y    9.
2 2 2 2
Figure 9.7 shows the solution.

12
A(1, 11)
10
M(2, 9)
8
B(3, 7)
6

x
FIGURE 9.7 0 1 2 3 4

◆
We might also want to determine a point at some other fraction of the distance
from A to B. To do so, we simply extend the preceding argument to determine a
gord.3896.09.pgs 4/24/03 10:15 AM Page 588

588 CHAPTER 9 Geometric Models

point P at any distance from A to B. Suppose that we want the point one-quarter of
the way from A to B. We then have
x  x0  1x1  x0 2 and y  y0  1 y1  y0 2.
1 1
4 4

Think About This Verify that this quarter-distance formula is correct by using an argument compara-
ble to the one used for the midpoint formula. ❐

In general, if we want the point P at a fraction t of the distance from A to B, it

will be located at
x  x0  t . 1x1  x0 2
y  y0  t . 1 y1  y0 2 ,
as shown in Figure 9.8. Incidentally, if t  1, we get a point on the line beyond B,
and if t  0, we get a point on the line before A.

y
B(x 1, y 1)

P
t

A(x 0, y 0 )

x
FIGURE 9.8

E XAMPLE 3

a. Find the point P located three-fifths of the way from A at 11, 32 to B at 14, 132 .
b. Find the point Q located seven-fifths of the way from A to B.

Solution

a. For P with t  3>5,

x  x0  t . 1x1  x0 2  1  34  112 4  2 and
3
5
y  y0  t . 1 y1  y0 2  3  113  32  9.
3
5
Verify by plotting the points 11, 3 2, 14, 13 2, and 12, 9 2 that P is located three-fifths
the way from A to B.
b. Similarly, for Q with t  7>5,
x  x0  t . 1x1  x0 2  1  34  112 4  6 and
7
5
y  y0  t . 1 y1  y0 2  3  113  32  17.
7
5
Plot Q at 16, 17 2 and note that this point lies on the line through A and B and is be-
yond B.
◆
gord.3896.09.pgs 4/24/03 10:15 AM Page 589

9.2 Analytic Geometry 589

We could continue this example with different values of t to find other points
on the line. In fact, every value of t determines a unique point on the line joining A
at 1x0 , y0 2 and B at 1x1 , y1 2. Therefore the two equations for x and y give us a differ-
ent way of representing the line. They are known as a parametric representation or
parametric equations of the line, and the quantity t is called a parameter.

Parametric equations of the line through 1x0 , y0 2 and 1x1 , y1 2 are

x  x0  1x1  x0 2t and
y  y0  1 y1  y0 2t.

Note that this parametric form involves two interrelated equations for the line,
not a single equation as in the point–slope form. It is possible to eliminate the pa-
rameter t to produce a single equation for the line. We ask you to do so in the Prob-
lems at the end of this section. However, the parametric form can provide valuable
information.

The Equation of a Circle

We apply the concept of distance between two points in the plane to define a circle.
A circle is the set of all points in the plane at a fixed distance from a fixed point.
The fixed distance is called the radius, and the fixed point is called the center.
This definition allows us to find a general equation for any circle. Let r be the
radius and let point C with coordinates 1x0 , y0 2 be the center of a circle. A point P
with coordinates 1x, y2 lies on this circle provided that its distance from the center
C is r, as shown in Figure 9.9. The distance formula gives the equivalent expression
0 CP 0  21x  x0 2 2  1 y  y0 2 2  r.
We can eliminate the square root in this equation by squaring both sides and thus
obtain the standard form for the equation of a circle.

P(x, y)
r

C(x 0, y 0 )

x
FIGURE 9.9

The equation of the circle with radius r centered at 1x0 , y0 2 is

1x  x0 2 2  1 y  y0 2 2  r 2.
gord.3896.09.pgs 4/24/03 10:15 AM Page 590

590 CHAPTER 9 Geometric Models

For instance, the circle of radius 8 centered at 15, 2 2 has the equation
1x  5 2 2  1 y  2 2 2  82  64,
whereas the equation of the circle of radius 3 centered at 15, 0 2 is
1x  5 2 2  y 2  9.
As a special case, the equation of a circle of radius r centered at the origin is
x 2  y 2  r 2.
Note that the equation of a circle does not represent a function. Picture any
vertical line that passes through the circle but is not tangent to the circle—it inter-
sects the circle twice and so the circle fails the vertical line test. That is, each such
value of x has two corresponding values of y, which violates the definition of a
function, as shown in Figure 9.10.
y

C(x 0, y 0 )

x
x0 − r x0 x0 + r
FIGURE 9.10

We get a similar result algebraically. For example, the circle of radius 10 cen-
tered at the origin has the equation
x 2  y 2  100.
If we select x  6, say, then
36  y 2  100 so that y 2  64,
which has the solutions
y  8 or y  8.
Again, two values of y correspond to one value of x. Even though a circle does not
represent a function, it is nonetheless a very important curve. We discuss several
other curves that do not represent functions in Sections 9.3 and 9.4.
Let’s start with the equation of the circle of radius 8 centered at 15, 2 2:
1x  5 2 2  1y  2 2 2  82  64.
We expand the left-hand side and combine like terms to get
x 2  10x  25  y 2  4y  4  64
or
x 2  y 2  10x  4y  35  0,
gord.3896.09.pgs 4/24/03 10:15 AM Page 591

9.2 Analytic Geometry 591

which is an equivalent, although different, representation for the same circle.

Clearly, we could do the same with the equation of any circle,
1x  x0 2 2  1 y  y0 2 2  r 2,
which is centered at 1x0 , y0 2 with radius r. Expanding the left-hand side, we get
x 2  2x0 x  x02  y 2  2y0 y  y02  r 2
ˇ ˇ

or, equivalently,
x 2  y 2  2x0 x  2y0 y  x02  y02  r 2  0.
ˇ ˇ

Because x0 , y0 , and r are constants, we can write this equation in the alternative
form
x 2  y 2  Cx  Dy  E  0,
where we have introduced the new constants
C  2x0 , D  2y0 , and E  x02  y02  r 2. ˇ ˇ

Such an equation, known as the general equation of the circle, represents a cir-
cle for any choice of constants C, D, and E such that the radius of the circle is
positive.
Suppose that we start with an equation such as
x 2  y 2  10x  4y  35  0,
which we know from the preceding derivation is the equation of a circle. Exam-
ples 4 and 5 demonstrate how to work backward from this equation to determine
the center and radius of the circle.

E XAMPLE 4
Show that
x 2  y 2  10x  4y  35  0
is the equation of a circle by finding its center and radius.
Solution To solve this problem, we use the technique of completing the square (see Ap-
pendix A8) in both the x- and y-terms on the left-hand side and obtain
x 2  y 2  10x  4y  35  1x 2  10x 2  1 y 2  4y2  35
 3 1x 2  10x  25 2  25 4  3 1y 2  4y  42  4 4  35
 3 1x  5 2 2  254  3 1 y  2 2 2  44  35
 1x  5 2 2  1y  2 2 2  64  0.
Adding 64 to both sides, we get
1x  5 2 2  1 y  2 2 2  64  82.
This is the equation of the circle with radius 8 and center at 15, 22.
◆

E XAMPLE 5
Find the radius and the center of the circle whose equation is
x2  y2  8x  10y  8  0.
gord.3896.09.pgs 4/24/03 10:15 AM Page 592

592 CHAPTER 9 Geometric Models

Solution To solve this problem, we again complete the square in both the x- and
y-terms on the left-hand side and obtain
x 2  y 2  8x  10y  8  1x 2  8x 2  1 y 2  10y2  8
 3 1x 2  8x  16 2  16 4  3 1 y 2  10y  252  25 4  8
 3 1x  4 2 2  16 4  3 1y  5 2 2  254  8
 1x  42 2  1 y  5 2 2  49.
Therefore the original equation becomes
1x  4 2 2  1 y  5 2 2  49  0.
Adding 49 to both sides, we get
1x  4 2 2  1 y  5 2 2  49  72,
which is the equation of a circle with radius 7 and center at 14, 52.
◆
Note that not every equation of the form x 2  y 2  Cx  Dy  E  0 repre-
sents the equation of a circle. In Examples 4 and 5, the constant term we ended up
with on the right, either 64 or 49, was a positive number, so we could take the
square root to get a radius. However, the constant on the right can be a negative
number, in which case the equation does not represent a circle. In fact, even if the
constant on the right is 0, we would not have a circle because the radius would be
0. We ask you to investigate such cases in the Problems at the end of this section.

E XAMPLE 6
In Example 4 of Section 4.6, we found two points on the line through the Earth and the
moon at which the gravitational forces of the Earth and the moon on a spacecraft are ex-
actly equal in size numerically. One point is 216 thousand miles from the Earth toward
the moon and the other is 270 thousand miles from the Earth, which is 30 thousand
miles beyond the moon. Find all such points in the plane containing the Earth, the
moon, and the sun.
Solution We set up a coordinate system with the Earth at the origin and the moon on
the horizontal axis, 240 thousand miles to the right. Suppose that the two gravitational
forces are numerically equal at some other point P in the plane with coordinates 1x, y2,
as shown in Figure 9.11. (Technically, we should consider not only the size of the two
forces but also the directions in which they are exerted; that requires the notion of a vec-
tor quantity, which we discuss in Chapter 10.)

(x, y)
re rm

FIGURE 9.11 240

Earth Moon

The distance formula gives the distance re from the Earth to the point P as
re  2x 2  y 2 so that re2  x 2  y 2.
ˇ

Similarly, the distance rm from the moon to the point P is

rm  21x  240 2 2  y 2 so that rm2  1x  2402 2  y 2.
ˇ
gord.3896.09.pgs 4/24/03 10:15 AM Page 593

9.2 Analytic Geometry 593

Let m0 be the mass of the spacecraft, m1 be the mass of the Earth, and m2 be the mass of
the moon. Because the Earth is 81 times as massive as the moon, m1  81m2 . Based on
the universal law of gravitation, the size of the gravitational force Fe that the Earth exerts
on the spacecraft is
Gm0 m1 Gm0 . 181m2 2 81Gm0 m2
Fe  2  2 
re re re2
and the size of the gravitational force Fm that the moon exerts on the spacecraft is
Gm0 m2
Fm  .
rm2 ˇ

Equating these two expressions yields

81Gm0 m2 Gm0 m2
2  .
re ˇ rm2 ˇ

Dividing both sides by the constant Gm0 m2 gives

81 1
2  ,
re
ˇ rm2 ˇ

or cross-multiplying,
81rm2  re2.
ˇ ˇ

We substitute the expressions for the two distances to get

81 3 1x  240 2 2  y 2 4  x 2  y 2,
or
81 3x 2  480x  1240 2 2  y 2 4  x 2  y 2
81x 2  811480 2x  811240 2 2  81y 2  x 2  y 2.
Collecting like terms gives
80x 2  38,880x  811240 2 2  80y 2  0.
Dividing through by the common factor 80 yields
x 2  486x  58,320  y 2  0,
which suggests the equation of a circle. We complete the square on the x terms on the
left-hand side and get
486 2 486 2
x 2  486x  58,320  y 2  x 2  486x  a b  a b  58,320  y 2  0.
2 2
Therefore
1x  243 2 2  12432 2  58,320  y 2  0,
or
1x  243 2 2  y 2  1243 2 2  58,320  729.
That is, in the plane formed by the Earth, the moon, and the sun, the size of the gravitation-
al forces from the Earth and the moon are numerically equal at every point on a circle of ra-
dius 1729  27 thousand miles centered at a distance of 243 thousand miles from the
Earth on a line through the moon. In fact, the center of this circle is just beyond the moon.
◆
Incidentally, even though a circle does not fulfill the requirements of a function,
if we restrict our attention to either its upper half or its lower half, the resulting
gord.3896.09.pgs 4/24/03 10:15 AM Page 594

594 CHAPTER 9 Geometric Models

semicircle does represent a function. For the circle x 2  y 2  r 2, we obtain the

equations for the semicircles by solving for y:
y 2  r 2  x 2 so that y   2r 2  x 2 .
Thus the upper semicircle is the graph of the function y  f 1x 2  2r 2  x 2 , and
the lower semicircle is the graph of the function y  g1x2   2r 2  x 2 .
Problems
In Problems 1–6, find the distance between each pair of form for the equation of a circle. Use it to determine the
points. radius and center of the circle. Then draw the graph of
1. 12, 42 and 15, 82 the circle.
2. 12, 4 2 and 17, 162 16. x 2  y 2  4x  6y  3
3. 14, 12 and 10,4 2 17. x 2  y 2  4x  6y  12
4. 12, 5 2 and 10, 42 18. x 2  y 2  10x  4y  7
5. 11, 52 and 13, 72 19. x 2  y 2  10x  4y  71
6. 13, 12 and 15, 42 20. x 2  y 2  2x  6y  9
7. Find the midpoint of the line segment joining the 21. x 2  y 2  2x  6y  6  0
points in Problem 1. 22. Determine which of the following equations repre-
8. Find the midpoint of the line segment joining the sent a circle and which do not. Explain.
points in Problem 2. a. x 2  y 2  4x  6y  15  0
9. Find the point one-third the way from the first b. x 2  y 2  4x  6y  13  0
point to the second point in Problem 3. c. x 2  y 2  4x  6y  12  0
10. Find the point three-fourths the way from the first 23. The equations x  3  2t, y  4  5t form a
point to the second point in Problem 4. parametric representation of a line.
11. Find the equation of the circle that has center a. Construct a table of values for x and y corre-
15, 22 and passes through the point P at 18, 22 . sponding to t  2, 1, 0, 1, . . . , 5.
12. Find the equation of the circle that has center b. Plot these points and verify that they do seem to
13, 72 and passes through the point P at 12, 5 2. lie on a line.
13. Find the equation of the circle that has 12, 42 and
c. What is the slope of this line?
110, 42 as the endpoints of a diameter.
d. What is a point–slope form for the equation of
this line using t  1?
14. Find the equation of the circle that has 12, 3 2 and e. Use the midpoint formula to find the midpoint
14, 112 as the endpoints of a diameter. of each of the consecutive line segments deter-
15. Repeat Problems 13 and 14, using the facts that any mined by the entries in your table from part (a).
angle inscribed in a semicircle is a right angle and Then use the parametric representation of the
that perpendicular lines have slopes that are nega- line with t  1.5, 0.5, 0.5, 1.5, 2.5, 3.5, and
tive reciprocals. 4.5. How do the results compare? Explain.
24. Consider again the parametric representation of
the line x  3  2t, y  4  5t in Problem 23.
(x, y) Eliminate the parameter t from the two equations
by first solving the first equation for t in terms of x
and then substituting the result into the second
equation. How does this result compare to the re-
sult obtained in part (d) of Problem 23?
In Problems 16–21, complete the square in both the x- 25. Start with the general parametric equations of a line
and y-terms for each equation to obtain the standard x  x0  1x1  x0 2t, y  y0  1 y1  y0 2 t
gord.3896.09.pgs 4/24/03 10:15 AM Page 595

9.3 Conic Sections: The Ellipse 595

and algebraically eliminate the parameter t. Identify find values of the parameter t that satisfy the equa-
the equation you produce. tion of the circle.)
26. a. Find the slope of the line having the paramet- 28. The three points P at 10, 2 2, Q at 12, 4 2, and R at
ric representation 14, 0 2 are noncollinear and as such determine a circle.
x  1  2t, y  2  3t. Find an equation of this circle. (Hint: Substitute the
b. Sketch the graph of this line. coordinates of each point into the general equation of
the circle, x 2  y 2  Cx  Dy  E  0, and then
27. Find the points at which the line x  1  2t, y 
solve the resulting system of three equations in three
2  t intersects the circle x 2  y 2  25. (Hint: First
unknowns.)

Exercising Your Algebra Skills

In Problems 1–8, complete the square for each expression. 5. y 2  10y  26 6. y 2  10y  26
1. x 2  8x  25 2. x 2  8x  25 7. y 2  4y  12 8. y 2  4y  12
3. x 2  6x  5 4. x 2  6x  5

9.3 Conic Sections: The Ellipse

When we introduced functions and their graphs in Chapter 1, we said that not
every graph represents a function. In Section 9.2 we pointed out that a circle is not
a function. Several other important curves that have useful and interesting proper-
ties similarly are not functions. We investigate some of these curves in this section
and Section 9.4.
Consider any equation of the form
Ax 2  By 2  Cx  Dy  E  0,
where A, B, C, D, and E are constants, provided that at least one of A and B is not 0.
In particular, if A  B, we can divide through by this constant, so that the result is
the equation of a circle if the coefficients lead to a positive radius. Let’s see what
happens when A  B.
The graph of any equation of the form Ax 2  By 2  Cx  Dy  E  0 is
known as a conic section. To see why, consider a slice through the double right cir-
cular cone shown in Figure 9.12. If the slicing plane is horizontal, each slice is a
circle. However, if the slicing plane is inclined slightly from the horizontal, the curve
produced is oval in shape, rather than circular, and is an ellipse. (Imagine a diagonal
slice through a round salami.) In fact, the sharper the angle of the slice, the more
elongated the ellipse will be, as shown in Figure 9.13. If the angle of slicing is in-
creased further so that it is parallel to the “edge” of the cone, the resulting curve is a
parabola, as shown in Figure 9.14. If the angle of the slice is increased still further,
the slicing plane intersects both the upper and lower parts of the cone and produces
a pair of separated curves, known as a hyperbola, as shown in Figure 9.15.
In summary, there are three types of conic sections: (1) the ellipse, (2) the
parabola, and (3) the hyperbola. The circle is a special case of the ellipse.
The conic sections occur often in various applications of mathematics and sci-
ence. For instance, the orbits of the planets about the sun are ellipses. The paths of
many comets and meteoroids are hyperbolic. A cross section of the metallic reflec-
tor inside a flashlight or an automobile headlight is a parabola. The path of a
gord.3896.09.pgs 4/24/03 10:15 AM Page 596

596 CHAPTER 9 Geometric Models

FIGURE 9.12 FIGURE 9.13 FIGURE 9.14 FIGURE 9.15

Slicing planes Slicing planes Slicing planes Slicing planes
are horizontal are at slight parallel to the at steeper
angles side of the angles
cone

thrown object, such as a perfect “spiral” pass in football or a “line drive” in baseball,
is also a parabola.
Although we typically use formulas when working with the conic sections, we
define them formally from a purely geometric perspective. This approach is analo-
gous to the way we defined a circle in Section 9.2 as the set of all points at a fixed
distance from a single fixed point, its center. In this section, we study the ellipse and
consider the hyperbola and the parabola in Section 9.4.

The Ellipse
An ellipse is defined as the set of all points in the plane for which the sum of the
distances to two fixed points is a constant. The two fixed points are called the foci
(the plural of focus) of the ellipse. The midpoint of the line segment joining the
foci is the center of the ellipse.
When the two foci are far apart, the resulting ellipse is very elongated. When
the two foci are close together, the ellipse is close to circular and, in fact, when the
two foci merge into a single point, the ellipse is a circle.
For convenience, we assume that the center of an ellipse is at the origin and
that the two foci lie on the x-axis. Suppose that the foci are at F1 with coordinates
1c, 02 and at F2 with coordinates 1c, 0 2, as illustrated in Figure 9.16. A point P

P(x, y)

F1(c, 0)
x
F2(–c, 0) Center

FIGURE 9.16
gord.3896.09.pgs 4/24/03 10:15 AM Page 597

9.3 Conic Sections: The Ellipse 597

with coordinates 1x, y2 lies on the ellipse if the sum of the two distances 0 F1 P 0 and
0 F2 P 0 is some constant k. To make things easier, we write k  2a. That is,
0 F1 P 0  0 F2 P 0  2a,
or, equivalently,
21x  c2 2  1 y  0 2 2  21x  c2 2  1 y  02 2  2a
or
21x  c2 2  y 2  21x  c2 2  y 2  2a.
When we simplify this equation by eliminating both square roots (we leave the ac-
tual simplification for you to do as a problem at the end of this section), we even-
tually obtain
x2 y2
 1
a2 b2
as the equations of the ellipse, where b  a2  c 2 is a new constant. The three
2

constants a, b, and c are related by the equation

a2  b2  c 2, where a  b.
We investigate the meaning of the constants a and b below.
To determine where the ellipse intersects the x-axis, we set y  0. The equa-
tion of the ellipse then reduces to
x2
 1 so that x 2  a2
a2
from which we find that either
x  a or x  a.
This result indicates that a represents the distance from the center of the ellipse to
the two points where the ellipse crosses the x-axis, as illustrated in Figure 9.17.
Similarly, if x  0, the equation of the ellipse yields y 2  b2, from which either

b
Minor
axis
F2(–c, 0) F1(c, 0)
x
–a Center a
Major axis

–b
FIGURE 9.17

y  b or y  b.
Thus b represents the distance from the center of the ellipse to the two points
where the ellipse crosses the y-axis. The four points, 1a, 0 2, 1a, 0 2, 10, b2 , and
10, b 2, are the vertices of the ellipse; any one of them is a vertex. The lines con-
necting opposite vertices are called the axes of the ellipse. The longer axis, whether
horizontal or vertical, is called the major axis and always contains the two foci; the
shorter axis is called the minor axis.
gord.3896.09.pgs 4/24/03 10:15 AM Page 598

598 CHAPTER 9 Geometric Models

In summary, a represents the distance from the center to either of the two
more distant vertices along the major axis of the ellipse; b represents the distance
from the center to either of the two closer vertices along the minor axis; and c rep-
resents the distance from the center to either focus of the ellipse. Because a is half
the length of the major axis, we sometimes call a the length of a semi-major axis.
Similarly, b is sometimes called the length of a semi-minor axis.

E XAMPLE 1
Describe and sketch the graph of the ellipse
x2 y2
  1.
16 9
Solution This ellipse is centered at the origin. Its vertices occur when y  0, so x  4,
or when x  0, so y  3. Therefore a  4 and b  3, so that the major axis extends
horizontally from x  4 to x  4, and the minor axis extends vertically from y  3
to y  3. See Figure 9.18. Because a, b, and c are related by a2  b2  c 2 we have
c 2  a2  b2  16  9  7,
so that c   17 . Therefore the foci are located at 1 17 , 0 2 and 1 17 , 0 2, giving the
graph shown in Figure 9.18.

y
a=4
b=3
(0, 3) c = √7

(−4, 0) F2 (−√7, 0) a (4, 0)

x
c
F1 (√7, 0)

(0, –3)
FIGURE 9.18
◆
So far, we have considered an ellipse centered at the origin with foci along the
x-axis. If we consider the analogous situation where the foci lie along the y-axis, the
resulting equation for such an ellipse is

x2 y2
  1, a  b.
b2 a2

Note that the constant a is still measured along the major axis and b is measured along
the minor axis of the ellipse, so a  b. From the equation of an ellipse, we can identi-
fy its major axis immediately by observing which of the two denominators is larger.
We have only considered ellipses that are centered at the origin. In fact, an el-
lipse can be centered at any point 1x0 , y0 2. In such a case, we get the following
standard forms for the equation of an ellipse.
gord.3896.09.pgs 11/21/03 11:01 AM Page 599

9.3 Conic Sections: The Ellipse 599

The equation of an ellipse centered at 1x0 , y0 2 with its major axis parallel to
the x-axis is
1x  x0 2 2 1 y  y0 2 2
  1.
a2 b2
The equation of an ellipse centered at 1x0 , y0 2 with its major axis parallel to
the y-axis is
1x  x0 2 2 1y  y0 2 2
  1.
b2 a2
In each case,
a2  b2  c2,
where c is the distance from the center to either focus.

E XAMPLE 2
Describe and sketch the ellipse whose equation is
1x  2 2 2 1y  7 2 2
  1.
4 25
Solution The center of this ellipse is at the point 12, 72 . Because 25  4, the major axis is
parallel to the y-axis. In particular, the major axis is on the vertical line x  2, and the foci
also lie on this line. The minor axis is on the horizontal line y  7. Also, because a2  25
and b2  4, a  5 and b  2. Thus the length of the major axis is 2a  10, and the length
of the minor axis is 2b  4. Consequently, the maximum horizontal distance from the
center is 2 on either side of x  2, and the maximum vertical distance from the center is 5
above and below y  7. The ellipse therefore extends horizontally from x  0 to x  4
and extends vertically from y  2 to y  12. Figure 9.19 shows the graph of the ellipse.
To locate the foci, we use
c 2  a2  b2  25  4  21,
which gives c  121 . Because the foci are on the major axis of the ellipse, they are on
the vertical line x  2. Thus the foci are at the points 12, 7  121 2 and 12, 7  1212 .

(x − 2) 2 ( y − 7) 2
12 + =1
4 25

(2, 7) y=7
2 2

2
x=2
x
FIGURE 9.19 0 4

◆
gord.3896.09.pgs 4/24/03 10:15 AM Page 600

600 CHAPTER 9 Geometric Models

Suppose that we again consider the ellipse in Example 2,

1x  2 2 2 1 y  722
  1.
4 25
We now multiply both sides of the equation by 100 to eliminate the fractions, so that
251x  22 2  41 y  7 2 2  100.
Expanding the terms, we get
251x 2  4x  4 2  41 y 2  14y  49 2  100,
25x 2  100x  100  4y 2  56y  196  100,
25x 2  4y 2  100x  56y  196  0.
This last equation is an equivalent equation for the same ellipse. Often we start
with such an equation and have to rewrite it algebraically to uncover the key infor-
mation about the ellipse. We illustrate how to do so in Example 3.

E XAMPLE 3
Verify that the equation
25x 2  9y 2  50x  36y  164  0
represents an ellipse, and find its center, vertices, and foci. Use this information to sketch
the ellipse.
Solution We first collect the terms in x and y separately and then factor out the coeffi-
cients of x 2 and y 2:
25x 2  9y 2  50x  36y  164  25x 2  50x  9y 2  36y  164
 3 251x 2  2x 2 4  391 y 2  4y2 4  164.
Finally, we complete the squares on both x and y to obtain
253 1x 2  2x  1 2  1 4  9 3 1 y 2  4y  4 2  4 4  164
 25 3 1x  1 2 2  1 4  9 3 1 y  2 2 2  44  164
 251x  1 2 2  25  91 y  2 2 2  36  164
 251x  1 2 2  91 y  2 2 2  225.
Therefore the original equation is equivalent to
251x  12 2  91 y  2 2 2  225  0 or 251x  1 2 2  91 y  2 2 2  225.
Dividing both sides by 225 yields
1x  1 2 2 1 y  222
  1,
9 25
which is the standard form for the equation of an ellipse. The center is at 11, 2 2 . The
major axis is vertical because 25  9. Moreover, because a  5, the major axis extends
from y  2  5  3 to y  2  5  7. The minor axis is horizontal with b  3, so it
extends from x  1  3  2 to x  1  3  4. To find the foci, we solve
c 2  a2  b2  25  9  16,
gord.3896.09.pgs 4/24/03 10:15 AM Page 601

9.3 Conic Sections: The Ellipse 601

which gives c  4. Therefore the foci are 4 units above and below the center 11, 2 2 , so
they are at 11, 2 2 and 11, 6 2 . The graph of this ellipse is shown in Figure 9.20.

(x − 1) 2 (y − 2) 2
(1, 7) + =1
9 25

F(1, 6)

(–2, 2) (1, 2) (4, 2)

F(1, –2)

(1, –3)
FIGURE 9.20 x=1

◆
At the beginning of the section, we pointed out that the orbits of the planets
about the sun are ellipses. More specifically, these elliptical orbits all have the sun as
one of their two foci. A natural question to ask is: What is the equation of the el-
lipse for the orbit of the Earth? To answer it, we need two pieces of data used by as-
tronomers to describe the orbits of the planets. The perihelion is the smallest
distance from a planet to the sun, and the aphelion is the greatest distance, as de-
picted in Figure 9.21. For the Earth, the perihelion is approximately 147.1 million
kilometers, or 91.38 million miles, and the aphelion is approximately 152.1 million
kilometers, or 94.54 million miles. These two distances help identify the location of
the sun on the major axis of Earth’s elliptical orbit.

Aphelion Perihelion

FIGURE 9.21

E XAMPLE 4
Find an equation of the Earth’s orbit about the sun.
Solution We first set up a coordinate system with the sun, the other (phantom) focus,
and the major axis on the x-axis, as shown in Figure 9.22. Because the perihelion and
aphelion distances are almost the same, the two foci are quite close together and the
orbit of the Earth is nearly circular. From Figure 9.22, we see that the distance from one
vertex to the other, or 2a, is 91.38  94.54  185.92 million miles, so
a  185.92>2  92.96 million miles.
gord.3896.09.pgs 4/24/03 10:15 AM Page 602

602 CHAPTER 9 Geometric Models

1.58

94.54 91.38

FIGURE 9.22

To find b, we first have to determine c, using the following reasoning. At perihelion,

the Earth is 91.38 million miles from the sun, so the distance from the center of the el-
lipse to the sun (a focus) must be
c  92.96  91.38  1.58 million miles.
From a  b  c , we have
2 2 2

b2  a2  c 2
 192.962 2  11.58 2 2  8639.07,

so that
b  28639.07  92.95 million miles.
Consequently, the equation of the Earth’s orbit about the sun is
x2 y2
  1.
192.962 2 192.952 2
As we observed previously, the Earth’s orbit is very nearly circular.
◆

The table of planetary data on the following page lists the perihelion and aphe-
lion distances, in millions of miles, for the planets in the solar system. You can use
it to compare the Earth’s orbit to that of the other planets. You will use some of
these entries for the Problems at the end of the section.

Reflection Property of the Ellipse

One of the most fascinating properties of an ellipse is known as the reflection
property. Consider any line segment emanating from one of the two foci—say,
F1—as shown in Figure 9.23. It eventually intersects the ellipse and then reflects.
gord.3896.09.pgs 4/24/03 10:15 AM Page 603

9.3 Conic Sections: The Ellipse 603

y
Planet Perihelion Aphelion
Mercury 28.56 43.38

Venus 66.74 67.68

Earth 91.38 94.54 x

F2 F1
Mars 128.49 154.83

Jupiter 460.43 506.87

Saturn 837.05 936.37

FIGURE 9.23
Uranus 1700.07 1867.76

Neptune 2771.72 2816.42

Pluto 2749.57 4582.61

(According to physical principles, the angle of incidence with a tangent line

equals the angle of reflection.) Any such reflected line segment will pass through
the second focus F2 .
This property is significant because many physical phenomena, such as light
and sound, travel in straight lines and reflect off solid surfaces. Thus, if a light-
bulb is placed at one focus of a three-dimensional shell whose cross sections
containing the major axis are all ellipses, all its light rays will bounce off the in-
side surface of the shell and reflect back through the other focus. The effect is
similar with sound waves. Probably the best known example of this is the whis-
pering gallery effect in the U.S. Capitol in Washington, D.C. The dome of the
Capitol has the approximate shape of a three-dimensional ellipse, and there are
two foci near floor level. If you stand at one of the foci and whisper, your voice
is carried to the second focus across the hall and can be heard clearly by anyone
standing there.

E XAMPLE 5
The distance between the foci in the “whispering gallery” of the Capitol is 38.5 feet, and
the maximum height of the ceiling above ear level is 37 feet. Find the equation of an el-
liptical cross section of the gallery under the Capitol dome.
Solution We set up axes as shown in Figure 9.24. Because the distance between foci is
38.5 feet, c  12 138.5 2  19.25 feet. Also, from the maximum height of the dome,
b  37 feet. For an ellipse, we know that
a2  b2  c 2  137 2 2  119.252 2  1739.5625,

37
F2 F1
x
FIGURE 9.24 38.5
gord.3896.09.pgs 4/24/03 10:15 AM Page 604

604 CHAPTER 9 Geometric Models

so a  41.7 feet. Therefore the equation of an elliptical cross section of the Capitol whis-
pering gallery is
x2 y2
  1.
141.72 2 137 2 2
◆
The Average Distance from the Sun
We have stated that the orbit of each planet is an ellipse with one focus at the sun.
A natural question to ask is: What is the average distance of a planet from the sun
during a full orbit? The answer is particularly simple, yet surprising, as we demon-
strate in Example 6.

E XAMPLE 6
Show that the average distance from all points on any ellipse
x2 y2
 1
a2 b2
is precisely equal to a, the length of the semi-major axis.
Solution We begin with the ellipse shown in Figure 9.25, with foci at F1 and F2 . Let P1
be any point on the right half of the ellipse. From the geometric definition of the ellipse,
we know that the sum of the two distances 0 F1 P1 0 and 0 F2 P1 0 must equal the constant 2a,
or
0 F1 P1 0  0 F2 P1 0  2a.

P2 P1

d d

x
F2 F1

FIGURE 9.25

Using the symmetry of the ellipse, there is a comparable point P2 on the left half of
the ellipse so that
0 F2 P2 0  0 F1 P1 0 and 0 F2 P1 0  0 F1 P2 0 .
Therefore
0 F1 P1 0  0 F1 P2 0  0 F1 P1 0  0 F2 P1 0  2a.
gord.3896.09.pgs 4/24/03 10:15 AM Page 605

9.3 Conic Sections: The Ellipse 605

Hence the average of these two distances from F1 to P1 and from F1 to P2 is simply a. This
argument can be applied to every possible pair of matching points on the ellipse, so the
average distance from F1 to all points on the ellipse must be a.
◆
Now let’s apply this result to the orbits of the planets. For instance, the aphe-
lion distance for the Earth is 94.54 million miles and the perihelion distance is
91.38 million miles. These distances can be expressed as
Perihelion  a  c and Aphelion  a  c.
Their arithmetic average is

1perihelion  aphelion2  1a  c  a  c2  a;
1 1
2 2
that is, the average distance of the Earth (or any other planet) from the sun is just
the average of its perihelion and aphelion distances.

Problems
1. For the satellite whose elliptic orbit about the Earth 3. Suppose that the satellite in Problem 1 is in a rela-
is shown in the accompanying diagram, indicate tively low orbit about the Earth so that it encoun-
the location of the following points and give rea- ters the upper fringe of the Earth’s atmosphere.
sons for your answers. What will be the atmosphere’s effect on the satel-
lite’s path? Sketch the graph of the resulting trajec-
tory. What will happen eventually?
4. Suppose that the satellite in Problem 1 fires its booster
rocket to speed up at the point in its orbit where it is
closest to the Earth. Compare the graph of the new
orbit to the graph of the old one in a sketch. What
happens to the orbit if the booster rockets are ex-
tremely strong or continue firing for a long time?
5. Which of the nine planets in the solar system has the
most circular orbit? the least circular orbit? Explain.
a. The point at which the gravitational force F ex- 6. During the next few years, Pluto’s orbit takes it in-
erted by the Earth on the satellite is greatest; side the orbit of Neptune. Use the values in the
where it is least. (Recall Newton’s law of univer- table of planetary data in the text to explain why
sal gravitation: F  f 1r2  GmM>r 2, where r is this situation can occur.
the distance between the two objects.) 7. Use the fact that the perihelion and aphelion dis-
b. The point at which the speed of the satellite is tances for Mercury are 46.0 and 69.8 million kilo-
greatest. (Hint: Think of the satellite as always meters respectively to find the equation of the orbit
“falling” toward the Earth.) of Mercury about the sun.
c. The point at which the speed of the satellite is least. 8. A salami is 4 inches in diameter. When the deli clerk
2. Suppose that the satellite in Problem 1 fires its retro- slices it, however, the slices are at an angle of 65° to
rockets to slow down somewhat at the point in its the main axis of the salami. Consequently, each slice
orbit where it is closest to the Earth. Compare the will be in the shape of an ellipse with a minor axis
graph of the new orbit to the graph of the old one in of length 4 inches, as shown on the following page.
a sketch. Find the length of the major axis of each slice.
gord.3896.09.pgs 4/24/03 10:15 AM Page 606

606 CHAPTER 9 Geometric Models

In Problems 11 and 12, each equation represents an el-

4 in. lipse. In each case, identify the center, the vertices, and
4 in.
65° the foci and use the pertinent information to draw its
graph.
9. The Roman Coliseum is in the shape of an ellipse 11. 4x 2  9y 2  1
whose major axis measures 620 feet and whose
minor axis measures 513 feet. 12. 25x 2  4y 2  100
In Problems 13–18, each equation represents an ellipse.
V2 Complete the square for x and y in each case to obtain
the standard form for an ellipse. Then identify the cen-
ter, the vertices, and the foci of the ellipse and use the
513 ft pertinent information to draw its graph.
620 ft
13. x 2  4y 2  2x  8y  1
V1
F2 F1 14. x 2  4y 2  2x  8y  11
15. x 2  4y 2  20x  40y  100  0
16. 4x 2  y 2  24x  2y  4  0
17. 9x 2  y 2  54x  4y  76
a. What is the equation of this ellipse? 18. x 2  9y 2  6x  36y  36
b. How far apart are the foci? 19. Complete the derivation of the equation of the el-
c. How far apart are two adjacent vertices? lipse by simplifying the equation in the text by
10. Write formulas expressing the perihelion and eliminating the two square roots. (Hint: First isolate
aphelion of an elliptic orbit in terms of the one of the radicals, then square both sides, and fi-
semimajor axis length a and the focal distance c in nally eliminate the remaining radical.)
an ellipse.

9.4 Conic Sections: The Hyperbola and the Parabola

We now turn our attention to the two remaining conic sections, the hyperbola and
the parabola. We begin this section by investigating the properties and some appli-
cations of the hyperbola.

The Hyperbola
We defined an ellipse geometrically as the set of all points for which the sum of the
distances to two fixed foci is constant. In an analogous way, we define a hyperbola
in terms of the difference of the distances to two fixed points being constant. A hyp-
erbola is the set of all points for which the difference between the distances to two
fixed points is a constant. The two points are the foci of the hyperbola. The mid-
point of the line segment joining the foci is the center.

E XAMPLE 1
During a severe thunderstorm, two lightning bolts appear to strike simultaneously.
You hear the thunderclap from one lightning bolt exactly 1 second after the light-
ning strikes at point P, and you hear the thunderclap from the second lightning bolt
gord.3896.09.pgs 4/24/03 10:15 AM Page 607

9.4 Conic Sections: The Hyperbola and The Parabola 607

2 seconds after it hits at point Q, as depicted in Figure 9.26. Sound travels at a speed
of about 1100 feet per second.

1 second or 2 seconds or
1100 feet 2200 feet

FIGURE 9.26 P Q

a. Based on this information what can you conclude about the point A, where you are?
b. Two friends of yours also see the same two lightning strikes. From Becky’s location at
point B, the thunder from the lightning bolt at point P takes 2 seconds to reach her
and the thunder from the lightning bolt at point Q takes 3 seconds. From Carl’s loca-
tion at point C, the times are 4 and 5 seconds, respectively. What can you conclude
about the three points A, B, and C?
Solution

a. The lightning bolts hit at points P and Q, and you’re located at point A. Because it
takes 1 second for the sound of the lightning bolt at P to reach A and sound travels
at 1100 feet per second, the distance from P to A must be 1100 feet. Similarly, it
takes 2 seconds for the sound of the strike at Q to reach A so that distance must be
2 seconds 1100 feet>second  2200 feet.
b. Figure 9.27 shows the points A, B, and C. Reasoning as in part (a), you can conclude
that Becky is 2 seconds 1100 feet>second  2200 feet from P and 3 seconds
1100 feet>second  3300 feet from Q. Similarly, Carl is 4400 feet from P and
5500 feet from Q.

5500
4400
2200
3300

A
1100 2200
FIGURE 9.27 P Q

However, you can deduce one more piece of information: For all three, the difference
in time between the two thunderclaps is 1 second. That is, the difference in distance from
each of the three points A, B, and C to the points P and Q is a constant equal to 1100 feet.
But if the differences in the distances from these three points to the fixed points where
the lightning bolts hit are all equal, the three points A, B, and C must lie on a hyperbola
whose foci are at P and Q.
◆
gord.3896.09.pgs 4/24/03 10:15 AM Page 608

608 CHAPTER 9 Geometric Models

Think About This Explain why you cannot determine the distance from P to Q in Example 1 based on
the triangle APQ by using trigonometry. What additional information would you
need to be able to find that distance? ❐

The Equation of a Hyperbola We now determine the equation of a hyperbola

from the geometric definition. For convenience, we place the center of a hyperbola
at the origin and the foci on the x-axis at the point F1 with coordinates 1c, 0 2 and F2
with coordinates 1c, 0 2, as shown in Figure 9.28. A point P with coordinates
1x, y2 lies on the hyperbola if, for some constant k,
0 F2 P 0  0 F1 P 0  k.

P(x, y)

x
F2(–c, 0) F1(c, 0)

FIGURE 9.28

As with the equation of an ellipse, we let the constant k  2a for convenience.

Thus
0 F2 P 0  0 F1 P 0  2a,
so that

21x  c2 2  y 2  21x  c2 2  y 2  2a.

We simplify this equation by eliminating both square roots, as was done for the
equation of an ellipse (see Problem 19, Section 9.3) and eventually obtain
x2 y2
  1,
a2 b2
where
c 2  a2  b2.
The graph of this hyperbola is shown in Figure 9.29. Note that the hyperbola has
two distinct branches, which is what we should expect from the discussion in Sec-
tion 9.3 of slicing through a double right circular cone. The two points where
this hyperbola crosses the x-axis are called its vertices; they correspond to the
points where y  0 and thus represent the points where the two branches are
closest.
gord.3896.09.pgs 4/24/03 10:15 AM Page 609

9.4 Conic Sections: The Hyperbola and The Parabola 609

y
x2 y2
− =1
a2 b2

a a
x
F2(–c, 0) F1(c, 0)

FIGURE 9.29

At the vertices, we have y  0, so that

x2
 1,
a2
and therefore x  a. Thus a represents the distance from the center to a vertex,
whereas c represents the distance from the center to a focus. The line containing
the foci is the axis of the hyperbola.
Alternatively, we can place the foci for a hyperbola on the vertical axis, as
shown in Figure 9.30. The equation of such a hyperbola is
y2 x2
  1.
a2 b2

F1(0, c)

y2 x2
− =1
a a2 b2
x
a

F2(0, –c)
FIGURE 9.30

E XAMPLE 3
Describe the hyperbola whose equation is
x2 y2
  1.
16 9
Solution The form of the equation indicates that the hyperbola is centered at the ori-
gin and that its axis is horizontal. Because a2  16 and b2  9, we have a  4 and
b  3, so that
c 2  a2  b2  16  9  25,
gord.3896.09.pgs 4/24/03 10:15 AM Page 610

610 CHAPTER 9 Geometric Models

and so c  5. Thus the vertices are at x  4 and x  4, or the points 14, 0 2 and
14, 0 2. The foci are the points 15, 0 2 and 15, 0 2.
◆
Think About This Use your function grapher to see what the graph of the hyperbola in Example 3
looks like. To do so, you have to rewrite the equation by solving for y as a function
of x. In particular,
y2 x2 x2
  1 so that y 2  9 a  1b .
9 16 16
The upper and lower halves of the hyperbola are therefore given separately by the
two functions
x2 x2
y3  1 and y  3  1.
B 16 B 16

What are the domains of the two functions? ❐

More generally, we can consider a hyperbola as being shifted horizontally and/or

vertically so that its center is at the point P with coordinates 1x0 , y0 2 rather than at the
origin. We then have the following standard forms for the equation of a hyperbola.

The equation of a hyperbola centered at 1x0 , y0 2 with its axis parallel to the
x-axis is
1x  x0 2 2 1 y  y0 2 2
  1.
a2 b2
The equation of a hyperbola centered at 1x0 , y0 2 with its axis parallel to the
y-axis is
1 y  y0 2 2 1x  x0 2 2
  1.
a2 b2
In each case,
c 2  a2  b2.

Note that, in the equation of a hyperbola in standard form, the term with the
positive coefficient determines the orientation. If the x2-term is positive, the two
branches open about the x-axis; if the y 2-term is positive, the two branches open
about the y-axis. Also, be sure to distinguish between the equation c 2  a2  b2 re-
lating the constants for a hyperbola and the equation a2  b2  c 2 relating the
constants for an ellipse.

E XAMPLE 4
Verify that
x 2  y 2  8x  6y  2
is an equation of a hyperbola. Find the center, vertices, and foci of the hyperbola and
sketch its graph.
gord.3896.09.pgs 4/24/03 10:15 AM Page 611

9.4 Conic Sections: The Hyperbola and The Parabola 611

Solution We complete the square on both x and y so that the left-hand side becomes
x 2  y 2  8x  6y  3 1x 2  8x  16 2  16 4  3 1 y 2  6y  92  9 4
 1x  4 2 2  16  1 y  3 2 2  9
 1x  4 2 2  1 y  32 2  7.
The original equation therefore becomes
1x  4 2 2  1y  3 2 2  7  2 or 1x  42 2  1 y  3 2 2  9.
Dividing by 9, we obtain
1x  4 2 2 1y  3 2 2
  1.
9 9
Consequently, the center of the hyperbola is 14, 3 2 and a  b  3. Furthermore,
since the x2-term is the positive one, the axis of the hyperbola is parallel to the x-axis.
That is, the vertices and the foci lie on the horizontal line y  3 through the center,
and the hyperbola opens to the left and the right. Because a  3, the vertices are 3 units
left and right of the center, or at 17, 3 2 and at 11, 3 2. Also,
c 2  a2  b2  9  9  18,
so c  118  3 12 . Thus the foci are located at F1 at 14  3 12 , 3 2 and F2 at
14  3 12 , 3 2 , as shown in Figure 9.31.

F2 (–4, –3) F1 y = −3

FIGURE 9.31
◆
Based on their graphs and the vertical line test, hyperbolas (like ellipses) can-
not be functions because any x-value can lead to two distinct y-values. Of course, if
an x-value is beyond the limits of the ellipse, there is no corresponding y-value.
Similarly, if an x-value is between the two branches of a hyperbola whose axis is
horizontal, there is no corresponding y-value.

Applications of the Hyperbola Probably the most significant application of

the hyperbola has been the long range navigation (LORAN) system used around
the world by sailors to locate their positions before the advent of the global po-
sitioning system (GPS) that makes use of orbiting satellites. With LORAN, radio
transmitters along coastal waters emit simultaneous radio signals that are
picked up by electronic equipment on ships. As we demonstrated in Example 1
FIGURE 9.32 with sound waves, there will usually be a difference in the times at which a ship
gord.3896.09.pgs 4/24/03 10:15 AM Page 612

612 CHAPTER 9 Geometric Models

receives radio signals from different stations, and this difference is used to
“place” the vessel on a specific hyperbola. When the same procedure is used with
other radio transmitters in the LORAN network, the ship is simultaneously
“placed” on a second hyperbola. Finding a point of intersection of the two hy-
perbolas and locating the position of the vessel is then a relatively simple matter,
as illustrated in Figure 9.32.
In Example 5 we demonstrate the actual use of these ideas. To do so, we use the
fact that any radio wave travels at the speed of light, or about 186,300 miles per sec-
ond, or 300,000 kilometers per second.

E XAMPLE 5
A sailboat is out on Long Island Sound when a heavy fog moves in. To the south of the
sailboat, on Long Island’s shore, are two LORAN radio transmitters 60 km apart at
points P and Q. A third transmitter is to the north on Connecticut’s shore at point R,
which is 40 km directly north of P. Figure 9.33 depicts this situation.
a. The receiver on the boat receives signals from P and Q that arrive 0.00016 second apart.
Find an equation of the hyperbola having foci at P and Q on which the boat is located.
b. The receiver on the boat receives signals from P and R that arrive 0.0001067 second
apart. Find an equation of the hyperbola having foci at P and R on which the boat is
located, based on the same coordinate system used in part (a).

Connecticut
R

40 km

P Q
60 km
Long Island
FIGURE 9.33

c. Estimate the location of the boat based on the results of parts (a) and (b).
Solution

a. Suppose that we set up a coordinate system with a horizontal axis through P and Q and
the origin midway between them, as shown in Figure 9.34. In this system, the coordi-
nates of P are 130, 02 and the coordinates of Q are 130, 02. One focus of the hyperbo-
la is at Q, so we have c  30. Suppose that the sailboat is at S with coordinates 1x, y2 on
this hyperbola. The difference in times between receipt of the two signals is 0.00016 sec-
ond; using the speed of light as 300,000 km>second, this difference in times is equiva-
lent to a difference in distance of 300,000 km>second 0.00016 seconds  48 km.
That is, 2a  48, so a  24.
For a hyperbola c 2  a2  b2, so that
b2  c2  a2  130 2 2  124 2 2  324 or b  18.
gord.3896.09.pgs 4/24/03 10:15 AM Page 613

9.4 Conic Sections: The Hyperbola and The Parabola 613

Consequently, the equation of this hyperbola, which opens to the left and the right, is
x2 y2
  1.
124 2 2 118 2 2

b. We now consider the hyperbola with foci at P and R and center at the point
130, 202, as shown in Figure 9.35. We use a1 , b1 , and c1 to represent the parameters
for this hyperbola. Because the hyperbola opens upward and downward, its equation
has the form
1y  202 2 1x  30 2 2
  1.
a 12 b12

y y

R (−30, 40)

S (−30, 36)
(−30, 0) (30, 0)
x (−30, 20)
P (−24, 0) (24, 0) Q (−30, 4) S
x
P (−30, 0)

FIGURE 9.34 FIGURE 9.35

The distance between the foci is 40 km, so we have c1  20.

The difference in time for receipt of the two signals is 0.0001067 second, which is
equivalent to a difference in distance of 300,000 0.0001067  32 km, so 2a1  32
and a1  16. Therefore
b12  c12  a12  120 2 2  116 2 2  144 or b1  12,
ˇ ˇ ˇ

and the equation of this hyperbola is

1y  202 2 1x  30 2 2
  1.
116 2 2 112 2 2

c. To locate the position of the sailboat, we have to find the point of intersection of the
two hyperbolas. (Note that there can be as many as four points of intersection, but in
practice we would know which branch of each hyperbola the boat is on, based on the
strength of the signal, so the problem reduces to finding a single point of intersec-
tion.) Although we can find the point of intersection algebraically, to do so is rather
complicated, so instead we estimate the point graphically. Using the equation of the
first hyperbola from part (a), we have
y2 x2 y2 x2
  1 or  1
118 2 2 124 2 2 324 576
so that
x2
y 2  324 a  1b .
576
gord.3896.09.pgs 4/24/03 10:15 AM Page 614

614 CHAPTER 9 Geometric Models

Figure 9.36 shows that the sailboat is above the line on which the center and the foci
lie, so y must be positive (otherwise the boat would be on land). We therefore take the
positive square root and get

x2
y  18  1.
B 576

R (−30, 40)

S Q (30, 0)
x
P (−30, 0)

FIGURE 9.36

Similarly, starting with the equation of the second hyperbola, we have

1y  202 2 1x  30 2 2 1y  20 2 2 1x  302 2
  1 or  1
116 2 2 112 2 2 256 144

so that

1x  30 2 2
1y  20 2  256 c
2
 1d.
144

Because the sailboat, as shown in Figure 9.35, is closer to the transmitter at P than the one
at R, we need the lower branch of the hyperbola. Taking the negative square root gives

1x  30 2 2 1x  30 2 2
y  20  16  1 or y  20  16  1.
B 144 B 144

To locate the sailboat, we need to find the point where the curves corresponding
to these two equations intersect. From Figure 9.36 we estimate graphically that the
point of intersection occurs at about x  24.18 and y  2.21. That is, the sailboat
is located about 2.21 km off the north coast of Long Island at a position about
30  24.18  5.82 km east of the transmitter at point P.
◆

The Parabola
We have shown that the graph of any quadratic function y  ax2  bx  c is a
parabola opening upward or downward. However, the same parabolic shape can
open to the left or the right, as shown in Figure 9.37, although neither of these
graphs represents a function. To unify these ideas about parabolas, we consider the
gord.3896.09.pgs 4/24/03 10:15 AM Page 615

9.4 Conic Sections: The Hyperbola and The Parabola 615

parabola from a somewhat different perspective in terms of its geometric defini-

tion as a conic section. We define a parabola as the set of all points in the plane for
which the distance to a single fixed point is equal to the distance to a fixed line, as
depicted in Figure 9.38. The fixed point is called the focus of the parabola. The
fixed line is the directrix of the parabola.

y
y y

F(0, c)

c
x x
P
x

c
FIGURE 9.37
L y = −c

FIGURE 9.38

The Equation of a Parabola For convenience, we place the focus of the

parabola at the point F on the y-axis with coordinates 10, c2 and let the directrix
be the horizontal line y  c, as shown in Figure 9.38. The graph shown corre-
sponds to the case where c  0. The parabola consists of all points P having the
property that the distance from P to the focus F is equal to the vertical distance
from P to the directrix line L. Thus a point P with coordinates 1x, y2 lies on the
parabola if the distance from P to F, 2x 2  1 y  c2 2 , equals the distance from
P to the line L, which is y  c; that is,

2x 2  1y  c2 2  y  c.
We square both sides of this equation and get
x 2  1 y  c2 2  1 y  c2 2,
or, equivalently, when we expand the equation, we have
x 2  y 2  2cy  c 2  y 2  2cy  c 2.
We subtract y 2 and c 2 from both sides of this equation and obtain
x2  2cy  2cy.
Finally, we add 2cy to both sides and solve for y;
x2
y .
4c
This is the equation of a parabola with vertex (or turning point) at the origin. If
c  0, the parabola opens upward. If c  0, the parabola opens downward. The
vertical line through the vertex is called the axis of symmetry of the parabola.
gord.3896.09.pgs 4/24/03 10:15 AM Page 616

616 CHAPTER 9 Geometric Models

Alternatively, had we positioned the focus on the x-axis at F with coordinates

1c, 0 2 with a vertical directrix at x  c, then we would have obtained
y2
x
4c
as the equation for the parabola. This parabola likewise has its vertex at the origin;
it opens to the right if c  0 and opens to the left if c  0. Finally, its axis of sym-
metry is now the horizontal line through the vertex, as depicted in Figure 9.39.

y y2 y
x=
4c
c>0 c<0

x x

FIGURE 9.39

More generally, we can describe a parabola whose vertex is at 1x0 , y0 2 with the
following standard forms of the equation of a parabola.

The equation of a parabola with its vertex at 1x0 , y0 2 and opening vertically is
1x  x0 2 2
y  y0  .
4c
The equation of a parabola with its vertex at 1x0 , y0 2 and opening horizontal-
ly is
1y  y0 2 2
x  x0  .
4c

Reflection Property of the Parabola Just as the ellipse has a remarkable—and

useful—reflection property, the parabola has one that is even more commonly en-
countered. It can be shown that any ray coming into a parabola along a line paral-
lel to the axis of symmetry of the parabola will “reflect” off the curve and pass
through the focus, as shown in Figure 9.40. Alternatively, any ray emanating from
the focus will reflect off the parabola and continue on a path parallel to the axis of
symmetry. This reflection property is used, for example, in flashlights and in the
headlights of an automobile, where the light source is located at the focus and the
beams of light bounce off a parabolic reflector to concentrate more light in a par-
ticular direction. The reflection property is also used by satellite TV dishes, which
are constructed in such a way that every cross section containing the axis of sym-
metry of the dish is a parabola, as illustrated in Figure 9.41. The TV signals com-
ing from a satellite relay in orbit arrive at the dish along rays parallel to the axis of
the dish and its parabolic cross sections. They reflect off the dish and pass through
a receptor unit positioned at the focus. There the signal is collected and then
transmitted to the television set.
gord.3896.09.pgs 4/24/03 10:16 AM Page 617

9.4 Conic Sections: The Hyperbola and The Parabola 617

F(0, c)

c
x
FIGURE 9.40 FIGURE 9.41

Conic Sections in General

The general equation of a conic section has the form
Ax2  By2  Cx  Dy  E  0,
where A, B, C, D, and E are constants and at least one of A and B is nonzero. When
B  A  0, we divide both sides by A to get
C D E
x2  y2  x  y   0,
A A A
which is the equation of a circle, provided that certain conditions are satisfied that
lead to a positive radius. When B  0 and A  0, the resulting equation is quad-
ratic in x but only linear in y and so gives the equation of a parabola opening either
upward or downward. Similarly, when A  0 and B  0, we get a parabola open-
ing either left or right. If A and B have the same sign—say, both are positive—the
resulting curve is an ellipse, provided that certain conditions are satisfied. If A and
B have opposite signs, the curve is a hyperbola. Thus, for instance,
4x2  9y2  8x  36y  5  0
represents an ellipse, whereas
4x2  9y2  8x  36y  5  0 and 25y2  16x2  10y  8x  3  0
both represent hyperbolas (one of which opens left and right and the other opens up
and down). You have to be able to identify the type of curve from the given equation.
Finally, our discussions of conic sections have been restricted to their being in
standard position—that is, their axes are parallel to the x- and y-axes. However, the
same shapes can be rotated through some angle u about the x-axis. When that
occurs, the equation for the conic section—whether an ellipse, hyperbola, or
parabola—includes a term of the form xy. For the most part, we aren’t concerned
with such situations here except for the case
xy  k,
where k is any constant. If we solve for y, this equation is equivalent to the power
function
k
y  kx1.
x
gord.3896.09.pgs 4/24/03 10:16 AM Page 618

618 CHAPTER 9 Geometric Models

This function is not defined at x  0 and has two branches, the more familiar
one in the first quadrant when x  0 and a symmetric one in the third quadrant
when x  0. Together, they form the graph of a hyperbola that has been rotated
from standard position through an angle of 45° or p>4 (when k  0), as shown
in Figure 9.42.

45°
x

FIGURE 9.42

Problems
1. The small satellite TV dishes now on the market for 36 in.
home use have parabolic cross sections containing
the axis of the dish. The focus is located at a point
about 6 inches from the vertex of the parabola. 16 in.

x
18 in.
6 in.
1.5 in.

3. In this problem, we ask you to investigate the signif-

icance of the quantities a and b in the equation of a
a. Find an equation of a parabolic cross section. hyperbola
x2 y2
Assume that the dish is aimed directly upward—   1.
which makes sense only at the equator because a2 b2
the communications satellites are in orbit over Suppose that you zoom out far enough on the
the equator. graph of the hyperbola so that what you see appears
b. The rim of the dish is a circle with a diameter of to be a pair of lines that intersect at the origin.
18 inches at a height of about 1.5 inches above a. Explain why—when x and y both are very large,
the vertex. If the dish were extended, the rim either positive or negative—you can ignore the
would enlarge. Find the diameter of the rim if the number 1 in the equation.
rim reached the height, 6 inches, of the focus. b. Ignore the number 1 and solve for y in terms of x
2. Suppose that a satellite receiver is 36 inches across to find the equations of the two lines described.
and 16 inches deep (vertex to plane of the rim). c. What are the slopes of the two lines that the
How far from the vertex must the receptor unit be branches of the hyperbola approach?
located to ensure that it is at the focus of the para- In Problems 4–13 complete the square for x and y in
bolic cross sections? each equation to obtain the standard form for a conic
gord.3896.09.pgs 4/24/03 10:16 AM Page 619

9.5 Parametric Curves 619

section. In each case, identify the conic section and use and a y-intercept of 1. If you were to erase the
the pertinent information to draw its graph. curve, you would still see its outline from the tan-
4. x2  4y2  2x  8y  1 gent lines. String and wire art designers use this
idea to suggest a variety of curves by using line seg-
5. x2  4y2  2x  8y  11
ments made of the string or wire. The points on the
6. x2  4y2  2x  8y  7 axes are selected so as to follow the outline of a de-
7. x2  4y2  2x  8y  19 sired curve, such as the hyperbola.
8. 4x2  y2  24x  2y  4  0
y
9. 4x2  9y2  16x  18y  31
10. 9x2  16y2  90x  64y  17 5 1
y= x
11. 4x2  4y2  24x  16y  43  0
4
12. 9x2  4y2  18x  16y  8
13. 9x2  4y2  18x  16y  6 3

14. Explain why x2  y2  2x  2y  4 is not the

2
equation of a conic section.
15. In this problem, we ask you to look at the mathe- 1
matics of string and wire art designs. Start with the
hyperbola xy  1 or y  1>x and construct a series 0 1 2 3 4 5
x

of lines tangent to the curve, as shown in the ac-

companying figure. For instance, the line tangent to a. Find the slope m of each of the five tangent lines
the curve when x  1 crosses the x-axis at x  2 shown in the figure.
and crosses the y-axis at y  2. The line tangent to b. Find a formula for the slope m as a function of
the hyperbola when x  2 has an x-intercept of 4 the point of tangency x.

9.5 Parametric Curves

Throughout our discussion of functions, we have almost always considered expres-
sions for which the dependent variable is given in terms of the independent vari-
able. In some cases, however, introducing an additional variable, a parameter, can
provide more insight into what is happening.

Parametric Representations of a Line

There are many ways to write an equation of a line, including the point–slope
form, the slope–intercept form, and the normal form. However, certain questions
about a line can’t be answered with any of these forms. For instance, as we demon-
strated in Section 9.2, if we want to locate the point that is a certain fraction of the
way from the point P at 1x0 , y0 2 to the point Q at 1x1 , y1 2 , it is essential to use the
parametric form

x  x0  1x1  x0 2 t
y  y0  1y1  y0 2 t

for the line, where the parameter t takes on any value. With this form, each possible
value of t gives a corresponding point on the line through P and Q.
gord.3896.09.pgs 4/24/03 10:16 AM Page 620

620 CHAPTER 9 Geometric Models

E XAMPLE 1
Consider the parametric equations x  2  4t and y  5  3t.
a. Construct a table of values for x and y corresponding to t  1, 0, 1, . . . , 4.
b. Use the table to explain why the two equations give a linear function.
c. Plot the points and draw the line that passes through them.
Solution

a. When t  1, the parametric equations yield x  2  411 2  2 and y 

5  3112  8. Similarly, when t  0, the equations give x  2  4102  2 and
y  5  310 2  5. We show the results for the given values of t in the following
table:

t 1 0 1 2 3 4

x 2 2 6 10 14 18

y 8 5 2 1 4 7

b. Note that each value of the parameter t gives rise to a pair of values x and y, which in turn
produces a point 1x, y2. To show that these points lie on a line, we consider just the x and
y values in the table. Each successive x value increases by 4 units and, simultaneously,
each corresponding y value decreases by 3. Because there is a constant change in y when
the x’s are uniformly spaced, we conclude that these points lie on a line and so the para-
metric equations represent a line. In particular, the slope of this line is
y>
x   34 .
c. Figure 9.43 shows the plot of these points and the line through them.

(–2, 8)
t = −1
(2, 5)
t=0
(6, 2)
t=1
x
(10, –1)
–10 –5 t=2
(14, –4)
–5 t=3
(18, –7)
t=4
–10
FIGURE 9.43

◆
The pair of equations
x  2  4t and y  5  3t
giving x and y in terms of the parameter t is called a parametric representation or
the parametric equations of the line. More generally, if we have a curve instead of a
line, the pair of equations for x and y in terms of a parameter t is a parametric rep-
resentation of the curve.
E XAMPLE 2
Eliminate the parameter t from the pair of parametric equations x  2  4t and
y  5  3t and so find the slope–intercept form for the equation of this line.
gord.3896.09.pgs 4/24/03 10:16 AM Page 621

9.5 Parametric Curves 621

Solution We start with the parametric equation x  2  4t and solve for t:

4t  x  2 so that t  1x  2 2.
1
4
When we substitute this expression into the parametric equation for y, we get

y  5  3t  5  3 a b 1x  2 2  5  1x  2 2,
1 3
4 4
or
y  5   1x  2 2,
3
4
which is the point–slope form for the equation of a line with slope  34 that passes
through the point 12, 5 2 . This value for the slope is the same value we found in Exam-
ple 1. Figure 9.43 shows that the line clearly passes through the point 12, 52, which is
also a point in the table we created in Example 1.
◆

The Path of a Projectile

Another case of a parametric representation of a function is the path of a thrown
object, such as a football. The path, or trajectory, is a parabola of the form
y  ax2  bx  c.
However, for most real-world applications, the equation of the parabola by itself
is of little value. Far more important is knowing when the ball, or other object,
will reach a particular point. Therefore introducing time as a variable is neces-
sary, and we do so by writing both x and y, the coordinates of each point along
the parabola, in terms of a parameter t that represents time. In particular, if the
object is released at time t  0 from an initial height y0 with an initial velocity v0
at an initial angle a, as shown in Figure 9.44, then at any time t thereafter, it
turns out that

x  1v0 cos a2t and y   gt 2  1v0sin a 2t  y0 .

1
2
y

v0

α
y0

x
FIGURE 9.44

Each of these expressions can be thought of as a function of the parameter t, so we

rewrite them as
x1t2  1v0 cos a 2t and y1t2   gt 2  1v0 sin a2 t  y0 .
1
2
gord.3896.09.pgs 4/24/03 10:16 AM Page 622

622 CHAPTER 9 Geometric Models

Each value of t determines a corresponding value for x (the horizontal distance)

and a value for y (the vertical height), which produces a point 1x, y2 on the parabo-
la. Again, the pair of equations for x and y as a function of the parameter t is a para-
metric representation of the curve, and the two equations are the parametric
equations of the curve.

Parametric Representation of a Circle

The fact that the parametric representations of a line and of a parabola are so valu-
able suggests that parametric representations of other curves might also be useful.
There are two key steps: (1) to decide on an appropriate parameter t, and (2) to
find a way to express the usual variables x and y in terms of t.
We begin with a circle of radius r centered at the origin:
x 2  y 2  r 2.
Recall that we can express both x and y in terms of an angle u drawn from the cen-
ter of the circle, as shown in Figure 9.45. We write
x  r cos u and y  r sin u.

P(x, y)

r
y = r sin θ
θ
x
x = r cos θ

FIGURE 9.45

In retrospect, this equation is a parametric representation of the circle with the

angle u as the parameter. For each value of u, we can calculate x and y and so get
the point 1x, y2 on the circle. In fact, we can use any other letter, such as t, and
get
x  r cos t and y  r sin t
as a parametric representation of the circle.
If we start with a parametric representation of a curve, we can sometimes
eliminate the parameter to construct a single equation of the curve, as we did for
the line in Example 2. For the circle x  r cos t and y  r sin t, we eliminate the
parameter t as follows:
x 2  y 2  1r cos t2 2  1r sin t2 2  r 2 1cos2t  sin2t2  r 2.
Thus we are left with x 2  y 2  r 2, the usual equation for the circle.
gord.3896.09.pgs 4/24/03 10:16 AM Page 623

9.5 Parametric Curves 623

Parametric Representation of an Ellipse

Now let’s consider the ellipse centered at the origin with its major axis along the x-axis:
x2 y2
 2  1.
a2 b
Its graph is shown in Figure 9.46. The question is: What might be an appropriate
parameter to introduce to help describe this ellipse?

b P(x, y)

θ
x
–a a

FIGURE 9.46 –b

E XAMPLE 3
Find a parametric representation of the ellipse.
Solution Visualize a point moving around the ellipse shown in Figure 9.46. Although
we can locate each point P in terms of its x- and y-coordinates, we may also be able to lo-
cate it by using the angle u determined by P and the positive x-axis. How do we express x
and y as functions of u?
At first thought, you might be tempted to create a right triangle by dropping a per-
pendicular from P to the x-axis, as we did for the circle. The problem with this approach
is that the length of the hypotenuse would change along with u as the point P moves
around the ellipse, unlike a circle in which the lengths of the line segments from O to P
remain constant. Thus the angle u is not a good choice for the parameter.
Nevertheless, our experience with the circle can provide some guidance. The para-
metric representation of a circle of radius r is
x  r cos t and y  r sin t.
If we think of the ellipse as having a “radius” of a associated with x and a “radius” of b as-
sociated with y, we might write
x  a cos t and y  b sin t.
Let’s see if doing so makes sense. Suppose that 1x, y2 is any point on the ellipse so that x
and y must satisfy
x2 y2
  1.
a2 b2
If we substitute our conjectured expressions for x and y into this equation, we find that
1a cos t2 2 1b sin t2 2 a2cos2t b2sin2t
  
a2 b2 a2 b2
 cos2t  sin2t  1.
gord.3896.09.pgs 4/24/03 10:16 AM Page 624

624 CHAPTER 9 Geometric Models

Because these expressions for x and y as functions of t satisfy the equation of the ellipse,
we conclude that x  a cos t and y  b sin t form a parametric representation of the el-
lipse with parameter t.
◆

Using a Calculator
One of the options available on all graphing calculators is a Parametric mode.
To use it, you need to supply an expression for x in terms of the parameter t and an
expression for y in terms of t.1 You then have to define a window not only in terms
of x and y, but also in terms of an interval of values for the parameter t. Enter the
parametric representation
x  5 cos t
y  3 sin t
for the ellipse
x2 y2
 1
25 9
with a range of values for the parameter from 0 to 2p in radians. Verify that the
graph is indeed that of an ellipse.
To use this parametric representation, suppose that we want to know the point
on the ellipse
x2 y2
 1
25 9
corresponding to a value of the parameter—say, t  p>6. We find that

x  5 cos a b  4.330
p
6

y  3 sin a b  1.5.
p
6
Alternatively, suppose you are told that the point 14, 9>5 2 lies on the ellipse.
(Verify that it does.) To find the value of the parameter t for this point, we consider
x  5 cos t  4
9
y  3 sin t  .
5
The first of these equations gives
4
cos t  ,
5
from which
t  arccos a b  0.6435 radian.
4
5
Verify that this value of t also satisfies the second equation y  3 sin t  9>5.

1
Note that in Parametric mode, different calculators use t or T as the “generic” variable just as
different models use x or X as the “generic” variable in the usual Function mode.
gord.3896.09.pgs 4/24/03 10:16 AM Page 625

9.5 Parametric Curves 625

Parametric Representations of a Hyperbola

We now consider how to write a parametric representation of the hyperbola.

E XAMPLE 4
Show that the equations
a
x  b tan t and y 
cos t
are a parametric representation of the hyperbola
y2 x2
  1.
a2 b2
Solution If we substitute the expressions for x and y into the equation of the hyperbo-
la, we get
1a>cos t2 2 1b tan t2 2 a2 b2 tan2t
  
a2 b2 a2 cos2t b2
1
  tan2t.
cos2t
But, recalling the trigonometric identity
1 1
1  tan2u  2 or  tan2u  1,
cos u cos2u
we see that the previous expression equals 1. Thus the two equations for x and y satisfy
the equation of the hyperbola and therefore represent a pair of parametric equations of
the hyperbola.
◆

An alternative way to develop a parametric representation of the hyperbola re-

quires introducing two new functions known as the hyperbolic sine and the
hyperbolic cosine. We discuss them briefly in the Problems at the end of this section.

Parametric Representations of a Parabola

At the beginning of this section, we described how to find a parametric representa-
tion of the parabolic path of a projectile. We now consider the same situation geo-
metrically. Actually, we can introduce a parameter in an extremely simple way. If
the equation of the parabola is
y  ax2  bx  c,
we can let x  t, so that
y  at2  bt  c.
This approach may strike you as somewhat unfair (too easy!), but it is effective. In
fact, it can be used with any function y  f 1x 2. Let’s look at one of the advantages
of doing so.
If we restrict our attention to the right-hand side of the parabola, we know
that the curve is strictly increasing (if a  0) and so has an inverse f 1. We know
that the graph of the inverse function is the mirror image of the graph of f about
gord.3896.09.pgs 4/24/03 10:16 AM Page 626

626 CHAPTER 9 Geometric Models

the diagonal line y  x. However, in all but the simplest cases, finding an explicit,
or closed form, expression for the inverse f 1 is not easy, or even possible. Without
such a formula for f 1, constructing the graph of the inverse function would nor-
mally be almost impossible.
With parametric functions, however, this becomes a simple chore. Recall the
definition of a function and its inverse function. If y  f 1x 2, for each value of x,
the function determines a single corresponding value for y. The inverse function
undoes this process in the sense that, for each value of y, f 1 returns the value of x
that led to y. We can draw the graph of f in the parametric form
x  t and y  f 1t2
by using the Parametric mode of the graphing calculator. To produce the
graph of the inverse function, all we need do is reverse the roles of x and y. That
is, if we set
x  f 1t2 and y  t,
so that
y  t  f 1 1x 2,
the calculator will draw the graph of the inverse function! Try it with, say, the right
side of a parabola or with an exponential function, where you know what the func-
tion and its inverse should look like.

Other Parametric Curves

Many curves that cannot be represented simply, if at all, with y as a function of x
can be represented fairly readily with parametric equations. Suppose that your
friend has a reflector attached to the rim of her bicycle tire. As she rides past you at
a constant speed, you observe that the path of the reflector is a curve such as the
one shown in Figure 9.47. This curve, showing y as a function of x, is called a
cycloid. If the radius of the tire is a, the parametric representation of the cycloid is
x  at  a sin t and y  a  a cos t,

2a

x
FIGURE 9.47

where the parameter t represents time. We simply cite these equations, which are
typically derived in calculus, and only discuss their reasonableness here. Since the
variable is time, be sure that you graph all such curves in radian mode.
Let’s begin with the expression for the height y  a  a cos t of the reflector
as a function of time t. The constant term a is the vertical shift, so y oscillates
gord.3896.09.pgs 4/24/03 10:16 AM Page 627

9.5 Parametric Curves 627

above and below it as the midline. The amplitude also equals a, so the height y ac-
tually oscillates between 0 and 2a, which makes sense in terms of the physical
phenomenon.
What about the expression x  at  a sin t for the horizontal distance? Note
that this expression involves a sine term, which oscillates between a and a. This
term is subtracted from at, which grows linearly, which again should make sense.
The bicycle wheel is rolling along, so the horizontal distance traveled by the center
of the wheel is simply at. Because the reflector is rotating about the rim of the tire,
there must be an oscillatory adjustment to the linear distance covered.
In Problem 26 of Section 2.5, we raised the question about the shape of a water
slide along which a person would slide most rapidly from one point to another; it
is called the brachistochrone problem. From physical principles, the curve should be
decreasing and concave up, so that the person gains the greatest speed at the begin-
ning of the slide. It turns out that the specific curve along which the time needed is
a minimum is an upside-down cycloid.
Let’s consider another application involving a parametric representation of a
curve. You have likely seen a spirograph, a toy with which you can draw intricate
shapes by tracing curves as one plastic wheel rotates about another plastic wheel.
Suppose that you have a large wheel of radius b and a smaller wheel of radius a that
is rolling on the outside of the larger wheel, as shown in Figure 9.48. A fixed point
on the outer (rolling) circle describes a curve that is known as an epicycloid. A
parametric representation of the epicycloid is
ab
x  1a  b2 cos a b  a cos c a btd
at
b b
ab
y  1a  b 2sin a b  a sin c a btd.
at
b b

b a

FIGURE 9.48

Let’s see what the path of the fixed point on the rolling circle looks like. As the
outer circle rolls on the fixed inner circle, the point on it moves back and forth, get-
ting closer to and farther from the origin. It is closest to the origin—at a distance
b—at the points where the two circles touch. It is farthest from the origin when the
point is at the farthest possible position on the rolling circle, a distance of b  2a.
At any other time, the point is at an intermediate distance between b and b  2a.
gord.3896.09.pgs 4/24/03 10:16 AM Page 628

628 CHAPTER 9 Geometric Models

The actual shapes of epicycloids are often visually surprising and striking, as
shown in Figure 9.49 for a  11, b  28, and t between 0 and 421p. A much sim-
pler case is when the fixed inner circle has a radius of b  4 and the rolling circle
has a radius of a  1, which gives the epicycloid shown in Figure 9.50 for t be-
tween 0 and 8p  25.13; the same curve thereafter repeats with period 8p because
the identical points are repeatedly traced out. We also superimpose the inner fixed
circle to indicate how the epicycloid is traced out by the fixed point as the outer
(unseen) circle rolls around on the inner circle.

y
x
6

x
–6 –2 2 6
–2

–6

FIGURE 9.49 FIGURE 9.50

In the Problems for this section we ask you to experiment with the epicycloid
and other curves by using parametric equations. You will see some surprising shapes
if you simply try interesting combinations of functions. A favorite parametric curve
that you can try is the “snowman” function whose parametric representation is
1 1
xt sin 10t and y  5 sin t  cos 10t.
2 2

Problems
1. Consider the parametric representation of the line c. How does the slope of the line relate to the coef-
x  4  3t, y  2  5t. ficients in the parametric representation?
a. Construct a table of points that lie on this line d. Eliminate the parameter t algebraically by solv-
and find the slope of the line from the table. ing for t from the first equation and substituting
b. Use the slope and a point on the line to write an the result into the second.
equation of the line with y as a function of x.
gord.3896.09.pgs 4/24/03 10:16 AM Page 629

9.5 Parametric Curves 629

e. Eliminate the parameter t algebraically by solv- 7. Repeat Problem 5 with a  2 and b  3, 5, 7, and 9.
ing for t from the second equation and substitut- 8. Figure 9.47 shows the graph of a cycloid, which is
ing the result into the first. the path of a reflector mounted on the rim of a tire
2. Consider the parametric representation of the line: of radius a. Determine the coordinates of the points
x  7  3t, y  4  5t. where the curve touches the horizontal axis.
a. Based on the results of part (c) of Problem 1, 9. A hypocycloid is the curve generated by a fixed
what do you expect the slope of this line to be? point on a circle of radius a that rolls around the in-
b. Create a table of values for this line and use the side of a larger circle of radius b. The parametric
entries to sketch the graph of the line. equations for a hypocycloid are
c. Find a point–slope form of the equation of the ba
line. x  1b  a2cos t  a cos c a btd
a
d. Find an equation of the line by eliminating the
ba
parameter t algebraically. y  1b  a2sin t  a sin c a btd.
a
3. Consider the curve given in parametric form
x  t3  1, y  t2  2.
a. Create a table of values for this function by using
t  2, 1.5, 1, . . . , 2 and plot the points to
construct a rough sketch of the graph.
b. Draw the curve using the Parametric mode on
your function grapher. How does the result b a

compare to that in part (a)?

c. Eliminate the parameter t algebraically by first
solving for t in terms of x.
d. Graph the function you obtained in part (c) by
using the Function mode on your function
grapher. How does it compare to your graph in Use your function grapher to see the shapes that
part (b)? result when (a) a  1, b  2; (b) a  1, b  3;
4. Consider the curve with the parametric representa- (c) a  1, b  4; and (d) a  2, b  3.
tion x  t2  1, y  t2  2. 10. Suppose that a bicycle reflector is mounted partway
a. Create a table of values for this function, using along one of the spokes in a wheel at a distance
t  1, 0, 1, . . . , and plot the points to con- b  a from the center of the wheel. How should the
struct a rough sketch of the graph. What surpris- parametric equations for the cycloid be modified to
ing result do you get? reflect this new position?
b. Use your function grapher in Parametric
mode to verify that the result you obtained in
part (a) is correct.
c. In terms of x and y, what are the domain and
range for the curve you found in part (a)?
d. Eliminate the parameter t algebraically and ex-
b<a
plain why you got the shape you did.
5. Use the Parametric mode on your function graph-
er to draw the graph of the epicycloid with a  1 and
b  3. Repeat with b  4, b  5, and b  6 while
keeping a  1. Do you see any pattern in the periods 11. In the equations for the cycloid, x  at  a sin t
of these curves? Do you observe any pattern in the and y  a  a cos t, use the second equation to
number of loops that you get? Explain these patterns. solve for t and then substitute the result into the
6. Repeat Problem 5 with a  1 and b  6, 8, 10, and first equation to eliminate the parameter and ob-
12. How do the curves you get compare to the ones tain x as a function of y. What does the resulting
you obtained before? equation tell you about the path of the reflector?
gord.3896.09.pgs 4/24/03 10:16 AM Page 630

630 CHAPTER 9 Geometric Models

9.6 The Polar Coordinate System

As we discussed in Section 9.1, the polar coordinate system is based on the idea that
every point in the plane must lie on some circle centered at the origin. In this coor-
dinate system, the origin is known as the pole. To locate a point P in such a system,
we must indicate the radius r of the particular circle on which P lies, as shown in
Figure 9.51. That is, does P lie on a circle of radius r  3 or a circle of radius r  4
or a circle of radius r  4.2689? Knowing that P lies on a specific circle still does
not locate the point exactly. We must also specify where the point lies on that circle.
Is it at an angle of 30° or an angle of 45° or an angle of 263° measured counter-
clockwise from the horizontal?

FIGURE 9.51

To formulate these ideas more precisely, let’s develop some appropriate termi-
nology. We first introduce a horizontal axis starting at the pole and pointing to the
right. It is called the polar axis and serves as a reference. The distance from the pole
to the point P, which is equivalent to the radius of a circle centered at the pole, is de-
noted by the coordinate r. To locate a specific point P on this circle of radius r, we
must indicate how far around the circle P lies, starting from the polar axis. We meas-
ure this distance around the circle in terms of an angle coordinate u drawn counter-
clockwise, or in a positive direction, from the polar axis, as shown in Figure 9.52.
Thus we can locate any point in the plane if we know its distance r from the pole (to
determine a circle) and the angle u around this circle. The polar coordinates of the
point P consist of r and u, so we write the point as 1r, u 2.
For example, a point that lies 5 units from the pole at an angle of 60°, or p>3
radians, with the polar axis has polar coordinates 15, 60°2 or 15, p>32, as shown in
Figure 9.53. Similarly, the point Q that lies 3 units from the pole at an angle of
2p>3, or 120°, has coordinates 13, 2p>3 2 or 13, 120°2.
We can visualize the polar coordinates of a point P in the following alternative
way. Any point is located on a line passing through the pole, as measured by the
angle of inclination u, which is known as the polar angle. The particular location
of the point P along that line is determined by its distance r from the pole. Thus P
can be visualized as lying at the intersection of a line through the pole and a circle
centered at the pole.
This approach has an added geometric advantage. Recall from geometry that
the radius drawn to any point on a circle is perpendicular to the tangent line at that
point. Therefore the polar coordinates of a point are determined by the intersection
gord.3896.09.pgs 4/24/03 10:16 AM Page 631

9.6 The Polar Coordinate System 631

π
2π 2 π
3 3

P
P(r, θ ) π
6
r Q

θ
π θ =0
1 2 3 4 5 6 r
pole
(origin)

3π
2

FIGURE 9.52 FIGURE 9.53

of two curves that we can think of as being “perpendicular” at the point. This prop-
erty is analogous to what we do in rectangular coordinates where the vertical and
horizontal lines that determine a point are perpendicular to each other.
Although there are many advantages to working with a polar coordinate system,
it does have one disadvantage. In rectangular coordinates, every point has a unique
pair of coordinates. However, every point in polar coordinates has more than one
address. Consider the point 1 unit to the right of the pole on the polar axis. Accord-
ing to our discussion so far, you might conclude that its polar coordinates are r  1
and u  0. However, with a little thought, it should be evident that the address for
this point could also be r  1 and u  2p, or r  1 and u  4p, and so on. Thus
there are infinitely many polar coordinate representations of the same point.
In fact, there are still other ways to give the polar address of this point. In gen-
eral, any angle u measured counterclockwise from the polar axis is considered pos-
itive; any angle u measured clockwise from the polar axis is considered negative, as
illustrated in Figure 9.54. Thus our point on the polar axis could also be written as
11, 2p 2, for instance.
Furthermore, we encounter some situations in Section 9.7 in which an angle u
gives rise to a negative value for r. Let’s see what this means because a negative
value of r cannot represent the radius of a circle. If u  p>4 and r  3, we simply
measure a distance of 3 units along the terminal side of the angle u  p>4. How-
ever, if u  p>4 and r  3, we can locate the corresponding point by extending
the terminal side of the angle backward through the pole and measuring 3 units
along this extension, as illustrated in Figure 9.55.

Terminal side

θ = π4
θ >0
θ <0
3

π
(−3, 4
)

FIGURE 9.54 FIGURE 9.55

gord.3896.09.pgs 4/24/03 10:16 AM Page 632

632 CHAPTER 9 Geometric Models

Let’s look at this issue a bit more formally. When we draw any angle u, it deter-
mines a terminal side OP from the pole through some point P, as illustrated in Fig-
ure 9.56. The obvious polar coordinate representation for this point is 1r, u 2, where
r is positive because the distance is measured along the terminal side. However, we
can also represent that point by considering the angle u  p (corresponding to an
additional rotation of p radians or 180°) and measuring a distance r from the pole
in the opposite direction. In such a case, we think of r as negative and the polar co-
ordinates of the point as 1r, u  p2 . Thus, if a point P is located at u  p>4 and
r  3, we can consider the associated angle p>4  p  5p>4 and assign the coor-
dinates 13, 5p>4 2 to the point P as well.
With these ideas in mind, we can find even more ways to write our earlier
point with coordinates r  1 and u  0. For instance, we can obtain this
point when r  1 and u  p or when r  1 and u  2p or when r  1
and u  p, as illustrated in Figure 9.57.

P(r, θ )
P(–r, θ + π ) θ = −2 π

θ+π
P(–1, – π )
θ = 2π P(1, 0)
θ
P(1, 2 π )
P(–1, π )

θ = −π

FIGURE 9.56 FIGURE 9.57

Think About This Can you think of any other coordinates for this point when r is negative? when u is
negative? ❐

Hence any point in the polar coordinate system has infinitely many pairs of
coordinates. Even the pole, where r  0, has infinitely many representations be-
cause it can be thought of as corresponding to any possible angle u.

Transforming Between Polar and Rectangular Coordinates

Often, it is useful to think of the two coordinate systems, polar and rectangular,
as being superimposed. In such a case, the pole and the origin are the same
point; the polar axis and the positive x-axis coincide. The question then is: How
do the coordinates of a point P in one system relate to the coordinates of the
same point in the other system? That is, how do we transform the rectangular
coordinates 1x, y2 of a point into the equivalent polar coordinates 1r, u2 and vice
versa?
Suppose that we start with a point P having polar coordinates 1r, u2 and we
want to determine the corresponding rectangular coordinates x and y. From the
right triangle shown in Figure 9.58, it is clear that
gord.3896.09.pgs 4/24/03 10:16 AM Page 633

9.6 The Polar Coordinate System 633

x  r cos u and y  r sin u.

E XAMPLE 1
The point P has polar coordinates r  5 and u  p>3. Find the corresponding rectan-
gular coordinates x and y.
Solution Using the preceding two equations, we find that the rectangular coordinates are

 5 a b  2.5
p 1
x  r cos u  5 cos
3 2
and
23
 5a b  4.33.
p
y  r sin u  5 sin
3 2
◆
For the reverse problem, suppose that we start with a point P whose rectangu-
lar coordinates are 1x, y2, as shown in Figure 9.59. We now want to find the corre-
sponding polar coordinates r and u. First, we observe that r is the distance from the
pole (origin) to P. The Pythagorean theorem gives
r 2  x 2  y 2 so that r   2x 2  y 2 .

P(r, θ ) = P(x, y) P(x, y)

r = √x 2 + y2
r y
y = r sin θ

y
θ = arctan x
θ
x
x = r cos θ x

FIGURE 9.58 FIGURE 9.59

Next, observe that

y y
tan u  so that u  arctan .
x x

Thus, given the rectangular coordinates 1x, y2 of a point, we can find the polar co-
ordinates 1r, u 2 by using

y
r   2x 2  y 2 and u  arctan .
x
gord.3896.09.pgs 4/24/03 10:16 AM Page 634

634 CHAPTER 9 Geometric Models

However, these formulas have to be used with great care, as demonstrated in

Example 2.

E XAMPLE 2
If the rectangular coordinates of a point are x  3 and y  4, find one set of polar coor-
dinates for that point.
Solution We find one set of polar coordinates using
r   2x2  y2   232  42   225  5
and
4
u  arctan  0.927 radian,
3
or about 53.13°. These values give rise to infinitely many possible pairs of coordi-
nates, but not all of them are appropriate for the point P. The point P lies in the first
quadrant, as shown in Figure 9.60, so one possible polar representation is
1r, u2  15, 0.9272 . But the coordinates 15, 0.9272 are not correct because they lie in
the third quadrant. However, 15, p  0.9272 , which is 15, 4.0692 in radians or
15, 233.13°2, is another possible pair of coordinates for P. But the polar coordinates
15, p  0.9272 in radians or 15, 233.13°2 is not correct because it also is a point in the
third quadrant. Be sure to plot the point in order to decide which value of r to match
with which value of u.

y
(5, 53.13°)
(–5, 233.13°)
P
5
233.13°
53.13°
x

(–5, 53.13°)
FIGURE 9.60 (5, 233.13°)

◆
Think About This Explain why only even multiples of p are used for the u-coordinate of the different
representations for the point P in Example 2. ❐

Polar coordinates are particularly useful in representing situations in which

there is a single special point and all other ideas of interest are centered at that
point. For instance, in physics, the total mass of a body is often assumed to be at a
single point corresponding to the pole. Thus satellites can often be thought of as
moving in circular orbits about a planet located at the pole of a polar coordinate
system. Similarly, the magnetic field associated with a magnet can be thought of as
being centered at the pole of a polar coordinate system, and all related phenomena
are often best expressed in terms of polar coordinates.
gord.3896.09.pgs 4/24/03 10:16 AM Page 635

9.6 The Polar Coordinate System 635

Problems
1. For each of the points P, Q, R, and S shown, do the h. 13, 8 2
following. 4. Transform the polar coordinates 1r, u 2 in (a)–(i) to
π
equivalent rectangular coordinates. Indicate the lo-
2 π cation of each point graphically in the polar plane.
a. 15, 0 2
3

π b. 15, p>2 2
c. 15, p>22
P 6
Q
d. 15, 0 2
π θ =0 e. 13, p>3 2
f. 13, p>3 2
1 2 3 4 5 6 r

g. 12, 3p>2 2
R S

h. 12, 5p>4 2
i. 12, 5p>3 2
3π
2 5. A satellite is in a circular orbit about the equator at
a height of 22,800 miles above the surface of the
a. Write a polar coordinate representation with r
Earth. The radius of the Earth is about 4000 miles.
and u both positive.
The longitude line running north–south from the
b. Write a polar coordinate representation with r
north pole to the south pole through Greenwich,
positive and u negative.
England, serves as the 0° reference. Because the cir-
c. Write a polar coordinate representation with r
cumference of the Earth is about 24,000 miles, each
negative and u positive.
15° of longitude corresponds to about 1000 miles
d. Write a polar coordinate representation with r
along the equator.
and u both negative.
2. A merry-go-round at an amusement park has an
inner radius of 9 feet and an outer radius of 26 feet.
On the merry-go-round are five concentric circles
of horses, 3 feet apart, starting with the innermost 0°

circle 10 feet from the center. Let the polar axis ex-
tend from the center of the merry-go-round to the
entrance gate of the ride. 3000 mi
a. What are the polar coordinates of the horse in
Kampala
the outer, or fifth, circle that is one-third of the
way around the merry-go-round to the right 22,800 mi
from the gate?
b. What are the polar coordinates of the horse in
the second circle that is one-fifth of the way
around to the left from the gate?
3. Transform the rectangular coordinates 1x, y2 in
(a)–(h) to the equivalent polar coordinates. Sketch a. What are the polar coordinates of the Ugandan
the location of each point in the polar coordinate capital Kampala, which is 3000 miles east (posi-
plane. tive direction) of the Greenwich baseline?
a. 14, 42 b. What are the polar coordinates of the capital of
b. 14, 4 2 Borneo, which is about 7500 miles east of the
c. 14, 42 Greenwich baseline?
d. 14, 42 c. What are the polar coordinates of the satellite
e. 13, 4 2 when it passes over Quito, Ecuador, which is
f. 13, 42 5200 miles west of the Greenwich baseline?
g. 18, 32
gord.3896.09.pgs 4/24/03 10:16 AM Page 636

636 CHAPTER 9 Geometric Models

9.7 Families of Curves in Polar Coordinates

In Section 9.6, we introduced the notion of polar coordinates and considered coor-
dinates 1r, u 2 for individual points in such a system. A far more interesting and use-
ful question is: How do we represent curves and families of curves in polar
coordinates?
Recall that, in rectangular coordinates, the curve associated with an equation
y  f 1x2 consists of all points 1x, y2 whose coordinates satisfy the equation. We use
the comparable notion when working with polar coordinates but with the under-
standing that it is only necessary that one representation of a point in polar coor-
dinates satisfies the equation.
To begin, it is usually much simpler to think of the angle u as the independ-
ent variable and the distance r from the pole as the dependent variable. Thus for
most functions in polar coordinates, we write r  f 1u 2 for some set of values of
the angle u. Then for each allowable value of u, the function determines a corre-
sponding value for r and the pair 1r, u 2 represents the polar coordinates of a point
P in the plane. The totality of all such points determined by the equation consti-
tutes the graph of the function. Note that writing the polar coordinates of a point
as 1r, u2 reverses our usual notation of writing the independent variable first and
the dependent variable second, as is done in rectangular coordinates with 1x, y2.
Let’s begin with some particularly simple cases. First, consider the equation
r  f 1u 2  c,
where c is a constant. Thus, no matter what the angle u is, the distance r from the
pole is the constant c. The set of all points that satisfy this condition forms the cir-
cle of radius c centered at the pole, as shown in Figure 9.61.
Next, consider the equation u  a, where a is a constant; note that the dis-
tance r is not explicitly mentioned. No matter what distance we use, the correspon-
ding point is always located at the angle a as measured from the polar axis. The set
of all such points forms a line, inclined at the angle a which passes through the
pole, as depicted in Figure 9.62. For instance, u  p>4 represents the line through
the pole inclined at a 45° angle.

c c
c
r
c α
c

FIGURE 9.61 FIGURE 9.62

Shapes that are far more interesting and intricate than a circle and a line arise
from relatively simple polar equations. We investigate some types of shapes and
their underlying patterns for various families of polar coordinate curves. In the rest
gord.3896.09.pgs 4/24/03 10:16 AM Page 637

9.7 Families of Curves in Polar Coordinates 637

of this section, you should use your graphing calculator set in Polar mode or a
polar graphing program for a computer.
Working with polar coordinates often has a special advantage over working
with rectangular coordinates. Consider the simple curve shown in Figure 9.63,
which is known as an Archimedean spiral. Its equation in polar coordinates is
r  u. When u  0, we have r  0, so the curve starts at the pole. As u increases,
the distance r from the pole likewise increases, and as u loops repeatedly around
the pole, so does the curve to form the spiral shape shown.
y

x
–8 –6 –4 –2 2 4 8
–2

FIGURE 9.63 –6

Now, let’s find an equivalent equation in rectangular coordinates, using the

transforming equations we derived in Section 9.6:
y
r 2  x 2  y 2 and u  arctan ,
x
and
x  r cos u and y  r sin u.
Substituting the first pair of these expressions into the equation r  u, we get the
rectangular equation
y
2x 2  y 2  arctan ,
x
which is not particularly attractive. We can simplify this expression slightly by tak-
ing the tangent of both sides to eliminate the arctangent function:

tan 1 2x 2  y 2 2  .
y
x
Or, if we multiply through by x,
x tan 1 2x 2  y 2 2  y.
Neither of these expressions is any more attractive. Moreover, we can’t simplify any
of these expressions to write y as a function of x or to write x as a function of y. (In
fact, recall that such a curve does not represent a function.) Furthermore, none of
these rectangular expressions gives any insight into the behavior of the curve,
whereas the polar representation r  u was very helpful in understanding the
spinal curve shown in Figure 9.63.

Think About This What happens to the spiral if u  0? ❐

gord.3896.09.pgs 4/24/03 10:16 AM Page 638

638 CHAPTER 9 Geometric Models

E XAMPLE 1
Consider the polar function r  f 1u 2  cos u, whose graph is shown in Figure 9.64. By
eye, the curve appears to be circular, appears to pass through the pole, and appears to be
symmetrical about the polar axis. Show that this curve is a circle.

r = cos θ

FIGURE 9.64

Solution To prove that the curve is a circle, we can try to express it in rectangular coor-
dinates where the equation of a circle would be recognizable. Although we could attempt
to substitute the transforming expressions for r and u into the equation r  cos u, using
a little trick is much easier. We multiply both sides of the given equation r  cos u by r
to get
r2  r cos u,
which is equivalent to the rectangular equation
x2  y2  x or x2  x  y2  0.
To determine whether this is the equation of a circle, we complete the square in the x-terms:
1 2 1 2
1x2  x 2  y2  c x2  x  a b  a b d  y2
2 2

 ax2  x  b   y2
1 1
4 4
1 2
 ax  b  y2   0,
1
2 4
so that
1 2
ax  b  y2  .
1
2 4
This is the equation of a circle with radius 12 centered at the (rectangular) point 1 12 , 0 2. This
circle is indeed symmetrical about the horizontal axis and does pass through the pole.

Think About This 1. Describe the graph of r  5 cos u.

2. Describe the graph of r  a cos u, for any multiple a  0.
3. Describe the graph of r  a sin u, for any multiple a  0. ❐
gord.3896.09.pgs 4/24/03 10:16 AM Page 639

9.7 Families of Curves in Polar Coordinates 639

The Family of Rose Curves

Let’s consider some related curves in polar coordinates. When we graph the equa-
tion r  cos 2u, we obtain the result shown in Figure 9.65. This graph corresponds
to angles u ranging from 0 to 2p. If we extend the values beyond this interval, in ei-
ther direction, the same points repeat, so the result is a periodic function with pe-
riod 2p. If you experiment with your polar function grapher, you will notice that
the graph shown is traced repeatedly when you take a large range of values for the
angle u.
Don’t just look at the completed shape, but rather consider this curve and other
polar coordinate curves we discuss in a dynamic manner. How are the curves pro-
duced or traced? Think of the cursor on the calculator or the computer screen as a
moving point that traces the curve and observe carefully how the curve is generated.
Note that the graph shown in Figure 9.65 for r  cos 2u consists of four loops
of equal size. (Actually, depending on the calculator or computer graphics pack-
age you use, there may be some distortion and the loops may not appear to be pre-
cisely the same size even though they are.) To get a better feel for how the
particular shape evolves, watch carefully as the curve is traced in Figure 9.66. Note
that it starts at the far right (corresponding to u  0 where r  1) and then loops
around (portion ①) until it passes through the pole (corresponding to u  p>4,
where r  cos 21p>4 2  cos1p>22  0). It then starts to form a second loop
(portion ②) as r takes on negative values. Eventually, it completes the loop (por-
tion ③) before it again passes through the pole, this time at an angle of u  3p>4
so that again r  0. It then begins to form the third loop (portion ④ ) and com-
pletes that loop (portion ⑤) when it passes through the pole, where u  5p>4. It
then forms a fourth loop (portions ⑥ and ⑦), for u between 5p>4 and 7p>4. It fi-
nally completes the original loop (portion ⑧) as u progresses to 2p. This curve is
known as a four leaf rose.

5π 7π
4
to 4

1
6
r = cos 2θ 7
4 1 0 to π4

3π 6π
4
to 4
θ =0

–1 –0.5 0.5 1 5 8
7π
4
to 2π
2 3

π 3π
–1 4
to 4

FIGURE 9.65 FIGURE 9.66

Think About This 1. What shape is produced if you graph r  a cos 2u for any multiple a?
2. Describe the graph corresponding to r  a sin 2u for any multiple a. How
does it compare to the graph of the cosine function in part (1)? ❐
gord.3896.09.pgs 4/24/03 10:16 AM Page 640

640 CHAPTER 9 Geometric Models

Let’s now make a relatively simple change and consider r  cos 3u instead of
the four-leaf rose r  cos 2u.

E XAMPLE 2
Describe the graph of r  cos 3u.
Solution The resulting graph is shown in Figure 9.67, but we need to observe carefully
how the curve is traced. First, we observe that the curve now consists of only three loops,
and they are traced for values of u between 0 and p. For any angles outside the interval
30, p 4, the same points are produced, so the polar curve is periodic with period p, even
though the function f 1u 2  cos 3u is periodic with period 2p>3. Next, we observe that
the curve starts when u  0 and r  1 to produce the point at the far right. It then forms
a half loop and passes through the pole when u  p>6. The lower left full loop is traced
for values of u between p>6 and p>2. The upper left full loop is traced as u ranges from
p>2 to 5p>6. The bottom half of the right-hand loop is completed as u ranges from
5p>6 to p. This curve is known as a three-leaf rose.

r = cos 3θ

–0.6 1

–1
FIGURE 9.67

◆
In general, the family of curves given by r  cos nu or r  sin nu for any posi-
tive integer n are called rose curves. Figures 9.68(a) and (b) show the graphs of
r  cos 4u and r  cos 5u; note that they contain eight and five loops, or petals,
respectively.

1 r = cos 4θ r = cos 5θ
1

0.5

–1 1 –1 –0.5 1

–0.5

–1 –1

(a) (b)

FIGURE 9.68
gord.3896.09.pgs 4/24/03 10:16 AM Page 641

9.7 Families of Curves in Polar Coordinates 641

Think About This 1. Investigate some other cases using your polar function grapher until you
can devise a rule to predict the number of petals in the rose curve
r  cos nu for any positive integer n. Are there any numbers of petals that
cannot occur in this family of rose curves? If so, what are they?
2. What can you conclude about the number of petals in the related family of
rose curves given by r  sin nu? ❐

The Family of Cardioids

Let’s consider another family of polar coordinate curves, those given in the form
r  a11  cos u 2 . Figure 9.69 shows the graph of r  1  cos u, with a  1. The
heart-shaped appearance of this curve suggested its name, a cardioid.

r = 1 + cos θ
1.5
1

–0.5 0.5 1 1.5 2

–1

FIGURE 9.69 –1.5

E XAMPLE 3
Describe how the graph of the cardioid r  1  cos u is traced.
Solution We start with u  0 so that r  2. The curve begins at the point at the far
right. As u increases to p>2, the curve arches upward. For u between p>2 and p, the curve
bends downward and eventually inward to the pole; the resulting point at u  p is called
a cusp. As the angle u increases from p to 2p, the curve traces the mirror image of the
upper half of the cardioid; this cardioid is symmetric about the polar axis with period 2p.
◆
Those of you who have read Section 8.4 on chaos have seen that the primary
central portion of the Mandelbrot set is a cardioid.

Think About This a. What is the effect of a multiple a on the shape of the curve
r  a11  cos u 2?
b. Describe the graph of the related equation r  1  cos u. How does it
compare with the cardioid r  1  cos u? ❐

Think About This Sketch some graphs of the related equations r  1  sin u. Identify an
axis of symmetry for them. ❐

Think About This Suppose that you combine the ideas on the rose curves and the cardioids
to consider the class of polar equations of the form r  1  cos nu for dif-
ferent positive integers n. Determine a pattern regarding their shapes. ❐
gord.3896.09.pgs 4/24/03 10:16 AM Page 642

642 CHAPTER 9 Geometric Models

The Family of Limaçons

An extension of the cardioid known as the limaçon is defined by the equations
r  a  b cos u and r  a  b sin u. In particular, we consider two cases: a  b
and a  b. When a  b, both equations reduce to that of a cardioid. Figure 9.70
shows the graph of r  3  4 cos u.

5 r = 3 + 4 cos θ

3
θ ≈ 221.4°

–1 1 3 5 7

θ ≈ 138.6°
–3

FIGURE 9.70 –5

The curve starts at the far right, where u  0 and r  7. The curve then traces
around the upper arch and eventually bends inward to pass through the pole. After
passing through the pole, the curve traces the small inner loop and then passes
through the pole again. It then traces the large outer loop below the polar axis,
which is a mirror image of the large loop above the polar axis. The resulting curve
is called a limaçon with a loop. (It comes from the Greek word limax, for snail, be-
cause the first half of the curve traced from u  0 to u  p resembles a snail-like
shape.) For u  2p, the curve precisely repeats this behavior.

E XAMPLE 4
At what angles does the limaçon curve r  3  4 cos u pass through the pole?
Solution The graph of the limaçon in Figure 9.70 shows two such angles—one in the
“second quadrant” and the other in the “third quadrant”. To find these angles, we use the
fact that the pole corresponds to r  0. Therefore, if we set r  0, we get the equation
3
3  4 cos u  0 so that cos u   .
4
Thus one angle at which the limaçon passes through the pole must satisfy is

u  arccos a b  2.419 radians, or 138.59°.

3
4
We use the symmetry of the cosine function to find the second solution at u  3.864 ra-
dians, or 221.41°. These values agree with the visual estimates that can be made by look-
ing at Figure 9.70.
◆
Figure 9.71 shows the graph of the polar curve r  5  4 cos u. This curve is
known as a limaçon without a loop, or a dimpled limaçon. It is similar in appearance
to a cardioid, but it does not reach the pole.
gord.3896.09.pgs 4/24/03 10:16 AM Page 643

9.7 Families of Curves in Polar Coordinates 643

r = 5 + 4 cos θ
6

–2 2 4 6 8 10
–2

–6
FIGURE 9.71

Think About This Can you account for the fact that the curve r  5  4 cos u never passes
through the pole (where r  0 ) and so never produces a loop? ❐

Think About This Devise criteria based on the values of a and b in r  a  b cos u so that
you can determine whether there is a loop. Be sure that you graph a variety
of limaçons using your polar function grapher to collect enough informa-
tion to know that you are correct. ❐

Think About This Describe the shape of limaçons given byr  a  b cos u. ❐

Think About This What happens in the related family of limaçons given by r  a  b sin u? ❐

We urge you to experiment with the curves generated by polar coordinate

equations. You can get some incredibly striking effects just by creating strange
combinations of different functions. For instance, the graph of the polar function
r  sin5u  8 sin u cos3u
is the butterfly shape shown in Figure 9.72.

2
r = sin5θ + 8 sin θ cos3θ
1.5

–2.5 –0.5 0.5 2.5

–1

FIGURE 9.72 –1.5

Think About This Explore systematically some family of polar functions—say, r  sinnu for various
integers n. You may well discover some fascinating new patterns and add some new
items to the literature of mathematics. ❐
gord.3896.09.pgs 4/24/03 10:16 AM Page 644

644 CHAPTER 9 Geometric Models

Problems
In Problems 1–11, graph each polar curve using your a. By experimenting with different combinations
polar function grapher. For each, use a variety of inter- of a and b, can you determine any rules for pre-
vals for the angle u until you obtain a “good” picture of dicting the number of (overlapping) loops that
the graph. will result? If so, state them.
4 sin2u b. Can you determine any rules for predicting the
1. r  (Cissoid of Diocles) interval of angles u needed to trace one complete
cos u petal of this curve? If so, state them.
1 c. Can you determine any rules for predicting the
2. r   2 (Conchoid of Nicomedes)
sin u interval of angles u needed to trace the entire
3. r  4 sin u cos2u (Bifolium) curve? If so, state them. (Hint: Consider different
cases, depending on whether a and b are odd or
4. r  5 a4 cos u  b (Trisectrix)
1
even.)
cos u
24. Consider the family of polar curves given by
3 sin u r  sinnu.
5. r  (Cochleoid)
u a. After graphing the curves corresponding to
4 n  1 and n  2, what shape do you expect for
6. r  (Lituus)
2u n  3? for n  4?
8 b. Account for the fact that the shapes are not what
7. r  (Cruciform) you expected.
sin 2u
c. Determine a pattern for the number of loops
10 that will correspond to any value of n.
8. r  (Ellipse)
3  2 cos u d. What interval of angles corresponds to a com-
9. r 2  4 cos 2u (Lemniscate of Bernoulli) plete curve? Do the same conclusions apply to
(Caution: Be sure to restrict your attention to values of r  cosnu?
u that cause the right-hand side to be positive.) 25. Consider the family of generalized lemniscates given
10. r 3  4 cos 3u (Generalized Lemniscate) by r 2  cos nu. Can you find any pattern for the
number and location of the loops that will result for
(Caution: Some programs and calculators are not able
any n? (Caution: When you try to graph these
to evaluate the cube root of a negative number.)
curves, you must take into account intervals of an-
4 gles for which the function is well defined.)
11. r 
sin u 26. Consider the family of generalized lemniscates given
12–22. Repeat Problems 1–11 by changing some of the by r n  cos nu. Can you determine a pattern for the
terms. What happens to the shape you produced if number and location of the loops that will result for
you use different values for the coefficients? What any n? If so, what is it?
happens if you interchange sines and cosines? What 27. Consider the family of generalized limaçons given
happens if you change the multiples of u? Keep a by r  a  b cos nu. Can you find a pattern for the
record of what you do and of your findings. number and location of the loops that will result
23. Consider the family of “hybrid rose curves”1 given for any n? If so, what is it?
by r  cos1 ab u2 for any rational number a>b.

Chapter Summary

In this chapter we introduced and discussed a variety of topics related to coordinate

systems in general and several specific coordinate systems in particular. This includes:

1
These curves were studied in detail by a student, Kenneth Gordon, in the article, Investigating the
petals of hybrid roses, Mathematics and Computer Education, 1992, 26, 66–73.
gord.3896.09.pgs 4/24/03 10:16 AM Page 645

Review Problems 645

◆ What a coordinate system is.

◆ How to find the distance between points in the plane.
◆ How to find the midpoint of a line segment.
◆ How to find a point at any given distance along the line through two points.
◆ The parametric equations of a line.
◆ The equation of a circle.
◆ The equation of an ellipse, including finding its center, vertices, and foci.
◆ The reflection property of an ellipse and its applications.
◆ The equation of a hyperbola, including finding its center, vertices, and foci.
◆ Applications of the hyperbola.
◆ The equation of a parabola, including finding its vertex, focus, and directrix.
◆ The reflection property of a parabola and its applications.
◆ The parametric representation of curves in the plane.
◆ What the polar coordinate system is and how to transform between polar
and rectangular coordinates.
◆ The behavior of families of curves in polar coordinates.

Review Problems
1. Find an equation of each ellipse shown. 2. Find an equation of each hyperbola shown.

y y

1
x
1
1
x
1

a. a.

y y

1
x
x 1
1
–1 (–√13, –1) (√13, –1)

b. b.
gord.3896.09.pgs 4/24/03 10:16 AM Page 646

646 CHAPTER 9 Geometric Models

3. Find an equation of each parabola shown. 13. Find the equation of the hyperbola that has vertices
10, 4 2 and passes through the point 16, 180 2.
a. y 14. Find the equation of the set of points P with coor-
dinates 1x, y2 in the plane such that the sum of the
distance from 112, 02 to P and the distance from
112, 02 to P is 30.
15. An ellipse passes through the point P at 1 13>2, 22
and has foci at 13, 0 2 and 13, 02. Use the geomet-
ric definition of an ellipse to find the equation of
this ellipse.
1 16. The ceiling of a whispering gallery is built so that
x the highest point of the structure is 16 feet above
1
the floor. The floor has vertices 40 feet apart. Where
along the axes should each person stand to be able
y to get the “whispering effect”? Ignore the height of
b.
the two people.

) ))

)))
)))
)

))))
)))
)))

))))
)))
)))

)
)))

)
16 ft

)
)))

) ))
)))

))
)))

))) )
))

x 40 ft
2

17. A lithotripter is a medical device used by doctors to

break up kidney stones by bombarding them with
intense bursts of sound waves, using the reflection
4. Identify the conic whose equation is xy  5 and property of an ellipse. The device is situated so that
sketch the curve. the sound waves emanate from one focus, reflect off
In Problems 5–9, determine the equation for the stan- an elliptic-shaped bowl, and come together to strike
dard form of the conic section. Identify the conic and the kidney stone at the other focus. The distance
sketch the curve. Wherever applicable, give the focus between the two foci is 23 cm, and the distance
(foci), vertex (vertices), and center for each conic. from the source focus to the vertex on the elliptic
5. x2  6x  y  34  0 reflector bowl is 3 cm. Find the equation of an ellip-
6. x2  y2  8x  6y  9  0 tic cross section of the lithotripter bowl.
7. 2x2  3y2  20x  12y  28  0 18. Let x  4  t and y  2  3t. Graph the points
1x, y2 for t  2, 1, 0, 1, and 2. Find the function
8. 3x2  4y2  6x  24y  45  0 determined by the parametric equations.
9. 3y2  2x2  12y  12x  24  0 19. Sketch the parametric curve given by
10. Find the equation of the ellipse with foci 18, 0 2 x  3t and y  t 2  1, for 2  t  2.
and 18, 02 and with vertices 110, 02 and 110, 0 2.
20. Let x  t  3 and y  t  1.
2 3
11. Find the equation of the ellipse centered at 16, 32
with one focus at 10, 32 and the minor axis 4 units a. Graph the curve for 4  t  4.
long. b. Eliminate t and write an expression for the curve
in x and y.
12. Find the equation of the hyperbola centered at c. At what value of x is y  0?
12, 32 with one focus at 12, 72 and the correspon-
ding vertex at 12, 62.
gord.3896.09.pgs 4/24/03 10:16 AM Page 647

Review Problems 647

21. Sketch the curve a. 13, p>3 2

x  1  log t and y  log t, for 1  t  10. b. 13, p>4 2
c. 14, 3p>2 2
d. 14, 5p>4 2
a. Eliminate the parameter to find the expression
e. 15, 5p>6 2
for y as a function of x.
f. 15, 2 2
b. What is the largest possible domain for this
function?
27. Using polar coordinates, sketch the curve
22. Graph the equations
1 1
x  sin 2t and y  cos t, r .
2 1  cos u
for 2p  t  2p.
Convert the polar expression to rectangular coordi-
23. Compare the graph in Problem 22 to the graphs of nates and find the equation of the conic.
a. x  sin 4t, y  cos 2t; 28. The polar equation of a well-known family of
b. x  sin 6t, y  cos 2t; curves is
c. x  sin 6t, y  cos t. 1
d. Determine the period of each graph in parts r .
2
(a)–(c). cos u sin2u
 2
24. Use appropriate trigonometric identities to elimi- B a2 b
nate t and write the following expressions in terms What are these curves?
of x and y: In Problems 29–32, compare the graphs of each set of
x  cos 2t and y  sin t. equations.
25. Transform each point from rectangular coordinates 29. r  cos u, r  cos 2u, and r  cos 4u.
1x, y2 to an equivalent point in polar coordinates. 30. r  cos u, r  cos 3u, and r  cos 5u.
a. 13, 32 31. r  sin u, r  sin 2u, and r  sin 4u.
b. 11, 3 2 32. r  sin u, r  sin 3u, and r  sin 5u.
c. 14, 1 2
d. 10, 62.
26. Transform each point from polar coordinates 1r, u 2
to rectangular coordinates.
gord.3896.09.pgs 4/24/03 10:16 AM Page 648
Gord.3896.10.pgs 4/28/03 3:29 PM Page 649

10
Matrix Algebra and
Its Applications
10.1 Geometric Vectors
In the early 1800s, physicists found that most physical quantities could be catego-
rized in one of two ways: those that have only size, such as length, time, or mass,
and those that have both size and direction, such as force, velocity, or acceleration.
Quantities that have only size, or magnitude, are called scalars; those that have
both magnitude and direction are called vectors. We use lightface, italic lowercase
letters, such as a, m, or x to denote scalars; we use boldface, roman lowercase letters
such as b, v, or x to denote vectors.
Although vectors may represent physical (as well as other) quantities, we will
think of them geometrically in this section. In two dimensions, we visualize a vec-
tor as an arrow connecting two points, as shown in Figure 10.1. The length of the
arrow represents the magnitude of the vector. The slope of the line through any
two points on the arrow, along with the arrowhead, gives the direction of the vec-
tor v. In three dimensions, we likewise visualize a vector as an arrow connecting
two points, as shown in Figure 10.2.

y
z

v Q
e
ud
nit
ag
M

v
θ
O y
x P
O x

FIGURE 10.1 FIGURE 10.2

Any vector starting at the origin is known as a position vector because it gives the
position of the arrowhead with respect to the origin. A vector connecting two points
P and Q, sometimes written v  PQ, is called a displacement vector; it indicates how

649
Gord.3896.10.pgs 4/28/03 3:29 PM Page 650

650 CHAPTER 10 Matrix Algebra and Its Applications

to get from P to Q by moving a given distance from P in the desired direction. Ob-
viously, a position vector is also a displacement vector, indicating how to move from
the origin to point Q. But a displacement vector is not a position vector if it starts at
any point P other than the origin.
If we know the coordinates of the initial point and the final point of the arrow,
we can write the vector simply. First, we consider the position vector v from the
origin to the point 13, 42 shown in Figure 10.3. It involves moving 3 units to the
right and 4 units upward from the origin, and we write the vector v as either a
row vector v  33 4 4 or as a column vector v  B R . The numerical entries 3 and
3
4
4 in the vector are called its components. The decision to write the vector as a row vec-
tor or as a column vector is usually a matter of choice, so long as you are consistent.
We cover several specific cases later in this chapter in which the choice of column
vectors is essential; in this section we primarily use row vectors for convenience.
The magnitude of the vector v  33 4 4 is the length of the arrow, as shown in
Figure 10.3. It is the distance from the origin to the point 13, 4 2, and so is 5, using
the Pythagorean theorem.
Next, we consider the displacement vector w from the point 16, 15 2 to the
point 111, 3 2, as shown in Figure 10.4. It involves a move of 5 111  62 to the
right and a move of 12 13  152 vertically. We therefore write this vector

either as the row vector w  3 5 12 4 or as the column vector w  B

5
R . Note
12
that, in a displacement vector, the components are the differences in the coordinates
of the points defining the vector. The magnitude of the vector w  3 5 124
equals the distance from one point to the other, or
Magnitude  2111  6 2 2  13  152 2  252  112 2 2
 225  144  2169  13.

y
y (6, 15)
(3, 4)

v= 3 4 13
–12

5 4
w = 5 –12

(11, 3)
5

FIGURE 10.3 x FIGURE 10.4 x

O 3 O

We write the magnitude of a vector v as 7 v7 . In general, in two dimensions, we

have the following.

If v  3 a b4 , then
7 v7  2a2  b2 .
Gord.3896.10.pgs 4/28/03 3:29 PM Page 651

10.1 Geometric Vectors 651

For example, if v  37 4 4 , then 7 v7  272  14 2 2  265 .

Similarly, in three dimensions we can write a vector v in terms of three compo-
nents a, b, and c as v  3a b c4. The magnitude of such a vector is defined anal-
ogously with the Pythagorean theorem.

If v  3a b c4 , then
7 v7  2a2  b2  c2 .

However, specifying the direction of a vector in space is considerably harder than

in the plane, and we don’t go into it here.

E XAMPLE 1
Find the magnitude of the vector from the point 11, 2, 42 to the point 13, 3, 8 2.
Solution We first write this vector in terms of its components, which are the differences in
each of the three coordinates. Therefore v  33  1 3  2 8  4 4  32 5 4 4,
and its magnitude is
7 v7  222  15 2 2  42  24  25  16  245 .
◆
We say that two vectors v and w are equal, written v  w, if all their corre-
sponding components are equal. For instance, 33 4 7 4  33 116 7 4, but

1 1
C8S  C0S
2 2 .
Geometrically, two vectors are equal if they have the same magnitude and the same
direction. Figure 10.5 shows that v  y (they have the same magnitude and the
same direction); but v  x (they are parallel and have the same direction, but have
different magnitudes) and v  w (they have the same magnitude, but do not have
the same direction because they are not parallel).

y
y
w

x
v

FIGURE 10.5 x
O
Gord.3896.10.pgs 4/28/03 3:29 PM Page 652

652 CHAPTER 10 Matrix Algebra and Its Applications

A Constant Multiple of a Vector

If the vector v  33 4 4, it seems reasonable to assume that two times v is just
2v  2 . 3 3 4 4  3 6 8 4 .
Does this make sense geometrically? Figure 10.6 shows v as the displacement vec-
tor from an arbitrary point 1x, y2 to the point 1x  3, y  4 2. Note that the mag-
nitude of v is 5. We also show the vector w  36 8 4 starting from the same point
1x, y2; it extends 6 units to the right and 8 units up, so it ends at the point
1x  6, y  8 2. From the Pythagorean theorem, the magnitude of w is
7 w 7  262  82  2100  10.

y
(x + 6, y + 8)
w = 2v

10
(x + 3, y + 4)
v= 3 4 8

5 4

3
(x, y)
6
x
FIGURE 10.6 O

Thus the magnitude of w is twice the magnitude of v. So multiplying v by 2 pro-

duces a vector w  2v that is twice the length of v, but in the same direction.
In general, for any vector v  3a b4 and any scalar multiple m,
m . v  3ma mb4
is a vector m times as long as v (shorter if 0  m  1) that points in the same di-
rection if m  0. If the multiple m  0, the resulting vector is parallel to v, but it
points in the opposite direction.
Moreover, this definition of the multiple of a vector suggests the following im-
portant and useful fact.

Two vectors v and w are parallel if and only if one is a multiple of the other.

Unit Vectors
EXAMPLE 2
Find a vector u of length 1 that is in the same direction as the vector v  36 8 4.
Solution Because the vector u we want to find is in the same direction as v, it will be
parallel to v and so must be some multiple of v, u  m . v, as shown in Figure 10.7. The
problem is to find the appropriate multiple m. The magnitude of v is
7 v7  216 2 2  82  236  64  2100  10.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 653

10.1 Geometric Vectors 653

v = –6 8

8 10

x
FIGURE 10.7 –6 O

The vector u we seek is to have length 1, so it must be one-tenth of v. That is,

u a bv  c d.
1 6 8
10 10 10

Any vector whose length is 1 is called a unit vector. In general, if v is any

nonzero vector, a unit vector u in the same direction as v is
v
u
7 v7
.

In two dimensions, the two most important unit vectors are the coordinate vec-
tors along the horizontal and vertical axes. The unit coordinate vector pointing to
the right is denoted by i  3 1 0 4 , and the unit coordinate vector pointing up-
ward is denoted by j  30 1 4.

The Sum of Two Vectors

We add vectors—whether they are two row vectors or two column vectors—by
adding the corresponding components. For instance, if v  34 9 4 and
w  37 34 are row vectors, their sum is the row vector
v  w  34 9 4  3 7 3 4  3 4  7 9  3 4  311 6 4.

In general, if v  3v1 v2 4 and w  3 w1 w2 4 are any two row vectors,

v  w  3 v1  w1 v2  w2 4 .

Geometrically, adding vectors involves “adding” the arrows, which can be

thought of in two ways. First, in Figure 10.8 vector w is “moved” so that it starts at
the end of vector v and still points in the same direction. (Equivalently, vector w is
replaced by an equal vector that starts at the end of vector v.) Then v  w is the
vector from the start of v to the end of w. The sum of the two vectors is the third
side of the triangle formed by the vector v and the shifted vector w.
Alternatively, in Figure 10.9, v and w form adjacent sides of a parallelogram. The
upper side has the same length as v and is parallel to v; therefore it equals v. Similar-
ly, the right side of the parallelogram has the same magnitude as w and is parallel to
w, so it equals w. The sum v  w then is the long diagonal in the parallelogram.
Gord.3896.10.pgs 11/21/03 11:05 AM Page 654

654 CHAPTER 10 Matrix Algebra and Its Applications

y y

v
v+w
w w w w
v+w

v v
x x
O O

FIGURE 10.8 FIGURE 10.9

The Coordinate Vectors i and j

One of the advantages of the coordinate vectors i  31 0 4 and j  3 0 1 4 is that
any vector in the plane v  3a b4 can be written in terms of i and j. In particular,
as shown in Figure 10.10, the vector v can be thought of as the sum of a horizontal
vector with magnitude a and a vertical vector with magnitude b. We write the hor-
izontal vector with magnitude a as ai and the vertical vector with magnitude b as
bj. Consequently, v  3 a b4  ai  bj.

v= a b
= ai + bj bj

ai
x
FIGURE 10.10 O

All operations with vectors, such as addition, can be done in terms of i and j.

E XAMPLE 3
Given the vectors v  3i  5j and w  7i  4j, find (a) their sum and (b) 4 times the
first vector.
Solution
a. The sum of the two vectors is v  w  13i  5j2  17i  4j2  10i  j.
b. 4v  413i  5j 2  12i  20j.
◆
Applications of Vectors
We next look at several examples involving physical situations that use vector
addition.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 655

10.1 Geometric Vectors 655

E XAMPLE 4
Tom is trying to open a window that is stuck. He exerts a force of 30 lb at an angle of 20°
with the wall. What is the effective vertical force that he exerts upward against the window?

Solution We start by drawing a sketch of the situation, as shown in Figure 10.11. The
force that Tom exerts is a vector F whose magnitude is 30 and whose direction is at a 20°
angle with the wall. The force actually consists of two components—one vertically up-
ward 1Fy 2 , which represents the effective force that he exerts to raise the window, and the
other 1Fx 2 perpendicular to the window, which doesn’t have any effect on moving the
window vertically. Thus the total force F is just the sum of the two vectors Fx and Fy , and
what we seek is the vertical vector Fy .

20°

30 pounds Fy

70°
FIGURE 10.11 Fx

The angle at the upper vertex of the triangle is 20°, so the angle at the starting point
of vector F is 70°. The length of the hypotenuse of the triangle is just the magnitude of
the force vector, or 7 F7  30. Using trigonometry, we have
opposite
sin 70°  ,
hypotenuse

where the length of the hypotenuse is 7 F7  30. Therefore the length of the vertical side
of the triangle is 7 F7 sin 70°  30 sin 70°  28.19, and the corresponding vertical vector
is Fy  3 0 28.194 . Consequently, the effective force that Tom exerts upward to move
the window actually is 7 Fy 7  30 sin 70°  28.19 pounds.
◆

E XAMPLE 5
A flock of Canadian geese is trying to fly due south for the winter with a constant veloc-
ity of 12 mph. A stiff wind is blowing at a constant rate of 20 mph from a direction 35°
west of north. Find the actual direction that the geese end up flying and their actual
speed with respect to the ground.
Solution We begin with a sketch of the situation, as shown in Figure 10.12. Each goose
is trying to fly due south, so there is one velocity vector, g, for the goose having magnitude
12 and pointing vertically downward. In addition, each goose is pushed by the wind,
which is coming from a northwesterly direction. The wind is represented by a second ve-
locity vector, w, having magnitude 20 and pointing from a direction 35° west of north.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 656

656 CHAPTER 10 Matrix Algebra and Its Applications

20 sin 35°
20 mph

35° 20 cos 35°

θ
12 mph
g+w
FIGURE 10.12 g

The actual velocity vector for the goose is the sum of these two vectors. To find the
direction that a goose actually flies—and then the speed at which it flies—we need to
find the components of g  w. Because the goose is trying to fly due south, g has only a
vertical component of 12, so g  30 12 4 . The velocity vector for the wind has both
a horizontal and a vertical component. Using trigonometry, we see that the horizontal
component of w is 20 sin 35°  11.47. Similarly, the vertical component of w is
20 cos 35°  16.38 (it is negative because it is directed downward). Consequently,
the wind vector w is

w  311.47 16.38 4.

Thus the sum of the two vectors is

g  w  30 12 4  311.47 16.384  3 11.47 28.384.

The actual speed with which the goose flies is the magnitude of this vector g  w, which is

Speed  2111.472 2  128.382 2  2131.56  805.42  2936.90  30.61.

Next, to find the direction in which the goose flies, we need to find the angle u that
the vector g  w makes with the vertical. From the large right trangle in Figure 10.12,
we find

20 sin 35°
tan u 
20 cos 35°  12
11.47
  0.404
28.38
so that
u  arctan 0.404  22°.
Thus the geese actually end up flying in a direction 22° east of south instead of due south.
◆

The Difference of Two Vectors

We define the difference of two vectors v  3v1 v2 4 and w  3 w1 w2 4 to be
v  w  3 v1  w1 v2  w2 4 ;
that is, we simply take the difference of corresponding components.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 657

10.1 Geometric Vectors 657

For instance, if v  3 14 3 4 and w  3 6 10 4, then v  w  38 7 4. Alter-

natively, in terms of the unit vectors i and j, v  14i  3j and w  6i  10j, so that

v  w  114i  3j2  16i  10j2  8i  7j.

Now let’s interpret the difference of two vectors geometrically. Consider the
vector x in Figure 10.13, which connects the end of v to the end of w. We know
from the sum of two vectors that

v  x  w so that x  w  v.

w x

v
x
FIGURE 10.13 O

In general, the difference of two vectors always connects the end of the second vec-
tor to the end of the first. The only question is: Which way does the difference vec-
tor point? The easiest way to decide that is to draw a sketch such as Figure 10.13.
Another way of looking at the difference v  w is to think of it as
v  w  v  1w 2 , where
 w  11 2 3w1 w2 4  3 w1 w2 4,
a vector with the same length as w but pointing in the opposite direction.

Problems
1. Plot each position vector as an arrow in the xy- c. Adding t to s.
plane from the origin to the point having the ap- d. Adding u to v.
propriate coordinates. 3. Using the vectors in Problem 1, plot each vector in
4 2 a–c on the same graph.
a. r  B R b. s  B R a. The result of adding one-half of r to one-half
0 4
2 3 of s.
c. t  B R d. u  B R b. The result of adding one-quarter of r to three-
4 4
quarters of s.
1 c. The result of adding three-quarters of r to one-
e. v  B R
2 quarter of s.
2. Using the vectors in Problem 1, plot the result of: d. Plot r and s. Draw a straight line joining these
a. Adding vector u to vector r. two vectors. Which of the vectors drawn in parts
b. Adding vector r to u. Compare your answer to (a), (b), and (c) lie on this line?
the vector obtained in part (a).
Gord.3896.10.pgs 11/21/03 11:05 AM Page 658

658 CHAPTER 10 Matrix Algebra and its Applications

4. Determine the magnitude of the position vectors c. Repeat part (a) with a force of 40 lb at an angle
from the origin to the following points. of 20°.
a. 13, 42 b. 112, 52
c. 13, 22 d. 17, 3 2
e. 11, 2, 22 f. 12, 3, 42
5. Determine the magnitude of the displacement vec-
tor from point A to point B for each pair of points.
a. A  11, 22, B  15, 52
b. A  12, 12, B  14, 12
c. A  11, 32, B  13, 42
30 pounds

d. A  11, 2, 32, B  14, 5, 32

e. A  11, 6, 32, B  13, 1, 22 25°

6. Determine the vector that is the given multiple of

the vector 31 24 . 13. a. If a jet plane is flying on a heading of due east at
a. 2 b. 7 600 mph and the wind is blowing due south at
c. 2 d. 0 100 mph, what are the actual direction and
7. Determine the vector that is the given multiple of speed of the plane?
the vector 33 1 24. b. Repeat part (a) if the plane is flying due east at
a. 3 b. 10 600 mph and the wind is blowing in a direction
c. 7 d. 12 that is 40° south of east at 100 mph.
c. Repeat part (a) if the plane is flying southwest (45°
8. Find the unit vector that points in the same direc-
south of west) at 300 mph and the wind is blowing
tion as the given vector.
in a direction 50° south of west at 100 mph.
a. 3 3 44 b. 3 1 14
c. 3 0 54 d. 31 14
e. 3 1 2 24 f. 31 1 1 4 600 mph
100 mph
9. Express each vector as a sum of multiples of the co-
ordinate vectors i  3 1 04 and j  30 1 4.
N

a. 3 2 14 b. 3 1 34
W E

c. 3 2 2 4
1 1
d. 3 3 04 S

10. Express each vector as a sum of multiples of the three

coordinate vectors i  31 0 04 , j  3 0 1 0 4,
14. Suppose that a boat is moving at 10 mph in the di-
and k  3 0 0 14.
rection of 20° north of east across a bay and the tide
is moving the water in the bay at 4 mph in the di-
a. 31 2 34 b. 32 0 34 rection of 40° west of south. What are the actual di-
c. 3 1 3 14 d. 3 0 12 04 rection and speed of the boat? (Hint: Express both
11. Refer to Example 4 in the text. What is the upward the boat’s vector and the tide’s vector in terms of
force on the window if Tom exerts the two coordinate vectors i and j).
a. a force of 30 lb at an angle of 25°?
b. a force of 20 lb at an angle of 30°?
c. a force of 40 lb at an angle of 15°? 40°
12. a. A sliding door is difficult to open. If Claire exerts
20°
a horizontal force of 30 lb at an angle of 25° to
the sliding door, what is the effective force on the N
door in the direction in which the door slides?
W E
b. Repeat part (a) with a force of 25 lb at an angle
of 40°. S
Gord.3896.10.pgs 4/28/03 3:29 PM Page 659

10.2 Linear Models 659

10.2 Linear Models

The real-world problems to which people apply mathematical models often involve
large and very complex situations. For instance, one might want to analyze the effect
that imposing a 50¢ per gallon tax on gasoline would have on the national economy
with its thousands of interdependent businesses and industries. An airline must have
a reservations system that takes into account all its aircraft, the cities it serves, flight
schedules, dates, and different fare structures in effect. A company may need a battery
of tests that can predict how well applicants will perform at a given job. The data in
such problems usually come in the form of a rectangular array of numbers, called a
matrix. For example, if four students, Ann, Bob, Carol, and Dan, take exams in
French, mathematics, and sociology, the set of exam results could be displayed in the
matrix
Ann Bob Carol Dan
French 84 73 82 85
Mathematics C 88 78 94 92 S
Sociology 76 81 83 78
Thus, for instance, Bob received an 81 in sociology and Carol a 94 in math.
Matrix algebra provides a systematic way of working with such arrays of num-
bers. In this chapter, we develop the basic language and methods of matrix algebra
that will allow you to use matrices to solve various problems involving systems of
linear equations.
We refer to a rectangular array of numbers as an m  n matrix when it has m
rows horizontally and it has n columns vertically. A 3  4 matrix thus has 3 hori-
zontal rows and 4 vertical columns, as in the preceding matrix of exam scores. We use
boldface capital letters, such as A, to denote matrices in print. (When writing matri-
ces by hand, you may find it convenient to use a wavy line under the letter, as in A.)
Two more examples of matrices are
5 1 1
1 0 3
A  C1 7 2S and N  B R.
1 7 4
6 5 0
Here A is a 3  3 matrix and N is a 2  3 matrix because it has two rows across
and three columns vertically.
We denote the entry in row i and column j of matrix A by aij . Thus in matrix A,
a13  1 because 1 is the entry in the first row and the third column, whereas
a31  6 because 6 is the entry in the third row and first column. Similarly, in ma-
trix N, n23  4 because 4 is the entry in the second row and the third column.
When a matrix has only one row or only one column, we call it a row vector or
a column vector, respectively, or simply a vector. A vector having three numbers is
called a 3-vector, whereas a vector consisting of n numbers is called an n-vector. As
noted in Section 10.1, we use boldface lowercase letters, such as b or x, to denote
vectors. Note that any vector is also a matrix. Some examples of vectors are
2
b  31 3 0 5 4 and x  C 4 S .
1
Gord.3896.10.pgs 4/28/03 3:29 PM Page 660

660 CHAPTER 10 Matrix Algebra and Its Applications

Here b is a row 4-vector and x is a column 3-vector. We write b1 for the first entry
in vector b, b2 for the second entry in b, and bi for the ith entry in b. Thus for the
vectors b and x, we have b2  3 and x2  4.
Recall from Section 10.1 that two vectors v and w are equal, written v  w, if
all their corresponding entries, or components, are equal. Similarly, two matrices
are equal if all their corresponding components are the same.
Any list of numbers can be thought of as a column vector or a row vector.
Whether we choose a column or row format if only vectors are involved usually
doesn’t matter, but when a vector and a matrix are multiplied, it is important to
distinguish clearly whether the vector is a row vector or a column vector. For rea-
sons that will be clear shortly, we usually treat most vectors as column vectors. Note
that a column n-vector is an n  1 matrix and a row n-vector is a 1  n matrix.
An m  n matrix A can be thought of as a set of n column m-vectors or as a set
of m row n-vectors. In the case of students and their test results, each column vec-
tor of the matrix gives the scores for one student in all these courses, whereas each
row vector gives the scores in one course for all these students.
We use the following notation to refer to rows and columns in a matrix:
a j denotes the jth column vector in A; and
a i œ denotes the ith row vector in A.
For instance, in the matrix

5 1 1 1
A  C1 7 2S , a2  C7S and a 1œ  35 1 1 4.
6 5 0 5

A Geometric View of Vectors

In Section 10.1, we presented vectors geometrically as positions and displacements
in coordinate space. As we pointed out there, vectors can be used to represent
points in space. In two-dimensional space, we use a 2-vector; in three-dimensional
space, we use a 3-vector. The point 15, 2 2 in the plane can be thought of as the
2-vector B R . Similarly, the point 13, 2, 72 in three-dimensional space with coordi-
5
2
nates x  3, y  2, z  7, or equivalently, x1  3, x2  2, x3  7, can be written
as the 3-vector
3
C2S .
7
Thus the coordinates of a point become the components of a position vector.
We next consider how matrices in general and vectors in particular occur in
applied problems from many different fields.

A Clothes Production Model

A textile company runs three clothing factories. Each factory produces three types
of women’s clothing: vests, pants, and coats. For simplicity, we assume that one size
fits all. Suppose that the first factory produces 20 vests, 10 pants, and 5 coats from
Gord.3896.10.pgs 4/28/03 3:29 PM Page 661

10.2 Linear Models 661

each roll of cloth. The second and third factories produce different amounts of
these three products, as described in matrix A.
Factory 1 Factory 2 Factory 3
Vests 20 4 4
Pants C 10 14 5S  A
Coats 5 5 12
Each column of A is a vector of clothing produced by a factory from one roll of
cloth. For instance, Factory 3 has the output vector
4
a3  C 5S ,
12
which indicates that it makes 4 vests, 5 pants, and 12 coats from each roll. Each row
of A is a vector of factory production of one particular type of clothing from one
roll of cloth. The row vector for coats is a 3œ  35 5 12 4, which indicates that
Factory 1 produces 5 coats, Factory 2 produces 5 coats, and Factory 3 produces 12
coats from each roll.
Let x1 denote the number of rolls of cloth used by the first factory; similarly, x2
and x3 denote the numbers of rolls used by the second and third factories, respective-
ly. Suppose that the company gets an order for 500 vests, 850 pants, and 1000 coats.
This triple of numbers is called the demand, which we write as a column vector
500
C 850 S .
1000
Then x1 , x2 , and x3 need to satisfy the system of linear equations
vests: 20x1  4x2  4x3  500
pants: 10x1  14x2  5x3  850
coats: 5x1  5x2  12x3  1000.
In words, the vests equation says: The number of vests produced by Factory 1, 20x1
(this expression is 20 vests per roll times the x1 rolls used by Factory 1), plus the
number of vests produced by Factory 2, which is 4x2 , plus the number of vests pro-
duced by Factory 3, which is 4x3 , must equal the demand of 500 vests.
As we demonstrate in Section 10.3, we can write this system of linear equations
as the matrix–vector equation
20 4 4 x1 500
C 10 14 5 S C x2 S  C 850 S .
5 5 12 x3 1000
We can also write this as a single vector equation in the column vectors of the ma-
trix as

20 4 4 500
x1 C 10 S  x2 C 14 S  x3 C 5 S  C 850 S .
5 5 12 1000
Gord.3896.10.pgs 4/28/03 3:29 PM Page 662

662 CHAPTER 10 Matrix Algebra and Its Applications

If b is the demand vector on the right side of the matrix–vector equation and x
is a (column) vector of the xi’s, matrix algebra gives us a way to write the system of
linear equations concisely in terms of A, b, and x as Ax  b. We discuss how to do
this in Section 10.3. In Section 10.5, we extend these ideas to solve any system of
three equations in three variables by using matrix algebra techniques.
The expression 2x1  4x2  12 x3 is a linear expression in three variables; it in-
volves only the first power of the variables x1 , x2 , and x3 . Other cases are the expres-
sions on the left side of the preceding system of linear equations. More formally, a
linear expression is one that involves a sum of terms made up of constants multiply-
ing individual variables that are raised only to the first power. In contrast, a nonlinear
expression involves one or more variables that are raised to various powers (different
from 1), or exponential, logarithmic, trigonometric, or other more complex expres-
sions. (This terminology is similar to that used to describe linear difference equa-
tions.) The term linear is used to indicate that a “line-like” graph is associated with
each variable in the expression. For example, the vest expression 20x1  4x2  4x3 is
a linear expression; if x2 and x3 are fixed—say, x2  x3  3—with only x1 remaining
as a free variable, the resulting expression 20x1  4132  413 2, or 20x1  24, de-
fines a function y  20x1  24 whose graph is a line.
The clothing production equations form what is called a linear model because
the equations involve only linear expressions (linear equations). In the following
sections, we will return to this and other models introduced here as we develop the
mathematical methods needed to analyze linear models.

A Markov Chain Model for the Stock Market

We next develop a linear model for the behavior of the stock market. Here we show
how matrix methods can be used to represent a situation in which the values of a
number of variables at one stage of a process are related to their values at the pre-
ceding stage.
Each business day, the stock market goes up, goes down, or stays the same.
Suppose that historical studies show that if the market goes up one day—say
today—the probability is 41 that it will go up tomorrow, the probability is 21 that it
will go down tomorrow, and the probability is 14 that it will stay the same tomor-
row. If the market goes down today, there are three other observed probabilities for
tomorrow’s market performance. Similarly, if the market stays the same today,
there is a third set of three probabilities for what will happen tomorrow. We can
conveniently display all nine of these probabilities in a matrix A:

Market Today
Up Down Same
1 1 1
Up 4 2 4
Market
Down C 12 1
4 2S
1
 A.
Tomorrow
Same 14 1
4
1
4

The probabilities in this matrix are called transition probabilities because they give
us information about how to relate one stage of a process to the next. The matrix A
is called a transition matrix. Each column corresponds to a type of market move-
ment today, and each row corresponds to a type of market movement tomorrow.
The matrix entry a23 , which equals 12 , is in the “down tomorrow” row and in the
Gord.3896.10.pgs 4/28/03 3:29 PM Page 663

10.2 Linear Models 663

“same today” column. The value 12 represents the probability that the market will
go down tomorrow given that it stays the same today.
Note that the probabilities in each column of the transition matrix must add to
1 because they include all possible outcomes for tomorrow given a particular type
of market behavior today. A mathematical model such as this with given transition
probabilities is known as a Markov process, or Markov chain (named after Russ-
ian mathematician Andrei Markov, who first developed these ideas). A convenient
way to display the information in a Markov chain is with a transition diagram, such
as the one shown in Figure 10.14. In this diagram, there are three nodes, one for
each type of market movement: go up (U), go down (D), or stay the same (S).
These are the possible states for the system. Note also that we indicate each transi-
tion probability with an arrow.

1
4

1 1
2 1 1 2
2 4

1
4
U S
1
4
1 1
FIGURE 10.14 4 4

E XAMPLE 1
Suppose that, before the stock market opens today, we believe that there is a 50–50
chance of the market going down or staying the same, but no chance of its going up. Use
the values in the preceding transition matrix A to compute the probabilities of the mar-
ket being in any of the three states tomorrow—up, down, or the same—based on the
probabilities of the market being up, down, or the same today.
Solution Let p1 , p2 , and p3 denote today’s probabilities of the market being up, down,
and the same, respectively, and let p1, p2, and p3 denote tomorrow’s probabilities of
being up, down, and the same, respectively. Let’s see how to compute p1. First, to com-
pute the probability of two successive events—such as (i) being in State 1 today (probabil-
ity p1) followed by (ii) switching from State 1 today to State 1 tomorrow (probability 14)—we
multiply the probabilities of the two events and get 41 p1 . Similarly, the probability of
(i) being in State 2 today (probability p2) followed by (ii) switching from State 2 to State 1
tomorrow (probability 12 ) is the product 12 p2 . Also, the probability of (i) being in State 3
Gord.3896.10.pgs 4/28/03 3:29 PM Page 664

664 CHAPTER 10 Matrix Algebra and Its Applications

today (probability p3) followed by (ii) switching from State 3 to State 1 tomorrow
(probability 14 )is 41 p3 .
To get the total probability p1 that the stock market will go up tomorrow, we add
these three values to get
1 1 1
p1  p1  p2  p3 .
4 2 4
In the same way, we calculate the probabilities p2 and p3 that the stock market goes
down or stays the same tomorrow and so obtain a set of three equations based on the
transition matrix A:
1 1 1
p1  p1  p2  p3
4 2 4
1 1 1
p2  p1  p2  p3 (1)
2 4 2
1 1 1
p3  p1  p2  p3 .
4 4 4
Note that the coefficients in this system of linear equations come directly from the en-
tries in the transition matrix A.
Believing that p1  0, p2  12 , and p3  12 are today’s probabilities of the market
being up, down, and the same, respectively, we calculate the probability p1 that the mar-
ket goes up tomorrow, using the first of Equations (1), as follows:
1 1 1 1 1 1 1 1 3
p1  p1  p2  p3  . 0  .  .  .
4 2 4 4 2 2 4 2 8
In the same way, using Equations (1), we obtain the probabilities for the other two mar-
ket outcomes tomorrow. Thus tomorrow’s probabilities p1 , p2 , and p3 are
1 1 1 1 1.1 1 1 3
p1  p1  p2  p3  . 0   .  ,
4 2 4 4 2 2 4 2 8
1 1 1 1 1.1 1 1 3
p2  p1  p2  p3  . 0   .  ,
2 4 2 4 4 2 2 2 8
1 1 1 1 1.1 1 1 1
p3  p1  p2  p3  . 0   .  .
4 4 4 4 4 2 4 2 4

◆
E XAMPLE 2
Use the equations for p1 , p2 , and p3 to predict the market probabilities p1 , p2 , and
p3 two days ahead. Then predict the market probabilities farther into the future.
Solution We repeat the process in Example 1, using p1  38 , p2  38 , and p3  14 to
obtain
1 1 1 1 3 1 3 1 1 11
p1  p1  p2  p3  .  .  .  ,
4 2 4 4 8 2 8 4 4 32
1 1 1 1 3 1 3 1 1 13
p2  p1  p2  p3  .  .  .  ,
2 4 2 2 8 4 8 2 4 32
1 1 1 1 3 1 3 1 1 8 1
p3  p1  p2  p3  .  .  .   .
4 4 4 4 8 4 8 4 4 32 4
Gord.3896.10.pgs 4/28/03 3:29 PM Page 665

10.2 Linear Models 665

From these probabilities for 2 days hence, we can predict the market 3 days ahead,
and so on indefinitely, so long as the probabilities of the market going up, going down,
or staying the same continue to hold. In Section 10.4, we introduce a far simpler way to
perform these calculations based on matrix algebra. For now, we simply indicate the re-
sults in the following table, assuming that today’s market probabilities are 0 for going up,
1 1
2 for going down, and 2 for staying the same.

Up Down Same
1 1
Today 0
2 2
3 3 1
Tomorrow
8 8 4
11 13 1
2 days ahead
32 32 4
45 51 1
3 days ahead
128 128 4
5 days ahead 0.35 0.40 0.25

10 days ahead 0.35 0.40 0.25

100 days ahead 0.35 0.40 0.25

◆
Each triple of probabilities for any day—say p1 , p2 , p3 for the first day— can be
thought of as the components of a 3-vector of probabilities for that day. Also, note
that on any given day, the sum of the probabilities is always 1 because one of the
possibilities (up, down, or the same) must occur.
The sequence of all the successive vectors p, p, p, p, . . . associated with
any transition matrix A is called a Markov chain because the successive vectors are
linked by the matrix A. Eventually, the successive probabilities in this Markov chain
stabilize at 0.35 for the market going up, 0.40 for the market going down, and 0.25
for the market staying the same. That is, the probabilities converge over time to
these limiting values. This behavior occurs regardless of the initial values we used
for today’s probabilities. Later, we formulate a system of three linear equations in
three variables and solve it to determine these stable probabilities directly.

A Population Growth Model

The following model relates populations of hares and wolves from one week to the
next. To make the numbers work out conveniently, we measure the hare population
in groups of 10 hares and the wolf population in single wolves. Suppose that the
number of groups of hares H grows by 20% per week when no wolves are present,
so the population of hares next week H would be 1.2H. But W wolves are present
and the wolves eat the hares at the rate of each wolf eating 3 hares each week, which
is 30%  0.3 of a group of 10 hares. Thus the hare population is reduced by 0.3W
per week, giving H  1.2H  0.3W as next week’s hare population.
Next, without hares present, the wolf population decreases at a rate of 30%
per week, so the wolf population next week W would be 0.7W. But the wolf
population grows each week when hares are present at the rate of one wolf for
Gord.3896.10.pgs 4/28/03 3:29 PM Page 666

666 CHAPTER 10 Matrix Algebra and Its Applications

every 50 hares or every 5 groups of 10 hares, which is 15  0.2 wolf per group of 10
hares. Hence W  0.2H  0.7W.
Together, these two equations are our model for the hare and wolf population
over time:
H  1.2H  0.3W
W  0.2H  0.7W.

E XAMPLE 3
Suppose that we start with 1000 groups of hares and 800 wolves. Use the preceding ex-
pressions for H and W to calculate the populations of hares and wolves over time.
Solution If we start with H  1000 and W  800, the hare–wolf model predicts that
H   1.2110002  0.318002  960
W   0.2110002  0.718002  760
as the populations after the first week. We now use these values to predict the two popu-
lations after the second week:
H   1.2H   0.3W   1.219602  0.31760 2  924
W   0.2H   0.7W   0.219602  0.71760 2  724.
Extending these values from one week to the next, we obtain the following table for the
sizes of the hare and wolf populations over time.

Weeks Groups of hares Wolves

0 1000 800
H, W
1 960 760
Number of each species

1000
2 924 724 800
Hares
3 892 692 600
Wolves
400
10 739 539
200
20 649 449 n

50 602 402 Months

100 600 400 FIGURE 10.15

Note that over time the populations converge to 600 groups of 10 hares and 400 wolves.
Figure 10.15 shows the graphs of both populations as functions of time.
◆
We can visualize this situation another way: We can think one population de-
pends on the other. That is, the number of wolves W can be viewed as a function of
the number of hares H. If we plot the number of wolves versus groups of hares, we
find that they fall in a straight line. In particular, the linear function that fits the points
in the preceding table is W  H  200. The graph in Figure 10.16 shows several traj-
ectories for the populations (in hundreds) of groups of 10 hares and wolves. One tra-
jectory shown starts from the initial point 110, 8 2 and leads to the point 16, 42.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 667

10.2 Linear Models 667

10
(10, 8)

Wolves (hundreds)
8

6 (9, 6)
(6, 5)
4
(6, 4) (7, 4)
2
(3, 2)
H
0 2 4 6 8 10 12
FIGURE 10.16 Groups of hares (hundreds)

Suppose we start with a different set of initial values for the two populations—
say, H0  700 and W0  400. Then

H  1.21700 2  0.31400 2  720

W  0.21700 2  0.714002  420

and so on. The resulting points all lie on a line starting at the point 1700, 4002 and
converge to the point 1900, 600 2. Similarly, if we start with H0  600 and W0  500,
the resulting points all lie on a line and converge to 1300, 2002, as also shown in Fig-
ure 10.16. In each case, the limiting values for H and W satisfy W  23 H. That is, the
limiting points lie on the line W  23 H, as shown in Figure 10.17. In fact, for any ini-
tial pair of population values 1H0 , W0 2, the points of successive pairs 1Hn , Wn 2 all lie
on some line, and in each case the successive points are converging (as indicated by the
arrows) toward a limiting point on the line W  23 H, as illustrated in Figure 10.18.
Thus under this model, all populations, regardless of the initial values, converge over
time to populations in which the number of wolves is two-thirds the number of
groups of 10 hares.

W 2
Stable Values W = H
3
10 (15, 10)
(10, 8)
W 8

10 6
Wolves (hundreds)

(10, 8) (6, 4)
Wolves (hundreds)

8 2 4
W= H
3
6 (9, 6) 2
(6, 5)
4 H
(7, 4) 2 4 6 8 10 12 14 16
(6, 4)
2
(3, 2)
H
0 2 4 6 8 10 12
Groups of hares (hundreds) Groups of hares (hundreds)

FIGURE 10.17 FIGURE 10.18

Gord.3896.10.pgs 4/28/03 3:29 PM Page 668

668 CHAPTER 10 Matrix Algebra and Its Applications

Note that the pair of equations we used to define this model can be rewritten
as a pair of difference equations:

Hn1  1.2Hn  0.3Wn

Wn1  0.2Hn  0.7Wn .
We present a more detailed analysis of population models based on systems of
difference equations, including a more sophisticated predator–prey model, in
supplementary Section 12.6.

Problems
1. Ted, Carol, and Alice took tests in German, physics, Factory 1 Factory 2 Factory 3
theater, and politics. Ted’s test scores in these sub- Vests 6 4 2
jects were 64, 73, 86, 85; Carol’s scores were 82, 69, Pants C 4 8 4S
77, 91; Alice’s were 82, 84, 81, 83. Construct a matrix Coats 3 2 8
of these test results. Label the columns and rows.
2. For the matrix Suppose that the company has a demand for 400
vests, 800 pants, and 500 coats. Write a system of
1 5 3 equations whose solution would determine pro-
6 1 7 duction levels to yield the desired numbers of
AD T,
6 9 5 vests, pants, and coats. As in the clothes produc-
0 2 8 tion model, let xi be the number of rolls of cloth
processed by the ith factory.
write the following row and column vectors and
5. Three oil refineries (1, 2, and 3) produce the follow-
entries.
ing amounts, in thousands of gallons, of heating oil,
a. a 1œ diesel oil, and gasoline from each shipment of
b. a 3 crude petroleum.
c. a 4
d. a12 Refinery 1 Refinery 2 Refinery 3
e. a23 Heating Oil 8 5 3
f. a41 Diesel Oil C 2 5 5S
3. In the matrix of letters Gasoline 3 7 6
H R B I Suppose that demand is for 6200 thousand gallons
N S O A of heating oil, 4000 thousand gallons of diesel oil,
AD T,
E T Y L and 4700 thousand gallons of gasoline. Write a
M G D I system of equations whose solution would de-
termine production levels to yield the desired
spell the words represented by the following se- amounts of heating oil, diesel oil, and gasoline. Let
quences of entries. xi be the number of shipments processed by the ith
a. a11 a23 a12 a21 refinery.
b. a41 a23 a32 a11 a31 a12 6. The staff dietitian at the California Institute of
c. a31 a24 a12 a34 a33 Trigonometry has to make up a meal with 600 calo-
d. a33 a31 a22 a32 a31 a12 a43 a24 a33 ries, 20 grams of protein, and 200 mg of vitamin C.
4. A clothing company’s three factories (1, 2, and 3) The three food types that the dietitian can choose
produce the following numbers of vests, pants, and from are gelatin, fish sticks, and mystery meat. They
coats from each roll of cloth. have the following nutritional content per unit.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 669

10.2 Linear Models 669

Gelatin Fish Sticks Mystery Meat 11. The Pins, a bowling team, plays in a bowling league
Calories 10 50 200 each week. If they win this week’s game, they have a
2
Protein C 1 3 0.2 S 3 chance of winning next week’s game. If they lose
this week’s game, they have a 21 chance of winning
Vitamin C 30 10 0
next week’s game.
Construct a mathematical model for this situation, a. Construct a Markov chain for this situation; give
based on a system of three linear equations. the matrix of transition probabilities and draw
7. A company has a budget of $280, 000 for comput- the transition diagram.
ing equipment. The types of equipment available b. If there is a 50–50 chance of the Pins’ winning
are microcomputers at $2000 each, terminals at this week’s game, what is their chance of win-
$500 each, and workstations at $5000 each. There ning next week’s game?
should be five times as many terminals as micro- c. If they won this week, what is their chance of win-
computers and twice as many microcomputers as ning the game 2 weeks from now?
workstations. Write a system of three linear equa- 12. Consider a weather Markov chain having two states:
tions to describe this situation. sunny and cloudy. If today is sunny, there is a 34 prob-
8. In the clothes production model in the text, suppose ability that tomorrow will be sunny and a 41 probabil-
that Factory 1 processes 15 rolls of cloth, Factory 2 ity that tomorrow will be cloudy. If today is cloudy,
processes 20 rolls, and Factory 3 processes 60 rolls. there is a 41 probability that tomorrow will be sunny
For which product, vests, pants, or coats, does pro- and a 34 probability that tomorrow will be cloudy.
duction deviate the most from the demand for 600, a. Write the transition matrix for this Markov
800, 1000? chain and draw its transition diagram.
9. Refer to the stock market Markov chain in Example 1. b. In this weather Markov chain, starting with the
Determine the set of probabilities for tomorrow’s 1
vector of probabilities B R (a sunny day), compute
market for each set of probabilities that the market 0
will be up, down, or the same today. and plot the vectors of probabilities for four suc-
a. p1  1, p2  0, p3  0 cessive days.
b. p1  0, p2  12 , p3  12 c. Repeat the process starting with the probability
c. p1  12 , p2  0, p3  12 0
vector B R (a cloudy day). Can you guess the val-
d. p1  14 , p2  12 , p3  14 1
e. p1  0.35, p2  0.40, p3  0.25 ues of the equilibrium state to which your prob-
ability vectors are converging?
10. The copy machine at the student union breaks
down according to the following pattern. If it is 13. The following model for learning a concept over a set
working today, it has a 70% chance of working to- of lessons identifies four states of learning: I  igno-
morrow (and a 30% chance of breaking down). If rance, E  exploratory thinking, S  superficial
the copy machine is broken today, it has a 50% understanding, and M  mastery. If you are now in
chance of working tomorrow (and a 50% chance of state I, after one lesson you have a probability of 21 of
being broken again). still being in I and a probability of 12 of being in E. If
you are now in state E, after one lesson you have a
a. Construct a Markov chain for this situation; give
probability of 41 of being in I, 12 in E, and 41 in S. If you
the matrix of transition probabilities and draw
are now in state S, after one lesson you have a prob-
the transition diagram.
ability of 41 of being in E, 21 in S, and 14 in M. If you are
b. If there is a 50–50 chance of the copy machine’s
in M, you always stay in M (with a probability of 1).
working today, what is the chance of its working
tomorrow? a. Construct a Markov chain for this learning model.
c. Based on the situation in part (b), what is the b. If you start in state I, what is your probability
chance that the copy machine is working the day vector after two lessons? After three lessons?
after tomorrow? 14. a. In Example 3, if initially there were 800 groups
d. If the copy machine is working today, what is the of 10 hares and 300 wolves, how many hares and
chance that it is working the day after tomorrow? wolves would there be after 1 month, 3 months,
Gord.3896.10.pgs 11/21/03 11:05 AM Page 670

670 CHAPTER 10 Matrix Algebra and Its Applications

5 months, and 10 months? What do the limiting a. To find the equation of this limiting line, assume
values for the two populations appear to be? that H  H and W  W in the two equa-
b. Repeat part (a) with an initial population of 600 tions defining the system and solve these two
groups of 10 hares and 800 wolves. equations in two unknowns.
c. Repeat part (a) with an initial population of 900 b. For any given starting populations—say,
groups of 10 hares and 600 wolves. H0  1000 and W0  800—calculate the next
15. Consider the following cattle–sheep models in point 1H, W 2 on the trajectory. What is the
which the two species compete for common graz- equation of the line through 1H0 , W0 2 and
ing land. In each case, compute the populations 1H, W 2?
after 1 month, 2 months, and 3 months if the initial c. You now have the equation of the limiting line
populations are 50 cattle and 100 sheep. and the equation of the trajectory. Describe how
you would use them to find the final population
a. C   1.2C  0.3S
values for H and W. What are the values for initial
S   0.2C  1.2S
population values of H0  1000 and W0  800?
b. C   1.2C  0.1S
S   0.5C  1.4S 18. a. The population models in Example 3 and
Problems 15–17 all involve linear expressions
16. Consider the rabbit–fox model
in H and W. Suppose that the equations for H 
R  1.1R  0.2F and W  contained nonlinear expressions in H
F  0.2R  0.6F. and W. How might such expressions affect the
Plot the following on the same graph. trajectory?
b. We develop a more sophisticated mathematical
a. The trajectory of populations starting from
110, 152.
model for two species, known as the predator–
prey model, in supplementary Section 12.6. It is
b. The trajectory of populations starting from
110, 302.
based on equations such as
c. The trajectory of populations starting from H  1.2H  0.003HW
120, 102. W  0.2W  0.005HW.
17. In Example 3 we found that the solution to the system If the initial population values are H0  200 and
H  1.2H  0.3W W0  100, calculate and plot the population val-
W  0.2H  0.7W ues over the first 3 months. What do you observe
about the trajectory?
converged to a point on the line W  23 H for any
starting values H0 and W0 .

10.3 Scalar Products

In this section, we explain how to multiply two vectors in what is called a scalar
product. For instance, suppose that a family normally eats three vegetables—
asparagus, beans, and corn. Suppose that the costs per pound for these vegetables
are $0.80 for asparagus, $1.00 for beans, and $0.60 for corn. We can then form a vec-
tor p  30.80 1.00 0.604 for the prices of these three vegetables. Suppose further
that the family consumes 2 lb of asparagus, 5 lb of beans, and 3 lb of corn each week,
so we can also form a vector d  32 5 3 4 of the family’s weekly demand for these
vegetables. The total cost of the family’s weekly demand for the vegetables is
2  0.80  5  1.00  3  0.60  1.60  5.00  1.80  $8.40
because the cost for asparagus is 2  $0.80, the cost for beans is 5  $1.00, and the
cost for corn is 3  $0.60. This result suggests a natural way to define the product
Gord.3896.10.pgs 4/28/03 3:29 PM Page 671

10.3 Scalar Products 671

of the two vectors d and p. We write this product as d . p and call it the scalar prod-
uct of d and p. Here the scalar product d . p is
d . p  32 5 3 4 . 3 0.80 1.00 0.604
 210.802  511.00 2  310.60 2
 1.60  5.00  1.80  8.40,
or $8.40. Note that the scalar product involves multiplying the corresponding en-
tries in each position of the vectors and adding the results. Each vector in a scalar
product can be either a row or a column vector, but multiplication makes sense
only when the two vectors have the same size (that is, they have the same number
of entries). The scalar product of two vectors of different sizes can’t be formed be-
cause there would be terms that do not match. For instance, the scalar product of
31 2 3 4 5 4 and 3 20 40 60 4 can’t be formed.
Recall that the word scalar means a single number, as opposed to a vector or
matrix. The scalar product of two vectors is so named because its result is a single
number—a scalar. The scalar product also is known as the dot product and the
inner product, but we will not use either of these terms.
The product of two vectors can also be defined in a different way, known as the
vector product, which produces a vector instead of a scalar, or number, as the result.
However, we don’t consider it here.
More formally we have the following definition of the scalar product.

Scalar Product
Let
a  3 a1 a2 ... an 4 and b  3 b1 b2 ... bn 4

be vectors of the same size n. Then the scalar product a . b of a and b is the
single number (a scalar) equal to the sum of the products,
a . b  a1 b1  a2 b2  . . .  an bn .

Figure 10.19 will help you visualize how to calculate the scalar product of two vec-
tors a and b.

a1b1

a2b2 b1
b2
..
a1 a2 ... an .
bn
FIGURE 10.19 anbn

For instance, the scalar product of the vectors a  3 3 2 5 44 and b 

3 6 3 2 0 4 is
a . b  316 2  213 2  152 12 2  410 2  18  6  10  0  2.
We now see how the scalar product arises in a variety of situations.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 672

672 CHAPTER 10 Matrix Algebra and Its Applications

E XAMPLE 1
Suppose that peaches cost 30¢ each, pears cost 20¢ each, apples cost 35¢ each, and grape-
fruits cost 50¢ each. Amy wants to get 5 peaches, 3 pears, 2 apples, and 2 grapefruits, and
Bill wants to get 3 peaches, 4 pears, 3 apples, and 3 grapefruits.
a. Write vectors to represent the prices of the fruits and the amount of each fruit that
Amy and Bill will purchase.
b. Write the total costs of their fruit purchases, using vector methods.
Solution
a. We form the price vector p  3 0.30 0.20 0.35 0.504 for the prices of the respec-
tive fruits and the two demand vectors, a  3 5 3 2 2 4 for Amy and b 
3 3 4 3 3 4 for Bill.
b. The scalar products a . p and b . p give the total costs of Amy’s and Bill’s purchases:
a . p  35 3 2 2 4 . 30.30 0.20 0.35 0.50 4
 510.302  310.20 2  210.352  210.502
 1.50  0.60  0.70  1.00  3.80;
b p  33 4 3 3 4 . 30.30 0.20 0.35 0.50 4
.
 310.302  410.20 2  310.352  310.502
 0.90  0.80  1.05  1.50  4.25.
Thus the cost of fruit was $3.80 for Amy and $4.25 for Bill.
◆

A Geometric View of Scalar Products

An interesting special case of the scalar product involves the use of coordinate vectors,
which we introduced in Section 10.1. In two dimensions, they are the unit vectors
3 1 04 along the horizontal axis and 30 1 4 along the vertical axis. See Figure 10.20.
In three dimensions, the coordinate vectors are 31 0 0 4 , 30 1 0 4, and
3 0 0 1 4 . They lie along three mutually perpendicular axes, as illustrated in Fig-
ure 10.21.

z
y
1
1
0
0
0
1
0 1
0
1 y
1
x 1
1 1
1 0
0 x 0

FIGURE 10.20 FIGURE 10.21

Any vector can be built from its components. For instance, for the 3-vector
3 2 5 3 4 , we use the components 2, 5, and 3 and the coordinate vectors to write
32 5 3 4  2 3 1 0 0 4  5 3 0 1 0 4  3 30 0 1 4.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 673

10.3 Scalar Products 673

In general, for any 3-vector a  3a1 a2 a3 4 , we can write

a  a1 31 0 0 4  a2 30 1 0 4  a3 30 0 1 4.
In many geometric uses of vectors, we need to know whether two vectors
“point” in the same general direction or in opposite directions. When two vectors
point in approximately the same general direction, as shown by the arrows in Fig-
ure 10.22, their scalar product will be positive. When two vectors point in approxi-
mately opposite directions, as shown in Figure 10.23, their scalar product will be
negative. Most interestingly, when two vectors form a right angle, as shown in Fig-
ure 10.24, their scalar product will be zero.

y y y
2 2 2
a= a= a=
4 4 4
4 4 –4 4
b=
4 2
b=
2

x x x
–4 4 –4 4 –4 4

–4 –3 –4 –4
b=
–3

FIGURE 10.22 FIGURE 10.23 FIGURE 10.24

These assertions follow from the fact that, if u is the angle between the vectors a
and b, as shown in Figure 10.25, we have the following relationship.

a.b
cos u 
2a . a 2b . b

θ a

x
FIGURE 10.25

(This formula is based on the Law of Cosines introduced in Section 6.5; we ask you
to derive this formula in a problem at the end of the section.) Note that, if
a  3a1 a2 4 is a 2-vector, a . a  a12  a22. We interpret a . a geometrically by
ˇ ˇ
Gord.3896.10.pgs 4/28/03 3:29 PM Page 674

674 CHAPTER 10 Matrix Algebra and Its Applications

using the Pythagorean theorem to represent the square of the hypotenuse in a right
triangle with sides a1 and a2 , as shown in Figure 10.26. Thus, in the formula for cos u,
1a . a is the length of the vector a. Similarly, if a is a 3-vector, a  3 a1 a2 a3 4, then
a . a  a12  a22  a32 ˇ ˇ ˇ

and so 1a . a is the length of a. Similarly, 1b . b is the length of b. Therefore we

can rewrite the formula for cos u as
a.b
cos u 
1length of a2 1length of b2

a2
a

x
FIGURE 10.26 a1

Clearly the length of a and the length of b are both positive. Thus, when the
vectors point in roughly the same direction and the angle u is between 0° and 90°,
cos u is positive and so is a . b. When the vectors point in roughly opposite direc-
tions and u is between 90° and 180°, cos u is negative and so is a . b. When u is 90°,
so that a and b are perpendicular, cos u is zero and so a . b  0.

E XAMPLE 2
Find the angle between the lines y  2x and y  3x.
Solution We know that both lines pass through the origin, as illustrated in Figure 10.27.
To find the angle u between the lines at the origin, we need to find vectors along each line.
Suppose that we arbitrarily choose x  1. On the first line y  2x, the corresponding
value of y is 2, so the point 11, 22 is on that line and the vector from the origin to that

y = 3x y = 2x
4

1
b=
3 3

2 1
a=
2
θ
1

x
FIGURE 10.27 0 1 2 3
Gord.3896.10.pgs 4/28/03 3:29 PM Page 675

10.3 Scalar Products 675

point is a  31 2 4. Similarly, using x  1, we find the corresponding point 11, 3 2 on the

second line y  3x, so the vector from the origin to that point is b  3 1 3 4. The angle u
between the lines is the same as the angle between the vectors, so
a.b 11 2 112  12 2 13 2
cos u  
2a . a 2b . b 211 2 11 2  122 12 2 211 2 11 2  132 13 2
7
  0.98995.
25 210
Consequently, the angle between the two lines is
u  arccos10.989952  8.13°  0.1419 radian.
◆

The Clothes Production Model Using Scalar Products

In Section 10.2 we introduced a mathematical model for production in three
clothing factories to meet a demand for 500 vests, 850 pants, and 1000 coats. If xi
denotes the number of rolls of cloth used by the ith factory 1i  1, 2, 32, the xi’s
must satisfy the system of linear equations
vests: 20x1  4x2  4x3  500
pants: 10x1  14x2  5x3  850
coats: 5x1  5x2  12x3  1000.
A key property of scalar products is that a . b is a linear combination of the entries
in each vector. (Similarly, you can think of a polynomial as being a linear combina-
tion of power functions.) Conversely, any linear combination of variables or num-
bers always can be interpreted as a scalar product of two vectors.

E XAMPLE 3
Rewrite the three linear equations for the clothes production model using scalar prod-
ucts of vectors.
Solution Consider the first linear equation of the clothes production model
20x1  4x2  4x3  500.
The left-hand side of this equation is a linear combination of the three variables. If
x1
a  320 4 4 4 and x  C x2 S ,
x3
(we explain shortly the reason for writing a as a row vector and x as a column vector), we
can write the left-hand side of this equation as the scalar product
x1
a . x  320 4 4 4 . C x2 S  20x1  4x2  4x3 .
x3
Gord.3896.10.pgs 4/28/03 3:29 PM Page 676

676 CHAPTER 10 Matrix Algebra and Its Applications

The first equation is then a . x  500. Note that the vector a consists of the entries in the
first row of the matrix

20 4 4
A  C 10 14 5S
5 5 12
that we used in the clothes production model of Section 10.2.
Using vectors b and c to represent the second and third rows of matrix A, we can
write the other two linear equations in the same way as b . x  850 and c . x  1000.
◆
Any system of linear equations can be written in terms of a system of scalar
products. For example, the left-hand sides of the equations for the clothes produc-
tion model are

x1
vests: 20x1  4x2  4x3  320 4 44 C x2 S  a 1œ . x,
.
x3
x1
pants: 10x1  14x2  5x3  310 14 5 4 £ x2 §  a 2œ . x,
.
x3
x1
coats: 5x1  5x2  12x3  35 5 12 4 C x2 S  a 3œ . x,
.
x3

where a i œ is the ith row of the clothes production coefficient matrix

Factory 1 Factory 2 Factory 3

Shirts 20 4 4
Pants C 10 14 5 S  A.
Coats 5 5 12

The Matrix–Vector Product

Although we encounter many important uses of single scalar products in matrix
algebra, their most important use is as a building block for defining the product of
a matrix and a vector and, in Section 10.4, the product of two matrices. We define
the matrix–vector product as follows.

Matrix–Vector Product
The product of an m  n matrix A and a column n-vector c is a column m-
vector of scalar products
Ac  a i œ . c (each row a i œ of A multiplies c).
Gord.3896.10.pgs 4/28/03 3:29 PM Page 677

10.3 Scalar Products 677

The diagram in Figure 10.28 will help you visualize this definition. Think of each
row in the first matrix A as a row vector (or think of A as consisting of a collection
of row vectors). The scalar product of each row vector in A and the column vector
c creates a new column vector. In order for this definition to make sense, the entries
in each row of the matrix A (equivalently, the number of columns in matrix A)
must equal the size of the vector c.

a11b1 + a12b2 + ... + a1nbn

a11 a12 a13 ... a1n b1

a21 a22 a23 ... a2n b2
.. .. .. .. ..
. . . . .
am1 am2 am3 ... amn
bn
FIGURE 10.28 am1b1 + am2b2 + ... + amnbn

For instance, if A is a 2  3 matrix and c is a column 3-vector,

3
2 1 0
AB R and c  C 4 S ,
5 3 6
2
then

213 2  114 2  012 2

3
2 1 0 10
Ac  B R C4S  B R  B R.
5 3 6 513 2  314 2  612 2 39
2
As we said previously, each row of matrix A is treated as if it were a row vector, and
it is used to form a scalar product with the column vector c. However, we cannot
multiply
3
2 1 0 4
AB R and c  D T
5 3 6 2
5
because the number of entries in each row of A does not match the number of en-
tries in C.
We can recast this illustration symbolically as follows. If
c1
a a12 a13
A  B 11 R and c  C c2 S ,
a21 a22 a23
c3
then
a 1œ . c a11 c1  a12 c2  a13 c3
Ac  B œ. R  B R.
a2 c a21 c1  a22 c2  a23 c3
Gord.3896.10.pgs 4/28/03 3:29 PM Page 678

678 CHAPTER 10 Matrix Algebra and Its Applications

The only other allowable way to multiply a vector and a matrix is to multiply a
row vector by a matrix. In that case, the matrix must have the same number of rows
as the row vector has entries. That is, we can multiply
2 0
2
c  31 4 2 34
6
and A  D T
3 5
1 5
because there are four entries in the row vector, which matches the number of rows
(vertical entries per column) in the matrix A.

E XAMPLE 4
Consider again the set of equations discussed in Example 3 for the clothes produc-
tion model:
vests: 20x1  4x2  4x3  500
pants: 10x1  14x2  5x3  850
coats: 5x1  5x2  12x3  1000.
Rewrite the system of equations as a matrix–vector equation.
Solution If we make vectors of the left-hand sides of the three equations, we have
20x1  4x2  4x3 320 4 4 4 . x a 1œ . x
C 10x1  14x2  5x3 S  C 310 14 5 4 . x S  C a 2œ . x S  Ax.
5x1  5x2  12x3 35 5 12 4 . x a 3œ . x
If we let
500
b  C 850 S
1000
represent the column vector of demands for the different products, the system of equa-
tions becomes
20x1  4x2  4x3 500
Ax  b or C 10x1  14x2  5x3 S  C 850 S .
5x1  5x2  12x3 1000
◆
As Example 4 suggests, any system of linear equations can be written as a
matrix–vector equation. As we develop the tools for working with and solving such
equations, we demonstrate that having such a formulation has significant advan-
tages, especially for systems that are larger than three equations in three unknowns.

The Fruit Purchase Model Revisited In Example 1, we computed the scalar prod-
ucts for the costs of fruit purchased by Amy and Bill. Recall that Amy wanted 5
peaches, 3 pears, 2 apples, and 2 grapefruits, whereas Bill wanted 3 peaches, 4 pears,
3 apples, and 3 grapefruits. Also, we know that peaches cost 30¢ each, pears 20¢
each, apples 35¢ each, and grapefruits 50¢ each.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 679

10.3 Scalar Products 679

E XAMPLE 5
Write a matrix–vector equation to represent the costs of the fruit purchases by Amy
and Bill.
Solution Let’s make a matrix A of fruit purchases. The columns represent the different
fruits, the first row gives Amy’s fruit shopping list, and the second row gives Bill’s list. We
also make a (column) vector p of the costs, in cents.

Peaches Pears Apples Grapefruits 30 Peaches

5 3 2 2 Amy 20 Pears
AB R pD T
3 4 3 3 Bill 35 Apples
50 Grapefruits

We can now write the costs of Amy’s and Bill’s fruit purchases as the matrix–vector product

30
5 3 2 2 20 51302  3120 2  2135 2  2150 2
Ap  B RD T  B R
3 4 3 3 35 31302  41202  3135 2  3150 2
50
380 $3.80
B RB R.
425 $4.25

This result is the same as we obtained in Example 1.

◆
Markov Chain for the Stock Market Revisited Recall the Markov chain intro-
duced in Example 1 of Section 10.2. The equations for determining the probabili-
ties p1 , p2 , and p3 that the stock market goes up, goes down, or stays the same
ˇ ˇ ˇ

tomorrow given the probabilities p1 , p2 , and p3 of its going up, going down, or stay-
ing the same today were
1 1 1
p1  p1  p2  p3
4 2 4
1 1 1
p2  p1  p2  p3
2 4 2
1 1 1
p3  p1  p2  p3 .
4 4 4

E XAMPLE 6
Write the preceding Markov chain model for the stock market as a matrix–vector equation.
Solution Let
p1 p1ˇ

p  C p2 S and p   C p2 S ,
ˇ

p3 p3ˇ

respectively, be the vectors of today’s and tomorrow’s probabilities and let the matrix of
transition probabilities be A:
Gord.3896.10.pgs 4/28/03 3:29 PM Page 680

680 CHAPTER 10 Matrix Algebra and Its Applications

Market Today
Up Down Same
1 1 1
Up 4 2 4
Market
Down C 12 1
4 2S
1
 A.
Tomorrow
Same 14 1
4
1
4

Then the preceding set of probability equations can be written simply as p  Ap because

4 p1  2 p2  4 p3
1 1 1 1 1 1
4 2 4 p1
Ap  C 21 1
4 2S
1
C p2 S  C 12 p1  14 p2  12 p3 S .
4 p1  4 p2  4 p3
1 1 1 1 1 1
4 4 4 p3
◆
Similarly, the probabilities p for the day after tomorrow can be found from
p  Ap
and so on.
In Section 10.4 we present a concise way of writing these probability vectors.

A Geometric View of Matrix–Vector Products

In geometric terms, when we multiply a vector v by some matrix A, the vector v is
transformed by the multiplication into another vector w  Av. Because of the di-
mensions of A and v, this product makes sense only in the order Av (rather than
vA); we thus call this operation pre-multiplication of v by A. We can view pre-
multiplication by A as defining a function: w  f 1v2, where f 1v2  Av. This type
of transformation is used in computer graphics, for instance, to produce creative
lettering and moving images (animation) on the screen. It also provides a way to
visualize the effect on a vector of multiplying it by a matrix.
Rather than using arrows, here we simply represent 2-vectors as points in
the plane.

E XAMPLE 7
Consider the vectors

0 0 2 2
v1  B R v2  B R v3  B R v4  B R
0 2 2 0
0 1 2 1
v5  B R v6  B R v7  B R v8  B R .
1 2 1 0

Here, v1 , v2 , v3 , and v4 point to the corners of a square whose sides are of length 2 and v5 , v6 ,
v7 , and v8 point to the midpoints of the sides of this square, as shown in Figure 10.29(a).
Describe the effects of pre-multiplying each of these eight vectors by the matrix
1 1
AB R.
1 1
Gord.3896.10.pgs 4/28/03 3:29 PM Page 681

10.3 Scalar Products 681

y y

v2 v6 v3 w2
2 2

v5 v7 w5 w6
1 1

v1 v8 v4 w1 w3
x x
0 1 2 3 4 0 1 2 3 4

–1 –1 w8 w7

–2 –2
w4
FIGURE 10.29 (a) (b)

Solution When we pre-multiply the v’s by the matrix A, we obtain the following eight
vectors wi  Avi .
1 1 0 0 1 1 0 2
w1  B RB R  B R w2  B RB R  B R
1 1 0 0 1 1 2 2
1 1 2 4 1 1 2 2
w3  B RB R  B R w4  B RB R  B R
1 1 2 0 1 1 0 2
1 1 0 1 1 1 1 3
w5  B RB R  B R w6  B RB R  B R
1 1 1 1 1 1 2 1
1 1 2 3 1 1 1 1
w7  B RB R  B R w8  B R B R  B R.
1 1 1 1 1 1 0 1
When we examine these eight new vectors, shown in Figure 10.29(b), we observe that
they also form a square, but it has a different orientation and size. Each of the four ver-
tex vectors v1 , v2 , v3 , and v4 of the original square was transformed into a vertex vector
w1 , w2 , w3 , and w4 of the new square. Each of the midpoint vectors v5 , v6 , v7 , and v8 in
the original square was transformed into a corresponding midpoint vector w5 , w6 , w7 ,
and w8 in the new square. Also, in the original square, the sides were of length 2 and each
diagonal was 2 12, from the Pythagorean theorem. In the transformed square, each di-
agonal has length 4, so that the sides have length 2 12 . Thus pre-multiplying the vectors
v1 , v2 , v3 , and v4 forming the original square by A does three things to the square.
1. It rotates the square clockwise through 45° or p>4.
2. It increases the length of each side by a factor of 12 .
1 2
3. It moves the center of the square from B R to B R .
1 0
◆
Example 7 demonstrates that one square can be transformed into another by
multiplying a set of vectors by an appropriate matrix. The same principle applies to
any shape whose corners are determined by a set of vectors.
If we pre-multiplied the vectors forming the original square by other (appro-
priately chosen) matrices, we could get rotations of the square through any desired
angle, increase or decrease the lengths of the sides by any desired multiple, and
Gord.3896.10.pgs 4/28/03 3:29 PM Page 682

682 CHAPTER 10 Matrix Algebra and Its Applications

place the center in any desired location. However, the origin will always be un-
0 0
affected because B R multiplied by any matrix yields B R .
0 0
1
0
For example, using B  B R rotates the square counterclockwise
1 0
1
through an angle of p>2 and transforms the center to B R . We ask you to verify this
result in the Problems at the end of this section. 1
In general, pre-multiplying by the matrix

cos u sin u
RB R
sin u cos u

rotates the square counterclockwise through an angle of u.

E XAMPLE 8
a
Show that the effect of pre-multiplying any vector v  B R by the matrix R is to rotate v
through an angle u. b

Solution We start with an arbitrary vector v, as shown in Figure 10.30, that is inclined
at an angle a from the horizontal, so that
b
tan a  .
a

y
w = Rv

v
φ
b
α
x
FIGURE 10.30 O a

Also, we let the angle associated with the vector w  Rv be f. We want to show that the
difference f  a in the two angles must equal u.
When we pre-multiply vector v by R to form vector w, we get

cos u sin u a a cos u  b sin u

w  Rv  B RB R  B R.
sin u cos u b a sin u  b cos u
Consequently, the tangent of angle f is
a sin u  b cos u
tan f  .
a cos u  b sin u
Gord.3896.10.pgs 4/28/03 3:29 PM Page 683

10.3 Scalar Products 683

We factor a out of both the numerator and denominator and divide both the nu-
merator and denominator by cos u:

a. a  b
sin u b b
tan u 
cos u a a
tan f   .
a . a1  b
b sin u b
1  tan u
a cos u a
However, we know that tan a  b>a, so that this expression reduces to
tan u  tan a
tan f  .
1  tan a tan u
When we compare this expression to the sum identity for the tangent from Problem 37
of Section 8.1, we see that
tan f  tan 1u  a2 so that f  u  a.
Therefore the effect of pre-multiplying any vector v by the matrix R is to rotate the re-
sulting vector through an angle of u.
◆
What if we want to change the length of a vector by using matrix multiplica-
tion? We note that the lengths of the sides of the square in Example 8 are un-
changed when we use this rotation matrix R. If we want to enlarge or contract the
square, we must use an appropriate scalar multiple of the rotation matrix, so that
we multiply every entry in the matrix by a constant amount. Thus, if
1 1 1 1 2 2
AB R , then 2A  2 B RB R.
1 1 1 1 2 2

In Example 7, to make each side of the square grow by a factor of 2 instead of a fac-
tor of 12 , we would multiply A by 12 because 12 . 12  2. Incidentally, not
every matrix can be a rotation matrix; a special form is necessary.
Any transformation that takes a 2-vector v into the 2-vector w  Av, for any
2  2 matrix A, always transforms or maps lines into lines. Because of this linearity
property, a mapping of the form v S w  Av is called a linear transformation.
Suppose that we start with any figure comprising (very short) line segments.
Each segment can be interpreted as a vector, and appropriate matrices can be con-
structed to create any kind of transformation—a shift, a stretch, or a rotation—that
we desire. This method is the mathematical foundation of the computer graphics
animation that appears in movies, on television, and on computer screens.

Problems
1. Let Compute the scalar products.
3
a  B R,
0
b  B R,
5
c  B R, a. a . c b. b . c c. b . d
5 2 1 .
d. a d e. c . d f. b . e
1 4 g. c . e
dB R, e  B R.
0 5
Gord.3896.10.pgs 4/28/03 3:29 PM Page 684

684 CHAPTER 10 Matrix Algebra and Its Applications

2. Let 6. Let a, b, c, and d be as in Problem 3 and let

4 2 4 3 1 2 0 1 3
a  B R, bB R, cB R,
2 4 2 A  C0 1 5S , B  C 1 5 1S ,
4 2 0 4 3 4 1 6
dB R , and e  B R .
2 4
0 1 2
Plot the pairs of vectors and compute their scalar 3 4 5
products. CD T.
6 7 8
a. a, b b. a, c c. a, d
9 10 11
d. b, d e. c, d f. d, e
What geometric pattern do you find in pairs of vec- Find the matrix–vector products.
tors whose scalar products are zero? What numeri-
a. Aa b. Ab c. Bc
cal pattern do you find in the ratios of components
d. Bd e. Cb f. Cc
in these pairs?
7. Let a, b, c, and d be as in Problem 3 and let
3. Let
2 0 1 0 0 0 0 1
a  C5S , b  C1S , A  C0 1 0S , B  C0 1 0S ,
1 0 0 0 1 1 0 0
7 2 1 0 0
c  C 2 S , d  C 3 S . C  C0 1 0S .
1 1 0 0 2

Compute the scalar products. Compute the matrix–vector products and describe
a. a . c b. b . c c. b . d in words the effect of multiplying a vector by the
d. a . d e. a . a f. c . d particular matrix.
4. Let a. Ac b. Ad c. Ba
d. Bd e. Cb f. Ca
3 0
8. Explain why it is not possible to multiply the matrices
a  C1S , b  C 2S ,
2 2 1 2 3 4
AB R and b  B R.
1 2 4 5 6 7
c  C4S , d  C1S . 9. Write each vector–matrix equation as a system of
8 5 equations. Here x denotes a column vector of vari-
ables x1 , x2 , . . . , where the number of variables
Compute the scalar products. equals the number of columns in A.
a. a . c b. b . d c. c . d 5 1 2
d. 1a  c2 a . e. a 1c  d2
. a. Ax  b, where A  B R and b  B R
f. 1b  a2 . 1a  c2
4 3 5
b. Ax  b, where
5. Let a, b, c, and d be as in Problem 1 and let
1 4 4
2 1 AB R and b  B R
1 7 5 2 2 3 9
AB R, BB R, C  C 4 2 S .
4 2 1 0 c. Ax  b, where
3 1
5 2 1 1
Compute the matrix–vector products.
A  C4 1 6S and b  C 5 S
a. Ab b. Ac c. Ba
d. Bb e. Cc f. Ca 3 1 0 2
Gord.3896.10.pgs 4/28/03 3:29 PM Page 685

10.3 Scalar Products 685

d. Ax  b, where 1 1 3
a. v1  B R , v2  B R , v3  B R ,
2 1 5 0 1 3 3
A  C3 1 2S and b  C 0 S 3
v4  B R
5 1 3 0 1
10. Write each system of equations in matrix notation. 0 1 0
b. v1  B R , v2  B R , v3  B R ,
Define any matrix or vector that you use. 0 1 2
a. x1  2x2  6 1
2x1  6x2  4 v4  B R
1
b. 5x1  2x2  4x3  6
0 1
x1  3x2  2x3  2 15. Repeat Problem 14 with the matrix B  B R.
2x1  5x2  5x3  5 1 0
c. 2x1  5x2  2x3  0 In Example 8, we proved that the mapping of v to
3x1  8x2  4x3  0 w  Bv acts to rotate a square counterclockwise
x1  x2  7x3  0 through an angle of p>2. Do your results confirm
this outcome?
11. Write in matrix notation the systems of linear equa-
tions obtained in the following Problems in Section 16. Transform each square in Problem 14 by using the
10.2. Define any matrix or vector that you use. 1 2
a. Problem 4 b. Problem 5 matrix C  B R . Does the mapping of v to
3 4
c. Problem 6 d. Problem 7
12. Write in matrix notation the systems of linear equa- w  Cv transform the square into a square?
tions obtained in the hare–wolf population model cos u sin u
in Example 3 of Section 10.2. Define any matrix or 17. Consider the rotation matrix R  B R.
sin u cos u
vector that you use.
a
13. Suppose that you will need 10 hero sandwiches, 6 Show that, for any vector v  B R , the vector w  Rv
quarts of fruit punch, 3 pounds of potato salad, and b
2 plates of hors d’oeuvres for a party. The matrix has the same length as v.
shows the cost per unit of these supplies from three 18. Use the matrix R from Problem 17 to determine the
different caterers. a
effect that the matrix 2R has on the vector v  B R .
Caterer A Caterer B Caterer C b
Hero sandwich $4 $6 $5 19. The rotation matrix R in Problem 17 acts to ro-
Fruit punch $2 $1 $0.85 tate any nonzero vector counterclockwise through
D T
Potato salad $1.50 $2 $2.50 an angle u.
Hors d’oeuvres $6 $5 $7 a. Modify matrix R to produce a matrix that rotates
any nonzero vector clockwise through an angle u.
a. Express the cost of catering the party by each b. Write a matrix that will rotate any nonzero vec-
caterer as a matrix–vector product. tor counterclockwise through an angle of 30°.
b. Determine the cost of each caterer. 5
14. Plot a square with corners determined by the vectors c. Plot the position vector v  B R , pre-multiply it
2
v1 , v2 , v3 , and v4 in (a)and (b). Then, for the matrix by the matrix you created in part (b), and then
1 1 plot the resulting vector w on the same graph.
AB R , plot the transformed corners d. Pre-multiply the vector w by the matrix you cre-
1 1
w1  Av1 , w2  Av2 , w3  Av3 , and w4  Av4 . ated in part (a) with u  30°. What is the result
Confirm that the midpoints of the sides of the orig- of this operation?
inal square are mapped to the midpoints of the 20. The five points A, B, C, D, and E at 13, 2 2, 17, 1 2,
sides of the transformed square. 19, 1 2, 110, 32 , and 16, 6 2 determine a five sided figure
Gord.3896.10.pgs 11/21/03 11:05 AM Page 686

686 CHAPTER 10 Matrix Algebra and Its Applications

having two right angles. Use vectors to determine vectors with the smallest angle between them.
which of the five angles in the figure are the right Which college did she choose?
angles. 24. (Continuation of Problem 23) Suggest some other
21. Find the acute angle between each pair of vectors. ways that Susan could have decided which school was
a. 33 54 and 32 44 closer to Ivy Tech? Would you necessarily make the
b. 3 1 44 and 32 54 same decision as to which school to choose? Explain.
c. 3 1 4 54 and 3 2 3 24 25. (Derivation of the formula for cos u.) In the triangle
d. 3 6 4 14 and 35 3 24 shown, let a  3a1 a2 4 and b  3b1 b2 4. The
22. Rewrite each polynomial as the scalar product of
a vector of numbers and a vector of power func- lengths of a and b can be written 7 a7  2a12  a22
tion terms of the form x  3 1 x x2 4 or and 7 b7  2b12  b22 , so that 7 a7 2  a . a and
x  31 x x2 x3 4 .
7 b7 2  b . b.
a. 5  3x  2x2
b. 8  7x  3x2 y
c. 4  2x  6x2  5x3
23. When Susan was applying to college, she was
turned down by her top choice, Ivy Tech, but she c
was accepted by State Tech and the Hawaii Institute
of Technology. To decide on which school to attend, b
she rated each school (on a scale of 0 to 10) on five
important criteria and then selected the one that
was closer to Ivy Tech. She used her knowledge of θ a
vectors to decide how to interpret closer—the two
x
O

a. Write the vector c in terms of the vectors a and b.

Ivy State Hawaii b. Use the expression from part (a) to form the
Location 7 6 10 scalar product c . c.
c. Use the law of cosines to write an equation giv-
Size 6 3 7 ing 7 c7 2.
d. Compare the expressions from part (b) and
part (c) and use the facts that a . a  7 a 7 2,
Campus 8 4 7

Faculty 9 6 4 b . b  7 b7 2, and c . c  7 c7 2 to show that

Programs 5 7 6 a.b
cos u 
7 a7 7 b7
.

10.4 Matrix Multiplication

In Section 10.3, we introduced the concept of the scalar product a . b of two vectors

a and b. For example, if a  32 1 4 and b  B R , then a . b  2142  116 2  14.

4
6
We used this scalar product of vectors to define the matrix–vector product Ab of a
matrix A times a vector b. For instance, if

2 1 4
AB R and again b  B R ,
0 3 6
Gord.3896.10.pgs 4/28/03 3:29 PM Page 687

10.4 Matrix Multiplication 687

then

2 1 4 214 2  1162 14
Ab  B RB R  B R  B R.
0 3 6 014 2  3162 18

Thus Ab is a column vector consisting of the scalar products of each row of A with b.
In this section, we extend this process to define the product of two matrices A
and B. In particular, in the product AB we think of the second matrix B as consist-
ing of a series of column vectors and pre-multiply each of them by the matrix A.
That is, we multiply the first column of B by A, then multiply the second column of
B by A, and so on, as illustrated in Figure 10.31. The result of each product is a col-
umn vector, so the product AB will be a matrix having the same number of
columns as there are in B. We can think of the product AB as a matrix whose
columns are a sequence of matrix–vector products 3Ab1 Ab2 . . . Abn 4.

a11 a12 ... a1n b11 b12 ... b1p

a21 a22 ... a2n b21 b22
...
b2p
.. .. ..
. . . .. .. ..
. . .
am1 am2 ... amn
bn1 bn2 ... bnp
FIGURE 10.31

Think About This In the matrix product AB, you can also think of the first matrix A as consisting of
a series of row vectors and the second matrix B as consisting of a series of col-
umn vectors and then take the scalar product of each of the row vectors making
up A with each of the column vectors making up B. Draw a sketch comparable to
Figure 10.31 to illustrate this interpretation. ❐

To demonstrate the product of two matrices, consider again the matrix

2 1 4 7 1
AB R and let B  B R.
0 3 6 5 9

We think of each of the three columns of matrix B as a column vector. Note that
4 4
the first column b1  B R of B is precisely the column vector b  B R we used
6 6
above. The corresponding first column in the product matrix AB is then Ab1 ,
14
which we computed above to be B R .
18
Similarly, we take the matrix–vector product of the matrix A with the second
7
column of B thinking of it as the vector b2  B R to produce the second column
5
of the product AB. That gives

2 1 7 217 2  115 2 19
Ab2  B RB R  B R  B R.
0 3 5 017 2  315 2 15
Gord.3896.10.pgs 4/28/03 3:29 PM Page 688

688 CHAPTER 10 Matrix Algebra and Its Applications

Finally, we take the matrix–vector product of A with the third column of B,

1
which is the vector b3  B R, to produce the third column of the product AB.
9
The complete product AB is therefore

2 1 4 7 1 214 2  1162 217 2  115 2 211 2  119 2

AB  B RB RB R
0 3 6 5 9 014 2  3162 017 2  315 2 011 2  319 2
14 19 7
B R.
18 15 27
In general, we have the following definition for the product of two matrices.

Matrix Multiplication
Let A be an m  r matrix and B be an r  n matrix. The number of columns
in A must equal the number of rows in B.
The matrix product AB is the m  n matrix obtained by forming the scalar
product of each row a i œ in A with each column bj in B. That is, the (i, j)th
entry in AB is a i œ bj . Thus
a 1œ b1 a 1œ b2 ... a 1œ bn
a œb a 2œ b2 ... a 2œ bn
AB  D 2 1 T.
o o o o
a mœ b1 a mœ b2 ... a mœ bn

In summary, there are three ways to interpret matrix multiplication.

1. We can think of AB as the scalar product of each row of A with each col-
umn of B. Figure 10.32 illustrates this interpretation. The product requires
that the number of entries in each row of A must equal the number of en-
tries in each column of B.

a11 a12 ... a1n b11 b12 b1p

b21 b22 ... b2p
a21 a22 ... a2n
.. .. ..
... ... ..
. . . .
am1 am2 ... amn bn1 bn2 bnp

FIGURE 10.32

2. We can think of AB as a sequence of matrix–vector products—that is, the

product of A with each of the column vectors making up B, so that
AB  A3b1 b2 ... bn 4  3Ab1 Ab2 ... Abn 4.
Gord.3896.10.pgs 11/21/03 11:05 AM Page 689

10.4 Matrix Multiplication 689

For example, for the matrices A and B above, we verify that the first col-
umn of AB is the matrix–vector product Ab1 . Then

2 1 4 214 2  116 2 14
Ab1  B RB R  B R  B R.
0 3 6 014 2  316 2 18

3. We can view AB as a set of vector-matrix products a i œ B (defined analo-

gously to matrix–vector products), where each row of A, considered as a
row vector, pre-multiplies the matrix B. Thus

a1œ a1œ B
a2œ a2œ B
AB  D T B  D T.
o o
amœ amœ B
For instance, for the matrices A and B, we verify that the first row of AB is
the vector–matrix product a 1œ B:
7 1
a 1œB  32 1 4 B
4
R
6 5 9
 3 214 2  116 2 217 2  115 2 211 2  119 2 4  314 19 7 4.
For the matrix product AB to make sense, the number of columns in A must
equal the number of rows in B. Thus, if A is m  n (m rows and n columns) and B
is n  k (n rows and k columns), the product AB is m  k. For instance, the prod-
uct of a 3  5 matrix and a 5  8 matrix will be a 3  8 matrix. But, the product
of a 5  8 matrix and a 3  5 matrix is not defined—the numbers of columns and
rows do not match, so it is not possible to perform the multiplication.
Note that our definition of a matrix–vector product in Section 10.3 is a special
case of matrix multiplication because, in the matrix–vector product Ab, the col-
umn vector b can be interpreted as an n  1 matrix. Then Ab is the matrix prod-
uct of an m  n matrix A and an n  1 matrix b. The result Ab is an m  1
matrix (a column m-vector).
In general, except in rather unusual circumstances, matrix multiplication is
not commutative; that is, AB  BA. In fact, unless A and B are both square matri-
ces with the same size, only one at most of AB and BA is defined. For instance, if A
is 3  5 and B is 5  2, we can form AB, but not BA.

E XAMPLE 1
Given

2 1 3 1
AB R and B  B R,
0 3 1 4

find AB and BA and decide whether matrix multiplication is commutative.

Solution Using the given matrices, we find that

2 1 3 1 213 2  111 2 211 2  1142 7 2

AB  B RB RB RB R,
0 3 1 4 013 2  311 2 011 2  314 2 3 12
Gord.3896.10.pgs 4/28/03 3:29 PM Page 690

690 CHAPTER 10 Matrix Algebra and Its Applications

whereas
3 1 2 1 312 2  11 2 10 2 311 2  112 13 2 6 0
BA  B RB RB RB R.
1 4 0 3 112 2  410 2 111 2  413 2 2 13
Thus, for these two matrices, AB  BA, so matrix multiplication is not, in general, a
commutative operation.
◆
We also need a way to add two matrices. Recall how we added vectors in Exam-
ple 3 of Section 10.1: We simply added the entries in the corresponding positions.
For instance, the sum of the row vectors c  32 1 5 4 and d  3 4 3 0 4 is
c  d  3 2 1 5 4  3 4 3 0 4  32  4 1  3 5  0 4  36 4 5 4.
We add matrices in the same way by simply adding all corresponding entries.
However, for addition to make sense, the two matrices or the two vectors must be
of the same size.

E XAMPLE 2
Find the sum of the two matrices
2 1 3 1
AB R and B  B R.
0 3 1 4

Solution Matrix addition involves adding the entries in the corresponding positions,
so we have
2 1 3 1 23 1  112 5 0
ABB RB RB RB R.
0 3 1 4 01 34 1 7
◆
E XAMPLE 3
Find the sum of the 3  3 matrices
4 2 3 2 6 1
E  C2 1 3S and F  C 4 0 2S .
0 5 4 7 3 8

Solution Again, by adding the entries in the corresponding positions, we find that
4 2 3 2 6 1 42 26 3  112 6 8 2
E  F  C2 1 3S  C4 0 2S  C2  4 10 3  2 S  C6 1 5S .
0 5 4 7 3 8 07 53 48 7 8 12

◆
Subtraction of matrices is defined in the analogous way—we simply take the
difference of the corresponding entries in each position. However, the quotient of
two matrices can’t be defined.
We now summarize the laws of matrix algebra for matrix addition and multi-
plication. In each case we assume that the matrices have the appropriate sizes so
that the operations make sense.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 691

10.4 Matrix Multiplication 691

Basic Laws of Matrix Algebra

Associative Law
Matrix addition and matrix multiplication are associative:
(A  B)  C  A  (B  C) and (AB)C  A(BC).

Commutative Law
Matrix addition is commutative:
A  B  B  A.
Matrix multiplication is not commutative (except in special cases):
AB  BA.

Distributive Law
A(B  C)  AB  AC and (B  C)A  BA  CA

Law of Scalar Factoring

k(AB)  (kA)B  A(kB),
where k is a scalar constant.

With the exception that matrix multiplication is not commutative, these laws
are basically the same as the laws used in algebra for working with real numbers.
However, because the objects now are arrays and the operation of matrix multipli-
cation is much more complicated than real-number multiplication, it is not at all
obvious that these matrix laws should be true. Some effort is required to verify
them (but this is beyond the scope of this chapter).
Because a vector is just a 1  n matrix or an n  1 matrix, these laws also
apply to vectors. However, scalar products of vectors are commutative.
Matrix calculations are standard features on most calculators and in many
software packages. Typically, you have to give a name for a matrix, such as A, then
specify the dimensions of the matrix (the number of rows by the number of
columns)—say, 3  4—and then enter the values in the appropriate positions.
Once you have entered the matrices, you can use the calculator to perform any of
the allowable operations such as sums, differences, and products. See your instruc-
tion manual for details.

The Fruit Purchase Model Revisited

In Example 1 of Section 10.3, we computed the scalar products of fruit costs and
quantities of fruit to be purchased by Amy and Bill. Recall that Amy wanted 5
peaches, 3 pears, 2 apples, and 2 grapefruits, whereas Bill wanted 3 peaches, 4 pears,
3 apples, and 3 grapefruits. Also, peaches cost 30¢ each, pears 20¢ each, apples 35¢
each, and grapefruits 50¢ each. In Example 6 of Section 10.2, we constructed a ma-
trix A of the fruit shopping lists of Amy and Bill and made a column vector p of the
fruit costs (in cents).
Gord.3896.10.pgs 4/28/03 3:29 PM Page 692

692 CHAPTER 10 Matrix Algebra and Its Applications

Peaches Pears Apples Grapefruits 30 Peaches

5 3 2 2 Amy 20 Pears
AB R pD T
3 4 3 3 Bill 35 Apples
50 Grapefruits

380 $3.80
The matrix-vector product Ap  B RB R gave the cost of the fruit pur-
425 $4.25
chases of Amy and Bill.

E XAMPLE 4
Suppose now that there are two other stores at which Amy and Bill can shop for fruit. In-
stead of a vector of fruit prices, we now have a matrix P of fruit prices (whose first col-
umn is the original store’s set of prices).
Store 1 Store 2 Store 3
30 25 30 Peaches
20 25 25 Pears
PD T
35 40 30 Apples
50 60 45 Grapefruits
Construct a matrix giving the cost to Amy and Bill of their fruit purchases at each store.
Solution We need to compute the matrix product AP, which is well defined becasue A
is a 2  4 matrix and P is a 4  3 matrix:
30 25 30
5 3 2 2 20 25 25
AP  B RD T
3 4 3 3 35 40 30
50 60 45
5130 2  3120 2  21352  2150 2 5125 2  31252  21402  21602 51302  3125 2  21302  21452
 B R
31302  4120 2  31352  31502 31252  41252  3140 2  3160 2 31302  4125 2  31302  31452

380 400 375

B R.
425 475 415
Alternatively, had we entered the cost of fruit in the form $0.30 instead of 30¢, say, we
would get

$3.80 $4.00 $3.75

AP  B R.
$4.25 $4.75 $4.15
◆
Note that we would have gotten the same results using a calculator to multiply
the two matrices.

Powers of Markov Chain Transition Matrices

Next, we consider a more substantial use of matrix multiplication—one that great-
ly expands the power of the Markov chain model for the stock market introduced
in Section 10.2.
Gord.3896.10.pgs 4/28/03 3:29 PM Page 693

10.4 Matrix Multiplication 693

In the Markov chain model, the equations for determining the probabilities
p1 , p2 , and p3 of the market going up, going down, or staying the same tomor-
row given the probabilities p1 , p2 , and p3 of the market going up, going down, or
staying the same today were
1 1 1
p1  p1  p2  p
4 2 4 3
1 1 1
p2  p1  p2  p
2 4 2 3
1 1 1
p3  p1  p2  p3 .
4 4 4
If
p1 p1
p  C p2 S and p   C p2 S
p3 p3

are the vector of today’s probabilities and the vector of tomorrow’s probabilities,
respectively, and the matrix A of transition probabilities is

Market Today
Up Down Same
1 1 1
Up 4 2 4
Market
Down C 21 1
4
1
2 S,
Tomorrow
Same 41 1
4
1
4

then the given system of transition probabilities can be written simply as p   Ap,
or, equivalently, as
p1 1
4
1
2
1
4 p1 4 p1  2 p2  4 p3
1 1 1

C p2 S  C 21 1
4 2S
1
C p2 S  C 12 p1  14 p2  12 p3 S .
p3 4
1 1
4
1
4 p3 1
4 p1  4 p2  4 p3
1 1

E XAMPLE 4
Use matrices to find the probabilities of the stock market going up, going down, or stay-
ing the same tomorrow if

0
p  C 12 S
1
2

is the vector of probabilities of the market going up, going down, or staying the same
today, as in Example 1 of Section 10.2.
Solution The vector of probabilities for the stock market tomorrow is p   Ap, so we get
1 1 1 3
4 2 4 0 8
p   Ap  C 21 1
4 2S
1
C 12 S  C 38 S .
1 1 1 1 2
4 4 4 2 8
Gord.3896.10.pgs 4/28/03 3:29 PM Page 694

694 CHAPTER 10 Matrix Algebra and Its Applications

Alternatively, using a calculator where the entries in the matrix A and the vector p are
given as decimals instead of fractions, we get
0.25 0.5 0.25 0 0.375
p   Ap  C 0.5 0.25 0.5 S C 0.5 S  C 0.375 S .
0.25 0.25 0.25 0.5 0.25
◆
Recall that a Markov chain can be extended farther into the future. Just as to-
morrow’s probability vector p can be computed by the matrix expression
p  Ap, so too the probability vector p for the day after tomorrow is given by
p  Ap  A(Ap)  (AA)p  A2p,

where we have written A2 to represent AA, the square of the matrix A. Note that
A2  AA is possible provided that A is a square matrix. (Don’t confuse “square of
a matrix” and “square matrix”, where the number of rows equals the number of
columns.) Similarly, the vector of probabilities p three days hence is given by
p  Ap  A(A2p)  A3p.

In writing these two matrix equations for p and p, we made use of the fact
that matrix algebra is an associative operation, so A(Ap)  (AA)p  A 2 p, and
A(A2p)  (AA2)p  A3p, where A3 can be thought of as AAA.

E XAMPLE 5
Compute A2 and A3 for the stock market Markov transition matrix A.
Solution We first do the calculations by hand:
1 1 1 1 1 1
4 2 4 4 2 4
A2  C 21 1
4 2S
1
C2
1 1
4 2S
1

1 1 1 1 1 1
4 4 4 4 4 4
1
4
. 14  12 . 12  14 . 14 1
4
. 12  12 . 14  14 . 14 1
4
. 14  12 . 12  14 . 14
 C 12 . 14  14 . 12  12 . 14 1
2
. 12  14 . 14  12 . 14 1
2
. 14  14 . 12  12 . 14 S
1
4
. 14  14 . 12  14 . 14 1
4
. 12  14 . 14  14 . 14 1
4
. 14  14 . 12  14 . 14
3 5 3
8 16 8
 C 38 7
16 8S .
3

1 1 1
4 4 4

Similarly, we compute A3 as
1 1 1 3 5 3
4 2 4 8 16 8
A3  AA2  C 21 1
4 2S
1
C 38 7
16 8S
3

1 1 1 1 1 1
4 4 4 4 4 4
1
4
. 38  12 . 38  14 . 14 1
4
. 165  12 . 167  14 . 14 1
4
. 38  12 . 38  14 . 14
 C 12 . 38  14 . 38  12 . 14 1
2
. 165  14 . 167  12 . 14 1
2
. 38  14 . 38  12 . 14 S
1
4
. 38  14 . 38  14 . 14 1
4
. 165  14 . 167  14 . 14 1
4
. 38  14 . 38  14 . 14
Gord.3896.10.pgs 4/28/03 3:29 PM Page 695

10.4 Matrix Multiplication 695

11 23 11
32 64 32
 C 13
32
25
64 32 S .
13

1 1 1
4 4 4

Alternatively, we can find these matrices by using the matrix capabilities of a calcu-
lator or computer package, though likely with the entries in decimal form. In that case,
we would define the 3  3 matrix A and then refer to it by name to form
0.375 0.3125 0.375
A2  A^2  C 0.375 0.4375 0.375 S
0.25 0.25 0.25
and
0.34375 0.359375 0.34375
A3  A^3  C 0.40625 0.390625 0.40625 S .
0.25 0.25 0.25
You can verify that these decimal entries are equivalent to the fractions we calculated by
hand initially.
◆
The entries in A2 are the transition probabilities for 2 days from now and the
entries in A3 are the transition probabilities for 3 days from now. For instance, the
value 38  0.375 in position 12, 1 2 of A2 (the entry in the second row, first column)
means that, if we are now in State 1 (Up—first column), the chance is 38 that in 2 days
we will be in State 2 (Down—second row). Similarly, the value of 13>32  0.40625
in entry 12, 12 of A3 indicates that, if the market is now up, the probability is 13>32
that in 3 days it will go down.
The values we obtained in computing A2 and A3 look reasonable. In particular,
the numbers in each column of A2 and A3 sum to 1 (the sum of all probabilities for
the market on any given day must be 1). Note that all the entries in each of the
three rows of A3 have roughly the same numerical value; the entries in the first row
are slightly larger than 13 , those in the second row are all about 0.40, and all the en-
tries in the last row are 0.25.

E XAMPLE 6
If the probabilities of the stock market going up, going down, or staying the same today
are given by the vector
0
p  C 12 S ,
1
2

use matrices to find the probabilities p of the market going up, going down, or staying
the same the day after tomorrow and the probabilities p for the day after that.
Solution We know that p  A2p and p  A3p. Multiplying, we get
0.34375 0.3515625
 
p  A p  C 0.40625 S
2
and p  A p  C 0.3984375 S .
3

0.25 0.25
You can easily verify that these entries are the same values we obtained in Section 10.2.
◆
Gord.3896.10.pgs 4/28/03 3:30 PM Page 696

696 CHAPTER 10 Matrix Algebra and Its Applications

If you refer back to the table that gave the market probabilities over many days
near the end of Example 2 in Section 10.2, you will observe that the probabilities
shown there are similar to the entries in the columns of A3. That is, the probabilities
after 3 days are actually fairly close to the long-term probabilities, so this Markov
chain converges to a limiting state rather quickly.
Example 6 illustrates how, with concise notation, matrix algebra allows us to
express quite complex expressions.
We can visualize what happens in a Markov chain graphically in the case of a
0.6 0.3
two-by-two transition matrix—say, A  B R . (Again, notice that the sum
0.4 0.7
of the entries in each column is 1.) Let’s start with an initial vector of probabilities
0.75
p  p0  B R . Any vector, including this vector of probabilities, can be inter-
0.25
preted geometrically, as shown in Figure 10.33. When we pre-multiply the vector p 0
by the transition matrix A, we obtain another vector of probabilities
0.525
p  p 1  B R , which we can interpret geometrically, as shown in Figure 10.33.
0.475
Note that the resulting vector p1 points in a direction quite different from that of
the initial vector p0 . We can think of the matrix A as transforming the probability
vector p0 into the vector p1  Ap0 .
When we multiply the vector p1 by the matrix A to form the next probability
0.4575
vector p2  Ap1  B R , we get a vector p2  A2p0 that points in still another
0.5425
direction. What happens when we continue the Markov process to get p3  A3p0 ,
p4  A4p0 , . . . ? (We have moved from using the notation p, p, and so on, to
use subscript notation because the use of multiple ’s quickly becomes too un-
wieldy.) In the following table, we show the results of continuing this process nu-
merically for the two components p1 and p2 for each successive probability vector
p. As in Example 6, the probabilities seem to converge to a pair of limiting values—
approximately 0.42857, which is about 3>7, and approximately 0.57143, or 4>7.

n 0 1 2 3 4 5 6 7 ...
p1 0.75 0.525 0.4575 0.43725 0.43118 0.42935 0.42881 0.42864 ... 0.42857

p2 0.25 0.475 0.5425 0.56275 0.56883 0.57065 0.57119 0.57136 ... 0.57143

Figure 10.34 shows the corresponding geometric behavior. Note how the se-
quence of vectors also converges, getting closer and closer to a single limiting vector.

y y
p3
p2
p1 p1

p0 p0

x x
FIGURE 10.33 FIGURE 10.34
Gord.3896.10.pgs 4/28/03 3:30 PM Page 697

10.4 Matrix Multiplication 697

0.2
If we start with any other vector of probabilities, say, p 0  B
R, then the corre-
0.8
sponding sequence of vectors will also converge to this same limiting vector
3
B74R . This special limiting vector is called an eigenvector of the matrix A. A similar
7
type of convergence occurs with most other transition matrices, whether a two-by-
two matrix or larger.

A Geometric View of Powers of a Matrix

cos u sin u
We have shown that any matrix of the form R  B R acts as a rotation
sin u cos u
matrix to rotate a vector v through an angle u. We now examine the effect of apply-
ing R2 to a vector.
E XAMPLE 7
Show that the matrix R2 is a rotation matrix that will rotate any vector through an angle 2u.
Solution We have
cos u sin u cos u sin u
R2  B RB R
sin u cos u sin u cos u
cos2u  sin2u sin u cos u  sin u cos u
B R
sin u cos u  sin u cos u sin2 u  cos2 u
cos2u  sin2u 2sin u cos u
 c R.
2 sin u cos u cos2u  sin2u
If we now apply the double angle identities for the sine and cosine from Chapter 8, we
cos 2u sin 2u
get R2  B R, so that R2 indeed is a rotation matrix with an associated
sin 2u cos 2u
angle 2u.
◆

Problems
1. Let a. A  B b. B  C
0 2 4 2 c. A  B  C d. 2A  3C
AB R, BB R, 3. Let
1 4 1 1
5 0 5 1
3 1 AB R, BB R,
CB R. 1 4 0 2
1 3
1 2
Compute the matrix products. 1 5 2
C  C3 4S , DB R.
a. AB b. BA 3 0 2
c. CB d. BC 5 6
e. AC f. A2 Compute the products, if possible.
g. B2 a. AB b. CB c. BC
2. For the matrices in Problem 1, compute the sums d. AD e. DA f. CD
and linear combinations of matrices. g. DC h. C2
Gord.3896.10.pgs 4/28/03 3:30 PM Page 698

698 CHAPTER 10 Matrix Algebra and Its Applications

4. For A, B, and C in Problem 1, and Job 3 (doing a student’s homework assignment).

a. compute AB and then compute (AB)C. There are three families, the Joneses, the Smiths, and
b. compute BC and then compute A(BC). Does the Madonnas. Matrix A gives the times in hours it
(AB)C equal A(BC)? takes each robot to do each job. Matrix B tells how
many jobs of each type are required by each family.
5. For B, C, and D in Problem 3,
Compute with A and B to find a matrix showing
a. compute CB and then compute (CB)D. how long it will take each robot to do each family’s
b. compute BD and then compute C(BD). Does set of jobs.
(CB)D equal C(BD)?
Matrix of Times
6. Let
Job 1 Job 2 Job 3
1 3 2 0 1 0 3 4 2 Supremo
3 4 1 1 0 1 5 7 3 Ultramatic
AD T, BD T, AD T
4 0 1 0 1 0 1 2 1 Maxiumus
4 0 1 1 0 1 3 3 3 Gandalf
4 1 0 3
C  C 2 1 6 3S . Matrix of Jobs
Jones Smith Madonna
3 0 2 0
6 2 4 Job 1
Compute the products, if possible. If not possible, B C 8 5 4 S Job 2
explain why.
10 5 4 Job 3
a. AB b. BA c. AC
d. CA e. BC f. CB 11. Suppose that you are given the following matrices
g. A2 involving the costs of fruit at different stores, the
7. Pre-multiply each matrix by the matrix A in Prob- amounts of fruit that professors and engineers typ-
lem 6 and in each case describe how the columns of ically want, and the number of each type of person
the product matrix compare to the columns of A. in two towns.
0 0 1 0 1 0 Store 1 Store 2
a. C 0 1 0S b. C 0 0 1S 0.15 0.20 Bananas
1 0 0 1 0 0 A  C 0.25 0.15 S Peaches
1 0 0 0.20 0.25 Pears
c. C 0 3 0 S
Bananas Peaches Pears
0 0 2
6 12 4 Professors
8. For A, B, and C in Problem 6, compute the entry in B B R
6 8 5 Engineers
the third row, third column in (BC)A. Explain why
you do not have to multiply BC fully to determine
Professors Engineers
this entry in (BC)A.
2000 800 Town 1
9. Perform the matrix multiplication for Amy’s and C B R
Bill’s fruit purchases at the three different stores if 1500 1200 Town 2
the matrix of Amy’s and Bill’s needs are a. Compute a matrix product to find how much each
Peaches Pears Apples Grapefruits type of person’s fruit purchases cost at each store.
Amy 7 2 4 5 b. Compute a matrix product to find how many of
B R. each fruit will be purchased in each town.
Bill 2 5 1 8
c. Compute a matrix product to find how much
10. Suppose that you have four robots: Supremo, Ultra- was spent by each town at each store.
matic, Maximus, and Gandalf, and three types of 12. Consider the following population model for the
jobs: Job 1 (washing clothes), Job 2 (walking the dog), numbers of goats (G) and sheep (S) from year to year.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 699

10.4 Matrix Multiplication 699

G  2G  2S b. Compute A3. If the copy machine is working

S  G  3S today, what is the probability that it will be
G0 Gn working in 3 days?
Let x 0  B R be the initial vector and x n  B R be c. If it is working today, what is the probability that
S0 Sn
the vector of goats and sheep after n years. Let A be it will be working a week from today?
the matrix of coefficients in this system. (a) Write 15. Consider a weather Markov chain with 2 states,
an expression for xn in terms of A and x0 . (b) Find sunny and cloudy. If today is sunny, there is a 34
A2 and A3, and use them and your formula to de- probability that tomorrow will be sunny and a 14
termine x2 and x3 , if the starting population, in probability that tomorrow will be cloudy. If today is
2 cloudy, there is a 14 probability that tomorrow will
thousands, is x0  B R . (Check your answer by ap- be sunny and a 34 probability that tomorrow will be
5
cloudy. Write the transition matrix A for this Markov
plying the model equations for 2 and 3 years.)
chain.
(c) What does the model predict for the two popu-
lations in 10 years? a. Compute A2. What probability does the entry in
13. Consider the following population model for the position 11, 2 2 in A2 represent?
numbers of goats (G), sheep (S), and bears (B) from b. Compute A3. If today is cloudy, what is the prob-
year to year. ability that it will be sunny in three days?
c. If today is cloudy, what is the probability that it
G  G  S  B will be sunny a week from today?
S  G  2S  B 0.2 0.4
B  2G  S  B 16. Consider the transition matrix A  B R
0.8 0.6
Let 0.7
G0 and an initial vector p  B R.
0.3
x 0  C S0 S a. Calculate pand p by hand.
B0 b. Use the matrix features of your calculator to cal-
be the initial vector and let xn denote the vector of culate the next four iterates p, p, and so
goats, sheep, and bears after n years. Let A be the ma- on, corresponding to A3, A4, A5, and A6. Create a
table listing the entries in the vectors that result.
trix of coefficients in this system. (a) Write an expres-
Do they appear to be converging?
sion for xn in terms of A and x0 . (b) Find A2 and A3.
c. Plot the vectors you found in parts (a) and (b).
(c) Determine x2 and x3 , if the starting population,
Do they appear to be converging?
in hundreds, is
d. Repeat parts (a)–(c) if the initial vector is
1 0.5
x0  C 1 S . p  B R.
0.5
2 e. Repeat parts (a)–(c) if the initial vector is
(Check your answer by applying the model equa- 0.25
pB R.
tions for 2 and 3 years.) (d) How long does it take 0.75
for one of the populations to die out? f. What do you observe about the three sequences
14. The copy machine at the student union breaks of vectors from parts (b), (d), and (e)?
down as follows. If it is working today, it has a 70% 17. In Problem 22 of Section 10.3, we described how
chance of working tomorrow (and a 30% chance of to write a polynomial as a scalar product of a
breaking down). If the copy machine is broken vector of coefficients and a vector of power func-
today, it has a 50% chance of working tomorrow tions. Suppose that you now have two cubic poly-
(and a 50% chance of being broken again). Write the nomials P1x2  x3  4x2  7x  2 and Q1x2 
Markov chain transition matrix A for this scenario. 4x 3  5x 2  8x  17.
a. Compute A2. What probability does the entry in a. Express the sum of the two polynomials in terms
position 11, 22 in A2 represent? of vectors.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 700

700 CHAPTER 10 Matrix Algebra and Its Applications

b. Express the difference of the two polynomials in 0.9272 0.3746

terms of vectors. 20. The matrix R  B R is a rotation
0.3746 0.9272
18. Explain why you can’t have a power of a non-square matrix. What is the associated angle of rotation?
matrix A.
19. Prove that, if R is the 2  2 rotation matrix associ- cos 1° sin 1°
21. Consider the matrix R  B R.
ated with an angle u, then R3 is a rotation matrix as- sin 1° cos 1°
sociated with an angle 3u. (Hint: Write R3  RR2 Describe the effect of successively applying R, R2,
and use appropriate trigonometric identities to R3, R4, . . . , R90 to any nonzero vector v.
simplify the result of the multiplication.)

10.5 Gaussian Elimination

In this section we develop a procedure known as Gaussian elimination for solving
any system of linear equations. The elimination method was devised by Carl Friedrich
Gauss in about 1820 to solve systems of linear equations in astronomical and land-
surveying computations. For our purposes here, when we speak of a system of linear
equations, we assume that the number of equations equals the number of variables.
Suppose that we start with the system of linear equations
3x  5y  4z  10
3x  2y  5z  11
6x  2y  2z  10.
Gaussian elimination involves two stages. The first stage transforms the given
system of equations, or equivalently the associated matrix of coefficients, into
upper triangular form, with only 0’s below the main diagonal of the matrix:
3x  5y  4z  10
3y  z  1
z  1.
(We show how to do this shortly.) Once we have transformed the original system of
equations to upper triangular form, solving it is quite simple. The second stage of
the process uses “back substitution” to obtain values for the variables. That is, know-
ing from the third equation that z  1, we can solve for y in the second equation:
3y  11 2  1 so that y  0.
Now, knowing y and z, we can solve for x from the first equation:
3x  510 2  4112  10 so that x  2.
We can simplify this procedure by using vectors and matrices. Instead of using
x, y, and z as the variables, we use x1 , x2 , and x3 . We then write the upper triangular
form for the preceding system of equations in matrix form as Ax  b, where
3 5 4 x1 10
A  C0 3 1S , x  C x2 S , and b  C 1 S .
0 0 1 x3 1
To apply Gaussian elimination to any system of linear equations, we transform
the original coefficient matrix into upper triangular form by applying the follow-
ing three elementary row operations repeatedly.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 701

10.5 Gaussian Elimination 701

Elementary Row Operations

◆ Multiply or divide any row of a matrix (or an equation) by a nonzero
number.
◆ Add a multiple of one row (or equation) to another row (or equation).
◆ Interchange two rows (or two equations).

Realize that performing any of these operations on the rows of a matrix is

equivalent to performing the same operation on the original equations. Becasue the
operations do not materially change the equations, the equivalent operations do not
materially change the matrix, so we will get the same solution.
Because the numbers, not the variables, are what matter in the equations, we
work with the coefficient matrix A extended by a column for the right-hand side
vector b. This matrix [A  b] of coefficients along with the right-hand side vector
is called the augmented coefficient matrix. For instance, if the system of equa-
tions is
2x  3y  7
4x  5y  3
so that
2 3 7
AB R and b  B R ,
4 5 3

then the augmented matrix is

2 3 7
[A  b]  B 2 R.
4 5 3
In Example 1 we show the steps involved in applying Gaussian elimination to
both the system of linear equations and the associated augmented matrix.
E XAMPLE 1
Use Gaussian elimination to solve the following system of two equations in two unknowns.
2x  y  7 (1)
x  2y  4 (2)
Solution This system is equivalent to the augmented matrix
2 1 7
B 2 R.
1 2 4
To eliminate the x-term from Equation (2), we add  12 times Equation (1) to Equation
(2) (the second elementary row operation). The result is a new second Equation (2 ):

12 2 x  2y  4
 12 11 2 x  0.5y  3.5
12 2  12 2  12 11 2 0  2.5y  7.5

Equivalently, we perform the identical operation on the corresponding rows of the

Gord.3896.10.pgs 4/28/03 3:30 PM Page 702

702 CHAPTER 10 Matrix Algebra and Its Applications

augmented matrix: Add  12 times the first row to the second row. Our new system of
equations and the new augmented matrix become

2x  y  7 (1)
2.5y  7.5, (2')
and
2 1 7
B 2 R.
0 2.5 7.5
Note that any solution to Equations (1) and (2) is also a solution to Equations (1) and 12 2
because we can reverse the step that created Equation 12 2. That is, 12 2  12 2  12 11 2 im-
plies that 122  12 2  12 11 2 . Thus Equation (2) is formed from Equation 12 2 and a mul-
tiple of Equation (1), and so any solution to Equations (1) and 12 2 is also a solution to
Equations (1) and (2). But Equation 12 2 is trivial to solve, and gives y  3. Substituting
y  3 into Equation (1), we get
17  3 2
2x  3  7, so x   2.
2
Verify that x  2 and y  3 satisfy the two original Equations (1) and (2).
◆
When solving a system of linear equations, you should always check your re-
sult by substituting the values for the variables into the original equations, not the
transformed equations, in case you made a mathematical error along the way.
Note also that the work we did on the system of equations exactly parallels
what happens with the augmented matrix. Eventually, we will dispense with the
equations altogether and work exclusively with the augmented matrix because it
eliminates writing and keeping track of the variables at every step.

A Geometric Interpretation
If there are only two equations in two unknowns, you can also solve the system
graphically by plotting the two lines. The first equation represents all points on one
line and the second equation represents all points on the second line. Thus the so-
lution to the system of equations corresponds to the point of intersection and you
can approximate this point with your graphing calculator.
There is a comparable geometric interpretation for a system of three linear
equations in three unknowns. Just as an equation of the form ax  by  d repre-
sents a line in the two-dimensional coordinate plane, an equation of the form
ax  by  cz  d represents a plane in three-dimensional space. When you have
three equations in three unknowns, you actually have three different planes in
space, as suggested in Figure 10.35. Visualize, for instance, the ceiling, the wall in
front of you, and the wall to your left. These three planes intersect at a single point
in the upper corner of the room to your left. This point of intersection also is the
solution of the system of three equations. Unfortunately, graphing calculators
cannot yet use this geometric interpretation to solve such a system of equations.
Also, there can be some complications: You know that two lines can be parallel, so
the resulting system of two equations in two unknowns will not have a solution. Sim-
ilarly, three planes don’t necessarily have a common point of intersection. Visualize
Gord.3896.10.pgs 4/28/03 3:30 PM Page 703

10.5 Gaussian Elimination 703

FIGURE 10.35

the ceiling, the floor, and the wall in front of you—they don’t meet at a single point.
Or, picture a long triangular prism made up of three flat sides that likewise have no
single point of intersection. Alternatively, three planes can all pass through a common
line of intersection, as with a revolving door. In that case the system of linear equa-
tions has infinitely many solutions. We consider such cases later in Example 3.

The Clothes Production Model and Gaussian Elimination

Recall the clothes production model introduced in Section 10.2 with three clothing
factories whose raw material production levels had to be chosen to meet the de-
mands for vests, pants, and coats.
Vests: 20x1  4x2  4x3  500 (3)
Pants: 10x1  14x2  5x3  850 (4)
Coats: 5x1  5x2  12x3  1000 (5)

E XAMPLE 2
Set up the corresponding augmented matrix and use Gaussian elimination to solve the
system of equations.
Solution The augmented matrix is

20 4 4 500
C 10 14 5 3 850 S .
5 5 12 1000

To solve this system, we first use multiples of Equation (3) to eliminate x1 from Equa-
tions (4) and (5). Because 10x1 is half of 20x1 in Equation (3), we add  12 times Equation
(3) to Equation (4) to cancel the terms  12 120x1 2 and 10x1 and so get a new second
Equation 14 2 .

142 10x1  14x2  5x3  850

 12 13 2 10x1  2x2  2x3  250
14 2  14 2  1 12 2 13 2 0  12x2  3x3  600
Gord.3896.10.pgs 4/28/03 3:30 PM Page 704

704 CHAPTER 10 Matrix Algebra and Its Applications

Similarly, we add  14 times Equation (3) to Equation (5) to eliminate the x1-term from
Equation (5) and get a new Equation 15 2.

15 2 5x1  5x2  12x3  1000

 14 13 2 5x1  1x2  1x3  125
15 2  15 2  1 14 2 13 2 4x2  11x3  875

Our new system of equations is now

20x1  4x2  4x3  500 (3)
12x2  3x3  600 (4')
4x2  11x3  875, (5')
and the corresponding augmented matrix is

20 4 4 500
C 0 12 3 3 600 S .
0 4 11 875

Next we use Equation 14 2 to eliminate the x2-term from Equation 15 2 by adding

 13 times Equation 14 2 to Equation 15 2. We obtain a new Equation 15
2.

15 2 4x2  11x3  875

 13 14 2 4x2  1x3  200
15
2  15 2  1 13 2 14 2 10x3  675

Our final system of equations in upper triangular form then is

20x1  4x2  4x3  500 (3)
12x2  3x3  600 (4')
10x3  675 (5")
with the associated augmented matrix
20 4 4 500
C 0 12 3 3 600 S .
0 0 10 675
Any solution to the original system is also a solution to the new system. Further-
more, reversing the steps used in going from the original system to the final system (so
that the original system is formed from linear combinations of the equations in the final
system), we see that any solution to the new system is also a solution to the original sys-
tem. The final system is in upper triangular form, so we can solve it using back substitu-
tion. From Equation 15
2 , we have
675
x3   67.5.
10
Substituting this value into Equation 14 2 yields
12x2  3167.5 2  600, so 12x2  600  202.5, or x2  33.125.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 705

10.5 Gaussian Elimination 705

Next substituting x3  67.5 and x2  33.125 into Equation (3) gives

20x1  4133.1252  4167.52  500

so that
500  132.5  270
x1   4.875.
20
Therefore the vector of the number of rolls of cloth needed by each of the three clothing
factories is
x1 4.875 4 78
C x2 S  C 3.125 S  C 33 18 S .
x3 67.5 67 12
◆
Clearly in practice, cloths come in full rolls, so a more realistic solution would in-
volve rounding up to the next full roll of cloth.
Incidentally, you can apply the three elementary row operations in many differ-
ent ways to solve a particular system of equations. However, although the steps you
use might differ from those of others solving the same problem, you should all ob-
tain the same solution in the end, if none of you have made any mathematical errors.
Most graphing calculators have a built-in routine for applying Gaussian elimina-
tion to any set of linear equations. Typically, you would enter the coefficients, press
the SOLVE key, and the calculator will respond with the solution or will indicate that
either there is not a unique solution or no solution exists. (The specific key operations
differ from one machine to another, so check your manual for details.) From one
point of view, this approach is extremely simple because you get the answer instantly.
However, it does have the drawback of not letting you see how the method works or
understand what went wrong when the method fails, as we discuss below.

Systems of Linear Equations with Multiple Solutions

In our discussion regarding the geometric interpretation of systems of three equations
in three unknowns, we mentioned that it is possible that three planes can have many
points of intersection, as in a revolving door. We illustrate this type of situation in
Example 3 where we change the number on the right-hand side of the third equation
in Example 2 from 1000 to 325 and the coefficient of x3 from 12 to 2.

E XAMPLE 3
Apply Gaussian elimination to the system of linear equations
20x1  4x2  4x3  500 (6)
10x1  14x2  5x3  850 (7)
5x1  5x2  2x3  325. (8)
Solution The corresponding augmented matrix is

20 4 4 500
C 10 14 5 3 850 S .
5 5 2 325
Gord.3896.10.pgs 4/28/03 3:30 PM Page 706

706 CHAPTER 10 Matrix Algebra and Its Applications

After eliminating x1 from Equations (7) and (8), we have

162 20x1  4x2  4x3  500

17 2  17 2  12 16 2 12x2  3x3  600
18 2  18 2  14 16 2 4x2  1x3  200,

along with the associated augmented matrix

20 4 4 500
C 0 12 3 3 600 S .
0 4 1 200
Next we add  13 times Equation 17 2 to Equation 18 2 to eliminate the x2-term:
16 2 20x1  4x2  4x3  500
17 2 12x2  3x3  600
18
2  18 2  13 17 2 0x3  0
The corresponding augmented matrix is
20 4 4 500
C 0 12 3 3 600 S .
0 0 0 0
But this process eliminates the x3 term, and simultaneously the constant term on the
right becomes 0. Also, note that Equation 18 2 is equivalent to 13 times Equation 17 2. As
a consequence, we actually have only two equations in three unknowns. This system has
infinitely many solutions because we can choose any value for x3 and then use back sub-
stitution to determine x2 and x1 , based on our choice for x3 . For instance, if we choose
x3  0, Equation 17 2 gives 12x2  310 2  600, so x2  50, and then Equation (6) gives
20x1  41502  410 2  500, so x1  15. Alternatively, if we choose x3  40, eventually
Equation 17 2 gives x2  40, and so Equation (6) gives x1  9. That is, we have obtained
two very different solutions to the same system of equations. In fact, had we selected any
other positive value for x3 , we would have obtained still another solution to the system
(but not necessarily to the original problem). Thus, geometrically, the corresponding
planes in three dimensions all pass through a common line of intersection, and every
point on this line is a solution to the system.
◆
Incidentally, if you attempted to solve the system of equations in Example 3 on
a graphing calculator, say, you would get an error message of the form SINGULAR
MATRIX. We discuss what this message means later in this section.

Systems of Linear Equations with No Solutions

We pointed out in our discussion of the geometry of systems of equations that
three planes may have no common point of intersection, as with the floor, ceiling
and one wall in a room. We illustrate this situation in Example 4 by making anoth-
er minor change in the value on the right-hand side of the third equation in the
original system in Examples 2 and 3.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 707

10.5 Gaussian Elimination 707

E XAMPLE 4
Apply Gaussian elimination to the system of linear equations
20x1  4x2  4x3  500 (9)
10x1  14x2  5x3  850 (10)
5x1  5x2  2x3  1000. (11)
Solution The corresponding augmented matrix is

20 4 4 500
C 10 14 5 3 850 S .
5 5 2 1000
Eliminating x1 as before, we get
20x1  4x2  4x3  500 (9)
12x2  3x3  600 (10')
4x2  1x3  875. (11')
The associated augmented matrix is

20 4 4 500
C 0 12 3 3 600 S .
0 4 1 875
We now use Equation 110 2 to eliminate the x2-term in Equation 111 2 to get
(9) 20x1  4x2  4x3  500
110 2 12x2  3x3  600
111
2  111 2  1 13 2 110 2 0x3  675,

along with the augmented matrix

20 4 4 500
C 0 12 3 3 600 S .
0 0 0 675

The two Equations 110 2 and 111 2 are called inconsistent equations because they lead to
the impossible Equation 111
2: 0  675. Hence the original system has no solution.
Geometrically, this outcome indicates that the corresponding planes in three dimen-
sions do not have a common point of intersection.
◆
If you attempted to solve this system of equations on a graphing calculator, say,
you would again get an error message about a SINGULAR MATRIX.
In the real world, the inconsistency that occurred in Example 4 would be re-
solved by increasing one of the right-hand side demands (thus producing an excess
amount of one type of clothing).
We now summarize the steps of Gaussian elimination.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 708

708 CHAPTER 10 Matrix Algebra and Its Applications

Gaussian Elimination
1. Add multiples of the ith equation, for i  1, 2, . . . , n  1, to the
remaining equations to eliminate the ith variable from the other
equations.
2. Solve the resulting upper triangular system of equations, using back
substitution.

If, in the process, the coefficient of xi in the ith equation 1i  1, 2, . . . , n  1 2 is

zero and the coefficient of xi in one of the following equations is nonzero, simply
interchange the two equations.

Using the Inverse Matrix

We can use a related approach with matrices to solve systems of equations. Let’s
begin with the matrices
1 0 1 0
AB R and B  B 1 R.
4 2 2 2
Their product is
1 0 1 0 1 0
AB  B RB 1R B R.
4 2 2 2 0 1

1 0
The matrix B R is called the 2  2 identity matrix and is denoted by I2 .
0 1
For any 2  2 matrix A,
AI 2  I 2A  A,
so I 2 plays the same role for 2  2 matricies as the number 1 plays in arithmetic:
a . 1  1 . a  a. Similarly, the identity matrix for 3  3 matrices is
1 0 0
I3  C 0 1 0S ,
0 0 1
and so on. The preceding matrices A and B are said to be inverses of each other. We
write B  A1, and the product of the matrix and its inverse is the identity matrix

AA1  I2  A1A.

The inverse of a matrix can be extremely useful in solving systems of linear equations.
Not every matrix has an inverse. A matrix that does not have an inverse is called
a singular matrix. (Note that this is the same term that is used in the error message
on graphing calculators when a system of equations has no solution or multiple so-
lutions.) A square matrix A has an inverse if, when we use Gaussian elimination to
reduce it to upper triangular form, all the diagonal elements are nonzero. However, if
any of the diagonal elements are zero, no inverse exists and the matrix is singular.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 709

10.5 Gaussian Elimination 709

Suppose now that we have the matrix-vector equation Ax  b and that the
n  n matrix A is not singular, so its inverse A1 exists. Matrix A can be any size, so
we denote the corresponding identity matrix by I without designating its size.
Then we can left-multiply the equation by this inverse matrix and obtain
A1Ax  A1b
Ix  A1b
x  A1b,
which automatically gives the desired solution vector x.
Note that we haven’t discussed how to calculate the inverse of a matrix A. We
can do so by using an extension of the Gaussian elimination process, but that is
somewhat outside the scope of what we want to focus on here. Instead, note that
your calculator will do this for you. Enter a square matrix A, call it up and press the
x-1 key, and your calculator will give you the inverse matrix A-1 if it exists. If the in-
verse does not exist, you will get an error message. Multiply this inverse matrix and
the vector of constants b to get the desired solution vector x  A1b.
Some of you may have seen a method called Cramer’s rule, which uses
determinants for solving a system of linear equations. Although fairly effective for
solving systems of two or three equations, Cramer’s rule is very inefficient for larg-
er systems. Suppose, for instance, that you use a relatively slow computer that can
perform only 1 million operations per second. To solve a system of 20 equations in
20 unknowns with Cramer’s rule would take it about 77,000 years! The same com-
puter could solve that system in about 0.003 second using Gaussian elimination or
matrix methods.

Applications of Gaussian Elimination

We now consider a series of further applications involving systems of equations.
We begin by using Gaussian elimination to investigate the long-term behavior of
Markov chains.

Steady State of the Markov Chain for the Stock Market Consider again the
stock market Markov chain introduced in Section 10.2 with the transition matrix
Market Today
Up Down Same
1 1 1
Up 4 2 4
Market
Down C 2
1 1
4 2S
1
 A.
Tomorrow
Same 14 1
4
1
4

We noted that over many time periods, the successive probability vectors con-
verged to 0.35 for the market going up, 0.40 for the market going down, and 0.25
for the market staying the same. We confirm the fact that
0.35
p  C 0.40 S
0.25
is the steady state or the equilibrium state of the Markov chain by showing that if
the market ever reaches this state on one day, the vector p  Ap for the following
day will be the same as p. That is, at the steady state, there is no subsequent change
Gord.3896.10.pgs 4/28/03 3:30 PM Page 710

710 CHAPTER 10 Matrix Algebra and Its Applications

in the values for the probabilities. Using a calculator to perform the matrix–vector
product gives
1 1 1
4 2 4 0.35 0.35

p  Ap  C 21 1
4 2S
1
C 0.40 S  C 0.40 S  p.
1 1 1
4 4 4 0.25 0.25

In Example 5, we show how to find the equilibrium state, denoted by p*, for a
Markov chain exactly.

E XAMPLE 5
Find the equilibrium state p* for the preceding matrix A.
Solution We want a vector p* that satisfies the matrix-vector equation p*  Ap*. Let
the components of p* be denoted by p1* , p2* , and p3* . Note that p1*  p2*  p3*  1
ˇ ˇ ˇ ˇ ˇ ˇ

because the sum of all the probabilities associated with an event must sum to 1.
Because p*  Ap*, we have

4 p1*  2 p2*  4 p3*

1 1 1
p1*
C p2* S  C 12 p1*  14 p2*  12 p3* S .
4 p1*  4 p2*  4 p3*
1 1 1
p3*
Collecting the variables on the left, we get the following system of three equations in the
three unknowns p1* , p2 * , and p3 * :
ˇ

3
4 p1* ˇ  12 p2*  14 p3*  0 ˇ ˇ

 12 p1* ˇ  3
4 p2* ˇ  1
2 p3* ˇ 0
 14 p1* 
ˇ
1
4 p2* ˇ  3
4 p3* ˇ  0.
Solving this system by Gaussian elimination applied to the augmented coefficient ma-
trix, we obtain
3
4  12  14 0 3
4  12  14 0 3
4  12  14 0
C  12 3
4  12 3 0 S ⇒ C0 5
12  23 3 0 S ⇒ C 0 5
12  23 3 0 S .
 14  14 3
4 0 0  125 2
3 0 0 0 0 0
To eliminate the fractions, we multiply the first row by 4 and the second row by 12:

3 2 1 0
C0 5 8 3 0 S .
0 0 0 0
This augmented matrix is equivalent to the reduced system of equations
3p1*  2p2*  p3*  0
ˇ ˇ ˇ

5p2*  8p3*  0. ˇ ˇ

Because there are only two equations in the three unknowns, there is no unique solu-
tion. Instead, we can solve for any two of the variables in terms of the third—say, p3* . ˇ

The second equation then gives

8
p2*  p3* ˇ ˇ

5
Gord.3896.10.pgs 4/28/03 3:30 PM Page 711

10.5 Gaussian Elimination 711

and, when we substitute it into the first equation and solve for p1* , we have ˇ

2p2*  p3* 218>52 p3*  p3* 7

p1*    p3* .
ˇ ˇ ˇ ˇ

ˇ ˇ

3 3 5
Where is our unique vector of stable probabilities? In this solution we obtained infi-
nitely many solutions—one for each value of p3* . We now use the fact that the sum of
ˇ

the three probabilities must equal 1, or p1*  p2*  p3*  1. We substitute the expres-
ˇ ˇ ˇ

sions for p1* and p2* in terms of p3* to obtain

ˇ ˇ ˇ

7 8 20
p1*  p2*  p3*  p3*  p3*  p3*  p3*  4p3*  1,
ˇ ˇ ˇ ˇ ˇ ˇ ˇ ˇ

5 5 5
so that p3*  0.25. Therefore
ˇ

10.25 2  0.40 and p1*  10.25 2  0.35.

8 7
p2* 
ˇ ˇ

5 5
These probabilities are identical to those in the equilibrium state vector of the market
Markov chain given previously.
◆
Systems of linear equations of the form Ax  0 with a zero vector, 0, on the
right-hand side, such as the one in Example 5, occur frequently in matrix alge-
bra. They are called homogeneous systems of linear equations. In comparison, sys-
tems of equations of the form Ax  b, where b  0, are called nonhomogeneous
systems.
When solving a homogeneous system, we usually are interested in a nonzero
solution, as was the case here. Note, however, that x  0 is always a solution to any
homogeneous system Ax  0. Thus, if we are to get a nonzero solution, we must
have a case of multiple solutions.

Applications to Analytic Geometry Next, we apply Gaussian elimination to

solve some systems of linear equations that arise in analytic geometry.
One of the most basic objects in geometry is a line, whose equation can be
written as y  ax  b. Suppose that we have two points, P with coordinates 13, 1 2
and Q with coordinates 16, 7 2, and we want to find the equation of the line through
these points. We could solve this problem by finding the slope of the line segment
joining the two points and then using the point–slope formula. Here, however, we
solve this problem using systems of equations to illustrate some important mathe-
matical ideas that can be extended to answer considerably more complicated ques-
tions. The results either way are the same.

E XAMPLE 6
Use matrix methods to find the equation of the line through the points 13, 1 2 and 16, 72.
Solution First, let’s be clear about what we need to find. The equation of the line
y  ax  b is determined by two constants—the slope a and the vertical intercept b. Al-
though a and b are constants for a particular line, we can think of them as parameters
that distinguish one line from another. We need to determine appropriate values for a
and b so that the line y  ax  b passes through the points 13, 1 2 and 16, 7 2. According-
ly, a and b must satisfy the following two equations.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 712

712 CHAPTER 10 Matrix Algebra and Its Applications

For point 13, 1 2: 1  a13 2  b

For point 16, 72 : 7  a16 2  b
We rewrite these two equations in standard form to get

3a  b  1
6a  b  7.
Subtracting twice the first equation from the second equation yields

b  5, so that b  5.
Substituting back into the second equation gives
6a  15 2  7, so 6a  12, and a  2.
Therefore the equation of the desired line is
y  2x  5.
◆

We now extend the ideas in Example 6 to find the equation of a parabola

y  ax2  bx  c. We need to determine the values of the parameters a, b, and c.
To solve for these three unknowns, we need three linear equations. As with the
line, we get a linear equation associated with each point the parabola goes
through. For three equations, we need three points. Suppose that the three points
are 12, 1 2, 14, 3 2 and 11, 8 2 . Note that the three points must be noncollinear
(i.e., they cannot lie on a common line). Each point determines an equation
when we substitute its coordinates, x and y, into the parabola’s formula
y  ax2  bx  c.
For point 12, 12 : 1  a12 2 2  b12 2  c
For point 14, 32: 3  a14 2 2  b14 2  c
For point 11, 82 : 8  a11 2 2  b11 2  c
Hence we have the system of linear equations
4a  2b  c  1
16a  4b  c  3
a  b  c  8.
In Problem 17 we ask you to solve these three equations to determine a, b, and c.
Next, we apply this curve-fitting method to a more complicated situation—
namely, finding the equation of a circle. The familiar form of the equation of a
circle is
1x  a2 2  1y  b 2 2  c 2,
where 1a, b2 is the center of the circle and c is the radius. It turns out that, just as
three noncollinerar points determine a parabola, any three noncollinear points
also determine a circle. We now have three parameters a, b, and c to determine,
which should lead to three equations in the three unknowns. But the equations
arising from giving particular values to x and y in the above equation for the circle
Gord.3896.10.pgs 4/28/03 3:30 PM Page 713

10.5 Gaussian Elimination 713

will not be linear (there will be a2 and b2 terms), so we cannot use Gaussian elimi-
nation to determine a, b, and c based on the equation 1x  a2 2  1y  b2 2  c 2.
However, if we expand this expression algebraically, we get
1x  a2 2  1y  b 2 2  1x 2  2ax  a2 2  1 y 2  2by  b2 2
 x 2  y 2  2ax  2by  a2  b2  c 2.
Therefore we can transform the equation of the circle to the form
x2  y2  Cx  Dy  E  0,
where C  2a, D  2b, and E  a2  b2  c2 are three other parameters.
Using this form, we can set up three linear equations in these three unknowns and
solve for them with Gaussian elimination. Once we have their values, we can
rewrite the equation of the circle in the more familiar form and read the coordi-
nates of the center and the radius. We illustrate these ideas in Example 7.

E XAMPLE 7
Find the equation of the circle passing through the points 14, 2 2, 16, 2 2 and 15, 5 2 .
Solution Using the equation

x2  y2  Cx  Dy  E  0,
we create three linear equations that C, D, and E must satisfy.

For the point 14, 2 2: 42  12 2 2  C14 2  D12 2  E  0

For the point 16, 22 : 62  22  C16 2  D12 2  E  0
For the point 15, 52 : 52  52  C15 2  D15 2  E  0
Moving the constant terms to the right-hand side in each equation, we have
4C  2D  E  20 (12)
6C  2D  E  40 (13)
5C  5D  E  50. (14)
We now apply Gaussian elimination to this (nonhomogeneous) system of three linear equa-
tions in three unknowns. First, to eliminate C from Equations (13) and (14), we subtract 32 of
Equation (12) from Equation (13) and subtract 54 of Equation (12) from Equation (14).
(12) 4C  2D  E  20
113 2  113 2  32 112 2
1
5D  E  10
2
114 2  114 2  54 112 2
15 1
D  E  25
2 4
Next we eliminate D from the last equation by subtracting 3
of Equation 113 2 from
Equation 114 2.
2

(12) 4C  2D  E  20
1
113 2 5D  E  10
2
114
2  114 2  32 113 2
1
E  10
2
Gord.3896.10.pgs 4/28/03 3:30 PM Page 714

714 CHAPTER 10 Matrix Algebra and Its Applications

From Equation 114

2 , we have E  20. Substituting back into Equation 113 2 yields

120 2  10, so 5D  20 and D  4.

1
5D 
2
Then substituting back into Equation (12) yields
4C  214 2  120 2  20, so 4C  8 and C  2.
Therefore one form of an equation for our circle is
x2  y2  2x  4y  20  0.
We rewrite this equation in the more familiar form by completing the square on both x
and y (see Appendix A8):
3 1x 2  2x 2 4  3 1 y2  4y2 4  20  0
3 1x 2  2x  1 2  1 4  3 1 y2  4y  4 2  4 4  20  0
3 1x  1 2 2  1 4  3 1 y  2 2 2  4 4  20  0
1x  1 2 2  1y  2 2 2  1  4  20  25.
Thus our circle is centered at the point 11, 2 2 and has radius 5, as shown in Figure 10.36.
y

(5, 5)

(1, 2) (6, 2)

(4, –2)

FIGURE 10.36

◆
Think About This Substitute the coordinates of the three points in Example 7 into the usual equation
for a circle, 1x  a2 2  1 y  b 2 2  c2, to see the types of equations involving the
three parameters a, b and c that would result. ❐

Problems
1. Solve each system of equations using Gaussian variations may have no solution, some may have
elimination. multiple solutions (express such an infinite family
a. x  y  8 b. x  4y  3 of solutions in terms of x3), and some may have an
2x  3y  4 x  2y  9 unrealistic solution involving negative values.
c. 2x  3y  4 d. 3x  2y  3 a. 20x1  4x2  4x3  500
4x  y  1 2x  5y  2 8x1  3x2  5x3  850
e. 3x  2y  3 4x1  5x2  11x3  2050
4x  3y  2 b. 6x1  5x2  6x3  500
2. Use Gaussian elimination to solve each variation on 10x1  10x2  850
the clothes production model in Example 2. Some 2x1  12x3  1000
Gord.3896.10.pgs 4/28/03 3:30 PM Page 715

10.5 Gaussian Elimination 715

c. 6x1  2x2  2x3  500 c. 2x  5y  3

3x1  6x2  3x3  300 6x  15y  9
3x1  2x2  6x3  1000 d. x1  2x2  3x3  10
d. 8x1  4x2  3x3  500 2x1  x2  4x3  20
4x1  8x2  5x3  500 5x2  2x3  0
12x2  6x3  500 e. x1  5x2  4x3  10
3. Solve each system of equations using Gaussian x1  3x2  6x3  9
elimination. x1  x2  x3  5
a. x1  2x2  x3  3 f. x1  2x2  x3  0
x1  2x2  3x3  1 2x1  x2  3x3  0
2x1  x2  2x3  2 x1  x2  2x3  0
b. 2x1  x2  x3  2 6. For each augmented matrix, state whether the asso-
x1  2x2  2x3  3 ciated system of equations has a unique solution,
3x1  2x2  x3  0 multiple solutions, or no solution.
c. x1  3x2  2x3  1 3 3 1 0 1 2 2 3
5x1  4x2  6x3  12 a. C 0 1 5 3 1S b. C 0 5 1 3 6S
2x1  x2  3x3  4 0 0 0 2
0 0 1 2
d. 2x1  4x2  2x3  4
2x1  2x2  3x3  4 1 2 7 5 4 2 3 3
x1  x2  2x3  1 c. C 0 4 2 3 1S d. C 0 4 2 3 6S
e. x1  2x2  2x3  3 0 0 0 0 0 2 1 3
x1  2x2  3x3  11
2x1  x2  2x3  5 2 1 3 0
e. C 0 1 0 3 0S
f. x1  x2  2x3  2
3x1  2x2  4x3  2 0 0 1 0
3x1  x2  2x3  3
7. (Continuation of Problem 4 of Section 10.2) Con-
4. In each set of three equations, show that the third
sider the following clothes production model. There
equation equals some multiple r of the first equation
added to the second equation: 13 2  r . 11 2  12 2 .
are three clothing factories (1, 2, and 3) and from
each roll of cloth, the different factories produce the
a. 11 2 x  2y  4 following numbers of vests, pants, and coats.
122 3x  y  9
132 x  3y  1 Factory 1 Factory 2 Factory 3
b. 112 x  y  z  2 Vests 6 4 2
122 x  y  z  3 Pants C 4 8 4S
132 2x  4y  4z  3 Coats 3 2 8
c. 112 2x  y  2z  5
122 3x  y  z  8
Suppose that the demand is for 400 vests, 800 pants,
132 6x  0.5y  2z  0.5
and 500 coats. Write a system of equations whose
solution would determine production levels (rolls
5. Determine whether each system of equations has a of cloth needed by each factory) to yield the desired
unique solution, multiple solutions, or is inconsis- numbers of vests, pants, and coats. Find the solu-
tent (has no solution). tion using Gaussian elimination.
a. x  2y  3 8. (Continuation of Problem 5 of Section 10.2) From
3x  6y  9 each shipment of crude petroleum, Refineries 1, 2,
b. x  2y  3 and 3 produce the following amounts (in thousands
3x  6y  4 of gallons) of heating oil, diesel oil, and gasoline.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 716

716 CHAPTER 10 Matrix Algebra and Its Applications

Refinery 1 Refinery 2 Refinery 3 morrow. Construct a Markov chain for this problem
Heating Oil 8 5 3 and find the equilibrium state.
Diesel Oil C 2 5 5S 13. From past experience, the Pins bowling team knows
Gasoline 3 7 6 that if they win this week’s game, they have a 32
chance of winning next week’s game. If they lose
Suppose that the demand is for 6200 thousand gal- this week’s game, they have a 12 chance of winning
lons of heating oil, 4000 thousand gallons of diesel next week’s game. Construct a Markov chain for
oil, and 4700 thousand gallons of gasoline. Write a this problem and find the stable distribution.
system of equations whose solution would deter- 14. Find the equilibrium state for the Markov chains
mine production levels to yield the desired with the following transition matrices.
amounts of heating oil, diesel oil, and gasoline. Find
1 2 1 1 1 1
the solution using Gaussian elimination. 3 3 3 4 4 2
9. (Continuation of Problem 6 of Section 10.2) The a. C 0 0 3S
1
b. C 41 1
2 4S
1

staff dietitian at the California Institute of Trig- 2

3
1
3
1
3 2
1 1
4
1
4
onometry has to make up a meal with 600 calories, 1 1
20 grams of protein, and 200 mg of vitamin C. The 0 2 2
three food types that the dietitian can choose from c. C 21 0 1
2 S
are gelatin, fish sticks, and mystery meat. They have 1 1
2 2 0
the following nutritional content per ounce.
15. The following model for learning a concept over a set
Gelatin Fish Sticks Mystery Meat of lessons identifies four states of learning: I  Igno-
Calories 10 50 200 rance, E  Exploratory thinking, S  superficial
Protein C 1 3 0.2 S understanding, and M  Mastery. If you are now
Vitamin C 30 10 0 in State I, after one lesson you have probability 12 of
still being in I and probability 21 of being in E. If you
Describe the dietitian’s problem and create a are now in State E, you have probability 41 of being
mathematical model for it with a system of three in I, 12 in E, and 41 in S. If you are now in State S, you
linear equations. Find the solution using Gaussian have probability 41 of being in E, 12 in S, and 14 in M. If
elimination. you are in State M, you always stay in M (with prob-
10. (Continuation of Problem 7 of Section 10.2) A ability 1). Construct a Markov chain for this prob-
company has a budget of $280,000 for computing lem and find the equilibrium state.
equipment. The types of equipment available are 16. Find the equation of the line passing through each
microcomputers at $2000 each, terminals at $500 pairs of points, using Gaussian elimination.
each, and workstations at $5000 each. There should a. 11, 1 2, 12, 3 2 b. 11, 02, 10, 22
be five times as many terminals as microcomputers c. 12, 3 2, 11, 1 2 d. 12, 12, 13, 12
and twice as many microcomputers as worksta- e. 11, 1 2 , 12, 5 2 f. 14, 12, 11, 1 2
tions. Set up a system of three linear equations for
17. Solve the three equations for determining the param-
this situation. Find the solution using Gaussian
eters of the parabola y  ax2  bx  c that passes
through the three points 12, 1 2, 14, 3 2 and 11, 8 2.
elimination.

18. A parabola passes through the points 11, 42, 11, 2 2

11. Find the equilibrium state for the Markov chains
and 14, 72 . Set up a system of linear equations in a,
with the following transition matrices.
2 1 3 1 1
1 b, and c and then solve it, using Gaussian elimina-
a. B 13 2R
3
b. B 41 3R
4
c. B 1R
2

3 3 4 4 0 2 tion, to find the equation of the parabola.

12. The copy machine at the student union breaks 19. Find an equation of the circle passing through each
down with the following pattern. If it is working triple of points, using Gaussian elimination.
today, there is a 70% chance that it works tomorrow a. 11, 0 2, 10, 1 2, 11, 12
(and a 30% chance of breaking down). If it is bro- b. 12, 1 2, 12, 3 2, 10, 1 2
ken today, there is a 50% chance that it works to- c. 13, 1 2, 12, 5 2, 13, 6 2
Gord.3896.10.pgs 4/28/03 3:30 PM Page 717

Chapter Summary 717

d. 11, 12, 11, 3 2, 12, 12 n n n

e. 10, 0 2, 14, 1 2, 14, 1 2 a a xi2b a  a a xib b  a a xi yib .

i1 i1 i1
20. Just as two distinct points determine a line unique-
Recall that the summations indicate adding up all
ly, three noncollinear points determine a para-
the x-values, adding up all the y-values, adding up
bola. Moreover, four noncollinear points, all of
the squares of all the x-values, and adding up all the
which do not lie on a parabola, determine a cubic
products of the x and y-values. Note that all the
polynomial.
summation terms are known constants once the x’s
a. A parabola y  ax2  bx  c passes through and y’s have been given.
the points 11, 52, 11, 2 2, and 14, 62 . Set up a
a. For the set of data 11, 11 2, 12, 25 2, 13, 33 2,
14, 45 2, 15, 57 2, and 16, 652 , find the equation of
system of linear equations in a, b, and c and then
solve it, using Gaussian elimination, to find the
the regression line using your graphing calcula-
equation of the parabola.
tor or appropriate software package.
b. A cubic polynomial y  ax3  bx2  cx  d
passes through the four points 11, 4 2, 11, 2 2,
b. Calculate the sums needed in the two preceding
14, 7 2, and 15, 2 2. Set up a system of linear equa-
equations by completing the entries in the table.
tions in a, b, c, and d that could be used to find
the equation of the cubic. Then solve the system x y x2 xy
using Gaussian elimination.
c. How does your answer to part (b) compare to 1 11
the result you can obtain with your calculator 2 25
using the routine for fitting a cubic function to a
set of data? 3 33

1 0 4 45
21. Find the inverse of the matrix A  B R
4 2 5 57
algebraically. (Hint: Write the inverse as
a b 6 65
A1  B R and use the fact that AA1  I2 to
c d gx  gy  gx 2  gxy 
solve for a, b, c, and d.) How does your answer com-
pare with what your calculator shows using its ma-
c. Use the results of part (b) to write the system of
trix features?
linear equations in a and b that you can use to
22. a. Consider the rotation matrix determine the values for the unknown coeffi-
cos u sin u cients a and b in the regression equation.
RB R. d. Solve the system of equations in part (c) using
sin u cos u
Gaussian elimination. How do your results com-
Based on geometric principles, predict what the pare to what you obtained directly in part (a)?
inverse matrix R1 has to be. e. In Chapter 3 we suggested that you scale down
b. Prove algebraically that the matrix you construct- large numbers, such as the full years 2000, 2001,
ed in part (a) is indeed the inverse matrix for R. 2002, . . . , in a set of data. Based on your calcula-
23. The method of linear regression discussed in Chap- tions in part (b), explain why doing so is desir-
ter 3, in which the best fit line y  ax  b is con- able. In particular, what might happen if there
structed for a set of 1x, y2 data, is based on solving a were many data values, if the x’ s, say, consisted
system of linear equations for the unknown param- of full years and the y’ s were also large numbers?
24. Repeat parts (a)–(d) of Problem 23 for 10, 24 2,
eters a and b. It can be shown that, if the set of n data
points is 1x1 , y1 2, 1x2 , y2 2, 1x3 , y3 2 , . . . , 1xn , yn 2 , then
15, 21 2, 110, 172, 115, 142, 120, 122, and 125, 92.
a and b must satisfy the system of linear equations
n n
a a xib a  nb  a a yib
i1 i1
Gord.3896.10.pgs 11/21/03 11:05 AM Page 718

718 CHAPTER 10 Matrix Algebra and Its Applications

Chapter Summary

In this chapter we introduced vectors and matrices and some of their properties
and applications. In particular, we emphasized
◆ What a vector is geometrically, algebraically, and as an ordered pair or or-
dered triple of numbers.
◆ What a matrix is algebraically and as an array of numbers.
◆ How to use vectors and matrices to construct mathematical models, including
Markov chains and growth models, of a wide variety of real-world problems.
◆ How to use vectors and matrices to rotate geometric figures.
◆ How to add and subtract vectors and matrices.
◆ How to compute the scalar product of two vectors.
◆ How to multiply matrices and vectors, and matrices and other matrices.
◆ How to use matrix multiplication to solve applied problems, including
Markov chains and linear growth models.
◆ How to solve a system of n linear equations in n unknowns, using Gaussian
elimination.
◆ How to apply Gaussian elimination to solve various real-world problems
with a system of linear equations.
◆ How to find an equation of various types of curves, such as parabolas and
circles, that pass through a given set of points.

Chapter Review Problems

1. Determine the magnitude of the displacement vec- Table Chair Sofa
tors from point A to point B for each pair of points. Wood 4 1 3
a. A  16, 62, B  12, 32 Labor C3 2 5S
b. A  11, 12, B  14, 12 Upholstery 0 2 4
c. A  11, 2, 32, B  13, 3, 52
Write a system of equations whose solution would
2. a. If a plane is flying due south at 200 mph and the determine production levels to yield the desired
wind is blowing in a direction that is 40° north numbers of tables, chairs, and sofas.
of west at 50 mph, what are the actual direction 4. If the stock market goes up today, historical data show
and speed of the plane? that for tomorrow it has a 60% chance of going up, a
b. Suppose instead that the wind is blowing from a 20% chance of staying the same, and a 20% chance of
direction that is 40° north of west at 50 mph. going down. If the market is unchanged today, it has a
How do your answers to part (a) change? 20% chance of being unchanged, a 40% chance of
3. A furniture manufacturer makes tables, chairs, and going up, and a 40% chance of going down tomorrow.
sofas. In one month, the company has available If the market goes down today, it has a 20% chance of
1500 units of wood, 2300 units of labor, and 1800 going up, a 20% chance of being unchanged, and a
units of upholstery. The manufacturer wants a pro- 60% chance of going down tomorrow.
duction schedule for the month that uses all these a. Construct a Markov chain for this problem. Give
resources. The different products require the fol- A, the matrix of transition probabilities, and
lowing amounts of the resources. draw the transition diagram.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 719

Chapter Review Problems 719

b. If there is a 30% chance of the market going up be the price vector for the circuits (the cost in dollars
today, a 10% chance of being unchanged, and a of each type of circuit).
60% chance of going down, what is the probabil- a. Write an expression in terms of A, d, and p for
ity distribution for the market tomorrow? the total cost of the circuits needed to produce
1 1 2 the set of computers demanded; indicate where
the matrix–vector product occurs and where the
5. Let a  C 2 S , b  C 3 S , and c  C 5 S .
scalar product occurs.
3 1 8 b. Compute the total cost.
Compute: (a) a b . .
(b) b c (c) a 1b  c2
.
8. Let
(d) a . a
1 2 3 4
6. Let a, b, and c be as in Problem 5. Let
A  C2 4 6 8S ,
1 2 3 4 3 5 7 9
A  C2 4 6 8S ,
3 5 7 9 1 0 1
B  C2 2 0S ,
1 0 1 0 1 1
B  C2 2 0S ,
0 1 1 5 4 1
1 0 2
5 4 1 CD T.
3 2 1
1 0 2
CD T. 0 1 3
3 2 1
Compute each matrix product (if possible).
0 1 3
a. AB b. BA c. AC
Which of the following matrix calculations are well
d. CA e. CB
defined (the sizes match)? If the computation
makes sense, perform it. If necessary, a, b, or c may 9. Suppose that you are given the following matrices
be changed to row vectors. involving the costs of fruits at different stores, the
amounts of fruit different types of people want, and
a. aA b. bB c. cC
the numbers of people of different types in differ-
d. Aa e. Bb f. Cc
ent towns
7. Three different types of computers need varying
amounts of four different types of integrated cir- Store 1 Store 2
cuits. Matrix A gives the number of each circuit Apples 0.10 0.15
needed by each computer. Oranges C 0.15 0.20 S
Circuit Pears 0.10 0.10
1 2 3 4 Apples Oranges Pears
A 2 3 2 1 Person 1 5 10 3
Computers B C 5 1 3 2S  A B R
Person 2 4 5 5
C 3 2 2 2
Person 1 Person 2
Let d  310 20 304 be the computer demand Town 1 1000 500
vector. Let B R
Town 2 2000 1000
2
a. Compute a matrix that represents the cost of
5
pD T each person’s fruit purchases at each store.
1 b. Compute a matrix that represents the quantity
10 of each fruit to be purchased in each town.
Gord.3896.10.pgs 4/28/03 3:30 PM Page 720

720 CHAPTER 10 Matrix Algebra and Applications

10. a. For the Markov chain matrix A in Problem 4, b. x1  x2  x3  2

compute A2, A3, and A5. 2x1  2x2  4x3  4
b. What vectors do the columns of the powers of A x1  2x2  3x3  5
appear to be approaching? 12. Solve the system of equations obained for the furni-
11. Solve each system of equations, using Gaussian ture model in Problem 3.
elimination. 13. Find the stable distribution for the Markov chain in
a. 2x1  3x2  2x3  0 Problem 4.
x1  x2  x3  7 14. Use matrix methods to find the equation of the
x1  5x2  4x3  4 parabola that passes through the three points 11, 1 2,
12, 2 2, and 13, 52 .
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 721

Appendices

A Some Mathematical Moments to Remember

A.1 Absolute Value

Absolute value is used to transform any number, positive or negative, to the corre-
sponding positive value. We write the absolute value of a number x as 0 x 0 . Thus
0 5 0  5, 0 6 0  6, and 0 2.3 0  2.3.
In general, for any number x  0, 0 x 0  x and for any number x  0, 0 x 0  x.
We can write this as
if x  0
0x0  e
x
x if x  0.

A.2 Factorial Notation n!

Expressions of the form 4  3  2  1 involving the product of consecutive
positive integers starting with 1 are written using factorial notation. We write
such products as n! (and read it as n factorial) for any positive integer n. For
instance,
3!  13 2 12 2 11 2  6,
4!  14 2 13 2 12 2 11 2  24,
5!  15 2 14 2 13 2 12 2 11 2  120,
and so on. In general, for any positive integer n,
n!  n1n  1 2 1n  2 2 . . . 13 2 12 2 11 2.
Note that 10!  10  9  8  . . .  1  1019 . 8 . . . 1 2  1019!2 and, in general,
n!  n . 1n  1 2!,
for any n. For completeness, it is necessary to define 0!  1.

721
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 722

722 Appendices

A.3 Summation or Sigma Notation

Summation notation is used as a shorthand for writing expressions such as
1  2  3  . . .  100,
1  2  22  23  . . .  250,
and
1 1 1
1   ...  .
2 3 n
The Greek letter sigma 1 2 is used to denote summation and an index of summation—
say, k—is used to indicate which specific terms are included in the sum. Thus, to
add the squares of all the integers between k  1 and k  100, we write
100

ak
2
k1

because when k  1 we have 12, when k  2 we have 22, when k  3 we have 32,
and so on until k  100 when we have 1002. Therefore
100

ak 1 2 3 
2 2 2 2 . . .  1002.
k1

To add all the integers between 25 and 60, we write

60

a k  25  26  27 
. . .  60.
k25

Similarly, we write the sum of the powers of 2 from 20 to 250 as

50

a2 122 2 
k 2 3 . . .  250.
k0

Also, to add the reciprocals of all the integers between k  1 and some unspecified
upper limit k  n, we write
n
1 1 1 1 ... 1
a k1234  .
n
k1

The letter we use for the index of summation is immaterial; we could equiva-
lently write
n
1 1 1 1 ... 1
a i 1234 
n
or
i1

n
1 1 1 1 ... 1
a j 1234  .
n
j1

The numerical result of these summations is the same regardless of the letter
used as the index.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 723

Appendix A Some Mathematical Moments to Remember 723

A.4 Similar Triangles

Two triangles are similar if all three angles in one triangle are the same size as all
three angles in the other. Of course, the lengths of the sides of the two triangles
may be quite different. However, the key fact about similar triangles is:
Corresponding sides of similar triangles are proportional.
The right triangles ABC and ADE shown in Figure A.1 are similar because the
angle u is common to both, the angle f is the same in both, and both triangles have
a right angle. Therefore, we have a variety of ratios that are equal, including
AB BC AC AB AD BC DE
  and  and  .
AD DE AE AC AE AC AE

φ
C

θ
A B D

FIGURE A.1

A.5 Distance Between Points in the Plane

The distance from a point A with coordinates 1x0 , y0 2 to a point B with coordinates
1x1 , y1 2 is given by the distance formula

0 AB 0  21x1  x0 2 2  1 y1  y0 2 2 .
It is based on the Pythagorean theorem, as illustrated in Figure A.2. The distance
0 AB 0 is the hypotenuse of the right triangle formed by a base of x1  x0 (the hori-
zontal change) and a height of y1  y0 (the vertical change).

(x 1, y 1)
y1 B

|AB| = √(x 1 – x 0 ) 2 + (y1 – y 0 ) 2

y1 – y 0

A x1 – x 0
y0
(x 0, y 0 ) (x1, y 0 ) C

x
x0 x1

FIGURE A.2
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 724

724 Appendices

For instance, the distance between the two points A at 12, 5 2 and B at 16, 8 2 is

0 AB 0  216  2 2 2  18  5 2 2  216  9  225  5.

A.6 The Equation of a Circle

The equation of the circle with radius r and center at 1x0 , y0 2 is

1x  x0 2 2  1 y  y0 2 2  r 2.

For instance, the equation of the circle with radius 7 and center at 12, 5 2 is

1x  2 2 2  1 y  5 2 2  72  49,

as shown in Figure A.3. The equation of a circle is discussed in detail in Section 9.2.

2
1
x
–5 –4 –3 –2 –1 1 2 3 4 5 6 7 8 9
–1
–2
–3
–4
(2, – 5)
–5
–6
–7 r=7
–8
–9
–10
–11
FIGURE A.3 –12

A.7 The Equation of an Ellipse

The equation of the ellipse with center at 1x0 , y0 2 having its horizontal major (or
longer) axis of length 2a and its vertical minor (or shorter) axis of length 2b is

1x  x0 2 2 1 y  y0 2 2
  1,
a2 b2
as shown in Figure A.4.
For instance, the equation of the ellipse with center at 12, 5 2 , whose major
axis is horizontal and has length 12 12a2 and whose minor axis is vertical and has
length 8 12b 2 is
1x  2 2 2 1 y  522
  1 or
62 42
1x  2 2 2 1 y  522
  1,
36 16
as shown in Figure A.5. The ellipse is discussed in detail in Section 9.3.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 725

Appendix A Some Mathematical Moments to Remember 725

(x − x0) 2 (y − y0) 2
+ =1
a2 b2

a a
(x 0, y 0 )

FIGURE A.4

2
1
x
–5 –4 –3 –2 –1 1 2 3 4 5 6 7 8 9
–1
–2
–3 b=4
–4
a=6
–5
(2, – 5)
–6
–7
–8
–9
–10
–11
–12

FIGURE A.5

A.8 Completing the Square

The two quadratic functions y  x 2  6x  13 and y  1x  32 2  4 are alge-
braically equivalent because the second can be expanded to give
y  1x  3 2 2  4  1x 2  6x  9 2  4  x 2  6x  13.
You can also check this result by graphing the two functions on your function
grapher. Because their graphs are the same parabola, they are indeed the same
function. Although the first representation is more common, the second gives
more information about the behavior of the corresponding parabola: It is shifted 3
units to the right and 4 units up compared to the basic parabola y  x 2.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 726

726 Appendices

The process of transforming the first expression to the second is called

completing the square. We illustrate this procedure as follows.
◆ Start with y  x 2  6x  13.
◆ Take one-half of the coefficient of x: 12 16 2  3.
◆ Square this number: 13 2 2  9.
◆ Add and immediately subtract the resulting number 9 (we are actually
adding 0 and thus still have the equivalent of the original expression):
y  1x 2  6x  9 2  9  13  1x 2  6x  9 2  4.
◆ Recognize that the first three terms, 1x 2  6x  9 2, form a perfect square—
the square of 1x  3 2. Therefore we have
y  1x  3 2 2  4.
Note that the 3 in this expression is the same as half of the original coefficient of x.

E XAMPLE 1
Complete the square on the quadratic y  x 2  10x  11 and compare its graph to
that of the parabola y  x 2.
Solution We have
y  3 x 2  10x 4  11

 3 1x 2  10x  52 2  52 4  11 110 2  5
1
2
 1x  5 2 2  25  11  1x  5 2 2  36.
Graphically, the corresponding parabola is obtained by shifting y  x 2 to the left by 5
and down by 36.
◆
If the original quadratic expression has a leading coefficient other than 1, that
coefficient must first be factored out. We illustrate how to do so in Example 2.

E XAMPLE 2
Complete the square on the quadratic y  2x 2  16x  42 and compare its graph to
that of the parabola y  x 2.
Solution We begin by factoring out the leading coefficient 2 so that
2x  16x  42  2 3 1x 2  8x 2  21 4
2

 2 3 1x 2  8x  14 2 2 2  142 2  21 4 18 2  4

1
2
 2 3 1x  4 2 2  5 4  21x  4 2 2  10.
The corresponding parabola is obtained by doubling y  x 2 and then shifting it 4 units
to the right and 10 units up.
◆
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 727

Appendix B Solving Systems of Linear Equations Algebraically 727

Problems
In Problems 1–5, evaluate each number. 15. In the accompanying figure, triangle ABC is similar
1. 0 9  12 0 2. 0 6  10 0
to triangle ADE.

3. 0 7  3 0 4. 0 5 0  0 4 0
a. Find DE.
b. Find AE.
5. 0 5 0  0 4 0
6. Calculate the value of y  0 x 0 for x  4, 3,
E

2, . . . , 4. Plot the points and then connect C

them. How would you describe the graph of 9 6

y  0 x 0? 12 8

7. Repeat Problem 6, using y  0 x  3 0 for x  7,

A B D

6, 5, . . . , 1. How does the graph y  0 x  3 0 In Problems 16 and 17, complete the square to identify
compare to the graph of y  0 x 0 ? the center and radius of the circle.
In Problems 8 and 9, evaluate the expression. 16. x 2  y 2  2x  10y  55
17. x 2  y 2  8x  6y  11  0
5! 20!
8. 9. 18. Write an equation for the ellipse shown.
3! 3!17!
10. Rewrite the expression 1  4  9  16  . . .  y

100 in summation notation. 6

11. Rewrite the expression 1>5  1>6  1>7  . . .  5

1>40 in summation notation. 4

3
4
12. Find the value of g k . 3 2
k1
1

13. Find the distance between the points 12, 4 2 and x

15, 8 2.
–1 1 3 5 7 9
–1

14. Find the equation of the circle that has 12, 42 and
–2

15, 8 2 as endpoints on a diameter.

B Solving Systems of Linear Equations Algebraically

A pair of linear equations such as
x  5y  3 (1)
4x  3y  11 (2)
is a system of linear equations. Its solution is a pair of values, one for x and the other
for y, that satisfy both equations simultaneously. Geometrically, each of the two
equations represents a line and every pair (x, y) that satisfies each individual equa-
tion is a point on that line. A pair of values for x and y that simultaneously satisfy
both equations must be the point of intersection of the two lines, as shown in Fig-
ure A.6. The two lines seem to intersect at the point 12, 12 . We verify that this
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 728

728 Appendices

4
4x − 3y = 11
3
2
1
x
–3 –2 –1 1 2 3 4 5
–1
–2 x + 5y = −3
–3
FIGURE A.6 –4

point is indeed the solution by substituting x  2 and y  1 into the two origi-
nal equations. For the first, we have
12 2  511 2  3
and for the second,
412 2  311 2  11.
We can always use this kind of geometric approach to solve systems of two
equations in two unknowns, but if the solutions are not simple numbers, the best
we can get is a reasonably accurate estimate. Alternatively, we can solve such a sys-
tem algebraically using either of two methods.
1. The method of substitution: (a) Solve for one variable in terms of the
other, using one of the two equations. (b) Next, substitute the expression
for that variable into the other equation to eliminate that variable. (This
step is usually straightforward if the coefficient of one of the variables is 1
or 1.) (c) Then solve for the remaining variable. (d) Finally, substitute its
value back into one of the equations to find the value of the other variable.
2. The method of elimination: Add or subtract an appropriate multiple of
one of the equations to the other equation to eliminate one of the vari-
ables. (This method is discussed in detail in Section 10.5).

E XAMPLE 1
Solve the system of Equations (1) and (2) by using the method of substitution.
Solution The coefficient of x in Equation (1) is 1, so we use Equation (1) to solve for x
in terms of y:
x  5y  3. (3)
We substitute this expression into Equation (2) to get
415y  3 2  3y  11 so that 20y  12  3y  11.
Note that we eliminated the variable x and now have a single equation in y only. Next, we
collect like terms:
23
23y  23 so that y   1.
23
When we substitute y  1 back into Equation (3), we find that
x  511 2  3  5  3  2.
This solution is the same, x  2 and y  1, that we obtained graphically in Figure A.6.
◆
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 729

Appendix B Solving Systems of Linear Equations Algebraically 729

E XAMPLE 2
Solve the same system of linear equations
x  5y  3 (1)
4x  3y  11, (2)
using the method of elimination.
Solution To eliminate the variable x from the two equations, we multiply Equation (1)
by 4:
4x  20y  12 (4)
while
4x  3y  11 (2)
Note that the coefficients of x are numerically equal but of opposite sign. If we add
Equations (2) and (4), the x terms cancel, leaving
23y  23 so that y  1.
If we now substitute this value of y into either Equation (1) or (2)—say, Equation (1)—
we get
x  511 2  3 or x  5  3  2,
which again is the same solution.
Alternatively, we could eliminate the variable y from Equations (1) and (2). To do so,
we multiply Equation (1) by 3 and Equation (2) by 5 to get:
3 Eqn 11 2 3x  15y  9
5 Eqn 12 2 20x  15y  55.
We eliminate y by adding the two equations to get
23x  46,
which again leads to x  2 and hence y  1.
◆
If we have a system of three linear equations in three unknowns—say, x, y, and
z or x1 , x2 , and x3 or a, b, and c—or an even larger system of linear equations, the
method of substitution quickly becomes unworkable. The method of elimination
is almost always preferable. However, in practice, for systems larger than two-by-
two, calculators and computers are typically the method of choice instead of at-
tempting to solve the systems by hand.

E XAMPLE 3
Find the solution to the system of linear equations
2x  5y  4z  7 (5)
3x  y  5z  14 (6)
6x  2y  2z  4, (7)
using the method of elimination.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 730

730 Appendices

Solution We use Equation (7) to eliminate the variable x from the remaining two equa-
tions. To eliminate x from Equation (5), we first add 3 times Equation (5) to Equation (7):
Eqn 172 6x  2y  2z  4
3 Eqn 152 6x  15y  12z  21
 13y  14z  25. (8)
Similarly, to eliminate the variable x from Equation (6), we add 2 times Equation (6)
to Equation (7) to get
Eqn 17 2 6x  2y  2z  4
2 Eqn 16 2 6x  2y  10z  28
4y  12z  24. (9)
Equations (8) and (9) together are a system of two linear equations in the two unknowns
y and z:
13y  14z  25 (8)
4y  12z  24 (9)
We can solve this reduced system as before.
For instance, to eliminate the variable y, we multiply Equation (8) by 4 and multiply
Equation (9) by 13, so that
4 Eqn 18 2 52y  56z  100
13 Eqn 19 2 52y  156z  312.
We add these two equations to get
212z  212 so that z  1.
Substituting z  1 into Equation (9), say, gives
4y  1211 2  24,
so that
4y  24  12  12 and so y  3.
Substituting both of these values into Equation (5), say, gives
2x  513 2  4112  7
2x  15  4  7
2x  4 and so x  2.
Thus, the solution to the original system of three equations in three unknowns is x  2,
y  3, and z  1. Substitute these values into the three original equations to verify
that they satisfy all three equations.
◆
Graphing calculators have the capability of solving systems of up to 99 equa-
tions in 99 unknowns at the push of a button. On some calculators, there is a
SIMULT key for simultaneous equations; enter the number of linear equations,
then enter the coefficients and the constant terms, and finally press Solve to get
the solutions. On other calculators, you can solve a system of linear equations
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 731

Appendix C Solving Systems of Linear Equations with Matrices 731

with the Solve command; enter the list of equations and the list of variables and
then press enter. On still other calculators, you can solve systems of linear equa-
tions by using matrix methods, as discussed briefly in Appendix C and in detail
in Chapter 10. (Actually, all calculators use matrix methods for solving systems
of linear equations.)

C Solving Systems of Linear Equations using Matrices

Briefly, a matrix is any rectangular array of numbers, such as
4 0 5 1
£ 7 2 8 6§.
3 1 2 3

The size, or dimension, of a matrix is measured by the number of rows (horizontal-

ly across) and the number of columns (vertically down) in the array. The dimen-
sion of the preceding matrix is 3 by 4, which we write as 3  4.
Let’s look at the system of linear equations:
2x  5y  4z  7
3x  y  5z  14
6x  2y  2z  4.
(We considered this system in Example 3 of Appendix B, where we found the so-
lution algebraically to be x  2, y  3, and z  1.) We first construct the
coefficient matrix A, which consists of the coefficients from each of the equations:
2 5 4
A  £3 1 5 §.
6 2 2
This matrix has 3 rows and 3 columns, so its size is 3 by 3, or 3  3. We also con-
struct the matrix B of constants, consisting of the constants on the right-hand side
of the equations:
7
B  £ 14 § .
4
This matrix (also known as a vector) has three rows and one column, so its size is
3  1. Finally, we construct a 3  1 matrix X of variables:
x
X  £y§;
z
these are the unknowns that we want to determine.
The system of three linear equations in three unknowns is then equivalent to
the simple matrix equation
AX  B.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 732

732 Appendices

The solution to this matrix equation is found in terms of the inverse matrix A1 (if
it exists) of the matrix A:
X  A1B.
To solve this system on the calculator, you must “name” each of the matrices in
turn by giving its size: 3  3 for A and 3  1 for B. Then enter the values for each
position. Finally, by selecting the appropriate names of the matrices, you have the
calculator find A1 * B. The calculator displays the entries in the column matrix X:
2
£ 3 § .
1
See the instruction manual for your calculator for details on how to use these ma-
trix features.

E XAMPLE
Use matrices to solve the system of linear equations
5x  7y  4z  6
2x  4y  8z  13
3x  5y  9z  2.
Solution The coefficient matrix and the matrix of constants are
5 7 4 6
A  £2 4 8 § and B  £ 13 § .
3 5 9 2
After entering these matrices in the calculator and forming the expression A1 * B, we
find the corresponding matrix of variables is
x 3.791079812
x  £ y §  A B  £ 2.049295775 § .
1

z 0.347178404
That is, the solution to the system of equations is x  3.791, y  2.049, and z  0.347.
The identical values are obtained on calculators having simultaneous equations
(SIMULT or SOLVE) capabilities.
◆

D Symmetry
The notion of symmetry arises throughout mathematics in a variety of ways. We use
symmetry to describe the behavior of functions and other geometric objects when
one portion is a mirror image of another portion. We may describe a curve (whether
or not it represents a function) as being symmetric about a line, or symmetric with re-
spect to an axis, or symmetric with respect to the origin, or symmetric about a point P.
The ellipse shown in Figure A.7(a) is symmetric with respect to the x-axis be-
cause the lower half is the mirror image of the upper half. Similarly, as shown in
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 733

Appendix D Symmetry 733

y y y

x x x
d

P′

FIGURE A.7 (a) (b) (c)

Figure A.7(b), the ellipse is also symmetric with respect to the y-axis because the
left half is the mirror image of the right half. The ellipse is also symmetric with re-
spect to the origin because, for any point P on the ellipse, we can find the mirror
image P through the origin on the ellipse, as shown in Figure A.7(c).
The parabola y  x 2 is symmetric about the y-axis because the left and right
sides are mirror images of one another. In fact, every parabola of the form
y  ax 2  bx  c is symmetric about the vertical line through its turning point,
or vertex. However, the cubic y  x 3 shown in Figure A.8 is not symmetric about
the x-axis nor is it symmetric about the y-axis because the two portions of the
curve are not mirror images of each other about either axis. However, the curve is
symmetric with respect to the origin; if any point P with coordinates 1a, b 2 is on
the curve, so is the point P with coordinates 1a, b2, as shown in Figure A.8.
This condition is equivalent to one portion of the curve being rotated through an
angle of 180° to produce the other portion.
We summarize the key information about symmetry as follows.
1. A curve is symmetric about the x-axis if, when a point P at 1a, b 2 lies on
the curve, the point P at 1a, b 2 also lies on the curve, as shown in Fig-
ure A.9(a).

y y
y
(a, b) (−a, b) (a, b)
P′ P
P
(a, b)
P

x
x

x
P′

P′ (a, −b)
(−a, −b)
(a) (b)

FIGURE A.8 FIGURE A.9

Gord.3896.APP.pgs 4/29/03 10:10 AM Page 734

734 Appendices

2. A curve is symmetric about the y-axis if, when a point P at 1a, b2 lies on the
curve, the point P at 1a, b 2 also lies on the curve, as shown in Figure A.9(b).
3. A curve is symmetric about the origin if, when a point P at 1a, b2 lies on
the curve, the point P at 1a, b 2 also lies on the curve, as previously
shown in Figure A.8.
E XAMPLE
Show algebraically that the curve y  x 3 is (a) not symmetric about the x-axis, (b) not
symmetric about the y-axis, and (c) symmetric about the origin.
Solution
a. It is obvious from the graph of y  x 3 shown in Figure A.10 that the curve is not
symmetric about the x-axis; however, we illustrate how to prove this fact by applying
the principles of symmetry. Suppose that a point 1a, b 2 lies on the curve so that
b  a3. We now consider the point 1a, b 2. When we substitute x  a, we again get
a3  b, not b, so the point 1a, b2 is not on the curve and therefore the curve is not
symmetric about the x-axis.

27

18

x
–3 –2 –1 1 2 3

–9

–18

–27
FIGURE A.10

b. Again, suppose that a point 1a, b2 lies on the curve so that b  a3, and consider the
point 1a, b2. When we substitute x  a, we get 1a2 3  a3  b, not b, so the
curve is not symmetric about the y-axis.
c. Again, suppose that 1a, b 2 lies on the curve, so that b  a3, and consider the point
1a, b2. When we substitute x  a, we get 1a2 3  a3  b, so that 1a, b2
also satisfies the equation of the curve and the curve is symmetric about the origin.

E Arithmetic of Complex Numbers

Complex numbers were originally introduced to allow people to solve quadratic
equations such as x 2  4  0, which is equivalent to x 2  4. When we try to
solve this equation by taking square roots, we obtain

x 
24 ,
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 735

Appendix E Arithmetic of Complex Numbers 735

which has no solution among the real numbers.

To work with numbers such as these that involve the square root of a negative
number, we introduce the imaginary number i:

i  21 .
Using i, we can take the square root of any negative number. For instance,

225  22511 2  225 21  5i.

Similarly, the roots of the equation x 2  4  0 are therefore

x 
24 
24 21 
2i,
which is equivalent to x  2i and x  2i.

E XAMPLE
Find the roots of x 2  2x  10  0, using the quadratic formula

b
2b2  4ac
x .
2a

Solution Substituting a  1, b  2, and c  10 into the quadratic formula yields

12 2
212 2 2  411 2 110 2
x
211 2
2
24  40 2
236
 
2 2
2
6i
  1
3i.
2
Therefore the roots are x  1  3i and x  1  3i.
◆
Each of these roots is called a complex number. In general, we write a complex
number in the form z  a  bi, where a and b are both real numbers. Thus z con-
sists of a real number a and an imaginary number bi. We call a the real part of z and
b the imaginary part of z. Note that an imaginary number is just a multiple of i,
such as 4i; a complex number is the sum of a real number and an imaginary num-
ber, such as 3  7i.
In the equation x 2  4  0, the roots were x  2i and x  2i. In the above
example, the roots were x  1  3i and x  1  3i. As in both cases, complex
numbers typically arise in pairs of the form a  bi and a  bi, called complex
conjugates.

Powers of i
Because i  21 , we have
i2  1 21 2 2
 1;
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 736

736 Appendices

that is,

i 2  1.

From this relation we can find other powers of i:

i 3  1i 2 2 1i2  11 2 1i2  i,
i 4  1i 2 2 1i 2 2  11 2 11 2  1,
i 5  1i 4 2 1i2  11 2 1i2  i,
and so on. All higher powers simply cycle through these four values: i, 1, i, and
1.

Complex Arithmetic
The arithmetic of complex numbers, for the most part, is straightforward. Consid-
er the two complex numbers z  5  4i and w  3  11i.

Addition To add complex numbers, we add the real parts and the imaginary
parts separately. For instance,
z  w  15  4i2  13  11i2  8  7i.
This is totally analogous to how we collect like terms in algebra. In general,

If z  a  bi and w  c  di, then

z  w  1a  c2  1b  d2i.

Subtraction To subtract complex numbers, we subtract the real parts and the
imaginary parts separately. For our numbers z  5  4i and w  3  11i,
z  w  15  4i2  13  11i2
 15  3 2  14i  111i2 2  2  15i.
In general,

If z  a  bi and w  c  di, then

z  w  1a  c2  1b  d2i.

Multiplication To multiply complex numbers, we multiply them algebraically,

use the fact that i 2  1 to simplify any power of i, and collect like terms consist-
ing of the real part and the imaginary part. For instance,
z  w  15  4i2 13  11i2
 152 13 2  15 2 111i2  14i2 13 2  14i2 111i2
 15  55i  12i  44i 2
 15  43i  44 i2  1
 59  43i.
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 737

Appendix F Introduction to Data Analysis 737

In general,

If z  a  bi and w  c  di, then

z  w  1ac  bd2  1ad  bc2 i.

Note that, in the particular case where z and w are complex conjugates (say,
z  6  8i and w  6  8i) we have
z  w  16  8i2 16  8i2
 616 2  618i2  18i2 16 2  18i2 18i2
 36  48i  48i  64i 2
 36  6411 2  100. i2  1
In general,

If z  a  bi and w  a  bi are complex conjugates, then

z  w  a2  b2.

Consequently, the product of complex conjugates is always a real number.

Division of complex numbers is somewhat more difficult, but the ideas we de-
veloped in Section 8.3 provide a simple way to do it.

Problems
In Problems 1–4, simplify each expression. 9. 11  3i2 12  i2 10. 15  6i2 2
1. i 23 2. i 72 11. 115  2i2 115  2i2 12. 17  4i2 17  4i2
3. i 58 4. i 45 In Problems 13–18, find the roots of each equation.
In Problems 5 –12, perform each operation. Write each 13. x 2  25  0 14. 4x 2  9  0
answer in the form a  bi, where a and b are real 15. 5x 2  2x  1  0 16. x 2  4x  7  0
numbers. 17. 3x 2  4x  8  0 18. 5x 2  2x  4  0
5. 18  i2  16  3i2 6. 15  2i2  17  6i2
7. 110  i2  11  i2 8. 14  5i28i

F Introduction to Data Analysis

In all the cases presented in this book, we have used data relating two quantities and
sought to identify a relationship—linear or nonlinear—between them. Situations
also are common in which we have data only on a single quantity. For instance, your
professor may want to analyze the scores of all students in your class on an exam.
Alternatively, she may want to compare the set of scores that each student has on all
the exams in a course. In such cases, methods are needed to analyze the data to ex-
tract useful information.
Gord.3896.APP.pgs 11/21/03 11:10 AM Page 738

738 Appendices

Measuring the Center of a Set of Data

Given a set of data on a single quantity, it is usually most important to find the center
of the data to give a single number that describes the entire set of values. Probably the
most common method is to use the mean, or arithmetic average, of the data. Sup-
pose that we have n data values, x1 , x2 , x3 , . . . , xn . Their mean is given by
n
a g xk b
x1  x2  . . .  xn k1
x  .
n n
For instance, if Amy’s grades on her four math tests are 85, 91, 98, and 94, her mean
grade for the course is x  185  91  98  94 2 >4  92.
The mean also gives a simple way to compare one set of data to another. If Bret
had a mean of 87 on the same four tests, clearly he didn’t do quite as well as Amy
did. This type of comparison is usually applied by a professor in assigning final
grades in a course.
However, there are situations in which the mean gives misleading information
about a set of data, and other measures for the center for the data must be used.
The accompanying set of data shows the number of home runs for the 2001 season
for each of the starting players in the San Francisco Giants lineup. The mean num-
ber of home runs for the eight players was
37  22  6  73  15  5  6  8
x  21.5.
8
However, this value is unrepresentative of the team as a whole because Bobby
Bonds hit so many home runs that season. His record-breaking 73 home runs has
a disproportionate effect on the mean because his contribution is so far away
from the values for the other players.

Player Position Home runs

R. Aurillia SS 37
J. Kent 2B 22

B. Santiago C 6

B. Bonds OF 73
M. Benard OF 15

R. Martinez 3B 5

C. Murray OF 6
J. Snow 1B 8

Think About This Recalculuate the mean number of home runs of the Giants starting players when
you remove Bobby Bonds’s 73 home runs. ❐
For this set of data, the median is a far better measure of the center. To find the
median, you must list the data values in either increasing or decreasing order, as in
5, 6, 6, 8, 15, 22, 37, 73. If there are an odd number of entries, the median is defined
as the middle value when the data are in order. If there are an even number of en-
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 739

Appendix F Introduction to Data Analysis 739

tries, the median is defined as the average of the middle two values. In this way, the
median is always located at the center of the ordered list of data—there are just as
many values below it as there are above it. For the number of home runs hit by the
Giants during 2001, there are 8 entries, so the median is the average of the middle
two, 8 and 15. Thus the median is 18  15 2>2  11.5 home runs. This value is
more representative of all of the players; Bonds’s 73 home runs does not have the
disproportionate effect on the median that it has on the mean. In fact, if Bonds had
hit 103 home runs, instead of 73, the median would not change, but the mean cer-
tainly would change.

Think About This Recalculuate the median number of home runs of the Giants players when you re-
move Bobby Bonds’s 73 home runs. Did the value change as much as it did when
you recalculated the mean in the previous Think About This exercise? ❐
In general, the median is more representative of a set of data when there is
large variation among the data entries, especially if the data contain extreme val-
ues. In such a case, a handful of very large or very small values has a dispropor-
tionate effect on the mean, but not on the median.

Measuring the Spread in a Set of Data

Locating the center in a set of data often isn’t sufficient to give a full picture of the
data. For instance, consider two students, Carol and Doug, who had the following
scores on five tests in a course:
Carol: 78, 80, 85, 75, 82
Doug: 92, 75, 88, 63, 82.
Both have means of 80, but there is a clear difference in the overall scores of the two
students. Carol’s results are consistent; they are all quite close to the mean of 80. But
Doug’s results are very scattered, ranging from a minimum of 63 to a maximum of 92.
We therefore need a way to measure the amount of scatter, or spread, in a set
of data. One way to do so is to use the mean as the center of the data and calculate
the amount of spread about the mean. The result is a quantity called the standard
deviation.
There are two ways to calculate this quantity, depending on whether the data
represents a population (all possible values associated with a quantity) or just a
sample drawn from a much larger population. For a population, the standard devi-
ation is denoted by s (lowercase Greek letter sigma) and is found from
1x  x 2 2
s ,
B n
where n is the number of data points and x is the mean of the population. (In sta-
tistics texts, it is standard to write the mean of a population as m (the lowercase
Greek letter mu) instead of x. For a sample, the standard deviation is denoted by s
and is found from
1x  x 2 2
s ,
B n1
where n is the number of data points and x is the mean of the sample. (We won’t go
into the reason for the difference in the two formulas here; any introductory statis-
tics text gives a full discussion of the reason.)
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 740

740 Appendices

All these quantities, as well as several others that we discuss shortly, are routinely
obtained by using the statistical features of all calculators under the one-variable
option in the Statistics menu, as well as many software packages. In particular, for
Carol’s test scores, the calculator gives the values
X  80, s  3.808, and s  3.406;
and for Doug’s scores
X  80, s  11.467, and
(We write X as the calculator output; whether it represents a population mean m or
a sample mean x depends on the context.) Note that the value for the standard de-
viation for Doug is about three times as large as that for Carol, which indicates that
Doug’s test scores are much more widely spread about the mean than Carol’s are.
There is also a way to measure the spread in a set of data about the median as
the center of the data. We do so by partitioning the data values into four groups—
the top 25%, the next 25%, the 25% below the median, and the bottom 25%. These
are known as the quartiles. For instance, suppose that we have a set of data with 60
values that have been arranged in either increasing or decreasing order. Then the
number that separates the bottom 15 data values from the rest of the data values is
called the first quartile, Q1 . The second quartile, Q2 , separates the bottom 30 values
from the top 30 values and is the median. The number that separates the top 15
values from the rest is the third quartile, Q3 . In addition, it is standard to report the
minimum and maximum values in the data.
As with the standard deviation, these values are also included in the output of
all calculators with statistical features and in many software packages. In particular,
the values corresponding to Carol’s test grades are
Minimum  75, Q1  76.5, median  80,
Q3  83.5, and maximum  85.
Those corresponding to Doug’s grades are
Minimum  63, Q1  69, median  82,
Q3  90, and maximum  92.
These values give a clear picture of the way that the different sets of data values are
spread about the respective medians.

G 2002 World Population Data

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
WORLD 6,215 21 9 1.3 53.7 7,859 9,104 67 1.2
NORTH AMERICA 319 14 9 0.6 115.9 382 450 77 0.6
Canada 31.3 11 7 0.3 231.4 36.0 36.6 79 0.3
United States 287.4 15 9 0.6 115.9 346.0 413.5 77 0.6
Gord.3896.APP.pgs 11/21/03 11:10 AM Page 741

Appendix G 2002 World Population Data 741

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
LATIN AMERICA &
THE CARIBBEAN 531 23 6 1.7 41.1 697 815 71 0.7

CENTRAL AMERICA 140 27 5 2.2 31.9 188 225 74 0.5

Costa Rica 3.9 21 4 1.7 41.1 5.2 5.9 77 0.6
El Salvador 6.6 30 7 2.3 30.5 9.3 12.4 70 0.6
Guatemala 12.1 36 7 2.9 24.2 19.8 27.2 66 1.0
Honduras 6.7 33 6 2.8 25.1 9.6 12.2 66 1.6
Mexico 101.7 26 5 2.1 33.4 131.7 150.7 75 0.3
Nicaragua 5.4 34 5 2.8 25.1 8.6 11.6 68 0.2
Panama 2.9 23 4 1.9 36.8 3.8 4.3 74 1.5

CARIBBEAN 37 21 8 1.3 53.7 45 50 69 2.4

Antigua and Barbuda 0.1 22 6 1.6 43.7 0.1 0.1 71 —
Bahamas 0.3 18 5 1.3 53.7 0.4 0.5 72 3.5
Barbados 0.3 15 8 0.6 115.9 0.3 0.3 73 1.3
Cuba 11.3 12 7 0.5 139.0 11.8 11.1 76 0.1
Dominican Republic 8.8 26 5 2.1 33.4 12.1 14.9 69 2.7
Grenada 0.1 19 7 1.2 58.1 0.1 0.1 71 —
Guadeloupe 0.5 17 6 1.2 58.1 0.5 0.6 77 —
Haiti 7.1 33 15 1.7 41.1 9.6 11.9 49 6.1
Jamaica 2.6 20 5 1.5 46.6 3.3 3.8 75 1.2
Martinique 0.4 14 6 0.8 87.0 0.4 0.4 79 —
Netherlands Antilles 0.2 14 6 0.7 99.4 0.2 0.3 76 —
Puerto Rico 3.9 15 7 0.8 87.0 4.1 4.1 76 —
St. Kitts-Nevis 0.04 19 9 1.0 69.7 0.05 0.1 71 —
Saint Lucia 0.2 18 6 1.2 58.1 0.2 0.2 71 —
Trinidad and Tobago 1.3 14 8 0.7 99.4 1.4 1.4 71 2.5

SOUTH AMERICA 354 22 6 1.5 46.6 463 540 70 0.6

Argentina 36.5 19 8 1.1 63.4 47.2 54.5 74 0.7
Bolivia 8.8 32 9 2.3 30.5 13.2 17.1 63 0.1
Brazil 173.8 20 7 1.3 53.7 219.0 247.2 69 0.7
Chile 15.6 18 6 1.2 58.1 19.5 22.2 77 0.3
Colombia 43.8 22 6 1.7 41.1 59.7 71.5 71 0.4
Ecuador 13.0 28 6 2.2 31.9 18.5 22.9 71 0.3
Guyana 0.8 24 8 1.5 46.6 0.7 0.5 63 2.7
Paraguay 6.0 31 5 2.7 26.0 10.1 15.0 71 0.1
Peru 26.7 26 7 2.0 35.0 35.7 42.8 69 0.4
Suriname 0.4 24 7 1.7 41.1 0.5 0.4 71 1.2
Uruguay 3.4 16 10 0.7 99.4 3.8 4.2 75 0.3
Venezuela 25.1 24 5 1.9 36.8 34.8 41.0 73 0.5
EUROPE 728 10 11 0.1 — 718 651 74 0.4
NORTHERN EUROPE
Denmark 5.4 12 11 0.1 693.5 5.9 6.4 77 0.2
Gord.3896.APP.pgs 11/21/03 11:10 AM Page 742

742 Appendices

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
Estonia 1.4 9 14 0.4 — 1.2 0.9 71 1.0
Finland 5.2 11 10 0.2 346.9 5.3 4.8 78 0.1
Iceland 0.3 15 6 0.9 77.4 0.3 0.4 79 0.2
Ireland 3.8 14 8 0.6 115.9 4.5 4.5 77 0.1
Latvia 2.3 8 14 0.6 — 2.2 1.8 71 0.4
Lithuania 3.5 9 12 0.3 — 3.5 3.1 73 0.1
Norway 4.5 13 10 0.3 231.4 5.0 5.2 79 0.1
Sweden 8.9 10 11 0.0 — 9.5 9.8 80 0.1
United Kingdom 60.2 11 10 0.1 693.5 64.8 65.4 78 0.1

WESTERN EUROPE
Austria 8.1 9 9 0.0 — 8.4 8.2 78 0.2
Belgium 10.3 11 10 0.1 693.5 10.8 11.0 78 0.2
France 59.5 13 9 0.4 173.6 64.2 65.1 79 0.3
Germany 82.4 9 10 0.1 — 78.1 67.7 78 0.1
Liechtenstein 0.03 12 7 0.5 139.0 0.04 0.04 — —
Luxembourg 0.5 13 9 0.5 139.0 0.6 0.6 78 0.2
Monaco 0.03 23 16 0.6 115.9 0.04 0.04 — —
Netherlands 16.1 13 9 0.4 173.6 17.7 18.0 78 0.2
Switzerland 7.3 10 8 0.2 346.9 7.6 7.4 80 0.5

EASTERN EUROPE
Belarus 9.9 9 14 0.5 — 9.4 8.5 69 0.3
Bulgaria 7.8 9 14 0.5 — 6.6 5.3 72 z1
Czech Republic 10.3 9 11 0.2 — 10.3 9.4 75 z1
Hungary 10.1 10 13 0.4 — 9.2 8.1 72 0.1
Moldova 4.3 9 10 0.1 — 4.5 4.2 68 0.2
Poland 38.6 10 10 0.0 — 38.6 33.9 74 0.1
Romania 22.4 10 12 0.2 — 20.6 17.1 71 z
Russia 143.5 9 16 0.7 — 129.1 101.7 65 0.9
Slovakia 5.4 10 10 0.0 — 5.2 4.7 73 z
Ukraine 48.2 8 15 0.8 — 45.1 38.4 68 1.0

SOUTHERN EUROPE
Albania 3.1 17 5 1.2 58.1 4.1 4.7 74 z
Bosnia-Herzegovina 3.4 12 8 0.4 173.6 3.6 3.4 68 z
Croatia 4.3 10 12 0.2 — 4.1 3.6 74 z
Greece 11.0 10 10 0.0 — 10.4 9.7 78 0.2
Italy 58.1 9 9 0.0 — 57.5 52.2 80 0.4
Macedonia 2.0 15 9 0.6 115.9 2.2 2.1 73 z
Malta 0.4 11 8 0.3 231.4 0.4 0.4 77 0.1
Portugal 10.4 12 10 0.2 346.9 9.7 8.6 76 0.5
Slovenia 2.0 9 9 0.0 — 2.0 1.7 76 z
Spain 41.3 10 9 0.1 693.5 44.3 42.1 79 0.5
Yugoslavia 10.7 12 11 0.2 346.9 10.7 10.2 72 0.2
1
z  less than 12 %
Gord.3896.APP.pgs 4/29/03 10:10 AM Page 743

Appendix G 2002 World Population Data 743

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
AFRICA 840 38 14 2.4 29.2 1,281 1,845 53 6.6
NORTHERN AFRICA
Algeria 31.4 23 5 1.8 38.9 43.0 51.3 70 0.1
Egypt 71.2 27 7 2.0 35.0 96.1 115.4 66 z
Libya 5.4 28 4 2.4 29.2 8.3 10.8 75 0.2
Morocco 29.7 25 6 1.9 36.8 40.5 48.4 69 0.1
Sudan 32.6 36 12 2.4 29.2 49.6 63.5 56 2.6
Tunisia 9.8 17 6 1.2 58.1 11.6 12.2 72 z

WESTERN AFRICA
Benin 6.6 41 12 2.9 24.2 12.0 18.1 54 3.6
Burkina Faso 12.6 47 17 3.0 23.4 21.6 34.3 47 6.5
Cote d’lvoire 16.8 36 16 2.0 35.0 25.6 35.7 45 9.7
Gambia 1.5 42 13 2.9 24.2 2.7 4.2 53 1.6
Ghana 20.2 32 10 2.2 31.9 26.5 32.0 58 3.0
Guinea 8.4 45 18 2.7 26.0 14.1 20.7 48 1.5
Guinea-Bissau 1.3 45 20 2.5 28.1 2.2 3.3 45 2.8
Liberia 3.3 49 17 3.1 22.7 6.0 10.0 50 2.8
Mali 11.3 49 19 3.0 23.4 21.6 36.4 47 1.7
Mauritania 2.6 34 14 2.0 35.0 5.1 7.2 53 0.5
Niger 11.6 55 20 3.5 20.1 25.7 51.9 45 1.4
Nigeria 129.9 41 14 2.7 26.0 204.5 303.6 52 5.8
Senegal 9.9 38 12 2.6 27.0 16.5 22.7 53 0.5
Sierra Leone 5.6 49 25 2.4 29.2 10.6 14.9 39 7.0
Togo 5.3 40 11 2.9 24.2 7.6 9.7 55 6.0

EASTERN AFRICA
Burundi 6.7 43 21 2.2 31.9 12.4 20.2 41 8.3
Eritrea 4.5 43 12 3.0 23.4 8.3 13.3 56 2.8
Ethiopia 67.7 40 15 2.5 28.1 117.6 172.7 52 6.4
Kenya 31.1 34 14 2.0 35.0 33.3 37.4 48 15.0
Madagascar 16.9 43 13 3.0 23.4 30.8 47.0 55 0.3
Malawi 10.9 46 22 2.4 29.2 12.8 15.0 38 15.0
Mauritius 1.2 16 7 1.0 69.7 1.4 1.5 72 0.1
Mozambique 19.6 43 23 2.0 35.0 20.6 22.9 38 13.0
Rwanda 7.4 42 21 2.2 31.9 8.0 8.9 39 8.9
Somalia 7.8 48 19 2.9 24.2 14.9 25.5 47 1.0
Tanzania 37.2 40 13 2.7 26.0 59.8 88.3 52 7.8
Uganda 24.7 48 18 3.0 23.4 48.0 84.1 43 5.0
Zambia 10.0 42 22 2.0 35.0 14.3 20.3 37 21.5
Zimbabwe 12.3 29 20 0.9 77.4 10.3 10.1 38 33.7

MIDDLE AFRICA
Angola 12.7 48 20 2.9 24.2 28.2 53.3 45 5.5
Gord.3896.APP.pgs 11/21/03 11:10 AM Page 744

744 Appendices

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
Cameroon 16.2 37 12 2.5 28.1 24.7 34.7 55 11.8
Central African
Republic 3.6 38 18 2.0 35.0 4.9 6.4 44 12.9
Chad 9.0 49 16 3.3 21.3 18.2 33.3 51 3.6
Congo 3.2 44 14 3.0 23.0 6.3 10.7 51 7.2
Congo, Dem.
Rep. of 55.2 46 15 3.1 22.7 106.0 181.9 49 4.9
Gabon 1.2 32 16 1.6 43.7 1.4 1.8 50 4.2

SOUTHERN AFRICA
Botswana 1.6 31 22 0.8 87.0 1.2 1.2 39 38.8
Lesotho 2.2 33 15 1.8 38.9 2.4 2.8 51 31.0
Namibia 1.8 35 20 1.6 43.7 2.0 2.5 43 22.5
South Africa 43.6 25 15 1.1 63.4 35.1 32.5 51 20.1
Swaziland 1.1 41 20 2.0 35.0 1.4 2.0 40 33.4

ASIA 3,776 20 7 1.3 53.7 4,741 5,297 67 0.4

WESTERN ASIA
Armenia 3.8 8 6 0.2 346.9 3.7 3.2 72 0.2
Azerbaijan 8.2 14 6 0.8 87.0 10.2 13.0 72 z
Bahrain 0.7 22 3 1.9 36.8 1.7 2.9 74 0.3
Cyprus 0.9 12 7 0.6 115.9 1.0 1.0 77 0.3
Georgia 4.4 9 9 0.0 — 3.6 2.5 73 z
Iraq 23.6 35 10 2.5 28.1 41.2 60.1 58 z
Israel 6.6 21 6 1.5 46.6 9.3 11.0 78 0.1
Jordan 5.3 28 5 2.3 30.5 8.7 11.8 70 z
Kuwait 2.3 32 3 2.9 24.2 3.9 5.5 76 0.1
Lebanon 4.3 21 7 1.4 49.9 5.4 5.8 73 0.1
Oman 2.6 33 4 2.9 24.2 5.1 7.4 73 0.1
Palestinian Territory 3.5 40 4 3.5 20.1 7.4 11.2 72 —
Qatar 0.6 31 4 2.7 26.0 0.8 0.9 72 0.1
Saudi Arabia 24.0 35 6 2.9 24.2 40.9 60.3 72 z
Syria 17.2 31 6 2.6 27.0 26.5 34.4 70 z
Turkey 67.3 22 7 1.5 46.6 85.0 96.9 69 z
United Arab Emirates 3.5 17 2 1.5 46.6 4.5 5.1 74 0.2
Yemen 18.6 44 11 3.3 21.3 39.6 71.1 59 0.1
SOUTH CENTRAL ASIA
Afghanistan 27.8 43 19 2.4 29.2 45.9 67.2 45 z
Bangladesh 133.6 30 8 2.2 31.9 177.8 205.4 59 z
Bhutan 0.9 34 9 2.5 28.1 1.4 2.0 66 z
India 1049.5 26 9 1.7 41.1 1363.0 1628.0 63 0.8
Iran 65.6 18 6 1.2 58.1 84.7 96.5 69 0.1
Kazakhstan 14.8 15 10 0.5 139.0 14.7 14.0 66 0.1
Kyrgyzstan 5.0 20 7 1.3 53.7 6.5 7.5 69 z1
Gord.3896.APP.pgs 11/21/03 11:10 AM Page 745

Appendix G 2002 World Population Data 745

Percent
Births Deaths Projected Life of pop.
Population per per Growth Population Expectancy 15–49
Mid-2002 1,000 1,000 Rate Doubling (millions) at Birth with
(millions) Pop. Pop. (%) Time 2025 2050 (years) HIV/AIDS
Nepal 23.9 31 11 2.1 33.4 36.1 43.4 58 0.5
Pakistan 143.5 30 9 2.1 33.4 242.1 332.0 63 0.1
Sri Lanka 18.9 18 6 1.2 58.1 22.1 22.7 72 z
Tajikistan 6.3 19 4 1.4 49.9 7.8 8.5 68 z
Turkmenistan 5.6 19 5 1.3 53.7 7.2 7.9 67 z
Uzbekistan 25.4 22 5 1.7 41.1 37.2 38.6 70 z

SOUTHEAST ASIA
Brunei 0.4 22 3 2.0 35.0 0.5 0.6 74 0.2
Cambodia 12.3 28 11 1.7 41.1 18.4 21.9 56 2.7
East Timor 0.8 29 15 1.5 46.6 1.2 1.4 48 —
Indonesia 217.0 22 6 1.6 43.7 281.9 315.8 68 0.1
Laos 5.5 36 13 2.3 30.5 8.6 11.3 54 0.1
Malaysia 24.4 23 4 1.9 36.8 35.6 46.4 73 0.4
Myanmar 49.0 25 12 1.3 53.7 60.2 68.5 56 2.0
Philippines 80.0 28 6 2.2 319 115.5 145.7 68 z
Singapore 4.2 12 4 0.8 87.0 8.0 10.4 78 0.2
Thailand 62.6 14 6 0.8 87.0 72.1 71.9 72 1.8
Vietnam 79.7 19 5 1.4 49.9 104.1 117.2 68 0.3

EAST ASIA
China 1,280.7 13 6 0.7 99.4 1,454.7 1,393.6 71 0.1
China, Hong Kong 6.8 7 5 0.2 346.9 8.4 7.5 79 0.1
China, Macao 0.4 7 3 0.4 173.6 0.6 0.8 77 —
Japan 127.4 9 8 0.2 346.9 121.1 100.6 81 z
Korea, North 23.2 18 10 0.7 99.4 25.7 26.4 64 z
Korea, South 48.4 13 5 0.8 87.0 50.5 50.0 76 z
Mongolia 2.4 23 8 1.5 46.6 3.3 3.9 63 z
Taiwan 22.5 11 6 0.6 115.9 25.3 25.2 75 —

OCEANIA 32 18 7 1.0 69.7 40 46 75 0.2

Australia 19.7 13 7 0.6 115.9 23.2 25.0 80 0.1
Fiji 0.9 25 6 1.9 36.8 1.0 0.9 67 0.1
French Polynesia 0.2 21 5 1.6 43.7 0.3 0.4 72 —
Guam 0.2 24 4 2.0 35.0 0.2 0.3 77 —
New Caledonia 0.2 21 6 1.6 43.7 0.3 0.4 73 —
New Zealand 3.9 14 7 0.7 99.4 4.6 5.0 78 0.1
Papua-New Guinea 5.0 34 11 2.3 30.5 8.0 10.9 57 0.7
Samoa 0.2 30 6 2.4 29.2 0.2 0.2 68 —
Solomon Islands 0.5 41 7 3.4 20.7 0.9 1.5 67 —
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 746

Selected Answers

Chapter 1 17. BP

120
Section 1.1
110
1. a. Function, number of miles depends on number of gallons 100
b. Function, price depends on number of carats 90
e. Function, amount of rain depends on the day 80
c, d, and f are not functions.
x
3. a. (ii) c. (i) d. (iii) 0 2 4 6 8 10 12 14 16 18 20

Section 1.2 19. a. about 10 12 years

1. e is strictly increasing. b. 2000 and 2011
g and i are strictly decreasing
a, b, c, d, f, and h are neither
Section 1.3
3. y
Concave up Concave down Concave up 3. a. A 10-year-old child is 50 inches tall.
b. Increasing
Inflection
point c. Generally concave down because growth starts quickly
but usually slows down
Inflection 5. P1T2  kT for some constant k
point
k
7. F1d2  for some constant k
d2
x
9. a. 156 b. 210; 243; 270; 330
c. 150
5. 11. a. 348
b. 16 grams of peanut butter and 20 grams of jelly, or
25 grams of peanut butter and no jelly
c. Peanut butter
13. b. 164 feet; 104 feet
c. H 122  176 feet is the height of the ball after 2 sec-
onds, and H 132  156 feet is the height of the ball after
Inflection point
Turning point
3 seconds.
d. 1.875 seconds; 176.25 feet
e. 5.19 seconds
f. Domain: 0  t  5.19; range: 0  H  176.25
1 1 1 1 1
9. 15.  ;  ; ; ; ; x  2 would make the denominator
4 3 5 12 21
0; x  2; domain: x  2
17. 6, 20, 30, 110; no; nonnegative; domain: s  0
Inflection point

Section 1.4
1.(i) Function, with domain 3.5  x  4 and range
0  y  2; decreasing from 3.5 to 2.5 and from
1 to 1.7; increasing elsewhere; concave down from
15. a, b, c, d, e, h, i, and j are periodic. f and g are not 1.8 to 0 and concave up from 3 to 1.8 and
periodic. from 0 to 4

746
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 747

SELECTED ANSWERS 747

(ii) Function, with domain 0  x  5 and range 1  Section 1.5

y  2; increasing and concave down everywhere
5. a. 38 mph b. 49 mph c. 58 mph
(iii) Not a function
d. 69 mph e. 150 ft
(iv) Not a function
(v)Function, with domain 1  x  5 and range 7. a. y  f 1t2  80t  16t 2
1  y  2; decreasing and concave up everywhere b. y  f 1t2  v0 t  16t 2
5. x f (x)  x 2  3x  2
Review Problems
3 20 1. Independent variable: depth of tumor; dependent variable:
2 12 amount of radiation
1 6
0 2 3.
1 0
4
2 0
3 2 3
(16, 2.31) (20, 2.38) (24, 2.43)

Dosage
4 6 (8, 1.96) (12, 2.18)
5 12 2 (4, 1.6)
(0, 1)
fa b  ; fa b  ; f a b 
1 3 3 1 5 63 1 (12, 1.18) (16, 1.31) (20, 1.38)
(24, 1.43)
(8, 0.96)
2 4 2 4 2 4 (4, 0.6)

9. f 152 can be any number between 10 and 15. 4 8 12 16 20 24

Time (hours)
11. a. x1 x x3, x6 x x8, x11 x x14;
that is, 1x1, x3 2, 1x6, x8 2, 1x11, x14 2 4.
b. x4 x x6, x8 x x11; that is, 1x4, x6 2 , 1x8, x11 2
c. x4, x6, x8, x11 Population
d. x4, x8
e. x6, x11
f. x3, x5, x7, x9, x13
g. x1 x x3, x5 x x7, x9 x x13;
that is, 1x1, x3 2, 1x5, x7 2, 1x9, x13 2 . Time

h. x3 x x5, x7 x x9, x13 x x14;

that is, 1x3, x5 2, 1x7, x9 2, 1x13, x14 2.
5. a. Function b. Not a function
i. x12 x x14; that is, 1x12, x14 2 6. Overall, zoos with larger budgets have greater attendance.
j. x9 x x10; that is, 1x9, x10 2 7. b. 1990
k. x2, x10, x12 c. Most rapid change is between 1994 and 1995; slowest
13. a. g(x) b. f(x) c. h(x) change is between 1993 and 1994.
17. a. Increasing b. Concave up 8. 1, 2, 2.43, 2.0403, 34, f 1a2  3a2  2a  1
19. x f (x)  2x 9. a. $18,896.50 b. $24,174.10 c. about 6.7 years

0 0 10. a. 5.6 years b. 3.9 to 24.8 years

c. Increasing and concave down
1 1 d. 232 days e. 16.2 years
2 22  1.414 11. a. 10.375, 5.56252 b. y  5.5625
12. a. 10, 92 b. All real numbers
3 23  1.732
13. a. x  5 b. x  4 or x  4
4 2 c. x  3 d. All real numbers
5 25  2.236 14. a.
Weight W (oz) Cost ($)
6 26  2.449
0 W  1 0.37
fa b   0.707; f a b 
1 1 3 3
 1.225; 1 W  2 0.60
2 B2 2 B2 2 W  3 0.83

fa b 
5 5 3 W 4 1.06
 1.581; domain: x  0
2 B2 4 W  5 1.29
Gord.xxxx.ans.pgs 4/29/03 1:14 PM Page 748

748 SELECTED ANSWERS

15. a. From A to B the track is increasing and concave up. Exercising Your Algebra Skills
From B to C the track is increasing and concave down.
19 1 29
From C to D the track is decreasing and concave down. 1. x  3. x   5. y  7. x  1
From D to E the track is decreasing and concave up. 5 2 18
From E to F the track is increasing and concave up. 194
b. From A to B the car’s speed is decreasing at an 9. q   4.1 11. k  3
47
increasing rate.
From B to C the car’s speed is decreasing at a 19
13. x  15. t  2511.25
decreasing rate. 6
From C to D the car’s speed is increasing at an 2 8
increasing rate. 17. a. b. 4, 
3 3
From D to E the car’s speed is increasing at a
decreasing rate. 4 5 5
19. a.  b.  , 
From E to F the car’s speed is decreasing at an 7 4 7
increasing rate.
Section 2.3
Chapter 2 1. a. Yes, 47 b. No c. Yes, 137.1
Section 2.2 3. y  0.057x  1.329
1. a. (iii) b. (i) c. (v) d. (vi) 5. a. W  0.66t  26.5 b. 41.02 billion
e. (iv) f. (ii) c. W  26.5  0.661t  802 , 41.02 billion
3. a. y  5  71x  2 2, or y  7x  9 11. a. 3N  2G  30
b. y  2 c. Domain: 0  N  10; range: 0  G  15
c. y  0.626x  7.164 d. 3N  2G  60 e. 3N  G  30
5. a. CT  35.4625t  442.2 f. 6N  2G  30
G
b. CD  32.325t  865.7, where t  0 in 1990
c. The sales of cassette tapes are falling by 35.4625 million
per year the sales of CDs are rising by 32.325 million per 30
year. 3N + 2G = 60
d. Late 1983
e. B  3.1375t  1307.9
15 3N + G = 30
7. a. C  0.30t  0.40
b. Each minute costs 30¢, and it costs 40¢ to place the call. 6N + 2G = 30
c. $8.20
N
d. C  0.21t  0.28; each minute costs 21¢, and it costs 28¢ 0 5 10 15 20 25
to place the call; $5.74 3N + 2G = 30
9. a. DJ1: C  60t  120; DJ2: C  75t  100
c. DJ1 costs less than DJ2 if she is hired for longer than 13. a. For 3y  2x  12, the slope is 23 and the vertical inter-
1 13 hours or 1 hour 20 minutes. cept is at 4; for 4x  5y  20, the slope is  45 and the
11. a. T  3217.50  0.281I  21,4502 for 21,450  I  vertical intercept is at 4.
51,900 with a range of 3217.50  T  11743.50 c. 10, 42
b. There is a 28% tax on income for single taxpayers with 15. Possible: a, f, and i; Impossible: b, c, d, e, g and h
taxable income between $21,450 and $51,900.
17. a. PQ, PR, QR b. QR, PR, PQ
13. a. d  15x  300 b. about 1880
c. no
15. b. 30,300 feet Exercising Your Algebra Skills
1 26 A A 2K
17. a. y  5x  26 b. y   x  1. h  3. r  5. v 
5 5 b Bp Bm
19. b. 13, 22 c. 7x  2y  17 7. l  g . a
T 2
b 9. v 
rF
d. 9x  y  25 e. x  3 2p Bm

b. a2,  b
3 6 6 2 9 2
21. a. (i) and (iii) 11. y   x  6;  13. y   x  ; 
2 5 5 7 7 7
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 749

SELECTED ANSWERS 749

Section 2.4 d. y
A < 0, c < 1
1. a. A b. C c. B, C d. D x
e. B, D
5. a. I  49511.07942 t, where t  0 in 1990
b. I  0.51111.07942 t, where t  0 in 1900
c. I  14.59
1064 2 11.0794 2 t, where t  0 in year 0
e. $1557.19 billion; $1558.02 billion; $1557.95 billion
5. a. (vi) b. (v) c. (iv) d. (ii)
7. a. S  11911.0614 2 t, where t  0 in 1980 7. a. C  27,70010.98072 , where t  0 in 1980
t
b. $215.94 trillion c. $497.33 trillion d. During 2004 b. 17,352 cases c. 52.3 years
9. a. P1t2  21.811.0262 t, where t  0 in 1995 9. a. N  0.410.652 t b. 10.2 hours
b. 28.2 million c. 27 years
11. a. M  310.692 t
b. 0.68 mg c. 7.3 hours
11. a. C  1.3611.02012 t, where t  0 in 1990 13. About 96.3 years
15. a. I  I0 . 10.72 n, where I0 is the original level
b. 1.80 billion metric tons
13. a. 7.5 billion b. about 47 years b. 0.16807I0 , or about 16.8%
15. $106.14; $181.40 19. a. 64% b. S  0.956t c. 15.4 years
17. $1.24
10 , or $124 billion
11
21. a. T  442.210.87962 t b. C  865.711.03322 t
19. a. 1.125 b. 12.5% c. 36.4 c. Sales of cassette tapes are falling by 12.04% a year; sales
d. f 1x2  11.211.1252 x of CDs are rising by 3.32% a year.
d. In late 1985
21. D  96411.13862 t, where t  0 in 1981; 19,081
25. A  10, B  10, c  0.5
23. a possible; b and c impossible
25. In mid 1997 Exercising Your Algebra Skills
1. 2mn 3. 53x 5. 352a 7. 105x
9. 2 5x
11. 913 2 x
13. 10110 2  1011000x 2
3x
Exercising Your Algebra Skills
1. x8 3. a12 5. x2 7. r12 Section 2.6
9. w3 11. x1>4 13. z1 15. x15 3. a. P  34.611.0132 t b. 39.4 million c. 53.7 years
12 20
17. a b 19. a  2a b  b
6 3 2
5. 6.6 weeks
7. a. 23.45 years, 17.67 years, 14.21 years, 11.9 years,
Section 2.5 10.24 years
9. In mid 2051
1. a. Exponential: a, c, f; not exponential: b, d, e
11. 2137 B.C.
3. a. y
13. 780 years old
15. a. 78.5% b. 34 months, or 2 years 10 months
A > 0, c < 1 c. 114 months, or 9.5 years
17. 1000 times as strong
19. a. 106, or 1,000,000 times as loud
b. 1015 times more intense c. 1012 times more intense
d. 60 decibels e. 110 decibels
x 21. P1t2  61100.0065t 2; P1t2  6e 0.0149t

c. y
Exercising Your Algebra Skills
A > 0, c > 1 1. log x 6  6 log x 3. log (xy) 5. x 2
log 11 log 0.6
7.  1.23 9.  0.56
log 7 log 0.4
log 0.25 log15>4 2
 9.19 13.  23.32
log11.05>1.04 2
11.
log 0.86
x
15. 100 17. 6 19. 4
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 750

750 SELECTED ANSWERS

Section 2.7 f. No inverse g. No inverse

h. No inverse
1. a. Power function; k  0, p 0
c. Power function; k  0, 0 p 1 3. a. Domain: 0, 1, 2, 3, 4, 5; range: 1.12, 1.44, 1.84, 2.05, 2.48,
d. Power function; k  0, p  1 2.94
e. Power function; k 0, p 0 b.
b and f not power functions x 1.12 1.44 1.84 2.05 2.48 2.94
3. Exponential functions: b, c, g, and n; power functions: a, d, f (x) 5 4 3 2 1 0
e, h, j, and k; neither: f, i, l, and m
Domain: 1.12, 1.44, 1.84, 2.05, 2.48, 2.94; range: 0, 1,
5. a. y  ax 2 b. exponential c. y  bx 3 2, 3, 4, 5
9. 603.6 lb
11. a. f 1x2  3x1>2 b. f 1x2  3x 0.70752 7. p1 1t 2 
log t
 58.7084 log t
c. f 1x2  3x0.86848 d. f 1x2  0.55794x 2.2239 log 1.04
e. f 1x2  10x1>2 f. f 1x2  40x1 17. a. Each letter of the alphabet is matched with only one let-
13. a. 15 watts b. 120 watts ter of the alphabet.
c. 8 d. 1.875 watts; yes b. IS THIS MATH?
e. 9,645 windmills f. about 19 mph
17. a. T  442.2t 0.467, where t  1 in 1990 Exercising Your Algebra Skills
b. C  865.7t 0.119, where t  1 in 1990 25 log 0.20
c. In early 1989 1. 214  1.1113 3.  9.2309
log 0.84
19. About 43 miles log 1.75
8 32
5.  1.0823 7.  28.2597
21. The range is an additional 4 miles; the area increases by B 17 log 1.02
1246 square miles. 2>7
Q 4>3
9. a b 11. a b
y 81
23. About 7058 mph  4>3
12 27 Q
25.
H D  89 2H D(H)  2H  7920H 2
Review Problems
0.1 miles 28.14427 28.1427 1. a. Exponential, c 1
1 mile 89 89 b. Power function, 0 p 1
c. Exponential, c  1
10 miles 281.44 281.60
2. a. Exponential b. Power c. Logarithmic
100 miles 890 895.54 d. Power e. Power f. Exponential
g. Exponential h. Power i. Exponential
27. Slope of PQ is 1, slope of QR is 3, and slope of PR is 2. j. Power k. Power l. Linear
29. Slope of PQ is 2a  h, slope of QR is 2a  3h, and slope of m. Linear n. Exponential
PR is 2a  2h. 3. a. A b. D c. C d. B
e. E f. I g. H h. G
Exercising Your Algebra Skills
i. F j. J k. L l. K
1. 3 3. 16 5. x7 7. r4
4. a. (6) b. (5) c. (4) d. (3)
9. x 2 11. x1>2 13. a7>3 e. (1)
Section 2.8 6. F1t2  511.19352 t; 1.1935
3. x  1.098 7. a. P1t2  1.5  0.1t ; 1995
5. a. 4 b. 4 c. 3 d. 3 P
4
e. 4
3
7. a. 1 b. 2 c. 3
3
9. b. 1 c. About 0.414 d. About 0.707 P = 1.5 + 0.1t
2

Section 2.9 1 P = 0.95(1.1) t

1. a. Inverse b. Inverse t
0 2 4 6 8 10
c. No inverse, assuming that different students are the
same height b. P1t2  0.9511.102 t; during 1997
d. No inverse e. Inverse c. During 2000
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 751

SELECTED ANSWERS 751

102t  4
17. a. Domain: t  2; f 1 1t 2 
8. a. About 8.1 years b. About 20,424 eagles
c. during 2004 2
9. About 50.5°F b. Domain: all real numbers; g1 1x 2  1
3
x6
10. W  2t  141; 191 million 18. Exponential; F1x2  411.5 2 x
11. D  382t  2048; deaths are increasing by 382 deaths 19. a. g 1x2 b. h1x2 c. f 1x2
20. a. (i) f 1t2  0.2011.05652 , where t  0 in 1960 (ii) $2.37
per year; 10,834 deaths t

12. a. F1x 2  810.72492 x b. G1x 2  a b x  8

13
b. Ice cream c. 2093
c. H1x 2  2.722211.603572 x
3
d. $291.04 e. No
13. a. f 1t2  103.6t  1055 b. $2401.8 billion 21. 1166; 45 years
c. F1t2  105511.08322 t d. $2981.7 billion
22. 5.6 hours; 8.4 hours
e. Linear: $3127 billion; exponential: $5217 billion
14. a. I  1.135t  17.7 b. I  17.711.0442 t
c.
t Linear Model Exponential Model
Chapter 3
0 17.7 17.7 Section 3.2
5 23.375 21.952 1. a. (iii) b. (i) c. (iv)
d. (ii) e. (v)
10 29.05 27.226 3. a. Correlation coefficient must be between 1 and 1.
15 34.725 33.766 b. Slope and correlation coefficient must have the same
sign.
17 37 37 c. The smaller x is, the larger y will be.
25 46.075 51.938 5. r  0.9897; T  0.231s  4.06; 6.3 sec, 16.7 sec; the 45-
mph estimate
30 51.75 64.415
7. r  0.992; P  0.254t  10.682, where t  0 in 1900
9. a. H  1.78S  50.863; r  0.95.
b. same as (a)
d. Linear: $43,805; exponential: $47,652 c. H  1.96S  49.051
e. Power function
11. a. r  0.999; yes b. S  0.871T  67.508
15. a. Linear b. Linear c. 102.3
16. a. y
13. v  964.97t  0.32; cm>sec>sec
15. a. Length depends on mass
b. r  0.99998; yes
c. L  0.0404m  0.1
d. cm/gram; length increases by 0.0404 cm for each addi-
tional gram of mass.
e. L  0.0404m
17. a. A  28.23t  963.82, where t  0 in 1965
b. 2,234.17 million tons
y c. In late 2001
b.
19. a. C  0.022t  13.707
b. Cost per minute is 2.2¢; charge for service is $13.71
c. $57.71
21. a. 33
b. Increase slope and decrease vertical intercept

Section 3.3
x
1. i. Exponential ii. None iii. Power
iv. Exponential v. Power vi. None
c. No inverse vii. Logarithmic viii. Power
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 752

752 SELECTED ANSWERS

3. a. 9.3; 21.5 b. 0.3; 1.7 c. 151.2; 348.6 Section 3.4

d. 45.5; 197.9 e. Slower
1. a. P  3.000311.33032 t, where t  0 in 1780
5. 881 square miles b. Yes, since r  0.9989, whereas previously r  0.9982
7. a. 345 nanograms per milliliter 3. a. D  216.9711.04642 t, where t  0 in 1940
b. 335.5 minutes b. $4137.6 billion
9. a. N  1.506A0.3293 b. 1075.2 square kilometers 5.
Year t Average Debt
11. a. Linear: C  2.55t  11; exponential: C 
15.29 11.0735 2 t; power: C  6.76t 0.744, t  0 in 1975 1940 10 4.9431
b. 87.50, 128.37, 84.90
1950 20 1.7054
13. a. Linear: N  16.619t  16.726; exponential: N 
133.392 11.0428 2 t 1960 30 1.6230
b. Linear: 848; exponential: 1,084
1970 40 1.8741
c. Linear: in early 2009; exponential: in early 1998
d. 16.5 years 1980 50 4.0132
15. a. Linear: N  12.9203t  272.9816 1990 60 12.8951
exponential: N  26.017411.04052 t
power: N  0.3347 t 1.7269 2000 70 20.2061
logarithmic: N  1455.3584  501.7958 ln t
b. Linear: 1213 A  0.2520t  3.3284, where t  0 in 1930; $15.6 thousand
exponential: 2498 11. a. W  52.497211.3730 2 t, where t  0 in 1980
power: 1211 b. 2.2 years; amount of wind energy generating capacity
logarithmic: 926 doubles every 2.2 years.
d. Linear: in late 2065 c. 708,357 megawatts
exponential: in mid 1999
13. a. C  314.254211.00392 t, where t  0 in 1960
power: in late 2043
b. 178.1 years; it will take 178.1 years for the carbon diox-
logarithmic: in mid 2868
ide concentration to double.
e. 17.5 years
c. 381.8 parts per million
17. b. Linear: C  40.3864  0.335H d. In late 2021
exponential: C 52.729511.0011 2 H 15. a. N  91.109211.0602 2 t, where t  0 in 1960
power: C  0.2568H 1.014 b. 11.9 years; the number of telephones will double in
logarithmic: C  1402.8353  271.786 ln H 11.9 years.
c. College degrees increase by 335 for every 1000 addition- c. 3040 million
al high school diplomas. d. In late 2000
d. The number of college degrees is increasing with re- e. During 2040.
17. a. N  7.119512.0315 2 t, where t  0 in 1985
spect to the number of high school diplomas.
19. a. N  0.307211.329962 , where t  0 in 1979
t
b. 0.98; the number of computers connected to the Inter-
b. 1.32996; growth rate is about 33% per year. net doubles in less than a year.
c. 509.6 million c. 4.2
1011 thousand, or 420,000,000,000,000
d. Slower from 1984 to 1995 and then faster d. In late 1999
e. N  0.0657t 2.1432, where t  0 in 1979
f. 70.8 million Exercising Your Algebra Skills
3. y  11.827710.9499 2 x
g. Matches very well from 1980 to 1995 and then is slower
1. y  6.000711.05172 x
h. Power
5. y  2.2501112.14232 x 7. y  7.120310.05332 x
21. a. P  0.3378t  28.1705, where t  0 in 1945
b. 53.5%
Section 3.5
c. P  29.031411.00902 t, where t  0 in 1945
d. 56.8% 1. a. 166.3 feet b. 234.8 feet
e. P  20.5929 t 0.1907, where t  0 in 1945 c. 569.4 thousand pounds
f. 46.9% 3. a. D  0.000385n1.4865
23. a. T  4.4258  19.528 ln P; r  0.9950, so it seems a b. 9.2 cm; 15.8 cm c. 661
good fit 5. b. S  18.4490L1.1526 c. 45.4 cm
b. 40.1°C c. 371.8 kiloPascals d. 7470 cm>sec, or 245 ft>sec
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 753

SELECTED ANSWERS 753

7. a. S  10.2782L0.8600 b. 297 cm>sec Section 3.7

c. 205.0 cm 1. a. 0.9961; 0.9980; 99.61% of the variation
9. a. The number of species depends on area. b. y  6.284x1  0.435x2  34.561
b. N  5.5982A0.35209 c. 173.1 cm; within 1.1 cm
c. 109 species 3. a. 0.9157; 0.9569; 91.57% of the variation
d. 25,751 square miles b. y  0.7301x1  0.0547x2  0.1624
11. a. Angle depends on radius. c. 1.9 liters per second
b. 0° to 90° d. Vital lung capacity has a greater effect because its coeffi-
c. A  31.4584R0.8102 cient is larger than that of total lung capacity.
d. 2.49 meters 5. a. 0.3877; 0.6226; 38.77% of the variation
e. 22.6° b. y  0.1620x1  0.7117x2  5.891
13. a. T  0.0836R0.6216 c. 78.2, or close to 78.
b. 8.7 seconds d. 0.7386; 0.8594; 73.86% of the variation; y  0.1882x1 
c. 2200 meters 2.9607x2  1.9798x3  141.501; 86.266
d. 12.1 seconds e. 0.8518; 0.9229; 85.18% of the variation; y  0.1702x1 
1.2035x2  1.4929x3  1.8644x4  195.431; 72.8187
Exercising Your Algebra Skills
1. y  8.0002x1.5 3. y  12.0005x1.5 Review Problems
5. y  2.2501x 1.0843
7. y  7.1203x1.2733 1. A  0.1678B0.8757; r  0.7793
2. For all models, let t  0 in 1979; linear: H  51.79t 
Section 3.6 159.97; exponential: H  251.9211.0892 t; power: H 
1. a. (ii) b. (i) 218.35t 0.532; logarithmic: H  124.91  300.89 ln t;
c. (ii) d. (i) $1454.7 billion; $2123.0 billion; $1210.2 billion;
3. a. Residuals are 0, 0, 1, 1, 1; sum of squares is 3. $1093.4 billion
b. Sum of squares is 1.5, which is one-half the sum of 3. For all models, let t  0 in 1979; linear: H  13.90t 
squares for y  2x  1. 320.47; exponential: H  332.1111.0311 2 t; power: H 
c. y  2.3x  1; sum of squares is 0.3, or much less than 322.05t 0.1824; logarithmic: H  310.59  80.945 ln t;
for the preceding two models. $668.0 billion; $714.2 billion; $579.3 billion; $571.1 billion
5. a. S  5602.99E  43477.62; each additional year of edu- 4. a. Health expenditures are increasing by $51.8 billion
cation increases salary by $5,602.99. per year whereas public education expenditures are
c. 82,788,080.84 increasing by $13.9 billion per year.
7. a. S  111.5539E 2.1533 b. During 1983
b. 29,045,904.48 5. a. Health expenditures are increasing by 8.9% a year,
9. a. whereas public education expenditures are increasing by
Average speed 3.11% a year.
Planet Period t Distance D S  2PD>t b. In early 1984
6. H  3.68E  1012.33; Health expenditures increase by
Mercury 88 36.0 2.5704
$3.68 billion for every $1 billion increase in education
Venus 225 67.2 1.8766 expenditures.
Earth 365 92.9 1.5992 7. a. L  0.5432G0.6471
8. P  0.23w  0.14; an 8.5-oz letter would cost $2.10 by the
Mars 687 141.5 1.2941
model, but in reality it would cost $2.21.
Jupiter 4,329 483.3 0.7015 9. a. C  38.3991t  45.8474, where t  0 in 1970
b. $890.6 billion
Saturn 10,753 886.2 0.5178
c. In late 2007
Uranus 30,660 1782.3 0.3653 10. a. N  5.49111.02982 t, where t  0 in 1983
Neptune 60,150 2792.6 0.2917 c. In mid 2003
11. a. H  12.5877n  20.8676
Pluto 90,670 3668.2 0.2542
b. Height of the building increases by 12.5877 feet for each
story.
b. S  11.4233t0.3333 c. About 12 12 feet
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 754

754 SELECTED ANSWERS

12. f is linear, g is exponential, and h is logarithmic. 5. 8x 4  11x 3  3x  1 7. 3x 2  5x

y
9. 7x  3x 2 11. x 2  4x  3
5 f 13. x  x  6
2
15. x 2  25
3 17. x 2  4 19. x 2  2x  1
1
x
21. x 2  4x  4 23. 4x 2  24x  36
0 1 3

Section 4.2
3. a. (i) 3 (ii) 4 (iii) 4 (iv) 5
y b. (i) Positive (ii) Positive (iii) Negative (iv) Negative
g
5. a. (iv) b. (vi) c. (v)
5 d. (i) e. (ii) f. (iii)
7. y
3
600
500
1
x
0 1 3
x
y –7 –3 –1 1 3 4 6 7

h –300
3

–600
1
x
0 1 3 9. y

1000

13. a. t  1.03n  0.0186; each bounce will take about x

—7 —3 —1 1 3 5 7
1.03 seconds to return to the floor.
—1000
b. H  1.0889n  8.3482
c. H  9.000210.800072 n; the height of each bounce is —2000
decreasing by 20%. —3000
d. H  1.007t  8.695
e. The next bounce will reach a maximum height of 1.89 feet —4000
at 7.79 seconds and hit the floor at 8.26 seconds.
11. a. f 1x 2  1x  1 2 1x  3 2 1x  3 2
2
14. Yes.
15. If W  f 1S2, W  45.8 lbs; If S  f 1W2, W  57.4 lbs
9
b. f 1x 2   1x  1 2 2 1x  3 2
1
3
Chapter 4 13. a. (i) 2 real, 2 complex (ii) 3 real, 2 complex
(iii) 2 real, 4 complex (iv) 4 real, 2 complex
Section 4.1 b. (i) 3, 3 (ii) 2.386, 0.929, 5.108 (iii) 0.586, 3.414
1. a. 3 b. 4 c. 2 (iv) 0.209, 0.586, 3.414, 4.791
d. 8 e. 3 f. 6 15. a. 0.558, 0.769, and 2.040
3. 1, 0, 3 b. Increasing on 0.558
x
0.769 and 2.040
x
;
5. 3, 1, 2, 4 decreasing on 
x
0.558 and 0.769
x
2.040
c. 0 and 1.5
7. 2.165 d. concave up on 
x
0 and 1.5
x
; concave
9. 10.59, 1.732 and 11.41, 2.272 are the two turning points; down on 0
x
1.5
one inflection point e. 1.305 and 2.780
17. P1x 2  5x 2  10x
Exercising Your Algebra Skills
19. a. f 1x2  x 2  4x  12 b. x  2
1. 6x 3  10x 2  6x  8
c. f 1x2   x  5x  15 d. f 1x 2  x 2  5x  15
5 2 5
3. 5x 4  2x 3  9x 2  7x  1 4 4
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 755

SELECTED ANSWERS 755

2v0 9. 13 complete layers use only 819 grapefruit; 14 complete

21. a. 0 b. seconds
g layers use 1015 grapefruit.
v0 v02 11. 6565 inches
c. seconds d. cm
g 2g 13. a. 2y  2a b. 3y  6a
23. P1x 2  1x  0.872 1x  2.212 1x  3.662 c. 4y  24a d. 5y  120a

25. a. 3 b. Average value  3. 17. 105,625

c. Average value  8 d. Average value  9
Section 4.6
1 1
27. a. p1x2  1 b. p1x2  x 2  x  1 1. a. 56>5 b. 54>5 c. 11>5 d. 55
2 2
c. p1x2  ax 2  ax  1, a any real number e. 17>5 f. 1>11 g. 29 h. 5
d. p1x2  ax 2  ax  c, a and c any real numbers i. 3x  4  1>x j. 3x  4  1>x k. 3  4>x
l. 3x2  4x m. 3>x  4 n. 1>13x  42
o. 9x  16 p. x
Exercising Your Algebra Skills
3. a. 105  log 5 105  log 5 105log 5
1. 1x  42 1x  32 3. 1x  42 1x  32
b. c.
d. 105>log 5 e. 10log 5  5 f. log1105 2  5
5. 1x  42 1x  32 7. 1x  32 2 log1log 5 2
5
g. 1010  10100,000 h. i. 10x  log x
9. 1x  102 1x  102 11. x 1x  52 1x  42 j. 10x  log x k. 10x log x l. 10x>log x
x
m. x n. x o. 1010
13. x 1x  52 2
p. log1log x 2
5. c, d, and g are correct; while a, b, e, and f are incorrect.
Section 4.3
13. a. 2 b. 0
1. a. 806,991 cases. c. 2 d. 0
15. F1x 2  x  5, G1x2  x 4
c. 895,619 cases d. In early 2001

17. F1x 2  log x, G1x2  x  3

3. Answer depends on estimates.
7. T  0.00001873R2  0.0622R  323.5359
19. a. t  4.8 seconds b. t  5.6 seconds
9. P  0.03316t2  0.7830t  25.1921, t  0 in 1970; c. Domain: between 15 mph and 55 mph, which is between
18.84 million, 11.98 million 22 ft>sec and 81 ft>sec; range: 0
t
7 seconds
11. h  0.0001047d 2  0.4466d  496.9888 d. decrease the 20 and/or increase the 70
s 2  20s  1400
13. a. A  0.1437t 2  5.9982t  266.0148, t  0 in 1965 e. t 
b. 328.6 kg c. 244.9 kg 20s
3 5
15. a. D  0.000003644T 2  0.00005986T  1.0004 25. a. 2 b. c.
2 3
b. 0.97835 c. 43.2°C e. Yes, 1.618034
17. 3.4 seconds 1
27. a. V  S 3>2 b. S  6 2>3p1>3V 2>3
6 2p
Section 4.4 29. a. G b. L c. Q
d. X e. L
1. 50%
3. The discriminant becomes k 2 1b 2  4ac2; k  0 gives the
Section 4.7
same roots.
1.
5. a. 0.77 and 3.44 b. 0.14 and 4.81 f (x)  3 f (x  1)
x f (x) 5f (x) [ f (x)]2
c. 0.46 and 0.86 d. 0.44 and 0.94
e. None 3 5 25 8 UNDEF 25
7. a. Stable b. Unstable c. Stable 4 2 10 5 5 4
d. Unstable e. Stable
5 1 5 2 2 1
Section 4.5 6 3 15 6 1 9
3. The 344 should be 334.
7 8 40 11 3 64
5. P1x 2  x2  2x  1; P10.5 2   2.25; P132  16
9
4 5. a. y  12  m . 1x  52
7. a. 325 b. 5050 c. 500,500 b. y  y0  m . 1x  x0 2 , the point–slope form of a line
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 756

756 SELECTED ANSWERS

7. a. y 15.
1 n 1 2 3 4 5
x Turning point
1 2 3 4 1.4427 2.8854 4.3281 5.7708 7.2135
–1 for x  0

Regression equation: y  1.4427n; the linear fit is appro-

b. y priate; 2.1641
2
1 Exercising Your Algebra Skills
1 2 3 4
x 1. f 12x 2  4x 2  10x  3
3. f 14x 2  16x 2  20x  3
–1
–2
5. f 1x  1 2  x 2  3x  1
y 7. f 12x  12  4x 2  14x  9
c.
1 Section 4.8
1 2 3 4
x 1. T  8.65  45.201510.8292 2 t; worse
–1
3. T  70  128.106810.93952 t
–2
5. a. T  110  89.6910.9782 P
b. 31.87°C c. 129.77 kiloPascals
d. y
Section 4.9
1 1. a. Males: 176.8 cm; Females: 163.7 cm
x 192.91
1 2 3 4 5 b. Males: H  ,
1  1.876e0.166t
–1
171.52
Females: H 
1  1.73e0.211t
e. y c. Males: 192.91 cm; Females: 171.52 cm.
3. a. Turning point is farther to the right and higher; func-
1
tion eventually decays at the same rate.
x b. Turning point is farther to the left and lower; function
–1 1 2 3 4
eventually decays at the same rate.
–1
c. Turning point is farther to the left and lower; function
eventually decays at a faster rate.
y d. Turning point is farther to the right and higher; func-
f.
tion eventually decays at a slower rate.
2
Review Problems
1 1. y

x
1 2 3 4

x
–4 –2 –1 1 2 4
9.
x 3 4 5 6 7 8 9 10 11 f (x) = (x + 3)(x – 2)(x – 4)

y 31 23 18 15 16 22 34 46 42
2. y

g(x) = (2 – x)(x + 3)(x + 1)

11. a. x 2  3x  4  h
b. x 2  2xh  h2  3x  3h  4
x
c. 2xh  h2  3h d. 2x  h  3 –4 –1 1 3 4 5
e. 7.1; 7.01; 7.0001
13. T  350  31010.99472 t, where t is in seconds
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 757

SELECTED ANSWERS 757

3. y c. No zeros; turning point at x  0; no vertical asymp-

totes; end behavior as x approaches is 1 and end be-
havior as x approaches  is 1.
d. No zeros; turning point at x  0; vertical asymptotes at
x
x  3, x  3; end behavior as x approaches is 1
–3 –1 1 2 3 4 5 and end behavior as x approaches  is 1.
F(x) = (x + 2)(x – 3)(x – 4)(x – 1) 14. a. (i) f (ii) f

f(x) 3f (x) f (x)

4. y

x x
–4 –2 1 2 3 4 5 x

– f(x)
G(x) = (x + 3)(x – 2)(x – 4) 2
(iii) f (iv) f

5. P1x 2  1x  22 1x  32; the roots are x  2 and 3. f(x) f (x)

6. Q1x2  1x  52 12x  12 ; the roots are x  5 and x  12 .

7. R1x2  x1x  2 2 1x  12 ; the roots are x  0, 1, and 2
f (x − 3)

x x
9. 4 10. 20
f(x) − 4
11. 0 and 2.667
12. a. y

10 (v) f
8 x2 –5 f (x)
6
x 2 + 3x – 3 4
f(x + 3)
2
x x
–6 –4 4
–2
(vi) f
3x + 2 –6
f (x)

b. y
x
2x 3 + x2 +4
6
x2 – f (x − 4)

x
–3 –2 1 2 3 b. (i) f
2x 3 + 4
–2
f(x)

13. a. Zeros at x  2, x  2; turning point at x  0; no ver-

x
tical asymptotes; end behavior as x approaches is 1
and end behavior as x approaches  is 1.
b. Zeros at x  2, x  2; turning point at x  0; vertical
asymptotes at x  3, x  3; end behavior as x ap- – f(x)
proaches is 1 and end behavior as x approaches  is 1.
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 758

758 SELECTED ANSWERS

(ii) f (iii) f (v) f (vi) f

f(x)

3f(x) f(x) f(x)

f(x) − 4 f(x + 3)

x x x

– f(x − 4)

6
15. a. 19.4 b. 723 c.
7
f(x)
1
d.  e. 7.6 f. 47.5
4
31x 2  22 2x 2
h. 8x4  8x 2  3
1x  22
x g. i.
2
2x 2  3
(iv) f (v) f 1 2x 3  2x 2  x  1
j. k.
f(x) f(x − 3) f(x + 3) f(x)
x1 x2
2x 3  4x 2  x  2
l.
x1
x x 16. a. f 1g 102 2  2, f 1g 112 2  2, f 1g 122 2  3, f 1g 13 2 2  0
b. g 1 f 102 2  2, g 1 f 112 2  2, g 1 f 122 2  3, g 1 f 13 2 2  1
(vi) f
c. f 102  g 102  3, f 112  g 112  2, f 12 2  g 12 2  5,
f 132  g 132  3
f 102 f 11 2 f 122 3 f 13 2
f(x)

 2,  , 0
g 102 g 112 g 122 2 g 132
d. is not defined,
x
17. a. f 1g 112 2  3, f 1g 122 2  3, f 1g 132 2  1, f 1g 14 2 2  1
b. g 1 f 112 2  3, g 1 f 122 2  1, g 1 f 132 2  1, g 1 f 14 2 2  3
c. f 112  g 112  4, f 122  g 122  5, f 13 2  g 13 2  4,
f 142  g 142  2
– f(x − 4)

f 112 1 f 122 3 f 132 f 14 2

c. (i)  ,  ,  3, 1
g 112 3 g 122 2 g 132 g 142
f d.
f(x)
18. 0.106262
t
3.29374
19. 15150
x 20. a. 3 b. 5 c. 1 d. 5
e. Yes, 3
– f (x) 21. y  x 4  6x 2  x  4
22. a. y

(ii) f (iii) f 25
f(x) 20
3f(x)
15
10
x 5
x
f(x) − 4 –5 –3 –1 1 5 7
–5
–10

1x  32 1x  12 1x  12 1x  3 2 1x  6 2
f(x) (iv) f
b. R1x2 
x
1x  22 1x  22 1x  5 2
f(x)

f(x − 3)
23. a. (i) and (iv) are exponential; (ii), (v), and (vi) are power;
x (iii) is logarithmic.
b. There are many possible answers.
(i) f 1x2  2x  11 (ii) f 1x2  1x  5
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 759

SELECTED ANSWERS 759

1 x 7. a. Wn1  1.03Wn  50, W0  8000, n  0 in 1992

(iii) f 1x 2  log1x  4 2 (iv) f 1x 2  a b  7 b. W0  8000, W1  8190, W2  8386, W3  8587,
3
(v) f 1x 2  x1  9, x  0 W4  8795, W5  9009, W6  9229, W7  9456,
(vi) f 1x2  1x  5 2 3, x  5 W8  9690, W9  9930, W10  10,178, W11  10,433,
24. a. N  4.819611.38402 t, where t  0 in 1990 W12  10,696, W13  10,967
b. B  82.765010.91702 t, where t  0 in 1990 c. 240
c. R  398.894211.26912 t, where t  0 in 1990 9. a. Wn1  Wn  r . 1400  Wn 2 , where r is the propor-
d. R  398.890111.26912 t, where t  0 in 1990; almost tion learned
identical b. 0
r
1
11. 41 rabbits

Chapter 5
Section 5.3
Section 5.1 1. 10.15, 10.30, 10.45, 10.61, 10.76, 10.92, 11.08, 11.24, 11.40,
1. 5.5 mL; 1.9 mL; about 31.1 hours; about 82.7 hours 11.57; 40; about one-quarter of the way
3. a. 320, 320, 320, 320, 320, 320; same 3. 5.10, 5.20, 5.30, 5.40, 5.51, 5.61, 5.72, 5.83, 5.95, 6.06; 200;
7. 16.67 mg 9. No not close at all yet

11. 325 mg; virtually all washed out in 24 hours 7. b. 275, over estimate c. 950, over estimate

13. r3
r1
r2 9. a  0.05, b  0.000003125; 100, 104.97, 110.18, 115.65,
121.39
15. {0, 4, 8, 12, 20, …} 17. {0, 0.5, 1, 1.5, 2, 2.5, …}
19. 59, 2, 17, 54, 115, 206, . . .6
13. a. b will increase.
15. a. ∆Pn

21. e 1, , , , , , ... f
2 4 8 16 32
3 9 27 81 243 a, b > 0
1 1 1 1 1
23. 1, , , , , , . . .
2 3 4 5 6
25. 50, 0.8, 0.96, 0.992, 0.9984, 0.99968, . . .6
27. Diverges 29. Diverges 31. Diverges Pn
33. Converges to 0 35. Converges to 0
37. Converges to 1 39. Strictly increasing, no concavity
3 a
b. L 
41. Strictly increasing, no concavity Bb
43. Strictly increasing, concave up c. The population is increasing because Pn  0 for
45. Strictly decreasing, concave up 0
Pn
L.
47. Strictly decreasing, concave up d. The population is decreasing because Pn
0 for
Pn  L.
49. Strictly increasing, concave down
17. Logistic: Pn  0.1387Pn  0.00053456Pn2, r  0.2138;
51. a. 2, 2.25, 2.3704, 2.4414, 2.4883, 2.5216, 2.5465, 2.5658,
linear: P  3.425a  32.5278, r  0.9864; exponential: P 
2.5812, 2.5937; yes
38.373311.05322 a, r  0.9945; power: P  29.2962a0.3724,
b. e100  2.704814, e500  2.715569, e1000  2.716923,
r  0.9391; the exponential model seems to be the best fit.
e10,000  2.718146, e100,000  2.718268, e1,000,000 
2.718280; e  2.718281828 19. a. 255 cm, about 5 weeks b. a  0.8772, b  0.0037
c.
1
53. 55. 1
e t 1 2 3 4 5 6

Section 5.2 Actual H 17.9 36.4 67.8 98.1 131.0 169.5

1. a. $383,519 b. $506,055 Predicted H 17.9 32.4 57.0 94.9 144.9 194.3
3. About 7.6 months
bn1  11.06 2bn , b0  2000
t 7 8 9 10 11 12
5. a.
b. bn1  11.06 2bn  1000, b0  2000 Actual H 205.5 228.3 247.1 250.5 253.8 254.5
c. bn1  11.06 2bn  2000  10001n  12, b0  2000
d. bn1  11.06 2bn  200011.1 2 n1, b0  2000 Predicted H 225.0 235.1 236.8 237.0 237.077 237.080
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 760

760 SELECTED ANSWERS

 n is xn  1n  12!x1 , so the solution

21. If t0  0, t1  22, t2  46, etc., Pn  1.2751Pn  xn1
23. The solution to
0.3149Pn2, with r  0.9107. xn
 n , which is xn  3 1n  12!4 2x1 , grows much faster.
xn1 2
23. to
xn
n 1 2 3 4 5
Review Problems
Actual Pn 9.6 29.0 71.1 174.6 350.7
1. a. 51, 5, 11, 17, 23, . . . 6
Predicted Pn 9.6 25.3 65.5 162.7 361.0
b. e 3, , 9, , ,... f
9 81 243
2 4 5
n 6 7 8 9 10 c. 50, 0.7, 0.91, 0.973, 0.9919, . . . 6
Actual Pn 513.3 594.4 640.0 655.9 661.8 2. a. 52, 10, 18, 26, 34, . . . 6
b. 512, 4, 4, 12, 20, . . . 6
Predicted Pn 607.1 616.1 610.2 614.1 611.5
c. e 5, , , , , . . . f
5 5 5 5
Pn
 2.648010.99312 Pn, or Pn  2.6480Pn . 10.99312 Pn, 3 9 27 81
d. 510, 11, 8, 17, 10, . . .6
25.
Pn
with r  0.9506 3. a. 52, 7, 9, 16, 25, . . . 6 b. 53, 7, 10, 17, 27, . . . 6
27. a. a  2.792, b  0.0092 4. 167 mg 5. 21 mg 6. 150 pounds
b. 303.5 gigawatts c. between 1980 and 1985
d. Soon after 1980 7. 15,000; at year n  22.
8. b  0.000017; un1  1.2un  0.000017un2, u0  1000
Section 5.4 9. a. Pn1  1.3Pn  0.0004Pn2
1. n  7.9 hours, so time of death  1:06 A.M.; about 18 min- b. Pn1  1.3Pn  0.0004Pn2  2000
utes difference c. Pn1  1.3Pn  0.0004Pn2  400
d. Pn1  1.3Pn  0.0004Pn2  0.10Pn
3. n  60.7 minutes; about 2 minutes longer
10. a. Mn1  0.85Mn , M0  100%; there will be 0.34% maple
5. 36.3 minutes
syrup content.
7. 4.57 hours before 9 A.M. or about 4:26 A.M. b. Pn1  1.1Pn  2000, P0  50,000; John will have
9. a. Linear: d  0.5567T  34.8914, r  0.8465; $161,561.97.
exponential: d  622.709710.8999 2 T, r  0.9841; c. Pn1  Pn  20,000, P0  2,000,000; Tn1  1.10Tn ,
power: d 1,814,899,980T 5.1962, r  0.9907 T0  2,000,000; P5  $1,900,000, T5  $3,221,020; in-
b. Power function c. 17 days creasing roughly exponentially
11. b and c are possible; a and d are impossible 11. a. Qn1  0.75Qn  1, Q0  6
13. t1 t2 b. Qn  210.752 n  4
c. 2.4 weeks
Section 5.5 12. Increasing concave up 13. Increasing concave up
1. 1.999023438, 1.999999046, 1.999999999; the partial sums 14. Increasing concave down
are approaching 2. 15. Decreasing concave down
3. 4.57050327, 4.95388314, 4.99504824; the partial sums are 16. Increasing concave down
approaching 5.
17. Increasing concave up
5. 170.9951, 9973.7702, 575,251.1777; the partial sums do not
18. None of the above 19. Increasing concave up
converge.
7. A total of 29,997 cases from 1960 through 2000; no more
than 3 new cases between 2000 and 2010
Chapter 6
9. 64.5 billion kilowatt-hours
11. 653,722.5 metric tons Section 6.1
13. 1,900,648 pages 3. b  3.4409, c  12.4836
15. About 45.6 years 5. a  24, u  36.87°
17. 79.95 feet, 83.77 feet, 83.99 feet; 84 feet 7. The pole is 26.6 feet high.
19. 49.31 feet, 49.99 feet, 49.9998 feet; 50 feet 9. a. 571.5 feet b. 573.7 feet

a b 
25 1 25 11. a. 27 feet b. 26.1 feet
21.
100 1  1>100 99 13. 20.14 to 21.82 feet.
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 761

SELECTED ANSWERS 761

15. Jack is 27.3 feet from the base of the cliff. b.

17. a. θ
tan U 0 0.5 1 1.5 2 2.5 3
180°
U 0° 26.6° 45° 56.3° 63.4° 68.2° 71.6° 150°
120°
90°
Section 6.2 60°
3. a  11.5, b  3.3 5. u  15.47°, a  28.9 30°

7. c  135.915, b  129.263 cos θ

–1 –0.75 –0.5 –0.25 0.25 0.5 0.75 1
9. 9659 feet
c. Inverse cosine function
11. a. 5.71° c. 20.0998 miles
13. 507.6 feet 15. About 36.9°
Section 6.4
17. 41.81°
1. 0.954, 0.314; u  17.5° 3. cos u  0.8, tan u  0.75
3
19.  3.37 inches 21. 85.6 feet 3 4
cos 27° 5. sin u   0.6, cos u   0.8
5 5
23. 38.3 mph
7. 0.854, 0.609 9. 0.714, 0.980
25. a.
cos U 1 0.8 0.6 0.4 0.2 0 11. a. (ii) Is an identity c. 0° and 180° by the graphs

U 0° 36.9° 53.1° 66.4° 78.5° 90° Exercising Your Algebra Skills

1. sin x 3. sin2x
c. Inverse cosine function
5. 1  2 sin u cos u  1  sin 2u

Section 6.3 1
7. 9. tan2x
cos x
1. 0.707 3. 1 5. 0.5 7. 0.707
9. 0.5 11. 0.5
Section 6.5
13. Negative; Quadrant IV 15. Negative; Quadrant III
1. C  91°, a  5.90, and c  13.47
17. Negative; Quadrant IV 19. 0
3. C  80°, a  13.98, and b  22.09
21. Negative; Quadrant III 23. Negative; Quadrant II
5. B  50.47°, C  89.53°, and c  15.56; or B  129.53°,
25. Positive; Quadrant III 27. Positive; Quadrant IV C  10.47°, and c  2.83
29. Negative; Quadrant III 31. Positive; Quadrant III 7. There are various ways to describe the location of the third
33. a. Positive; negative; negative; ship. The ship is 37.9 miles in the direction of 54° south of
b. 0°; 90°; 180°; 270°; 360°; 450°; 540° east from the northernmost ship and 29.5 miles in the di-
c. rection of 41° north of east from the southernmost ship.
y 11. 32.2° 13. About 1332 miles
15. About 47.2 meters, 58°.
0.5 17. a. about 390 miles b. About 307 miles
19. About 329 meters 21. 2.4
0 x
0° 90° 270° 450° 540°
Review Problems
–0.5
75
1. a. d  b. 0°  u  72°
cos u
c. 72°  u  72°
d. Yes; 180° 2. a. x  75 tan u b. About 35 feet c. 0°  u  63°
35. a. 3. a. y  500 tan a b. 182 meters c. 420 meters
d. 76°
cos U 1 0.75 0.5 0.25 0 0.25 0.5 0.75 1
3 5
U 0° 41.4° 60° 75.5° 90° 104.4° 120° 138.6° 180° 4. a. 8.5° b. y  x c. y   x
20 12
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 762

762 SELECTED ANSWERS

5. 4.5 inches Section 7.2

6. a. 6.88 inches b. 9.83 inches 1. a. Periodic; period  2 b. Periodic; period  2
7. a. 14,424.3 cubic inches b. 17,819.1 cubic inches c. Periodic; period  2 d. Periodic; period  1
c. 13,82411  cos u2 sin u e. Periodic; period  2 f. Not periodic;
d. About 60° g. Not periodic
h. A constant function can be considered periodic, but a
8. a. 253.2 feet b. 93.4 feet
definite period cannot be determined.
9. 89.85° 10. About 6430 meters i. Periodic; period  2 j. Periodic; period  20
11. a. About 5 inches 3. a. 3.6 b. 365 c. 8.4 hours d. 15.6 hours
c. From 10 to 15 minutes after, from 15 to 20 minutes
5. H1t2  12  2.4 sin c 1t  802 d
after, and from 40 to 45 minutes after 2p
365
12. a. sin u  0.76822, cos u  0.64018
7. W1t2  25  25 sin a
6 5 2p
b. sin u   0.76822, cos u   0.64018 tb
11
261 261
9. a. 2 sin a b. 7.5 sin a
13. a. About 38 feet b. 37.4 feet p p
tb tb
c. The tree is 31.9 feet tall, and the top of the tree is 31.5 feet 40 150
above the ground.
11. a. T1t2  350  10 sin a tb
p
14. The fire is 9.5 miles from tower A, and 10.8 miles from 5
tower B.
13. a. f 1x2  51  21 sin c 1x  22 d
p
15. a. 13 yards b. 22.6° c. 37 yards d. 161.1° 12
16. a. 17.9 miles b. 163.5°
b. f 1x2  51  21 cos c 1x  8 2 d
p
c. About 1 hour; about 2.1 hours d. $18.00 12
17. a. 55.7° b. 3268 feet
15. H1t2  22  42 sin c 1t  1312 d
2p
18. a. 22° b. 5.2 centimeters c. 49.4 cm2 365
d. If the angle u is opposite side c, the area is 12 ab sin u.
17. Because the sine function is periodic there are infinitely
many correct formulas for each graph.
a. y  2 sin a b b. y  5 sin 2x
x
Chapter 7 3
c. y  4 cos 4x

d. y  3 sin a b e. y  1.5 cos a b f. y  2  cos 2x

x x
Section 7.1
5 4
1. The length of her fingernails is a periodic function with pe- g. y  5 cos 2x h. y  10  6 sin 8x
riod of 1 week. i. y  3  3 sin 4x j. y  3  3 cos 4x
k. y  sin a xb
5. a. 135° b. 144° c. 120° d. 85.9° p
l. y  4 cos px
e. 143.2° f. 171.9° g. 22.5° h. 300° 2
i. 270° j. 300°
19. Vertical shift is 59; amplitude is 12; period is 24 hours;
23 23
p
frequency is 12 ; T1t2  59  12 sin 3 12
p
1t  9 2 4
7. a. b. 1 c.  d. 1
2 2 23. About 12 times
23 1 22 25. a. For t in seconds,
e. f. g. h. 0.9749
2 2 2
y
9. Same graph
11. a. Roughly 10 years b. Roughly 10 years 250
c. Maxima: 1847, 1857, 1867, 1877, 1886, 1895, 1906, 200
1915, 1927, 1936; minima: 1852, 1862, 1872, 1881, 1891,
150
1901, 1908, 1920, 1930
d. Maxima: 1853, 1857, 1861, 1864, 1873, 1876, 1886, 100
1896, 1904, 1913, 1923, 1933; minima: 1847, 1855, 1859, 50
1862, 1868, 1882, 1890, 1899, 1908, 1918, 1928 t
e. The years in which a maximum (minimum) occur are 25 50 75 100 125 150
roughly 10 years apart.
f. Clear points of inflection near: 1850, 1858, 1863, 1866,
b. y  110  100 sin c 1t  7.52 d
1872, 1875, etc.; roughly halfway between a maximum p
and a minimum 15
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 763

SELECTED ANSWERS 763

c. x  100 sin a
p Exercising Your Algebra Skills
tb
15 3p
d. 0 t 7.5, 22.5 t 37.5, 52.5 t 67.5, etc. 1. u  0.90  np 3. u   np
4
e. x  100 sin a tb ; 7.5 t 22.5, 37.5 t 52.5,
p
3p
15 5. u  0.67  np 7. u   np
etc. 4
f. x2  1y  1102 2  1002
Review Problems
27. B1t2  4.00  0.35 sin c 1t  1.352 d
2p
1. cos x  sin ax  b and 5.70   7.2708.
5.4 p p
29. a. One such model might be 2 2
T  0.0042t  14.67  0.08 sin a tb .
p
2. a. u  120°, 420°, 480°, . . . ; u  240°, 300°, 600°,
25
660°, . . .
b. Yes c. 15.2°C
3. a. u  315°, 405°, 675° . . . ; u  45°, 315°, 405°,
Section 7.3 675° . . .
7p 9p 15p p 7p 9p
1. Eleven hours of daylight on day 55 (February 24) and again b. u  , , ...;u  , , ,
4 4 4 4 4 4
on day 287 (October 14); ten hours of daylight on January 23 15p
and November 16; San Diego always has more than 9 hours ...
4
of daylight.
3. a. 72° b. 66° Vertical Phase
c. Reaches 70° at 1.1 minutes after 9 A.M. and every 20 min- shift Amplitude Frequency Period shift
utes thereafter and also 8.9 minutes after 9 A.M. and every 2p
4. 325 10 9 9 0
20 minutes thereafter; reaches 67° at 12.3 minutes after
9 A.M. and every 20 minutes thereafter, and also 17.7 min- 5. 63 3 2p
25 0
25
utes after 9 A.M. and every 20 minutes thereafter.
2p
5. a. 1.13 hours, or 1 hour 8 minutes 6. 71 2 15 15 0
b. 0.35 hour, or 21 minutes p
7. 80 13 12 24 15
7. 99th day (April 9) and 346th day (December 12)
p
9. a. 1.2 and 2.8 months b. 4.7 and 11.3 months 8. 38 8 12 24 5
c. The average daytime high temperature never reaches 80°F. p
9. 100 25 36 72 0
Exercising Your Algebra Skills 10. 100 25 2p
97 0
97
1. No solution 3. No solution
2p
5. 0.85  2np and 3.99  2np 11. 145 40 83 83 0
7. 1.11  np and 2.03  np
20. a. y  5  2 cos 1 2p
3 t2 (where t is in radians), which starts
p at a maximum height and maximum displacement for-
9.  np, 0.30  2np, and 2.84  2np
2 ward of the crossbar.

b. x  6.93 sin a
Section 7.4 2p
tb
6
b. y  tan a xb c. y  3 tan x
1
1. a. y  tan 2x
2
d. y  2 tan x e. y  tan1x  30 2 21. D  160  4010.929572 t cos a tb
p
f. y  10  tan x 3

3. y  tan a xb
3 22. a. H  1.5  1.5 sin11000pt2
2
b. H  1.5  1.5 sin c 1000 p at  bd
1
5. The building is 37.5 meters tall; the smokestack is 45.1 me- 2000
ters tall.
23. H  9  5 sin a
2p
7. a. m  tan a, where a is the angle measured from the pos- tb
3
itive x-axis to the line
24. a. y  9  5 sin a
b. m  tan a, where a is the angle measured from the line p
xb
y  b to the line 10

11. a. b  arctan a b  arctan a b

6 1 4000p
b. About 2.4 feet 25. a. 1000p, or 3142 miles b. , or 4189 miles
x x 3
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 764

764 SELECTED ANSWERS

10,000p b. u  1.11  np c. u  1.43  np

c. , or 10,472 miles
3 33. a. x  tan 1.35  4.46
1000p b. x  sin 0.5  0.48
d. , or 1047 miles
3
26. a. 360p  1131 radians b. 720p  2262 feet
y
Chapter 8
28.
T S R
2 Section 8.1
1 π 3. Not an identity; 10, 12 5. Identity
7. Not an identity; 10, 02
π x
0 9. Identity
–1 3 2π
3 2π 3π 4π 11. Identity
–2
13. Not an identity; no points of intersection
29. 15. cos 3x  4 cos3 x  3 cos x
Phase
Frequency Period Amplitude Shift 17. cos 5x  16 cos5 x  20 cos3 x  5 cos x. There are many
3 8p
equivalent forms.
a. 4 3 2 0
1  cos 4x
3 8p 4p 21.
b. 4 3 2 3 8
c. p 2 2 3 23. a. Looks like y  sin x
4p
b. Looks like y  sin x
3 8 1
d. 4p 2 p
3 25. 7.15° or 82.85°
27. a. y  5 cos1t  126.87°2
4p 8np 20p 8np 31. a. Even b. Odd c. Even
30. a.  and 
9 3 9 3 d. Neither e. Neither f. Even
4p 8np 4p 8np g. Neither h. Even i. Neither
b.  and  
9 3 9 3 j. Neither
2 3 4 3 33. For x  p6 : If n  5, the sum is about 1.969. If n  10, the
c.   2n and   2n
3 4p 3 4p sum is about 1.999. Values approach 2. For x  p3 , more
8 1 8 16 1 8 terms are needed.
d.   n and   n
9 p 3 9 p 3 39. Not an identity; 10, 02
31. a. 41. Not an identity, 10, 12 , 1p, 1 2
43. Not an identity; 10.393, 0.172 2
y

45. Not an identity; 10, 0 2

1.5
1
0.5
Section 8.2
1. T2 102  1, T2 10.1 2  0.995,
x
–3π –2π –1π 1π 2π 3π

T2 10.22  0.98, T2 10.3 2  0.955, T2 10.4 2  0.92,

–0.5

T2 10.52  0.875, T2 10.62  0.82

–1
–1.5
3.
x T3 (x) sin x T3 (x)  sin x
b. y

1.5
0 0 0 0
1 0.1 0.09983 0.09983 8
108
0.5
0.2 0.19867 0.19867 3
106
x
–1 1
0.3 0.29550 0.29552 2
105
–0.5
–1 0.4 0.38933 0.38942 9
105
–1.5
0.5 0.47917 0.47943 3
104

32. a. u  0.98  np 0.6 0.56400 0.56464 6

104
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 765

SELECTED ANSWERS 765

5.
x  4p
25  3p
25  2p
25  25
p
0 p
25
2p
25
3p
25
4p
25

sin x 0.4818 0.3681 0.2487 0.1253 0 0.1253 0.2487 0.3681 0.4818

sin x  0.1640x 3  12.851

1014 2x 2
25. z2  169 c cos a2 arctan b  i sin a2 arctan b d
5 5
 0.9998x  12.886
1015 2; very close 12 12
7. sin1x 2 2  x 2  x 6 
1 1 10
x 27. z2  53923cos 11.024782  i sin 11.024782 4
29. z2  743 cos 14.382102  i sin 14.382102 4
6 120
y
y = x2 −
1 6
6
x +
1 10
120
x 31. z2  673 cos 16.709622  i sin 16.709622 4
2
33. z3  2035  828i 35. z3  14,625  5500i

37. z3  2197 c cos a3 arctan b  i sin a3 arctan b d

12 12
x
–π –0.5π –0.25π 0.25π 0.5π π 5 5
39. z3  15,6253 cos 16.642902  i sin 16.642902 4
y = sin(x 2 )
–2 41. z3  623.712273cos 11.076312  i sin 11.076312 4
Good match for x between 0.5p and 0.5p 43. z3  70.092803cos 12.267912  i sin 12.267912 4
x3 2 x2 x4 2 x6 x8 45. z0  1, z1  1  2i, z2  3  4i, z3  11  2i, and
9. ax  b  a1   b 1   1, z 4  7  24i
6 2 24 72 576
when x is close to 0 49. a. zw  1ac  bd2  1ad  bc2i
b. zw 
x2 x4 2a2  b2 2c2  d 2 3cos 1u1  u2 2  i sin 1u1  u2 2 4,
11. cos x  1   where tan u1  b>a and tan u2  d>c
2 32
c. zw  7 z7 7 w7 3 cos1u1  u2 2  i sin1u1  u2 2 4
13. a. 1
x2 x x 1x 2 2 d. (i) 51cos 0  i sin 0 2  5
b. 1   
2 2 6
23
(ii) 1 . acos  i sin b 
x2 p p 1
c. 1  , the quadratic approximation for cos x  i
2 6 6 2 2

51. a. 2z  2 7 z7 c cos a b  i sin a b d

u u
Section 8.3 2 2
1. 7 z7  5; u  arctan a b  0.64350 and 2z  2 7 z7 c cos a  pb  i sin a  pb d
3 u u
4 2 2
13 13
3. 7 z7  13; u  arctan a b  0.39479 i  11 acos  i sin b 
5 1 p p 1
b.   i,
12 B2 2 6 6 2 2
23
5. 7 z7  25392  73.43024; u  arctan a b  0.51239 and 21 acos b 
9 7p 7p 1
 i sin  i
16 6 6 2 2
 7 z7 1>n c cos a b  i sin a b d ,
u u
7. 7 z7  274  8.60233; u  p  arctan a b  2.19105
7 d. z1>n
5 n n
 u  2p
 7 z7 1>n c cos a b  i sin a b d,
u 2p
23
9. 7 z7  267  8.18535; u  p  arctan a b  3.35481
z1>n
n n
8
u  4p u  4p
z1>n  7 z7 1>n c cos a b  i sin a b d, . . .
11. 13 c cos aarctan b  i sin aarctan b d
12 12 n n
5 5
13. 25 3cos 12.214302  i sin 12.214302 4 Section 8.4
15. 273 3cos 10.358772  i sin 10.358772 4 1
1. a. The limiting value will be real whenever C  .
17. 217 3cos 10.755972  i sin 10.755972 4 4
19. z 2  7  24i 21. z 2  119  120i b. 50.5, 0.35, 0.2225, 0.14950625, 0.12235212, 0.11497004,
0.11321811, 0.11281834, 0.11272798, 0.1127076,
23. z 2  25 c cos a2 arctan b  i sin a2 arctan b d
3 3 0.112703, 0.112702, 0.112702, . . . 6
4 4 c. Diverges to infinity
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 766

766 SELECTED ANSWERS

1  11  4C 16. 2p 17. 2p 18. p 19. p

3. The equilibrium levels, , will be real when
2 20. 4p 21. 4p
1
C .
4
5. Every graph of the form y  x3  C intersects the line Chapter 9
y  x.
p
7. The limiting values are  np for any integer n. Section 9.2
2 1. 5 3. 241 5. 220  4.47

7. 13.5, 62 9. a , b
Review Problems 8 2
1. Not an identity 2. Identity 3 3
3. Not an identity 4. Not an identity 11. 1x  52 2  1y  2 2 2  25
5. Identity 6. Identity 7. Identity 13. 1x  62 2  1y  4 2 2  16
8. Identity 9. Not an identity 17. 1x  22 2  1y  3 2 2  25; center 12, 3 2, radius  5
19. 1x  52 2  1y  2 2 2  100; center 15, 2 2 , radius  10
11. With T3 1x2  3x  x 3, we have T3 10.22  0.564.
9
2 21. 1x  12 2  1y  3 2 2  4; center 11, 32, radius  2
9 3 y 23. a.
T3(x) = 3x − x
2
t 2 1 0 1 2 3 4 5
2

1
y = sin 3x x  3  2t 1 1 3 5 7 9 11 13
1.0
–1.0 –0.6 –0.4 –0.2 0.2 0.4 0.6
x y  4  5t 14 9 4 1 6 11 16 21
–π –π π π
3 6 6 3
–1 b. y

–2 12
8

12. With T5 1x2  3x  x 3  x 5, we have T5 10.22 

9 81 4
x
2 40 –1 1 3 5 7 9 11 13
0.564648.
–8
y
–12
–16
1
–20

d. y  1   1x  5 2
5 5
–1.0 –0.8 –0.6 –0.4 –0.2 0.2 0.4 0.6 0.8
y = sin 3x c. 
x 2 2
–π –π π π
e. a0, b , a2, b , a4, b , a6,  b , a8,  b ,
3 6 6 3 23 13 3 7 17
2 2 2 2 2
a10,  b , a12,  b
9 3 81 5
–1 27 37
T5(x) = 3x − x + x
2 40 2 2
13. Within 0.055556
t 1.5 0.5 0.5 1.5 2.5 3.5 4.5
14. a. z  10 3 cos 12.214302  i sin 12.214302 4
x  3  2t
w  229 3cos 10.380512  i sin 10.38051 2 4
0 2 4 6 8 10 12
b. z . w  53.851653cos 11.833792  i sin 11.833792 4 y  4  5t 11.5 6.5 1.5 3.5 8.5 13.5 18.5
 1.856953cos 12.594812  i sin 12.594812 4
z
y1  y0
w 25. y  y0  1x  x0 2; the two-point form of a line
x1  x0
 0.538523cos 12.594812  i sin 12.594812 4
w
z 27. 13, 42, 15, 0 2
15. a. z  31cos 52°  i sin 52°2
b. z  1.84698  2.36403i
Exercising Your Algebra Skills
c. z 5  2431cos 260°  i sin 260°2 1. 1x  4 2 2  9
d. 2z  1.732051cos 26°  i sin 26°2 3. 1x  32 2  4
and 1.732051cos 206°  i sin 206°2 5. 1 y  52 2  1 7. 1 y  22 2  16
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 767

SELECTED ANSWERS 767

Section 9.3 1x  32 2 1y  22 2
17.   1; center 13, 2 2; vertices 13, 12 ,
1. a. rightmost point; leftmost point 1 9
b. point on the right c. point on the left 13, 52 , 12, 22, 14, 22; foci 13, 2  2 122,
13, 2  2 122
3. The orbit will decay and the satellite spiral in.
5. Venus’ orbit is the most circular and Pluto’s the least circular.
Section 9.4
x2 y2
7.  1 x2
3352.41 3210.8 1. a. y  b. 24 inches
24
x2 y2 b b
9. a.  1 3. b. y   x c. 
96,100 65,792.25 a a
1x  12 2 1y  12 2
 1; ellipse with center at 11, 1 2 ,
b. 348.182 feet c. 402.358 feet

11. center 10, 02; vertices 1 12 , 02, 1 12 , 02 , 10,  13 2, and
5.
16 4
10, 13 2; foci 1 15
6 , 02 and 1 6 , 02
15 major axis parallel to x-axis, and a  4, b  2, so c  112 .
y 1x  12 2 1y  12 2
7.   1; hyperbola with center at
16 4
0.6 11, 12, axis parallel to the x-axis and a  4, b  2,
so c  120 .
1x  2 2 2 1y  1 2 2
0.3

  1; hyperbola with center at

14 2 1 389 2
x 9. 38
–0.6 –0.3 0.3 0.6
138
–0.3 12, 12 , axis parallel to the x-axis, and a  ,
138 1494 2
–0.6 b , so c  .
3 6
1x  32 2  1y  2 2 2  94 ; circle with center at 13, 2 2
1x  1 2 2 1y  1 2 2
11.
13.   1; center 11, 12 ; vertices and radius  32 .
1y  2 2 2 1x  12 2
4 1
13, 12, 11, 12, 11, 22, and 11, 02 ;   1; hyperbola with center at
1 12 2 2 1 13 2 2
13.
foci 11  13 , 12 and 11  13, 12
11, 22, axis parallel to the y-axis, and a  , b  .
1 1
y
2 3
9 4 1 1
1 15. a. 4,  , 1,  ,  b. m   2
4 9 4 x
x
–3 –2 –1 1
Section 9.5
–1
1. a.
t 2 1 0 1 2
–2
x  4  3t 10 7 4 1 2
–3
y  2  5t 12 7 2 3 8

1x  10 2 2 1y  52 2 5
 1; center 110, 52; vertices
m
15.  3
100 25
120, 52, 10, 52, 110, 02, and 110, 102;
5 14
b. y x
foci 110  5 13 , 52 and 110  5 13 , 5 2
3 3
c. The ratio of the coefficients of t in y and x.
y
5 14
d. y x
12 3 3
10 3 14 5 14
e. x  y  , so that y  x 
8 5 5 3 3
6 3. a.
4 t 2 1.5 1 0.5 0 0.5 1 1.5 2
2 x 7 2.375 0 0.875 1 1.125 2 4.375 9
x
–20 –16 –12 –8 –4 y 2 0.25 1 1.75 2 1.75 1 0.25 2
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 768

768 SELECTED ANSWERS

(b), (d) For a  2, b  5: period  10p

y y

9
3 7
1 5
x 3
–6 –4 –2 4 6 8
–2 x
–9 –7 –5 –3 –1 1 3 5 7 9

c. y  1x  12 2>3  2
–3
–5
–7
5. For a  1, b  3; period  6p
–9
y

5 For a  2, b  7: period  14p

y
3
1 11
x
–6 –3 –1 1 3 5 9
7
5
–5
3
x
For a  1, b  4; period  8p –11 –9 –7 –5 –3 –1 1 3 5 7 9 11
–3
y
–5
–7
4 –9
2 –11
x
–6 –2
–2
2 4 6 For a  2, b  9: period  18p
–4 y

For a  1, b  5; period  10p 10

8
y 6
4
2
5 x
–12 –8 –4 –2 4 8 12
–4
x –6
–5 5
–8
–5 –10

The period seems to be 2bp and the number of loops is b.

For a  1, b  6; period  12p
y 9. a. x  2 cos t, y  0
y

2
5

x x
–5 5 –2 2

–5 –2

b. x  2 cos t  cos 2t, y  2 sin t  sin 2t

y
The number of loops appears to be b.
7. For a  2, b  3; period  6p 3
y 2
1
5 x
–3 –2 –1 1 2 3
–1
x
–5 5
–2
–5
–3
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 769

SELECTED ANSWERS 769

c. x  3 cos t  cos 3t, y  3 sin t  sin 3t

e. r  5 and u  arctan a b
4
y
3

f. r  5 and u  arctan a b  p
4 4
3
2

g. r  173 and u  arctan a b

3
x 8
–4 –2 –1 2 4

h. r  173 and u  arctan a b

8
–2
3
–4
5. a. a4000, b
p
4
1 1
d. x  cos t  2 cos t, y  sin t  2 sin t
b. a4000, b
2 2 5p
y 8

c. a26800,  b
13p
3
30
2
1
Section 9.7
x
–3 –2 –1 1 2 3 1. 3.
–1 10 Cissoid of 2
–2
Diocles Bifolium
1
–3 5

–2 –1 1 2
ay ay
11. x  a arccos a b  a sin aarccos a bb
2 4 –1
a a –5 –2

Section 9.6 –10

P  a5, b Q  a3, b
p 2p
1. a.
6 3 5. 7. 20 Cruciform
R  a6, b S  a2, b
7p 5p 3
Cochleoid 10
6 3 2

P  a5,  b Q  a3,  b
11p 4p –20 –10 10 20
b. 1
6 3
–20
R  a6,  b S  a2,  b
5p p
–1 1 2 3
6 3
–1
P  a5, b Q  a3, b
7p 5p
c.
6 3 9. 11.
Lemniscate of
2 4
R  a6, b S  a2, b
p 2p Bernoulli 8 r=
sin θ
6 3 1
4

P  a5,  b Q  a3,  b
5p p –2 –1 1 2
d. –1
–10 –5 –2 5 10
6 3

R  a6,  b S  a2,  b
11p 4p –2

6 3
p Review Problems
3. a. r  132 and u 
4 1x  1 2 2 1y  22 2
3p 1. a.  1
b. r  132 and u  9 25
4 1x  12 2 1y  22 2
5p b.  1
c. r  132 and u  25 9
4
p y2 1x  22 2 x2 1y  1 2 2
d. r  132 and u  2. a.  1 b.  1
4 9 16 4 9
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 770

770 SELECTED ANSWERS

1x  4 2 2 1y  32 2 20. a. y
3. a. y  3 b. x  1 4
4 8
3
4. xy  5 is a hyperbola with axis on the line y  x. 2
1
x
y –2 –1 2 4 6
–2
10 –3
8 –4
6
4
b. x  1y  12 2>3  3 c. x  4
2

–10 –6 –2 2 4 6 8 10
x 21. a. Same as the graph of y  1  x, for 0  x  1.
–4 22. The graph has 4 “loops” and has period 4p.
–6 y
–8
–10 1.2

0.8
5. y  43  1x  32 2 is a vertical parabola, with vertex at
13, 432, focus at 13, 42.752, opening downward.
0.4

x
6. 1x  4 2 2  1 y  3 2 2  16 is a circle of radius 4 and center –1.2 –0.8 –0.4 0.4 0.8 1.2

at 14, 3 2.
–0.4

1x  5 2 2 1 y  222
–0.8

  1 is an ellipse with center at

1 343 2
7. –1.2
17
15, 2 2, major axis parallel to the x-axis and a  117 , 23. a. The graph has 2 “loops” and has period p.
y
b  2343.
1x  1 2 2 1y  3 2 2
8.   1 is a hyperbola with center 1
4 3
11, 3 2 whose axis is parallel to the x-axis.
x
1y  2 2 2 1x  32 2
–1 1

9.   1 is a hyperbola with center 13, 22

6 9
–1
whose axis is parallel to the y-axis.
x2 y2
10.  1 b. The graph has 3 “loops” and has period p.
100 36
1x  62 2 1y  3 2 2
y

11.  1
40 4 1
1y  3 2 2 1x  22 2
12.  1
9 7 x
–1 1
y2x2 x2 y2
13.  1 14.  1
16 9 225 81
–1
x2 y2
15. 2a  7.277,  1
13.24 4.24 c. The graph has 6 “loops” and has period 2p.
16. 12 ft from the point below the highest point, along the axis of y

the vertices.
x2 y2 1
 1
A 841
4 B
17.
78
18. The points are 16, 42, 15, 12, 14, 22, 13, 52 , 12, 82; the –1 1
x

function is y  3x  14.

x2 –1
19. Same as parabola y   1, between x  6 and x  6.
9
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 771

SELECTED ANSWERS 771

24. x  1  2y2 11. a. 27.2 pounds b. 17.3 pounds c. 38.6 pounds

25. a. a218 , b
p 13. a. 9.5° south of east at 608.3 mph
4 b. 5.4° south of east at 679.6 mph
b. 1 210 , 1.892552
c. 46.3° south of west at 399.7 mph

c. 1 217 , 0.2449792
Section 10.2
d. a6, b
p
2 64 73 86 85
1. C 82 91 S , rows for people, columns for subjects
26. a. 11.5, 1.5 232
69 77
82 84 81 83
3 22 3 22
b. a , b 3. a. HORN b. MOTHER c. EARLY
2 2 d. YESTERDAY
c. 10, 42 5. 8x1  5x2  3x3  6200
d. 12 22 , 2 22 2 2x1  5x2  5x3  4000
e. 12.5 23 , 2.52 3x1  7x2  6x3  4700
f. 12.08073, 4.546492 7. 2000x1  500x2  5000x3  280,000
5x1  x2 0
1 1
27. Parabola x    y2 x1  2x3  0
2 2
y
9. a. p1  1>4, p2  1>2, p3  1>4
b. p1  3>8, p2  3>8, p3  1>4
3 c. p1  1>4, p2  1>2, p3  1>4
2 d. p1  3>8, p2  3>8, p3  1>4
1 e. p1  0.35, p2  0.40, p3  0.25
x 2>3 1>2
–3 –2 –1 1 2 3 11. a. B R b. 7>12 c. 11>18.
–1 1>3 1>2
–2
1>2 1>4 0 0
–3
1>2 1>2 1>4 0
13. a. D T
28. ellipses 0 1>4 1>2 0
29. a circle along the polar axis; a 4-petaled rose along the axes; 0 0 1>4 1
an 8-petaled rose along the axes 3>8 5>16
30. a circle along the polar axis; a 3-petaled rose with one petal 1>2 15>32
b. D T , D T
along the polar axis; a 5-petaled rose with one petal along 1>8 3>16
the polar axis 0 1>32
31. a circle along the vertical axis; a 4-petaled rose between the
30 3 34.2
axes; an 8-petaled rose between the axes 15. a. B R, B R, B R
110 126 150.6
32. a circle along the positive vertical axis; a 3-petaled rose with
50 43.5 26.6
one petal along the negative vertical axis; a 5-petaled rose b. B R, B R, B R
with one petal along the positive vertical axis. 165 256 380.15

Section 10.3
Chapter 10 1. a. 10 b. 2 c. 0 d. 3
Section 10.1 e. 5 f. 10 g. 15
3. d. all lie on the line. 3. a. 3 b. 2 c. 3 d. 12
e. 30 f. 21
5. a. 5 b. 240 c. 217 d. 218
e. 266 14 2 25 4
5. a. B R b. B R c. B R d. B R
7. a. 39 3 64 b. 330 10 204
4 18 3 0
c. 321 7 144 d. 33>2 1>2 14 9 11
9. a. 2i  j b. i  3j c. 1
 1 e. C 22 S f. C 2 S
2i 2j
d. 3i 14 14
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 772

772 SELECTED ANSWERS

1 1 5 25 10
d. B R e. not possible
7. a. c b. d c. C 5 S d. C 3 S 13 5 6
2 2 7 5 2
6 10
0 2 f. C 15 15 2 S g. B R
13 18
e. C 1 S f. C 5 S 23 25 2
0 2 h. not possible

9. a. 5x1  x2  2 5 3 14 25 4
4x1  3x2  5 5. a. C 15 5 S , C 30 75 20 S
b. x1  4x2  4 25 7 46 125 36
2x1  3x2  9 14 25 4
c. 5x1  2x2  x3  1 2 25 12
b. B R , C 30 75 20 S
4x1  x2  6x3  5 6 0 4
46 125 36
3x1  x2 2
d. 2x1  x2  5x3  0 6.40 6.85 6.05
9. B R
3x1  x2  2x3  0 5.95 6.95 5.75
5x1  x2  3x3  0
4.70 4.00
11. Each has the form Ax  b 11. a. BA  B R
3.90 3.65
6 4 2 400
16800 30400 12000
a. A  C 4 8 4 S , b  C 800 S b. CB  B R
16200 27600 12000
3 2 8 500
12520 10920
8 5 3 6200 c. CBA  B R
b. A  C 2 5 5 S , b  C 4000 S 11730 10380
3 7 6 4700 13. a. x n  Anx 0
10 50 200 600 4 2 1 8 7 3
c. A  C 1 3 0.2 S , b  C 20 S b. A2  C 1 6 2 S , A3  C 3 15 7 S
30 10 0 200 3 1 4 10 3 8
2000 500 5000 280,000 8 21
d. A  C 5 1 0 S, b  C 0 S c. C 3 S , C 4 S d. between 4 and 5 years
1 0 2 0 10 23

10 3>4 1>4
4 2 6 1.5 15. B R
1>4 3>4
5T ≥ ¥
6
13. D6 1 2
3 5>8 3>8 9>16 7>16
5 0.85 7 2.5 a. B R b. B R , 7>16
2 3>8 5>8 7>16 9>16
b. $68.50, $82, $76.60 c. 0.496.
21. a. 57.53° b. 35.84° c. 81.39° d. 69.11° x3
x2
17. Let x  ≥ ¥ , p  3 1 4 7 24,
23. State Tech because the angle is smaller.
x
Section 10.4 1
q  34 5 8 174 : so P1x 2  p . x and Q1x 2  q . x
2 2 2 16 13 7
1. a. B R b. B R c. B
7 5
R a. 1p  q2 # x
b. 1p  q2 # x
8 6 1 6
10 2 2 6 2 8
d. B R e. B R f. B R 21. The repeated application of R to any vector successively ro-
2 2 1 11 4 18 tates the vector 1° at a time in the counterclockwise direc-
18 10 tion around the origin.
g. B R
5 3
Section 10.5
5 3
25 5 1. a. x  28>5, y  12>5 b. x  21, y  6
3. a. B R b. C 15 5S c. not possible
5 7 c. x  0.7, y  1.8
25 7
Gord.xxxx.ans.pgs 11/21/03 11:26 AM Page 773

SELECTED ANSWERS 773

d. x  19>11, y  12>11 7. a. dAp b. $2210

e. x  5, y  6 8. a. not defined
3. a. x1  6>5, x2  4>5, x3  1>5 2 3 4 5
b. x1  7>5, x2  5, x3  29>5 b. C 2 4 6 8 S
c. x1  22>3, x2  17>3, x3  13>3 5 9 13 17
d. x1  10>7, x2  5>7, x3  6>7 16 31 46 61
e. x1  3, x2  5, x3  8 16 14 20
7 12 17 22
f. x1  12>10, x2  37>10, x3  29>20 c. C 32 28 40 S d. D T
10 19 28 37
41 35 47
5. a. multiple solutions b. inconsistent 11 19 27 35
c. multiple solutions d. multiple solutions 13 7 4
e. unique solution f. unique solution 1 2 1
e. D T
7. Factory 1: 0; Factory 2: 78.57; Factory 3: 42.86 7 3 2
9. Gelatin 5.04; fish sticks 4.89; mystery meat 1.53. 2 1 3
$2.30 $3.05
1>2 1>2 1 9. a. BA  B R
11. a. B R b. B R c. B R $1.65 $2.10
1>2 1>2 0
7000 12500 5500
2>3 1>2 3>5 b. CB  B R
13. B R, B R 14000 25000 11000
1>3 1>2 2>5
0.48 0.4 0.32 0.43 0.4 0.37
10. a. A2  £ 0.2 0.2 § A  £ 0.2 0.2 §
1>2 1>4 0 0 0 3
0.2 0.2
1>2 1>2 1>4 0 0
15. D T, D T 0.32 0.4 0.48 0.37 0.4 0.43
0 1>4 1>2 0 0
0 0 1>4 1 1 0.405 0.4 0.395
A5  £ 0.2 0.2 0.2 §
17. y  x  4x  3
2
0.395 0.4 0.405
19. a. x 2  y2  x  y  0
b. x 2  y2  2x  2y  3  0 0.4
45 79 246 b. columns approaching C 0.2 S
c. x 2  y 2  x y 0 0.4
19 19 19
d. x  y  x  13>2 2y  9>2  0
2 2
11. a. x1  30, x2  14, x3  9
e. x 2  y2  17y  0 b. x1  1>3, x2  7>3, x3  0
1 0 12. 100 tables, 500 chairs, 200 sofas
21. A1  B R 0.4
2 0.5
13. £ 0.2 §
0.4
Review Problems
14. y  x 2  2x  2
1. a. 5 b. 229 c. 3
2. 77.1° south of west at 172.2 mph
3. 4x  y  3z  1500 Appendix A
3x  2y  5z  2300 1. 21 3. 10 5. 1
2y  4z  1800 7. The graph of y  0 x 0 shifted left 3 spaces.
40
0.6 0.4 0.2 0.34 1
9. 1140 11. a 13. 5
4. a. £ 0.2 0.2 0.2 § b. C 0.2 S k5 k

0.2 0.4 0.6 0.46 15. a. 10 b. 15

5. a. 2 b. 5 c. 38 d. 14 17. 1x  42  1 y  32  36; center 14, 32 , radius 6
2 2

6. a. not defined b. not defined

c. not defined d. not defined Appendix E
38 1. i 3. 1 5. 2  4i
0
18 7. 11  2i 9. 5  5i 11. 229
e. C 8 S f. D T
24
2 1 2 2 2 25
29 13. 5i 15.   i 17.   i
5 5 3 3
Gord.xxxx.ans.pgs 4/29/03 1:15 PM Page 774
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 775

Index

Absolute value, 722 Birds, wingspan vs. weight, 115, 120, 144 Complete solution of difference
Acceleration due to gravity, 190 Black thread method, 65–66, 162 equation, 368
Addition Bouncing ball, 423–425 Completing the square, 725–727
of complex numbers, 736 Brachistochrone problem, 627 Complex conjugates, 573, 735
of matrices, 691 Complex numbers
AIDS, 187–188, 195, 278–279 arithmetic of, 567, 734–737
Airplanes, wingspan vs. weight, 211–213 Calculators chaos and, 574–575
Amplitude, of sine function, 498 equation of line and, 165 complex conjugates and, 573
Analytic geometry. See also Geometry exponential functions and, 200 iteration processes applied to,
conic sections and, 595–618 graphs of polynomials on, 267–268 573–575
distance between points and, linear regression, 164–166 modulus of, 568
585–589 logarithms and, 202 powers of, 569–573
ellipse and, 595–605 nonlinear regression, 181–191 properties of, 566–569
equation of circle and, 589–594, 724 parametric representations on, 624 trigonometric form of, 568–569
explanation of, 585 sequences on, 360–361 Complex plane, 567
Gaussian elimination and, 711–714 solving systems of equations on, Complex root, 573
hyperbola and, 606–614 730–731 Composite functions
parabola and, 614–616 trigonometric functions on, applications of, 314–315
Angles 456–457, 485–486, 516 explanation of, 312–313
complementary, 429 Carbon dating, 120–121 Concave down, 11
of inclination, 431 Cardioid, 579, 641 Concave down growth functions, 136
reference, 457 Cartesian coordinate system Concave up, 10–11
Aphelion, 601, 605 explanation of, 25–26 Concave up growth functions, 136
Arccos functions, 522–527 transforming between polar and, Concavity, 10–12
Arcsin function, 517–522 632–634 Conic sections
Arctan function, 530–533 Cell phone, 341–343 ellipse, 596–605
Arithmetic of complex numbers, 567, Challenger disaster, 161–163, 334–336 explanation of, 595–596, 617–618
734–737 Chaos, 574–581 hyperbola, 606–614
Associative law of matrices, 691 complex numbers and Julia set, parabola, 614–616
Asymptote, vertical, 105 576– 579 Conjugate pair, 573
Asymptote, horizontal, 80 Mandelbrot set, 579–581 Constant of proportionality, 44
Augmented coefficient matrix, Circle Contaminant model. See Pollutant
701–702 equation of, 589–594, 724 model
Average rate of change, 136–139 parametric representations of, 622 Control systems, 290–291
Axis polar coordinates, 589 Convergence, 360
of ellipse, 597, 598 unit, 484–485, 489 Cooling model, 322–334, 408–412
explanation of, 25 Circular function. See Trigonometric Coordinate systems
of hyperbola, 609 function Cartesian, 25–26, 583
polar, 630 Clothes production model, 660–662, coding model, 332–334, 408–412
of symmetry, 615 675–676, 703–705 explanation of, 583–584
Coefficient polar, 584, 630–634
correlation, 169–170 rectangular, 25–26
Backward difference equations, 369 leading, 250 types of, 584
Base 10, 99–100 matrix, 701, 731 Coordinates, 25
Base e, 109 of determination, 240, 273, 274 Correlation coefficient
Bases, changing, 109–111 of polynomial, 250 critical values for, 171, 721
Best fit Column vectors, 650, 659–660 determining significant coefficient
curve, 181–191 Commutative law of matrices, 691 and, 170–176
line, 166–168 Comparing rates of growth, 131–136 explanation of, 169–170
Biological half-life, 363 Complementary angles, 429 fit and, 222

775
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 776

776 INDEX

method for finding, 170 interpreting sum of squares and, Newton’s laws of cooling and,
multiple, 240–241 227–228 407–412
Cosecant function, 533 linear models with several variables Newton’s laws of heating and,
Cosine function and, 236–243 413–414
applications of, 450–452 linear regression analysis and, point of inflection and, 390–392
approximating, 558–561 164–176 second order, 383
behavior of, 448, 492–493 linearizing, 196 solution sequence to, 360–369, 374,
explanation of, 448, 487–490 measuring the center of set of, 391
graph of, 462 738–739 Difference identities of sine and cosine,
inverse, 522–527 measuring the spread in set of, 543–545
law of cosines, 473–474, 673 739–740 Dimension of a matrix, 691, 731
modeling periodic behavior with, using shifting and stretching to ana- Directrix, of parabola, 615, 616
494–509 lyze, 331–345 Discriminant, of quadratic equations,
period of, 460 world population, 740–744 259–260
radian measure and, 490–492 Data analysis Displacement vector, 650
“special” angles and, 449–450 introduction to, 161–163, 737 Distance formula
Cotangent function, 533 Daylight model, 184–185, 200–202, explanation of, 585–586, 724
Cramer’s rule, 709 339–340, 349–352 use of, 586, 589
Cricket chirping, 27–28, 51–53, 64–65, Dead body model, 411 Distributive law of matrices, 690
505–506, 521–522 Decay factor, 88–89 Domain, of function, 20–23, 27
Critical values for correlation Decay rate, 88 Dot products. See Scalar products
coefficient, 171, 721 Decreasing function, 9–10 Double-angle identities, 541–542, 545,
Cubic functions Degree of polynomial, 250, 264–266 561
behavior of, 263–264 Delta 1 2, 46 Doubling time, 76–77
characteristics of, 261–264 DeMoivre, Abraham, 573 Dow Jones model. See Stock market
difference patterns, 292–294 DeMoivre’s theorem model
explanation of, 250 explanation of, 573 Drug doses, 357–363
fitting to data, 287–289 use of, 575, 576 Drug model
roots of, 287–289 Dependent variable, 19, 25 eliminating medication and, 87–88,
Curve fitting procedures, 213. See also Determinant, 709 102, 357–358
Fitting to data Descartes, Rene, 585 exponential decay and, 87–88
Cycloid, 626 Difference, of two functions, 305 half life, 90–93
Difference, of vectors, 656–657 maintenance level and, 258, 361,
Difference equations. See also Sequences 363–365
Data backward, 369 repeated dosage and, 358–361
application of fitting functions to, behavior of logistic function and,
228–234 392–394
capturing linear pattern in, 64–66 closed form solution and, 367 e base, 109
determining if set of data is linear, complete solution, 368 Earthquakes, 107–108
59–68 drug model and, 359–367 Earth’s orbit, 601–602
determining if set data is exponen- equilibrium, 393 Elementary row operations, 700–701
tial, 82–83 explanation of, 359–360, 367–369 Elimination method, 728–730
difference patterns in, 292–294 exponential growth and decay mod- Ellipse
explanation of, 1–2 els and, 371–383, 416 description and graph of, 599–601
fitting exponential and logarithmic Fibonacci model and, 382–383 equation of, 598–599, 724–725
functions to, 196–204 first order, 383 explanation of, 595, 596–598
fitting nonlinear functions to, forward, 369 foci, 596
181–191 geometric sequences and their sums graph of, 598–599
fitting polynomial functions to, and, 416–425 major and minor axes, 597
278–281 inhibited growth model. See logistic parametric representation of,
fitting power functions to, 208–218 model 623–624
interpreting correlation coefficient limit to growth, 388 planetary orbit and, 601–605
and, 222 logistic growth model and, reflection property of, 602–604
interpreting residuals and, 222–224 386 – 403 symmetry of, 604
interpreting scatterplot and, maximum sustainable population vertex of, 597
225–227 and, 389 Epicycloid, 627–628
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 777

INDEX 777

Equation of overview of, 70–72 connecting geometric and symbolic

circle, 589–594, 724 passing through two points, 80–82 representations of, 25–31
ellipse, 598–599 power functions vs., 132–134, 136 cosine, 448, 462, 487–490
hyperbola, 608–611 prediction with, 77–80 cubic, 250, 253, 261–266, 287–289
line, 49–51 rules for exponents and, 83–84 decreasing, 9–10
parabola, 615–617 Exponents, rules for, 83–84 difference of, 305
parametric, 622–628 Extrapolating, 168 domain and range of, 20–23, 27
quadrtic, 253 Extrapolation, 36 even/odd, 126–127, 524, 556, 561
roots of, 251 explanation of, 3, 22
systems of linear equations, 727–731 exponential, 183–185, 187, 196–202
Equilibrium, 393 Factoral notation n!, 722 exponential decay, 87–95, 131–134
state, 709–710 Factors and roots, 252, 258, 259, 260, exponential growth, 70–87,
Even function 262, 264, 265 131–134, 136–137
power, 126 Falling body, 19–20, 29, 145–146, families of, 43–44
trigonometric, 524, 561 174–175, 189–191, 273–276 formulas and equations of, 3
Even positive integer powers, 126 Fiber optic model, 93–94 function of, 311–313
Excel, multivariate regression in, 241–243 Fibonacci, 380 graphs of, 3–4, 26
Explanatory variables. See Independent Fibonacci difference equation, 382 implicit, 66–68
variables Fibonacci model increasing, 9–10
Exponential decay explanation of, 382, 383 inflection, point of, 11–12
determining if set of data is expo- for population growth, 380–383 inverse, 140–152
nential growth or, 94–95 Fibonacci sequence, 382 inverse cosine, 522–527
half-life for, 90–91 First order difference equations, 383 inverse sine, 517–522
models of, 371–372, 377–380 Fitting to data inverse tangent, 530–533
Exponential decay functions correlation coefficient and, 222 linear, 44–56, 59–68, 131
decay factor, 88–89 examples of, 228–234 logarithmic, 99–111, 132, 135–136,
decay rate, 88 exponential functions, 196–202 189, 196, 202–204
example of, 87–88 interpreting scatterplot and, 225 –227 logistic, 346–349, 392–394
explanation of, 88–90, 136 interpreting sum of squares and, mathematical models and, 34–38
formula for, 89 227–228 monotonic, 148
graphs of, 88 linear models with several variables, multiple of, 323–328
half-life and, 90–93 236–243 nonlinear, 181–191
radioactive decay and, 91–93 linear regression analysis, 164–176 periodic, 14–15, 460
Exponential functions logarithmic functions, 196, 202–204 point of inflection, 11–12
behavior of, 116, 122, 131 logistic functions, 397–403 polynomial, 249–269, 273–281,
combining sinusoidal and, 507–509 nonlinear functions, 181–191 285–291
determining if a set of data is expo- polynomials, 278–281 power, 113–127, 131–136, 185–186,
nential, 82–83, 94–95 power functions, 208–218 188, 190, 208–218
explanation of, 72 residuals and, 222–225 products of, 305–306
fit to data, 183–185, 187, 196–202 Florida population model, 28–29, quadratic, 250, 251, 255–260,
identities, 83–84, 100 70–72, 76, 77–79, 101, 104, 110, 285–287
logarithmic functions vs., 105–106 143–144 quartic, 251
passing through two points, 80–82 Focus quotient of, 306–307
rate of change of, 138–139 parabola, 615 range of, 20–23
Exponential growth ellipse, 596 rational, 307–310
applications of, 75–76 hyperbola, 606 represented symbolically, 18 –23
difference equation for, 372, 373 Forward difference equations, 369 shifting, 320–322
doubling time, 76–77 Fractal, 581 sine, 441–442, 485–486
linear vs., 74 Frequency, of sinusoidal function, sinusoidal, 496–497, 500, 501
models of, 371–377, 382–383 500–502 stretching and shrinking, 323–328,
population growth, 182–183 Fruit purchase model, 672, 678–679, 331–345
Exponential growth functions 691–692 sum of, 304
domain of, 79 Functions. See also specific types of surge, 349–352
doubling time of, 76–77 functions tables of, 4–5
explanation of, 72–73, 132 composite, 311–313 tangent, 432, 434, 527
formula for, 73–74 concavity of, 10–12 turning point of, 9–10
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 778

778 INDEX

Fundamental logarithmic-exponential Half-angle identities, 544–545 Interpolation, 36, 168

identities, 100 Half-life Interval notation, 20
explanation of, 363 Inverse cosine function, 522–527
for exponential decay process, 90–93 Inverse functions
Galileo, 3, 174, 273 Heating model, 324–325, 413–414 behavior of, 149–150
Gateway Arch, 280–281 Homogeneous systems, 711 cosine, 522–527
Gauss, Carl Friedrich, 700 Horizontal asymptote, 80 determining existence of, 147–149
Gaussian elimination Horizontal axis, 25 examples of, 141–147
applications of, 709–714 Horizontal line test, 149 explanation of, 140–141
clothes production model and, Horizontal shifts, 322, 340–344 exponential, 143–145, 149
703–705 Horizontal shrinking, 323 logarithmic, 145, 149
explanation of, 700–702 Horizontal stretching, 323, 327 method for finding, 150–152
geometric interpretation and, 702–703 Hours of daylight model. See Daylight power, 144–145, 149
inverse matrix and, 708–709 model resticted domain, 518, 523
systems of linear equations with Hyperbola sine, 517–522
multiple solutions and, 705–706 applications of, 611–612, 614 solving equations with, 515–517
systems of linear equations with no equation of, 608–611 Inverse matrix, 708–709, 732
solutions and, 706–708 explanation of, 606–608 Inverse sine function, 517–522
General equation of circle, 591 foci of, 606 Inverse tangent function, 530–533
Geometric sequences, 372, 417–420, graph of, 608–611 IRA account model, 374–375
423–425 parametric representations of, 625 Iteration, 575
Geometry Hyperbolic coordinate system, 584
analytic, 585–594 Hyperbolic cosine, 625
conic sections and, 595–618 Hyperbolic sine, 625 Julia, Gaston, 575–576
ellipse and, 595–605 Julia set, 576–579
families of curves in polar coordi-
nates and, 636–643 Identities. See also Trigonometric
hyperbola and, 606–614 identities Kepler, Johannes, 228
parabola and, 614–616 difference, 543–545 Kepler’s third law of planetary motion,
parametric representations and, double-angle, 541–542, 545, 561 228–234
619–628 explanation of, 100, 464, 539, 545
polar coordinate system and, 630–634 fundamental logarithmic-
Golden ratio, 175, 176 exponential, 100, 103 Law of cosines, 473–474, 673
Graphing calculators. See Calculators half-angle, 544–545 Law of sines, 470–473
Graphs involving tangent, 548 Law of Universal Gravitation, 141–142,
of cosine function, 462 Pythagorean, 465, 540, 545 310–311, 502–503
of cubic, 263–264 reflection, 540–541, 545 Leading coefficient, 250
of ellipse, 598–599 sum, 542–543, 545 Least squares line, 165
of exponential decay functions, Imaginary numbers, 735 Life expectancy, 3–4, 189, 203–204
88–92, 131 Implicit functions, 66–68 Limaçons, 642–643
of exponential growth functions, Inconsistent equations, 707 Limit to growth. See Maximum sustain-
72–74, 77–79, 131 Increasing functions, 9–10 able population
of function, 3–4, 26 Independent variables Linear approximation, 551–558
of hyperbola, 608–611 constant multiple of, 326 Linear correlation coefficient. See Cor-
of inverse functions, 149–150 explanation of, 19, 25, 237 relation coefficient
of linear functions, 45–46, 131 horizontal stretching and shrinking Linear equations, system of
of logarithmic functions, 106, 132 and, 323 algebraic solutions to systems of,
of logistic function, 392–394 Index of summation, 722, 723 727–731
of polynomials, 256, 261, 262, 264, 265 Inflection point. See Point of inflection calculators to solve, 730–731
of power functions, 115–119, 131–132 Inhibiting constant, 386, 387 matrices to solve systems of, 731–732
of quadratic functions, 255–257, 614 Inhibiting growth model. See Logistic with multiple solutions, 705–706
of sine function, 461–462, 486, 487 growth model with no solutions, 706–708
of tangent function, 527–530 Initial side, of angle, 457 Linear functions
symmetry and, 732–734 Initial condition of difference equation, behavior of, 131
Growth factor, 72–74 393 data and, 59–68
Growth rate, 72–74 Inner products. See Scalar products domain of, 79
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 779

INDEX 779

explanation of, 44 natural, 109 product of vector and, 676–680

facts about lines and, 56 nonlinear regression and, 204 singular, 708–709
graphs of, 45–46 properties of, 103–104 solving systems of linear equations
growth and, 74, 132, 136 use of, 100–102 with, 731–732
implicit, 66–68 Logistic curve subtraction of, 690
rate of change of, 137–138 explanation of, 346, 387 transition, 662–663
slope and, 46–47 limitating value of, 388–389 Matrix algebra, 659, 691
Linear growth, 74 point of inflection of, 388, 390–391 Matrix multiplication
Linear models Logistic function explanation of, 686–688
clothes production model, 660–662 behavior of, 392–394 fruit purchase model and, 691–692
explanation of, 659–660, 662 explanation of, 346–349 geometric view of powers of matrix
geometric view of vectors and, 660 Logistic growth model and, 697
Markov chain model for stock applications of, 394–397 Markov chain transition matrices
market, 662– 665 difference equation for, 386–403 and, 692–697
population growth model, estimating logistic parameters, methods to interpret, 686 –691
665 – 668 397–403 population growth models and,
with several variables, 236–242 explanation of, 386–387, 391 698–699
Linear regression analysis fitting to data, 397–403 Matrix-vector equation, 661, 679
correlation coefficient and, 168–176 inhibiting constant and, 386, 387 Matrix-vector product
examples of, 166–168 maximum sustainable population explanation of, 676–680
multiple, 238–240 and, 388–389 geometric view of, 680–683
overview of, 164–166 point of inflection and, 390–392, 394 scalar product of vectors and,
Linearizing data rabbit population and, 380–383 686–687
explanation of, 196, 199 U.S. Population and, 400–403 Maximum of a function, 9–10
exponential functions and, 200 LORAN system, 584, 611–612 Maximum sustainable population, 388,
power functions and, 208–209 389
Lines McAuliffe, Christa, 161
least squares, 165 Maintenance level of drug. See drug Mean, 738
normal form of, 67 model Median, 738–739
of best fit, 165 Mandelbrot, Benoit, 579 Midline, 497
parallel, 56 Mandelbrot set, 579–581 Midpoint formula, 587–588
parametric representations of, Markov chain model Mile run model, 166–168, 170–172
619–621 explanation of, 663 Minimum of a function, 9–10
perpendicular, 56 matrix multiplication and, 692–697 Models, 34. See also Mathematical
point-slope form for equation of, steady state of, 709–711 models
50–51 for stock market, 662–665, 679–680, Modulus, of complex number, 568,
slope intercept form, 49–50 709–711 570–572
slope of, 46–49 Mathematical models Monotonic functions, 148
that don’t pass through origin, 47–56 accuracy of, 37–38 Movie industry receipts, 164, 166,
that pass through origin, 44–46 with difference equations, 371–383 223–225
vertical intercept of, 47 explanation of, 35, 371 Multiple
Logarithmic functions expressed as functions, 35–37 of a function, 323–328
applications of, 106–109 linear, 659–668 of a vector, 652
behavior of, 104–105, 132 for Newton’s law of cooling, 409 Multiple correlation coefficient, 240–241
changing bases and, 109–111 parameters and, 37 Multiple regression. See Multivariate
explanation of, 99–100 periodic behavior with sine and co- regression
exponential vs., 105–106 sine functions as, 494–509 Multiplication
fit to data, 189, 196, 202–204 Matrix of complex numbers, 736–737
growth and, 135–136 applied problems using, 660–668 matrix, 686–697
indentities, 103 augmented coefficient, 701–702 matrix and vector, 676–680
inverse of, 145–149 coefficent, 701, 731 Multivariate regression
rate of change of, 139 dimension of, 691, 731 in Excel, 241–243
Logarithms explanation of, 659 explanation of, 238–240
to base 10, 99–100 inverse, 708–709, 732
to base e, 109 nonsingular powers of, 694–695
changing bases of, 109–111 product of matrices, 686–697 n-factorial, 722
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 780

780 INDEX

n-vector, 659 Periodic behavior with harvesting, 375–377

Natural logarithm, 109 of functions, 14–15 World, 371
Newton, Isaac, 3, 408 sine and cosine functions as models Position vector, 650
Newton’s law of cooling, 332–334, for, 494–509 Power functions
408–412 Periodic functions, 460. See also applications of, 120–122
Newton’s law of heating, 324–325, trigonometric functions behavior of, 116–119, 131–132
413–414 Periodic oscillatory effect, 483, 484 even positive integer, 126
Nonhomogeneous system, 711 Perpendicular lines, 56 explanation of, 113–116
Nonlinear function, 181–191 pH value, 106–107 fit to data, 185–186, 188, 190,
Nonlinear regression Phase shift, 502–503 208–218
correlation coefficient, 222 Planetary motion, 141–142, 228–234 fractional powers, 114–115
exponential function and, 196–202 Plate tectonics, 107 graph of, 115
logarithmic function and, 196, Point of inflection growth and, 132–136
202–204 examples of finding, 391–392 inverse of, 145
logistic function and, 397–403 explanation of, 11–12, 391 with integer powers, 126, 281
power function and, 208–218 of logistic curve, 388, 390–391 negative powers, 114
Normal form, of line, 67 polynomials and, 263, 265 odd positive integer, 126–127
turning point and, 263, 265 passing through two points, 122–125
Point-slope form, 50–51, 53 polynomials and, 249, 250
Odd function, 126–127, 556 Polar coordinate system Powers of i, 735–736
Odd positive integer powers, 126–127 explanation of, 630–632 Principal values, 518, 522, 525
Oil reserves, 420–423 families of curves and, 636–643 Probabilities, transition, 662–663
Origin rectangular coordinates and, 632–634 Product, of two functions, 305 –306
explanation of, 25 Pole, 630 of two matrices, 686–688
Oscillation, periodic, 483 Pollutant model, 377–380 Projectile motion, 314, 545–546,
Outliers, 227 Polynomial functions 621–622
approximating sine and cosine with, Proportionality, constant of, 44
550–565 Prozac model. See drug model
Parabola coefficients of, 250 Pythagorean identity
axis of symmetry of, 615 cubic, 250, 253, 261–266, 287–289 explanation of, 465, 490, 540, 545
directrix of, 615, 616 degree of, 250, 264–266 Pythagorean theorem
equation of, 615–617 difference patterns and, 292–294 distance formula and, 585–586, 724
explanation of, 250, 614–615 end behavior of, 267–269 explanation of, 429, 432–433, 444,
focus of, 615 explanation of, 249–251 568
parametric representations of, fit to data, 278–281 use of, 464–465
625–626 inflection points and, 263, 265 vectors and, 651, 674
quadratic functions and, 255–257 leading coefficient of, 250
reflection property of, 616 modeling with, 273–281
vertex of, 256 path of projectile and, 276 –277 Quadrants, 25
Parallel lines, 56 quadratic, 250, 251, 255–260, Quadratic equation, 251–252
Parallel vectors, 652 285–287 Quadratic formula, 252–253, 256, 267,
Parameters, 37, 589, 619 quartic, 251 285, 566
Parametric equations of line, 589 roots of, 285–291, 567 Quadratic functions
Parametric representations zeros of, 251–254 behavior of, 255–260
on calculators, 624 Polynomial patterns characteristics of, 260
of circle, 622 finding, 292–294 explanation of, 250, 251
of ellipse, 623–624 sums of integers and, 295–298 fitting to data, 273–281
of hyperbola, 625 sums of squares of integers and, formula for, 252–253, 256, 267, 285
of line, 619–621 298–301 graphs of, 255–257, 614
miscellaneous, 626–628 Population growth, 182–183, 375–377, linear factors of, 251–259
of parabola, 625–626 380–383, 665–668 turning point and, 260
path of projectile and, 276–277, Population models roots of, 257–260, 285–287
621–622 Florida, 28–29, 70–72, 76, 77–79, Quartic equations, 253
Path of projectile, 276 –277, 621–622 101, 104, 110, 143–144 Quartic functions, 251
Perihelion, 601, 605 U.S. Population, 182–183, 197–198, Quartiles, 740
Period, 460, 483 344–345, 348–349, 400–402 Quotient, of functions, 306–307
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 781

INDEX 781

Rabbit population, 380–383, 387 Scalar product vertical shift of, 497
Radian measure, 490–492 clothes production model using, Skydiving model, 336–338
Radioactive decay, 91–93, 112, 120–124 675–676 Slope, 46–49, 53
Radius, 589 explanation of, 649, 670–672 parallel lines, 56
Range, of function, 20–23, 27 geometric view of, 672–675, perpendicular lines, 56
Rate of change 680–683 Slope-intercept form, 49–50, 53
average, 136–139 matrix-vector product and, Solution sequence, 360–369
explanation of, 47 676–683, 686–687 Space shuttle. See Challenger
of exponential functions, 138–139 Scatterplot Species, number of, vs. area of habitat,
of linear functions, 137–138 examples of, 169 124–125, 185–186, 192, 210–211
of logarithmic functions, 139 explanation of, 162 Square, completing the, 725–727
Rational functions interpreting, 225–227 St. Louis Arch, 280–281
application of, 310–311 Secant function, 533 Standard deviation, 739
behavior of, 307–310 Second differences, 292 Standard error of the estimate, 228
explanation of, 307 Second order difference equation, 383 Steady plate. See equilibrium state
graph of, 300–310 Semi-log plot, 204 Stock market model, 662–665,
vertical asymptotes and, 308 Sequence. See also Difference equations 679–680, 709–711
Rectangular coordinate system, 25–26 convergent, 416 Stretching functions
polar coordinates and, 632–634 divergent, 416 analyzing data using, 344–345
Reference angles, 457 explanation of, 358 explanation of, 323–328
Reflection identities, 540–541, 545 exponential, 372 Substitution method, 728
Reflection property Fibonacci, 382, 383 Subtraction
of ellipse, 602–604 geometric, 372, 417–420, 423–425 of complex numbers, 736
of parabola, 616 limit of, 416 matrix, 690
Regression. See fitting to data notation for, 358 Sum identities, 542–543, 545
Regression, multivariate, 238–243 solution, 360–369, 374, 391 Sum of integers, 295–298
Regression line Shifting functions Sum of squares
equation of, 166–168, 401 analyzing data using, 331–344 of integers, 298–301
explanation of, 165 explanation of, 320–322 interpreting, 227–228
Regression plane, 238 Shrinking functions, 323–328 use of, 165
Residual plots, 222–223, 225–227 Sigma notation, 722–723 Sum of two functions, 304
Residuals Similar triangles, 723 Sum of vectors, 653–654
explanation of, 165 Sine function. See also sinusoidal function Summation notation, 722–723
fit and, 222–225 amplitude of, 498 Surge function, 349–352
Richter, Charles, 107 applications of, 444–448 Symmetry
Richter scale, 107, 108 approximating, 550–558 axis of, 615
Root-mean-square (RMS), 228 behavior of, 442, 492–493 explanation of, 732–734
Roots explanation of, 441–442, 485–486 parabolas and, 256
complex, 573 graph of, 461–462, 486, 487 Systems of linear equations
of cubic equations, 262, 263 inverse, 446, 517–522 algebraic solutions to, 727–731
double, 259–260 law of sines, 470–473 Cramer’s rule and, 709
explanation of, 251 modeling periodic behavior with, Gaussian elimination and, 700–702
and factors, 252, 258, 259, 260, 262, 494–509 matrices to solve, 731–732
264, 265 period of, 460, 501 with multiple solutions, 705–706
of polynomial functions, 285–291, radian measure and, 490–492 with no solutions, 706–708
567 “special” angles and, 443–444
of quadratic equations, 257–259, Singular matrix, 708–709
285–287 Sinusoidal function 3–vector, 659
real vs. complex, 259–260, 262–264, amplitute of, 498 Tables, representing functions with,
265 combining exponential and, 507–509 4–5, 26
using information on nature of, explanation of, 496–497 Tangent, of an angle, 430–439
290–291 frequency of, 500–502 Tangent function
Rose curve, 639–641 graph of, 504 applications of, 448
Rotation Matrix, 681–683 midline of, 497 behavior of, 434–435
Row vectors, 650, 659–660 period of, 501 explanation of, 432, 434, 527
Rumor model, 394–395 phase shift of, 502 graph of, 527–530
Gord.3896.Idx.pgs 5/14/03 11:42 AM Page 782

782 INDEX

identities involving, 548 double-angle, 541–542, 545, 561 angle between, 674
inverse, 530–533 explanation of, 539, 545 applied problems using, 660–668
period of, 460 half-angle, 544–545 column, 650, 659–660
Tangent ratio involving sine and cosine, 540 constant multiple of, 652
explanation of, 432, 527 law of cosines, 473–474 coordinate, 654, 672
use of, 435–437 law of sines, 470–473 difference of two, 656–657
Taylor, Brook, 556 Pythagorean, 465, 540, 545 displacement, 649–650
Taylor polynomial approximations reflection, 540–541, 545 explanation of, 452, 649–650
explanation of, 556 sum, 542–543, 545 geometric view of, 660
sin x and cos x using, 564–565 use of, 545–548 magnitude of, 650–651
Temperature oscillation model, 504, Trigonometry multiple of, 652
519–520 cosine function and, 448–452 parallel, 652
Terminal side, of angle, 457 cosine of an angle, 448 position, 649
Terminal velocity, 336–339 finding angle in triangle and, 437–439 rotation of, 682–683
Tide model, 503–504 graph of cosine function and, 462 row, 650, 659
Transformation approach graph of sine function and, 461–462 sum of two, 653–654
application of, 198–200 law of cosines and, 473–476 unit, 652–653
example of, 200–202 law of sines and, 470–473 Vertex
Transition matrix, 662–663 overview of, 429–430 of ellipse, 597
Triangle sine of an angle, 441–442 of hyperbola, 608–609
finding angle in, 437–439 sine function and, 441–448 of parabola, 256
right, 429 “special” angles and, 432–434 Vertical asymptote, 105
similar, 723 tangent function and, 434–435 Vertical axis, 25
Triangulation, 472 tangent of angle and, 430–432 Vertical intercept
Tribbles, 385 tangent ratio and, 435–437 examples of, 47–50, 52
Trigonometric functions. See also Co- Turning point of function, 9–10 explanation of, 47, 53
sine function; Sine function; Sinu- inflection point and, 263, 265 Vertical line test, 30
soidal function; Tangent function. Vertical shift, 321, 497
applications from physics for, Vertical shrinking, 323, 324
452–455 Unit circle, 484–485, 489 Vertical stretching, 323, 324
approximating, 550–565 Unit vectors, 652–653
complex numbers and, 566–575 U.S. Population, 197–198, 344–345,
inverse functions, 517–522, 522–527, 348–349 Wave phenomena, 546–548
530–533 exponential growth model, 182–183 Whispering gallery, 603–604
Julia set and, 576–579 logistic growth model, 400–403 World population, 371
Mandelbrot set and, 579–581 Words, representing functions with,
properties of complex numbers and, 5–6, 23
566–575 Variable
radian measure and, 490–492 dependent, 19, 25
reciprocals of, 533 independent, 19, 25, 237, 323, 326 y-intercept, 47
relationships among, 463–468 linear models with several, 236–243 Yeast growth model, 398–400
Trigonometric identities Vector product, 671
approximating sin x and cos x using, Vectors
561–563 applications of, 654–656 Zero of a function, 251–254
difference, 543–545
Gord.3896.Back.EP.pgs 4/29/03 12:47 PM Page B

Function Equation Behavior Graph

Sine y  sin x Passes through the origin y
Periodic with period
1
360°  2p radians
Oscillates between 1 and 1
θ
π π 3π 2π
2 2

–1

Cosine y  cos x Passes through (0, 1) y

Periodic with period 1

360°  2p radians
Oscillates between 1 and 1
θ
π π 3π 2π
2 2

–1

Sinusoidal y  D  A sin 1B 1x  C2 2 Centered vertically about the y

y  D  A cos 1B 1x  C2 2 midline 1vertical shift2  D

D+A
Amplitude  A  size of
oscillation above and below
D
the midline, so oscillates
between D  A and D  A
D−A
Frequency  B  number period
x
of cycles between 0 and
2p radians
Period  2p>frequency
Phase shift  horizontal shift  C

Tangent y  tan x Passes through the origin y = tan x

y

Periodic with period

180°  p radians
Vertical asymptotes at
x  90°  p>2 radians x
– 32π – π2 π
2
3π
2
Gord.3896.Back.EP.pgs 4/29/03 12:47 PM Page C

Critical Values for the Correlation Coefficient

n rn n rn n rn
3 0.997 21 0.433 39 0.316

4 0.950 22 0.423 40 0.312

5 0.878 23 0.413 42 0.304

6 0.811 24 0.404 44 0.297

7 0.754 25 0.396 46 0.291

8 0.707 26 0.388 48 0.284

9 0.666 27 0.381 50 0.279

10 0.632 28 0.374 55 0.267

11 0.602 29 0.367 60 0.254

12 0.576 30 0.361 65 0.245

13 0.553 31 0.355 70 0.236

14 0.532 32 0.349 75 0.227

15 0.514 33 0.344 80 0.220

16 0.497 34 0.339 85 0.213

17 0.482 35 0.334 90 0.207

18 0.468 36 0.329 95 0.202

19 0.456 37 0.325 100 0.195

20 0.444 38 0.320