Elementary Linear Algebra 9th Edition

Elementary Linear Algebra with ApplicationsElementary Linear Algebra with Applications Ninth Edition Bernard Kolman Drexel University David R. Hill Temple University Upper Saddle River, New Jersey 07458Library of Congress Catalogngin Publication Dat o le, Ftorial Director, Computer Science, Engineering, and Advanced Mathematics: Marca J Horton Senior Eaitor: Holly Stark Editorial Assistant Jennifer Lonschein Senior Managing Edito/Production Editor: Sco Disanno ‘An Director: Juan Lipez Cover Designor: Michael Frbeis ‘An Eahitor: Thomas Benfat ‘Mannfacturing Buyer: Lisa MeDowell ‘Marketing Manager Tim Galligan ‘Cover Image: (6) Willan T. Wiliams, Artist, 1969 Trane, 1969 Aerlic on cas, 108" x 84”, “Collection of The Studio Museur:in Harlem. Gift of Chares Cowles, New York (© 2008, 2008, 2000, 1996 by Pearson ucation, Ine Pearson Baveation, Inc Upper Sale River, New Jersey 07458 ane eine © 1991, 1986, 1982, by KT 1977, 1970 by Bens Koma ‘isis reserved. No part ofthis bok my be reproduced any frm ory any means, without person in wg rom he pais, Prin inthe Unit Stes of Americ woe? es asad ISBN 0-13-227654~-3 PeaeonBiscton, La, Landon Poaron Etncaon Auta PTY. Lied, Sydney Peateon EoneasonSnuapore, Pre Ltd Peatmoa Roteapon Nota Asia el Hong Kong Peas Eateston Cans, Lid oro Pearson Edusackn de Meni, S.A eC Pesron Education Japan, Tine Peaon Education Maly, Pes LatTo Lisa, Stephen, and to the memory of Lillie BK To Suzanne@ CONTENTS Preface xi To the Student — xvii ‘| Linear Equations and Matrices 1 1.1 Systems of Linear Equations 1 1.2 Matrices LL 1.3 Matrix Multiplication 21 1.4 Algebraic Properties of Matrix Operations 34 LS Special Types of Matrices and Partitioned Matrices 42 1.6 Matrix Transformations 55 1.7 Computer Graphies (Optional) 63 18 Correlation Coefficient (Optional) 73 Supplementary Exercises 80 2 Solving Linear Systems 86 2.1 Echelon Form of a Matrix 86, 22. Solving Linear Systems 95 2.3. Elementary Matrices: Finding A~! 117 24 Equivalent Matrices 126 25. LU-Fuetoiation (Optional) 130 Supplementary Exercises 137 3 Determinants 141 3.1 Definition 141 3.2 Properties of Determinants 146 3.3. Cofactor Expansion 157 3.4 Inverse of aMatrix 165 3.5 Other Applications of Determinants 169 3.6 Determinants from a Computational Point of View 172 applementary Exercises 174 AL Real Vector Spaces 177 4.1 Vectors inthe Plane and in 3-Space 177 4.2 VectorSpaces 188 43° Subspaces 197viii Contents 44 Span 209 45 46 47 48 Linear Independence 216 Basis and Dimension 229 Homogeneous Systems 244 Coordinates and Isomorphisms 253 4.9 Rank of aMatix 270 5 5. 52 53 Supplementary Exercises 285 Inner Product Spaces 290 Length and Direction in R? and R° 290 Cross Product in R° (Optional) 299 Inner Product Spaces 307 54 Gram-Schmidt Process 320 53. Orthogonal Complements 332 5.6 Least Squares (Optional) 350 Supplementary Exercises 358 6 Linear Transformations and Matrices 363 61 Definition and Examples 363 6.2 Ketnel and Range of a Linear Transformation 375 63. Matrix ofa Linear Transformation 389 6.4 Vector Space of Matrices and 65 ‘Vector Space of L Similarity 407 wear Transformations (Optional) 399 6.6 Introduction to Homogeneous Coordinates (Optional) 414 7 1 Supplementary Exercises 430 Eigenvalues and Eigenvectors 436 Eigenvalues and Eigemvoctors 436 7.2. Diagonalization and Similar Matrices 453 73 Diagonalization of Symmetric Matrices 463 Supplementary Exercises 477 8 applications of Eigenvalues and Eigenvectors 81 (Optional) 481 Stable Age Distribution in a Population; Markov Processes (Optional) 481 8.2 Spectral Decomposition and Singular Value Decomposition 488 83 Dominant Eigenvalue and Principal Component Analysis (Optional) S01Contents ix 84° Differential Equations 515 85 Dynamical Systems 525 8.6 Real Quadratic Forms 535 8.7 Conic Sections S44 88 Quadric Surfaces 552 9 Mamus for Linear Algebra 561 9.1 Input and Output in MATLAB 562 9.2. Matrix Operations in MATLAB 567 3 Matrix Powers and Some Special Matrices 570 9.4 Elementary Row Operations in MATLAB 573 9.5 Matrix Inverses in MATLAB 583 9.6 Vectorsin MATLAB 584 9.7 Applications of Linear Combinations in MATLAB 586 9.8 Linear Transformations in MATLAB 590 9.9 MATLAB Command Summary 592 10 Manas Exercises 597 Appendix A: Preliminaries A-1 AL Sets AL A2 Functions A-l Appendix B: Complex Numbers A-5 B.1 Complex Numbers A-S B.2_ Complex Numbers in Linear Algebra A-13 Appendix ¢: Introduction to Proofs A-22 C1 Logie 422 C2. Techniques of Proof A-27 Glossary for Linear Algebra A-31 Answers to Odd-Numbered Exercises A-37 Index 1-1 Page Index to Lemmas, Theorems, and Corollaries I-11 Photo Credits P-1Wl PREFACE Linear algebra is an important course for a diverse number of students for at least two reasons. First, few subjects ean claim to have such widespread applications in other areas of mathematics-multivariable calculus, differential equations, and probability, for example-as well as in physics, biology, chemistry, economics, fi- nance, psychology, sociology, and all fields of engineering. Second, the subject presents the student at the sophomore level with an excellent opportunity to learn how to handle abstract concepts, ‘This book provides an introduction to the basic ideas and computational techniques of linear algebra at the sophomore level. It also includes a wide variety of carefully selected applications. These include topics of contemporary interest, such as Google™and Global Positioning System (GPS). The book also introduces the student to working with abstract concepts. In covering the basic ideas of linear algebra, the abstract ideas are carefully balanced by considerable emphasis on the ‘geometrical and computational aspects of the subject. This edition continues to provide the optional opportunity to use MATLAB™or other software to enhance the pedagogy of the book. What's New in the Ninth Editien We have been very pleased by the wide acceptance of the first eight editions of this book throughout the 38 years ofits lif. In preparing this edition, we have carefully considered many suggestions from faculty and students for improving the content and presentation of the material. We have been especially gratified by hearing from the multigenerational users who used this book as students and are now using it as faculty members. Although a great many changes have been made to develop this major revision, our objective has remained the same as inthe first eight editions: 10 present the basic ideas of linear algebra in a manner that the student will find understandable. To achieve this objective, the following features have been developed in this edition: + Discussion questions have been added to the Chapter Review material. Many of these are suitable for writing projects or group activities. ‘+ Old Section 2.1 has been split into two sections, 2.1, Echelon Form of a Ma- trix, and 2.2, Solving Linear Systems. This will provide improved pedagosy for covering this important material, + Old Chapter 6, Determinants, has now become Chapter 3, to permit earlier coverage ofthis material ‘+ Old Section 3.4, Span and Linear Independence, has been split into two sec- jons, 4.3, Span, and 4.4, Linear Independence. Since students often have difficulties with these more abstract topics, this revision presents this material at a somewhat slower pace and has more examples.xii Preface * Chapter 8, Applications of Eigenvalues and Bigenvectors, is new to this edition in this form, It consists of old sections 7.3, 7.6 through 7.9, material from old section 7.5 on the transmission of symmetric images, and old sections 8.1 and 82. * More geometric material illustrating the discussions of diagons mettic matrices and singular value decompositions. * Section 1.7, Computer Graphics, has been expanded. ‘+ More applications have been added. ‘These include networks and chemical balance equations. ‘+ The exposition has been expanded in many places to improve the pedagogy and mote explanations have been added to show the importance of certain ‘material and results ‘A simplified discussion showing how linear algebra is used in global positioning systems (GPS) has been added. * More material on recurrence relations has been added, ‘More varied examples of vector spaces have been introduced. + More material discussing the four fundamental subspaces of linear algebra have been added. * More geometry has been added. * More figures have been added. ‘+ More exercises at all levels have been added, ‘+ Exercises involving real world data have been updated to include more recent data sets. ‘+ Mote MATLAB exercises have been added, jon of sym- EXERCISES The exercises form an integral part of the text. Many of them are numerical in nature, whereas others are of a theoretical type. New to this edition are Discus- sion Exercises at the end of each of the first seven chapters, which can be used for ‘writing projects or group activities. Many theoretical and discussion exercises, as Well as some numerical ones, call for a verbal solution. In this technological age, itis especially important to be able to write with care and precision; exercises of this type should help to sharpen this skill. This edition contains almost 200 new exercises. Computer exercises, clearly indicated by a special symbol ™. are of {wo types: in the fist eight chapters there are exercises allowing for discovery and ‘exploration that do not specify any particular software to be used for their soli- tion; in Chapter 10 there are 147 exercises designed to be solved using MATLAB ‘To extend the instructional capabilities of MATLAB we have developed a set of pedagogical routines, called scripts or M-fles, (0 illustrate concepts, streamline step-by-step computational procedures, and demonstrate goometric aspects of top ies using graphical displays. We feel that MATLAD and our instructional M-files provide an opportunity fora working partnership between the student nd the com puter that in many ways forecasts situations that will occur once a student joins the technological workforce. The exercises inthis chapter are keyed to topics rather than individual sections of the text. Short descriptive headings and references to MATLAB commands in Chapter 9 supply information about the sets of exercises.Preface xiii ‘The answers to all odd-numbered exercises appear in the back of the book. An In= structor’s Solutions Manual (ISBN: 0-13-229655-1), containing answers to all even-numbered exercises and solutions to all theoretical exercises, is available (to instructors only) from the publisher. ‘We have learned from experience that at the sophomore level, abstract ideas must be introduced quite gradually and must be based on firm foundations, Thus we begin the study of linear algebra with the treatment of matrices as mete array’ of numbers that arise naturally in the solution of systems of linear equations, a problem already familiar to the student. Much attention has been devoted from one edition to the next to refining and improving the pedagogical aspects of the exposition, The abstract ideas are carefully balanced by the considerable emphasis on the ‘geometrical and computational aspects of the subject. Appendix C, Introduction 10 Proofs can be used to give the student a quick introduction to the foundations of proofs in mathematics. An expanded version of this material appears in Chapter 0 Of the Student Solutions Manual In using this book, for a one-quarter linear algebra course meeting four times a ‘week, no difficulty hes been encountered in covering eigenvalues and eigenvectors, ‘omitting the optional material. Varying the amount of time spent on the theoretical material can readily change the level and pace of the course. Thus, the book can be used to teach a number of different types of courses. ‘Chapter 1 deals with matrices and their properties. In this chapter we also provide an early introduction to matrix transformations and an application of the dot product to statistics. Methods for solving systems of linear equations are discussed in Chapter 2. Chapter 3 introduces the basic properties of determinants and some of their applications. In Chapter 4, we come to a more abstract notion, real vector spaces. Here we tap some of the many geometric ideas that arise naturally. Thus we prove that an n-dimensional, real vector space is isomorphic to R®, the vector space of all ordered n-tuples of real numbers, or the vector space of all n x 1 matrices with real entries. Since R° is but a slight general RP and R?, two- and three-dimensional space are discussed at the beginning of the chapter. This shows that the notion of a finite-dimensional, real vector space is not as remote as it may have seemed when fist introduced. Chapter 5 cov- crs inner product spaces and has a strong geometric orientation. Chapter 6 deals ‘with matrices and linear transformations; here we consider the dimension theo- ems and also applications to the solution of systems of linear equations. Chapter 7 considers eigenvalues and eigenvectors. In this chapter we completely solve the diagonalization problem for symmetric matrices. Chapter 8 (optional) presents an introduction to some applications of eigenvalues and eigenvectors. Section 8.3, Dominant Eigenvalue and Principal Component Analysis, highlights some very useful results in linear algebra. It is possible to go from Section 7.2 directly to Section 8.4, Differential Equations, showing how linear algebra is used to solve differential equations. Section 8.5, Dynamical Systems gives an application of lin- ‘ear algebra to an important area of modern applied mathematics. In tis chapter we also discuss real quadratic forms, conic sections, and quadric surfaces. Chapter 9, MATLAB for Linear Algebra, provides an introduction to MATLAB. Chapter 10, MATLAB Exercises, consists of 147 exercises that are designed to be solvedxiv Preface ‘Marian SOFTWARE STUDENT SOLUTIONS MANUAL ACKNOWLEDGMENTS: using MATLAB. Appendix A reviews some very basic material dealing with sets and funetions. Tt can be consulted at any time as needed. Appendix B, on complex numbers, introduces in a brief but thorough manner complex numbers and their use in linear algebra, Appendix C provides a brief introduction to proofs in mathematics. ‘The instructional M-files that have been developed to be used for solving the exercises in this book, in particular those in Chapter 9, arc available on the following website: www.prenhall.com/kolman, ‘These M-fles are designed to transform many of MATLAB’ capabilities into courseware. Although the computational exercises can be solved using @ number of software packages, in our judgment MATLAB is the most suitable package for this purpose. MATLAB is a versatile and powerful software package whose comerstone is its linear algebra capabilities, This is done by providing pedagogy that allows the student to interact with MATLAB, thereby letting the student think through all the steps in the solution of a problem and relegating MATLAB o act as a powerful calculator to relieve the drudgery of tedious computation, Indeed, this is the ideal role for MATLAB (or any other similar package) in a beginning linear algebra course, for in this course, more than many others, the tedium of lengthy computations makes it almost impossible to solve a modest-size problem. ‘Thus, by introducing pedagogy and reining in the power of MATLAB, these M-files provide a working partnership between the student and the computer. Moreover, the introduction to a powerful tool such as MATLAB early in the student's college career opens the way for other software support in higher-level courses, especially in science and engineering. MATLAB incorporates professionally developed quality computer routines for linear algebra computation. The code employed by MATLAB is written in the C language and is upgraded as new versions of MATLAR are released, MATLAB is available from The Math Works Inc., 3 Apple Hill Drive, Natick, MA 01760, e-mail: info@mathworks.com, [508-647-7000]. The Student version is available from The Math Works at a reasonable cost. This Student Edition of MATLAB also includes a version of Maple™, thereby providing a symbolic computational capability ‘The Student Solutions Manual (ISBN: 0-13-229656-X), prepared by Dennis R. Kletzing, Stetson University, contains solutions to all odd-numbered exercises, both numerical and theoretical We are pleased to express our thanks to the following reviewers of the first eight editions: the late Edward Norman, University of Central Florida; the late Charles S. Dutis, and Herbert J. Nichol, both at Drexel University; Stephen D. Kerr, We- ber State College; Norman Lee, Ball State University; William Briggs, University of Colorado; Richard Roth, University of Colorado; David Stanford, College of William and Mary; David L. Abrahamson, Rhode Island College: Ruth Berger, Memphis State University; Michael A. Geraghty, University of lowa; You-Feng Lin, University of South Florida; Lothar Redlin, Pennsylvania State University, Abington; Richard Sot, University of Nevada, Reno; Raymond Southworth, Pro- fessor Emeritus, College of William and Mary; J. Barry Turett, Oakland Univer- sity; Gordon Brown, University of Colorado; Matt Insall, University of MissouriPreface xv at Rolla; Wolfgang Kappe State University of New York at Binghampton; Richard P. Kubelka, San Jose State University; James Nation, University of Hawaii; David Peterson, University of Central Arkansas; Malcolm J. Sherman, State Univer of New York at Albany; Alex Stanoyevitch, University of Hawaii; Barbara Tabak, Brandeis University; Loring W. Tu Tufts University; Manfred Kolster, McMaster University; Daniel Cunningham, Buffalo State College; Larry K. Chu, Minot State University; Daniel King, Sarah Lawrence University; Kevin Vang, Minot State University: Avy Soffer, Rutgers University; David Kaminski, University of Leth~ bridge, Patricia Beaulieu, University of Louisiana, Will Murray, California State University at Long Beach and of the ninth edition: Thalia D. Jefftes, Wichita State University, Manuel Lopez, Rochester Insitute of Technology, Thomas L. Scofield, Calvin College, Jim Gehrmamn, California State University, Sacramento, John M. Erdman, Portland State University, Ada Cheng, Kettering University, Juergen Ger- ach, Radford University, and Martha Allen, Georgia College and State University. ‘The numerous suggestions, comments, and criticisms of these people grcatly improved the manuscript. We thank Dennis R. Kletzing, who typeset the entire manuscript, the Student Solutions Manual, and the Instructor's Solutions Manual, He found and corrected ‘a number of mathematical errors in the manuscript. It was a pleasure working him, ‘We thank Blaise deSesa for carefully checking the answers to all the exercises. We thank William T. Williams, for kindly letting us put his striking image Trane on the cover of this edition. We also thank Lilian N. Brady and Nina Edetman, Temple University for crit- ically and carefully reading page proofs; and instructors and students from many institutions in the United States and other countries, for sharing with us their ex periences with the book for the past 38 years and offering helpful suggestions. Finally, a sincere expression of thanks goes to Scott Disanno, Senior Man- aging Editor; to Holly Stark, Senior Editor; to Jennifer Lonschein, Editorial As- sistant, and to the entire staff of Prentice Hall for their enthusiasm, interest, and Lnfailing cooperation during the conception, design, production, and marketing phases ofthis edition, It was a genuine pleasure working with them. BK, DRH@ TO THE STUDENT ‘This course may be unlike any other mathematies course that you have studied thus far in at least two important ways. Firs, it may be your initial introduction to abstraction. Second, itis a mathematics course that may well have the greatest impact on your vocation, Unlike other mathematics courses, this course will not give you a toolkit of {isolated computational techniques for solving certain types of problems. Instead, ‘we will develop a core of material called linear algebra by introducing certain def= initions and creating procedures for determining properties and proving theorems. Proving a theorem is 2 skill that takes time to master, so we will develop your skill at proving mathematical results very carefully. We introduce you to abstraction slowly and amply illustrate each abstract idea with concrete numerical examples and applications. Although you will be doing a lot of computations, the goal in ‘most problems is not merely to get the “right” answer, but to understand and be able explain how to get the answer and then interpret the result. Linear algebra is used in the everyday world to solve problems in other areas of mathematies, physics, biology, chemistry, engineering, statistics, economics, fi- nance, psychology, and sociology. Applications that use linear algebra include the {transmission of information, the development of special effects in film and video, recording of sound, Web search engines on the Internet, global positioning system (GPS) and economic analyses. Thus, you ean see how profoundly linear algebra affects you. A selected number of applications are included in this book, and if there is enough time, some of these may be covered in your course. Additionally, many of the applications can be used as self-study projects. An extensive list of applications appears in the front inside cover: ‘There are four different types of exercises in this book. First, there are computational exercises. These exercises and the numbers in them have been carefully chosen so that almost all of them can readily be done by hand. When you use linear algebra in real applications, you will find that the problems are much bigger in size and the numbers that occur in them are not always “nice.” This is not a problem hecause you will almost certainly use powerful software to solve them. A taste ofthis type of software is provided by the third type of exercises. ‘These are exercises designed to be solved by using a computer and MATLAB™, a powerful rmatrix-based application that is widely used in industry. The second type of ex- excises are theoretical. Some of these may ask you to prove a result or discuss an idea. The fourth type of exercises are discussion exercises, which can be used as {group projects. In today’s worl, it is not enough to be able to compute an answer; you offen have to prepare a report discussing your solution, justifying the steps in ‘your solution, and interpreting your results. These types of exercises will give you experience in writing mathematics, Mathematics uses words, not just symbols xvixviii To the Student How to Succeed in Linear Algebra * Read the book slowly with pencil and paper at hand. You might have to read ‘ particular section more than once. Take the time to verify the steps marked “verify” in the text. * Make sure to do your homework on a timely basis. If you wait until the prob- Jems are explained in class, you will miss leaning how to solve a problem by yourself. Even if you can't complete a problem, try it anyway, so that when {you see it done in class you will understand it more easily. You might find it helpful to work with other students on the material covered in class and on ‘some homework problems. ‘+ Make sure that you ask for help as soon as something is not clear to you. Each abstract idea in this course is based on previously developed ideas—much like laying a foundation and then building a house. If any of the ideas are fuzzy to you or missing, your knowledge of the course will not be sturdy enough for you to grasp succeeding ideas. ‘+ Make use of the pedagogical tools provided in this book. At the end of each scotion in the first eight chapters, we have a list of key terms; atthe end of each of the first seven chapters we have a chapter review, supplementary exercises, chapter quiz, and discussion exercises. Answers to the odd-numbered compu: tational exercises appear atthe end of the book. The Student Solutions Manual provides detailed solutions to all odd-numbered exercises, both numerical and theoretical. It can be purchased from the publisher (ISBN 0-13-229656-X) We assure you that your efforts to learn linear algebra well will be amply rewarded in other courses and in your professional career ‘We wish you much suecess in your study of linear algebra, [ermad Hoban anal R. Hob!Linear Equations and Matrices [RE systems of Linear Equations (One of the most frequently recurring practical problems in many fields of study— such as mathematics, physics, biology, chemistry, economics, all phases of enj neering, operations research, and the social sciences—is that of solving a system of linear equations, The equation b o Which expresses the real or complex quantity b in terms of the unknowns x1, x2, 1%» and the real or complex constants dy, 03, «dy, is called a linear equi tion. In many applications we are given b and must find numbers x4, X25. satisfying (1). ‘A solution to linear Equation (1) is a sequence of m numbers $1,825 +++. ns ayn bane to ah which has the property that (I) is satisfied when x1 = 51, ¥2 = $2, «+5 n substituted in (1). Thus x; = 2,.x2 = 3, and x3 = —4 is a solution to the linear equation 6x; 3m + 4x5 because 6(2) — 33) +4(-4) More generally, a system of m linear equations in rt unknowns, £1,212... ns or a linear system, is a set of m linear equations each in n unknowns. A linear ‘Note: Apex A reviews some very basic material dealing with sets and functions. Hean be ‘consul a an ime, a5 needed 12 Chapter 1 Linear Equations and Matrices system can conveniently be written as yxy + aang to + dint = by ayixy + amy +o 2 t aga8a +++ + aantn = by ‘Thus the ith equation is aust anne t+ inky Tn Q) the ay are known constants, Given values of by, ba... Dax We Want to find values of x}, x25... fy that will satisfy each equation in (2). A solution to linear system (2) is a sequence of m numbers s}, 93, Which has the property that each equation in 2) is satisfied when x, = 5), x» ssn = 5) ate substituted, IF the linear system (2) has no solution, itis said to be inconsistent; if it has, 4 solution, it is called consistent. If by = be = by = 0, then (2) is called @ homogeneous system, Note that x) =x) = in = 0 is always a solution to a homogeneous system; it is called the trivial solution. A solution to a homogeneous system in which not all of x1,.¥2,...,q are zero is called a nontrivial solution. Consider another system of r linear equations in n unknowns et team + cat + et + 8) nity + Graky boo + eky = Uy We say that (2) and (3) are equivalent if they both have exactly the same solutions. x) ~ 3m =-7 4) autos 7 ” has only the solution xy = 2 and x = 3. The linear system 8m -3n= 7 3u-2n= 0 6 10x; — 2p = 14 also has only the solution x; = 2 and x2 = 3. Thus (4) and (5) are equivalent, To find a solution to @ finear system, we shall use a technique called the ‘method of elimination; that is, we eliminate some variables by adding a multiple ‘of one equation to another equation. Elimination merely amounts to the develop- ‘ment of a new Linear system that is equivalent to the original system, but is much simpler to solve. Readers have probably confined their earlier work in this area toEXAMPLE 2 See) 1.1 Systems of Linear Equations 3 linear systems in which m = n, that is, linear systems having as many equations as unknowns. In this course we shall broaden our outlook by dealing with systems, in which we have m =n, m n. Indeed, there are numerous applica tions in which m # n, Ifwe deal with two, three, or four unknowns, we shall often ‘write them as.x, yz, and w. In this section we use the method of elimination as it ‘was studied in high school. In Section 2.2 we shall look at this method in a much more systematic manner. ‘The director of a trust fond has $100,000 to invest. The rules of the trust state that both a certificate of deposit (CP) and a long-term bond must be used. The director's goal is to have the trust yield $7800 on its investments for the year. ‘The CD chosen returns 5% per annum, and the bond 9% The director determines the amount x to invest in the CD and the amount y to invest in the bond as follows: Since the total investment is $100,000, we must have x + y = 100,000. Since the desired return is $7800, we obtain the equation 0.05x +0.09y wwe have the linear system xt y= 100,000 6 0.05x + 0.09y = 7800. To eliminate x, we add (—0.08) times the first equation tothe second, obtaining 0.04y = 2800, fan equation having no x term, We have eliminated the unknown x. Then solving for y, we have y = 70,000, and substituting into the first equation of (6), we obtain x = 30,000, ‘To check that x = 30,000, y = 70,000 is a solution to (6), we verify that these values of x and y satisfy each of the equations in the given linear system. ‘This solution is the only solution to (6); the system is consistent. ‘The director of the trust should invest $30,000 in the CD and $70,000 in the long-term bond. Hl Consider the linear system oO Again, we decide to eliminate x. We add (—2) times the first equation to the second one, obtaining 0=21, which makes no sense. This means that (7) has no solution; it is inconsistent. We could have come to the same conclusion from observing that in (7) the left side of the second equation is twice the left side of the first equation, but the right side of the second equation is not twice the right side of the first equation. 74 Chapter 1 Linear Equations and Matrices REY Consent near sytem ® ‘To eliminate x, we add (2) times the fist equation to the second one and (—3) times the first equation tothe third one, obtaining ty= 42-2 °) Sy — 10z = —20, o” ‘This isa system of two equations in the unknowns y and z. We multiply the second equation of (9) by (—2), yielding which we write, by interchanging equations, as y+2end ~Ty = 42 =2. “o We now eliminate y in (10) by adding 7 times the first equation tothe second one, to obtain 102 = 30, =3. ay Substituting this value of z into the first equation of (10), we find that y = ~2. ‘Then substituting these values of y and z into the first equation of (8), we find that x = 1. We observe further that our elimination procedure has actually produced the Tinear system DR os cer te tinear system e+ 2y 432 yr «ay obtained by using the first equations of (8) and (10) as well a (11). The importance of this procedure is that, although the linear systems (8) and (12) are equivalent, (12) has the advantage that it is easier to solve. . x42y «) Ww yokes 41.1 Systems of Linear Equations 5 Eliminating x, we add (2) times the first equation to the second equation to get ~3y 432 = 12. aay We must now solve (14). A solution is yer-4 where z can be any real number, Then from the first equation of (13), eand—2y 432 422-4) 432 acta Thus a solution to the Hinear system (13) is +4 yar-4 ny real number. ‘This means thatthe linear system (13) has infinitely many solutions. Every time wwe assign a value to we obtain another solution to (13). Thus, if and <= is a solution, while if z is another solution, . ‘These examples suggest that linear system may have @ unique solution, no solution, or infinitely many solutions. Consider next a linear system of two equations in the unknowns x and y ax tay byx + bay (as) ‘The graph of each of these equations is straight line, which we denote by €; and £5, respectively. If = si, y = 52 isa solution to the linear system (15), then the point (51, s2) lies on both lines é; and £2. Conversely, ifthe point (s1, 52) lies on both lines £, and ¢>, then x = s1, y = sy is a solution to the linear system (15). Thus we are led geometrically to the same three possibilities mentioned previously. See Figure 1.1 Next, consider a linear system of three equations in the unknowns x, y, and <: ax thy tera di dar thay tere = (16) agx + bay + est = ds ‘The graph of each of these equations is a plane, denoted by Pi, Ps, and Ps, respectively. AS in the case of a linear system of two equations in two unknowns,6 Chapter 1 Linear Equotions and Matrices (a) A unique solation, ()No solution.) Infinitely many solutions FIGURE 1.1 the linear system in (16) can have infinitely many solutions, a unique solution, ot no solution. These situations are illusirated in Figure 1.2. For a more concrete illustration of some of the possible cases, consider that two intersecting walls and the ceiling (planes) of a room intersect in a unique point, a comer of the room, so the linear system has a unique solution, Next, think of the planes as pages of book, Three pages of a book (held open) intersect in a straight line, the spine. ‘Thus, the linear system has infinitely many solutions. On the other hand, when the book is closed, three pages of a book appear to be parallel and do not intersect, so the linear system has no solution. (8) unigue solution. (6) Infinitely many solutions, (©) No solution * * : a bs i 5 Ps ‘es 5 Pe * Poy FIGURE 1.2 If-we examine the method of elimination more closely, we find that it involves three manipulations that can be performed on a linear system to convert it into an equivalent system, These manipulations are as follows: 1. Interchange the ith and jth equations. 2. Multiply an equation by a nonzero constant, 3. Replace the ith equation by c times the jth equation plus the ith equation, i # J. Thatis, replace A,X, + a.ak2 8 int = By by (cary, + 051)x4 + (aye + ai2)t2 +--+ (CAjn + Oin)XSree 1.1 Systems of linear Equations 7 not difficult to prove that performing these manipulations on a linear system leads to an equivalent system. The next example proves this for the second type of manipulation. Exercises 24 and 25 prove it for the first and third manipulations, respectively. ‘Suppose that the ith equation of the linear system (2) is multiplied by the nonzero constant c, producing the linear system. aks + ak HE ake = by yxy + rx + +++ + Gan = be 7 ayy + ematy + +++ + cane = cbs a Pn. Tay = 54,42 = $0, ++. = Sn is solution to 2), then it s a solution to all the equations in (17), except possibly to the ith equation. For the 7th equation we have yt H day + oot Oy (ass, + G08 +++ + dinSn) = Cb; or ajysy + 04282 +--+ Cann = ei ‘Thus the ith equation of (17) is also satisfied. Hence every solution to (2) is also, a solution to (17). Conversely, every solution to (17) also satisfies (2). Hence (2) and (17) are equivalent systems. . ‘The following example gives an application leading to a linear system of two equations in three unknowns: (Production Planning) A manufacturer makes three different types of chemical products: A, B, and C. Each product must go through two processing machines: X and Y. The products requite the following times in machines X and ¥: 1. One ton of A requires 2 hours in machine X and 2 hours in machine Y. 2. One ton of B requires 3 hours in machine X and 2 hours in machine ¥. 3, One ton of C requires 4 hours in machine X and 3 hours in machine ¥. Machine X is available 80 hours per week, and machine Y is available 60 hours per Week. Since management does not want to keep the expensive machines X and ¥ idle, it would like to know how many tons of each product to make so that the ‘machines are fully utilized, It is assumed that the manufacturer can sell as much of the products as is made. To solve this problem, we let x1, x2, and x; denote the number of tons of products A, B, and C, respectively, to be made. The number of hours that machine X will be used is 2xy +3m +4e3, which must equal 80, Thus we have xy + Bua + dis = 80.8 Chapler 1 Linear Equations and Matrices Similarly, the number of hours that machine ¥ will be used is 60, so we have Day + 2eo + 3x5 = 60. ‘Mathematically, our problem is to find nonnegative values of x1, x3, :and x3 so that 2xy + Sx + ds 2xy + 2x + 385 0 60, This linear system has infinitely many solutions. Following the method of Example 4, we see that all solutions are given by 20-45 9 x2 = 20-35 x4 = any real number such that 0 < x3 < 20, since we must have x1 > 0, x» = 0, and 13 > 0. When x5 = 10, we have =I, os while a when 15 = 7. The reader should observe that one solution is just as good as the other, There is no best solution unless additional information or restrictions are given. . ‘As you have probably already observed, the method of elimination has been described, so far, in general terms. Thus We have not indicated any rules for selec ing the unknowns to be eliminated. Before providing a very systematic description of the method of elimination, we introduce in the next section the notion of ‘matrix. This will greatly simplify our notational problems and will enable us to develop tools to solve many important applied problems. Key Terms Linear equation Unique solution Solution of ines equation Homogeneous system No solution Linear system “Trivial solution Infinitely many solutions Unknowns Nontival solution ‘Manipulations on linear systems Inconsistent system, Equivalent systems Method of elimination In Exercises I through 14, solve each given linear system by dort y the method of elimination. ayy 1 xt2ya8 22x —3y-¢de=—12 @ x4 ax dy=4 xodyt r= 5 de tay ae =2 Be yh221 9. nL crs 15, 1” 19. shan: ar + 8y— 22 xb y te De + 2y +60 B 3 a 4 1. de +3y— = 6 8 2e- y42 a8 “5 3r- yt 2=-7 iven the linear system den y=s 4x 2y et, (a) Determine a particular value of so that the system is consistent. (b) Determine a particular value oft so that the system is inconsistent, (@) How many different values of ¢ ean be selected in part (b)? Given the linear system ae bays r+ Bye, (a) Determine particular values for s and # so that the system is consistent (b) Determine particular values for s and 1 50 that the system is inconsistent. {@) What relationship between the values of sand ¢ will _Buarantoe that the system is consistent? Given the linear system x+ 2y=10 3x + (6+ =30, (a) Determine a particular value of so that the system has infinitely many solutions. (b) Determine & particular value of so that the system has unique solution, (©) How many different values of ¢ can be selected in part (by? Ts every homogeneous linear system always consistent? Explain. Given the linear system we3y- xa ay +5e a. 27. 28. 29. 1.1 Systems of Linear Equations 9 Lon —1is a solution. Ayn solution nba = ly = vty = 1 and Fag) 4 29 = 1 asoltion tthe linear system? (@) 83x, 3y, 32, where x, y, and £ are a in part (©), 8 solution to the linear system? ‘Without using the method of elimination, solve the Hinar system Qe yodeans at ‘Without using the method of elimination, solve the linear system 4s 8 2x + 3y Bx 4 5y— Is there a value of r so that x = 1, y = 2,2 = risa solution tothe following linear system? If there is, ind it Qee3y- r= x- yt2es7 det y-2e= 12 Is there a value of r so that r = r,y = 2.2 =1isa solution to the following linear system? If there is, find it Be xadytc tay t2= 9, Show that the linear system obtained by interchanging two equations in (2) is equivalent to 2) ‘Show tha the Finear system obtained by adding a multiple of an equation in (2) to another equation i equivalent Q). Describe the number of points that simultaneously lie in each ofthe three planes shown in each part of Figure 1.2. escre the number of points that simultaneously fe in ceach ofthe three planes shown in each part of Figure 1.3, Let Cy and C2 be circles in the plane, Deseribe the mum- ber of possible points of intersection of Cy and Cs. Tlus- wate each case witha figure. Let S; and 5; be spheres in space, Deseribe the number ‘of possible points of intersection of §, and S,. llustrate ‘each case with a figure10 Chopter 1 Linear Equations and Matrices @ AT > o FIGURE 1.3, 30. Anooil refinery produces low-sulfur and high-sulfur fuel Each ton of low-sulfur fuel requires 5 minutes in the blending plant and 4 minutes in the refining plant; each ton of high-sulfur fuel requires 4 minutes in the blending plant and 2 minutes in the refining plant. Ifthe blend ing plant is available for 3 hours and the refining plant is available for 2 hours, how many fons of each type of fuel, should be manufactured so thatthe plants are fully used? A plastics manufacturer makes two types of plastic: reg ‘lar and special. Each ton of regular plastic requires 2 hhours in plant A and 5 hours in plant B; each ton of special plastic requires 2 hours in plant A and 3 hours in plant B. If plant A is available 8 hours per day and plant Bis available 15 hours per day, how many tons of each, type of plastic can be made daily so that the plants are fully used? A dictcian is preparing a meal consisting of foods A, B, tnd C. Fach ounce of food A contains 2 units of pro“ tein, 3 units of fat, and 4 units of carbohydrate, Each ounce of food B contains 3 units of protein, 2 units of fat, and 1 unit of carbohydrate. Each ounce of food C ‘contains 3 units of protein, 3 units of fat, and 2 units of ‘carbohydrate. Ifthe meal must provide exactly 25 units Of protein, 24 units of fat and 21 units of carbobydrate, how many ounces of each type af food should be used? A manufacturer makes 2-minute, 6-minute, and 9-minute film developers. Each ton of 2-minute developer requires ‘6 minutes in plant A and 24 minutes in plant B. Bach ton ‘of 6-minute developer requires 12 minutes in plant A and 12 minutes in plant B. Each ton of 9-minute developer re- quites 12 minutes in plant A and 12 minutes in plant B. If plant A is available 10 hours per day and plant B is, 3 3 38. 236. 2a. Bas. 2x. available 16 hours per day, how many tons of each type of developer ean be produced so thatthe plants are fully used? Suppose that he dhree point (1, ~8),(—1, 1), and (2,7) lie onthe parabola p(x) = ax? Fb (@) Deiermine a near system of three equations in thee ‘unknowns that must be solved to find a,b, and (b) Solve the inear system obusined in part (a for a,b, and. An inheritance of $24,000 isto be divided among three ‘wuss, withthe second trstreeiving twice as mach as the fist trast, The three trusts pay interest annually at the rates of 9%, 10%, and 6%, respectively, and retum a total in interest of $2210 a the end of the first year. How ruch was invested in each ust? For the software you are using, determine the command that “automatically” solves a linear system of equations Use the command feom Exercise 36 to solve Exercises 3 and 4, and compare the output with the results you ob: tained by the method of elimination Solve the linear system ththe-3 sng yours Compu bcp sn with the exact solution x = 1, y = },2= 1 If your software includes access to a computer algebra system (CAS), use its follows: (@) For the linear system in Exercise 38, replace the fraction $ with its decimal equivalent 0.5, Enter this, system into your software and use the appropriate CAS commands to solve the system. Compare the solution with that obtained in Exercise 38. (b) Insome CAS environments you can select the number of digits to he used in the calculations. Perform, part (a) with digit choices 2, 4, and 6 to see what influence such selections have on the computed so- Intioa, 1 If your software incindes access 10 a CAS and you can seiect the number of digits used in calculations, do the following: Enter the linear system O7ix + 0.21y = 092 0.23 40.58) =081 into the program. Have the software solve the system with digit choices 2, 5, 7, and 12. Briefly discuss any the solutions generatedPeau 1.2 Matrices 11 1.2 | Matrices fwe examine the method of elimination described in Section 1.1, we ean make the following observation: Only the numbers in front of the unknowns x1, 225 ++» and the numbers by, by... Dy on the right side are being changed as we perform the steps inthe method of elimination. Thus we might think of looking for a way of writing a linear system without having to carry along the unknowns, Matrices enable us to do this—tha is, to write liner systems in a compact form that makes it easier to automate the elimination method by using computer software in order to obtain a fast and efficient procedure for finding solutions. The use of matrices, however, is not merely that of a convenient notation. We now develop operations ‘on matrices and will work with matrices according to the rules they obeys this will enable us to solve systems of linear equations and to handle other computational problems in a fast and efficient manner. Of course, as any good definition should do, the notion of a matrix not only provides a new way of looking at old problems, ‘but ulso gives rise to a great many new questions, some of which we study in this, book. Anm xn matrix A isa rectangular array of mm real or complex numbers arranged in m horizontal rows and 7 vertical columns: an an aie A= a) ay ¢ | annow 1 jt column ‘The ith row of A is [an a2 aie) sims thejth column of A is, ay ‘We shall say that A is m by m (written as m xm), If'm =n, we say that A is a square matrix of order m, and that the numbers @)1, 432... dy, form the main diagonal of A, We refer to the number a, which is in the ith row and jth column of A, as the i, jth element of A, or the (i,j) entry of A, and we often write (1) as, ay]2 Chapter 1 Linear Equotions and Matrices Ce Bee Let Then A isa2 x3 matrix with a = 2, ais 1; Bisa2x2 ‘matrix with bj, = 1-+i, bi2 = 4i, bap 3x1 matrix with e1y = 1, ca = 1, and es = 2; D is a3 x 3 matrix; E is a1 x 1 matrix; and F isa 1 x3 matrix. In D, the elements dj; = 1, dz2 = 0, and dss = 2 form the main diagonal : For convenience, we focus much of our attention in the illustrative examples and exercises in Chapters 1-6 on matrices and expressions containing only real snumbers. Complex numbers make a brief appearance in Chapter 7. An introdue- tion to complex numbers, their properties, and examples and exercises showing hhow complex numbers are used in linear algebra may be found in Appendix B. ‘Annx1 matrix is also called an n-veetor and is denoted by lowercase boldface letters. When m is understood, we refer to n-vectors merely as vectors. Vectors are discussed at length in Section 4.1 1 2 1 w=] _j | isadvectorand y= | =1 | isa 3-vector . 3 0. ‘The n-vector all of whose entries are zero is denoted by 0. Observe that if A is an m x1 matrix, then the rows of A are I xm matrices and the columns of A are m x 1 matrices. The set of all n-vectors with real entries is denoted by R". Similarly, the set of all n-vectors with complex entries is denoted by C”. As we have already pointed out, in the frst six chapters of this book we work almost entirely with vectors in R” (Tabular Display of Data) The following matrix gives the airline distances be- ‘ween the indicated cities (in statute miles): London Madrid New York Tokyo London 0 785-3469 5959 Madd | 785 0 3593. 6706 Now York | 3469 3593 06757 Toyo | 5959 6706-67570. . (Production) Suppose that a manufacturer has four plants, each of which makes three products. If we let aj; denote the number of units of product i made by plant1.2 Matrices 13 jin one week, then the 3 x 4 matrix Plant 1 Plant2 Plant3 Plant 4 Product! [ 560 360 3800 Product? | 340 450 420-80 Product} | 280 270 210 380 gives the manufacturer's production forthe week. For example, plant 2 makes 270 units of product 3 in one week. . ‘The windchill table that follows is a matrix. pease reense enn amsh 180-1 18-253 38 45 Lo -ID $24 31-39 6-53, A combination of air temperature and wind speed makes a body feel colder than the actual temperature. For example, when the temperature is 10°F and the wind is 15 miles per hour, this causes a body heat loss equal to that when the temperature =18°F with no wind, . By a graph we mean a set of points called nodes or vertices, some of which are connected by lines called edges. The nodes are usually labeled as Pj, Pa... Pas and for now we allow an edge to be traveled in either direction, One mathematical representation of a graph is constructed from a table. For example, the following table represents the graph shown’ PoP Ps my RO 1 0 0 BRoloo tt ce Poo 1 0 71 % Roo 1 tO a ‘The (i, j) entry = 1 if there is an edge connecting vertex P, to vertex Pj: otherwise, the (, j) entry = 0. The incidence.matrix A is the k x k matrix obiained by omitting the row and column labels from the preceding table, The incidence matrix for the corresponding graph is A meer ones 010 101 o10 od14 Chapter 1 Linear Equations and Matrices TIntemet search engines use matrices to keep track of the locations of information, the type of information ata location, keywords that appear in the information, and even the way websites link to one another} A large measure of the effective- ness of the search engine Google® is the manner in which matrices are used to determine which\Gites are referenced by other sites.) That is, instead of directly keeping track of the information content of an actual web page or of an individual search topic, Google's matrix structure focuses on finding web pages that match the search topic, and then presents alist of such pages in the order of their “importance.” Suppose that there are n accessible web pages during a certain month. A simple Way to view a matrix that is part of Google’s scheme is to imagine ann xn ‘matrix A, called the “connectivity matrix,” that intially contains all zeros. To build the connections, proceed as follows. When you detect that website j links to website i, set entry aiy equal to one. Since nis quite large, inthe billions, most entries of the connectivity matrix A are zero. (Such a matrix is called sparse.) If row 7 of A contains many ones, then there are many sites linking to site i Sites that are linked to by many other sites are considered more “important” (or to have a higher rank) by the software driving the Google search engine. Such sites would appear near the top of a list returned by a Google search on topics related to the information on site i. Since Google updates its connectivity matrix about every month, 1 increases over time and new links and sites are adjoined to the connectivity matrix. In Chapter 8 we elaborate a bit on the fundamental technique used for ranking sites and give several examples related to the matrix concepts involved. Further information can be found in the following sources: 1. Berry, Michael W., and Murray Browne. Understanding Search Engines— ‘Mathematical Modeling and Text Retrieval, 2d ed. Philadelphia: Siam, 2005. 2. www.google.com/technology/index. htm! 3. Moler, Cleve. “The World’s Largest Matrix Computation: Google’s PageRank Is an Eigenvector of a Matrix of Order 27 Billion,” MATLAB News and Notes, October 2002, pp. 12-13, Whenever new object is introduced in mathematics, we must determine when two such objects are equal. For example, inthe set of all rational numbers, the numbers 3 and are called equal, although they have different representa tions. What we have in mind is the definition that a/b equals c/d when ad = be. ‘Accordingly, we now have the following definition PER | Bo" x wmatices A = [ay] and B = [by ] are equal if they agree entry by entry, that is, if aiy = byy fori = 1,2,....m and j Sia Th matrices 1 2-1 12 A=|2-3 4] and pal2 x 0-4 5 y -4 are equal ifand only if w = —1, x = —3, y =0,and z = 5. .Poin mee eas 1.2 Matrices 15 1M Matrix Operations ‘We next define a number of operations that will produce new matrices out of given matrices. When we are dealing with linear systems, for example, this will enable us to manipulate the matrices that arise and to avoid writing down systems over and over again. These operations and manipulations are also useful in other applications of matrices Matrix Addition 1A =[a, Jand B = [b, ]are both mxn matrices, then the sum A+B isan mxn matrix C = [cy] defined by ey = aj +5y),i = 1,2,--.4m5 j= ‘Thus, to obtain the sum of A and B, we merely add corresponding entries. Let 0 21 Jeen[? 2] Then tapa[it® 242 341 10 4 AYP=lo41 -143 44¢-4]7[3 2 0 . (Production) A manufacturer of a certain product makes three models, A, B, and C. Bach model is partially made in factory F; in Taiwan and then finished in factory F> in the United States. The total cost of each product consists of the manufacturing cost and the shipping cost. Then the costs at each factory (in dollars) can be described by the 3 x 2 matrices Fy and Fo: Manufacturing Shipping cost cost 2 40] Model a. R= 30 80 | Model B 70 20 | Modelc Manufacturing Shipping cost ‘cost 40 60] Model A. = 50 50 | Model B 130 20} Moser c The matrix Fy + Fp gives the total manufacturing and shipping costs for each product, Thus the total manufacturing and shipping costs of a model C product are {$200 and $40, respectively. . IF x is an n-vector, then it is easy to show that x +0 = x, where 0 is the n-vector all of whose entries are zero, (See Exercise 16.) It should be noted that the sum of the matrices A and B is defined only when ‘A and B have the samé number of rows and the same number of columns, that is, only when A and B are of the same size.16 Chapter 1 Linear Equations and Matrices PE need ‘We now make the convention that when A + B is written, both A and B are of the same size. ‘The busie properties of matrix addition are considered in the next section and are similar to those satisfied by the real numbers. Scalar Multiplication IA = [ay ] is an m x n matrix and r isa real number, then the sealar multiple of Aby r.rA, isthe m x n matrix C= [i]. where cj = ray, = 1,2,..-.m and J = 1.2,-+.+m5 that is, the matrix C is obtained by multiplying each entry of A byr. We have [f 2 3)-[23 (-2)(-2) 9] 7-3 2" L-am Cae —-e) 3 4 6 =| 6 -4 . ‘Thus far, addition of matrices has been defined for only two matrices. Our ‘work with matrices will call for adding more than two matrices. Theorem 1.1 in Section 1.4 shows that addition of matrices satisfies the associative property: A+ (BHC) =(A4 BEC. If A and B are m xn matrices, we write A+ (—1)B as A ~ B and call this the difference between A and B. we 203-5 2-103 A [i 2 1] and 8 [i 3 al Then 2-2 341 -5-3]_[o 4 -8 acoefis ote ale 3a) Application Vectors in R" can be used to handle large amounts of data. Indeed, a number of computer software products, notably, MATLAB®, make extensive use of vectors. ‘The following example illustrates these ideas: (Unventory Control) Suppose that a store handles 100 different items. The inventory om hand at the beginning of the week can be described by the inventory vector win R™., The number of items sold at the end of the week can be described by the 100-vector v, and the vector uny represents the inventory atthe end ofthe week. Ifthe store receives a new shipment ‘of goods, represented by the 100-vector w, then its new inventory would be unvdw. .1.2 Matrices 17 We shall sometimes use the summation notation, and we now review this useful and compact notation, By Ya; we mean a; + ay + +++ +.dy, The letter i is called the index of summation; itis a dummy variable that can be replaced by another letter. Hence ‘Thus ‘ Vasatatata ‘The summation notation stistes the following properties 1 Die + s)a = Sra + Ysa, a) 2. Detnai) = ena £(Eu)-£(E4) Property 3 can be interpreted as follows: The left side is obtained by adding all the entries in each column and then adding all the resulting numbers. ‘The right side is obtained by adding all the entries in cach row and then adding all the resulting, numbers. TEAL, Aa... Ap are m xn matrices and ¢), ¢2,.-., €r are real numbers, then. aan expression of the form crAy + epAa toe tOAL @) is called a linear combination of Ay, Ao,-.., Au, and ¢},€2,..+¢e are called coefficients The linear combination in Equation (2) can also be expressed in summation notation as: Deas = cra teada te + eA ‘The following are linear combinations of matrices: 0 3S 5 2 3 323 4|/-s| 6 2 3 1-2-3 -1 203 2[3 -2]-3[5 o]+4[-2 5]. 1 On -0.5| -4| +04) -4 -6. 02.18 Chapter 1 Linear Equations ond Matrices te Pein med Using scalar multiplication and matrix addition, we can compute each of these linear combinations. Verify that the results of such computations are, respectively, -§ -10 0.46 3 8 [-17 16], and 04 Ios 3.08, 14.75 8.60. be a 3-vector that represents the current prices of three items at a store. Suppose tha the store announces a sale so that the price of each item is reduced by 20%. (a) Determine a 3-vector that gives the price changes for the three itera. (b) Determine a 3-vector that gives the new prices of the items, Solution (a) Since each item is reduced by 20%, the 3-vector (-0.20)18.95 3.79 3.79 0.20p = | (-0.20)14.75 2.95 | = — | 2.95 (-0.20)8.60, 1.72, 172 gives the price changes for the three items. (b) The new prices of the items are given by the expression 18.95] [3.79 15.16 p—0.20p = | 14.75 | — | 2.95 | = | 11.80 60} | 1.72 6.88. Observe that this expression can also be written as p—0.20p = 0.80p. . ‘The next operation on matrices is useful in a numberof situations IA = [aj] is an m x m matrix, then the transpose of A, AT = [af] is the 1 xm matrix defined by af, — a. Thus the transpose of Ais obtained from A by interchanging the rows and columns of A.1.2 Matrices 19 ‘Then 400 6 30 AT =| — . Bl= 2 +_fs 3 ec [i 2-3 . Key Terms Matix qual matrices Dierenoeof matrices Rous sevecto (or veto) Summation notation Coturans wees Index of summation Sine of «marx 0, zero vetor Liner combination Square mati Google Cooficients Main diagonal Matix ation Transpose Element or entry ofa matrix Sealar mip Le Let a 1000 boii a=[i 3 w a=|o 1000 01000 ad o1000 7 4 c=|-4 ath ctd]_fa 6 6 e-d a-b}~ lo 2 (a) What is a12,a22, 423? (b) Whatis by.» ? (©) Whats css.cn,688? 2. Determine the incidence matrix associated with each of the following graphs: fa) Bs (by Ps > 7 ee P (Es) - x( ) Dewa 18, emi he following exgrenion on teor false ie, pron ere ire pea comtranpl @ Darn= (E>) +n oy (5 ‘) =mn Lz oS (=) - | [=] 2 A tage sce! mantacer, who has 2500 employees, fh cach ployee say ev cmpnenc of eet ‘uin R®°, If an 8% across-the-board salary increase has Bean approved dan expen iolving that ies ation slr Di. A brokerage fi ead he ih no vale ofthe ‘pec ol sock cach ay Tenorio for given week is presented in two vectors, t and b, in R*, showing The Beh and fw vas, respectively. What expression bese averge daly vale ofthe rc of IBN ack faite cate sy est? ]atinerombinon of the ma- M22. Forthe software you are using, determine the commands tocnter a matrix, add matrices, multiply a scalar times a ‘maitin, and obtain the transpose of a matrix for matices with numerical entries, Practice the commands, using the linear combinations in Example 13. Determine whether the software you are using includes «computer algebra system (CAS), and if it does, do the following: (4) Find the command for entering a symbolic matrix. (This command may be different than that for enter- ig a numeric matrix.) (b)_Enter several symbolic matrices like vee Jw [E25] ‘Compute expressions like A + B, 2A, 34+ By A~2B, AT + BT, etc, (In some systems you must1.3. Matrix Multiplication 21 explicitly indicate scalar multiplication with an as- matrix. If there i, display the graphs for the incidence terisk.) ‘matrices in Exercise 3 and eompare them with those that B 24, For the software you are using, determine whether there you drew by hand. Explain why the computer-generated is a command that will display a graph for an incidence PED on ‘graphs need not be identical to those you drew by hand. EEE] Matrix Muttiptication In this section we introduce the operation of matrix multiplication. Unlike ma- ‘tix addition, matrix multiplication has some properties that distinguish it from multiplication of real numbers, ‘The dot product, or inner product, of the n-vectors in " a, by a by and b= a bn is defined as ‘The dot product is an important operation that will be used here and in later sections, ‘The dot product of 1 2] ~ 3 v=|73| and v=! > 4 1 uv = (DQ) +23) + BN-2) +4) = -6. . Solution We have 1 | (:] vache 4nd : Te dot product of vectors in C* is deine in Appendix B.222 Chapter 1 Linear Equations and Matrices RSE (Computing « Course Average) Suppose that an instructor uses four gras to determine a student's course average: quizzes, two hourly exams, and a final exam. ‘These are weighted as 10%, 30%, 30%, and 30%, respectively. Ifa student’s scores are 78, 84, 62, and 85, respectively, we can compute the course average by letting 0.10 18) 030} ag ga] 0.30 2 0.30. 85 and computing Wg = (0.10)(78) + (0.30)(84) + (0.30)(62) + (0.30)(85) = ‘Thus the studs 1's course average is 77.1 . 1 Matrix Multiplication 14 = [ay isan m x p matrix and B = [by] is a p xn matrix, then the product of A and B, denoted AB, is the m x n mattix C = [cy], defined by 4p = aby + ainboj +++ + dipbyy = Yay ( 1 matrix an an + ae] fe] fcowtayte mon ay | |e (row2(A))T -€ Ac= = ams aya = tan Lend [oom (A))? 6 @ anes ane +++ ain : dane; + dinta + °F anon nic) + O22 + -°- + Oe This last expression can be written as al te fete 8) = cyeoh (A) + excolg(A) + +++ + e,001,(A). Thus the product Ac of an m xn matrix A and an m x I matrix ¢ can be written as a linear combination of the columns of A, where the coefficients are the entries in the matrix e TirOUr study of linear systems of equations we shall see that these systems can be expressed in terms of @ matrix-vector product. ‘This point of view provides us with an important way to think about solutions of linear systems. 2-1 3 2 s-[2 1] --| ‘| 2-1 -3 2 2 1 3 5 fi 4] Ee GG) If Ais an m x p matrix and B is a p xn matrix, we can then conclude that the jth column of the product AB can be written as a linear combination of the Let1.3. Matrix Multiplication 27 ‘columns of matrix. A, where the coeflicients are the entries in the jth column of matrix B: c01)(AB) = Acol;(B) = by jcol\(A) + bajcola(A) + +++ + Bp/COlp(A), Wf A and B are the matrices defined in Example 11, then 125 47 6 Ap=| 30 4 [3 3 i] =| 6 17 6 5 WoT . 4 col,(AB) = [ ‘] = Acol(B) 17. T 1 2 coh(AB) = | 17 | = Acoh(B)=3] 3] +2] 4 7 -1 5 f cols(AB) = («| = Acols(B) Ml Linear Systems Consider the linear system of m equations in n unknowns, aun + ayaa bee + inte = 1 agin + ama to + doen = be w@ GyiX1 + aata °° Orne Now define the following matrices: ay ai. ay m1 by a-|@ @ en de Gy a A Xn bm28 Chapter 1 Linear Equations and Matrices ‘Then ay an ay] Px a) a am |) Ax= Ap A ann | Le 6 yxy + ane Foe + aint anX) + aay + + dank ayy1%1 + ygta t o> + Oren ‘The entries in the product Ax at the end of (5) are merely the left sides of the ‘equations in (4), Hence the linear system (4) can be written in matrix form as Ax=b, ‘The matrix A is called the coefficient matrix of the linear system (4), and the ‘matrix ay an aay | Bs a) daa ay | be Gigt yd» ann ry ‘obtained by adjoining column b to A, is called the augmented matrix of the linear System (4). The augmented matrix of 4) is writen as [A ib}. Conversely, any tnatrit-with_more than-one column can be thought of asthe augmented matrix of a linear system. The coefficient and sugmented matices play key roles in our method fr solving linear systems. Recall from Section 1.1 that f by by = + = Py =0 in (4), the linear system is called a homogeneous system. A homogeneous system can be written as Ax=0, where A is the coefficient matrix. EXAMPLE 14 Letting “|1.3 Matrix Multiplication 29 ‘we can write the given linear system in matrix form as Ax=b. ‘The coefficient matrix is A, and the augmented matrix is 2 0 «115 2 3-4/7]. 302 213 1: kun“: 2-1 314 300 215 2x ‘We can express (5) in another form, 28 follows, using (2) and (3) aunt + ayaa to danke aay + anny +--+ danke Ax= Gigi X1 + 28 Fo enka aux ana Ann ayy any Aan = + + att Ona Onn Xe au an an =x tall [tetal Ay Gn Onn 1 €01;(A) + x2 €01g(A) +++ xy COMQ(A). ‘Thus Ax is @ linear combination of the columns of A with coefficients that are the entries of x. It follows that the matrix form of a linear system, Ax = b, can be expressed as 251601 (A) + x2€01g(A) + +++ 60h (A) = b, 6 Conversely, an equation of the form in (6) always describes a linear system of the form in (4). PRESET Coie tinea tem As =p, where the eatiint mat 301 2 4-5 6 a bere30 Chapter 1 Linear Equations and Matrices Writing Ax = b as a linear combination of the columns of A as in (6), we have 3 1 27 74 4 -5 6 1 a} o(t@lo]+s13}=lo -1 2, ol 12 ‘The expression for the linear system Ax = b as shown in (6), provides an ‘important way 1o think about solutions of linear systems. cee annette emermincmee nat Ax = bis consistent if and only if b can be expressed as a linear combination ‘of the columns of the matrix A, ‘We encounter this approach in Chapter 2. Key Terms Dr provet (inner prod) Coefficient matrix Matrix-veetor product Augmented matrix In Exercises 1 and 2, compute a+b. Eh-L] w-E-L] SJ evel o-(p-l] sn ately wao-[] var -[i}e-[Jme Et Determine values of and y so that v= w= Oand yu = =A sae [12°] open L@ O) © 8, Find al values of x so that w= (a 1 ‘] 9, Final values os vob y= = 0y misae[} 2 sJuer-[3] 6 wrane[S Consider the following matrices for Exercises 11 through 15: Lo a-[h 2g} ofa f]. ba sag e-[s 1 3). o-[} 4]. ee 4s “2 ‘ f]. ow ref 3 2 33 1, If possible, compute the following: nd ad. (a) AB (b) BA © FE @ CBYD — @ AB+D*,where D = DD 12, If possible, compute the following: (@) DA+B () EC (CE @ EBY+F @ FC+D 13. Ifpossible, compute the following: (a) FD~3B (b) AB-2D (© FTB+D (@ 2F HAE) (©) BD+AE 14. possible; compute the following: (@) A(BD) @) (4B)D @ AC +E) (@) AC+AE (©) @ABY and (4B) ) A(C~3E) 15. If possible, compute the following: (a) at ant fe) (ABy (@ Brat (©) (CHE) Band CB + ETB (AB) and (4B) 16 Leta=[l 2 -3}B=[-1 4 2] and C=[-3 -0 1]. Hpossile, compute the following (a) AB’ (b) cat (© (BAT)C (a) ATB fe) ccT wcrc (@ BtCAAT 1.3. Matrix Multiplication 31 23 wisa=[-t i}eae=[) JJ] 03 conps ne wisn of A (a) the (1,2) entry (b) the (2,3) entry (©) the @,Lentey —@) the (3,3) entry on and [2D. wie 12 a-[} 3] os Show that AB # BA. TFA is the matrix in Example 4 und 0 is the 3 x 2 matrix ‘every one of whose entries is zero, compute AO. wine [} f]mo~[ 7 : Spesns ok In Bercises 21 and 22, let 1-1 32 4-2 201 1 0-1 2 B=|3 33 4 425 1 21. Using the method in Example 11, compute the following columns of AB: {a) the first column (b) the third column 22, Using the method in Example 11, compute the following columns of AB: (a) the second column (b) the Fourth column 23. Let 2-3 4 2 A=|-1 2 3] and e=|t Ss -1 2 4 Expeess Ac asa linear combination ofthe columns of A. 24, Let 1-2-1 1-1 A=|2 4 3] and B=|3 2 300 2 2 4 [Express the columns of AB as linear combinations of the columns of A. and.32 Chapter 1 Linear Equations and Matrices 3 saa ‘Jme-[3] (a) Verify that AB = 3ay + Sas + 2a, where a; isthe (ven ara = [AP] 26, (a) Find a value ofr so that AB” = 0, where =[r 1 -2}ade=[1 3-1} (b)_ Give an alternative way to write this product, 27, Finda value ofr anda value of sothat AB” = 0, where As[l r tans 28. (a) Let be anm xn matix with a row consisting en- trey of zeros. Show that if B isan np matin then AB has a row of zero. (©) Let bean m xn matsx with a column consisting entirely of zero and let B be pm. Show that BA has a column of zets, 321 _ ay vam [~} 3h] wita, =e jtheolamnot , F123. Verily chat ATA=| ala, afa: ala, alas aly 230, Consider the following linea system: dn tim-tnt ut ee 7 By In +s Qyt3n dy bt ut w= 5 (a) Find the coefticent matrix. (b)_ Write the linear system in matrix form, (©) Find the augmented matrix, 31, Write the linear system whose augmented matrix is 2-1 “32 10 300 32, Write the following linear system in matrix form: 2x +3 ay — 5 1. Weite the following near system in matrix form: 2m +3n <0 nt 0 Qn = a0 1. Write the linear system whose augmented matrix is 213 410 @ | 3-1 2 013) 21-8 312 213 4 a1 200 OM) 2 1-4 352 0 0 0 0i0. How are the linear systems obtained in Exercise 34 related? Write each of the following linear systems as a linear combination ofthe columns of the coeflicent matrix (a) 35 +20 + a mee; oy 7. Write each of the following linear combinations of columns asa linear system of the form in (4); oaleE ED of} E El Write cach of the following as a linear system in matrix form: oofjE-G)-U) «lol oE) ). Determine a solution to each of the following linear sys- ems, using the fact that Ax = b is consistent ifand only if bisa linear combination of the columns of A: [TELE “FER4 a 2. 4B 46. 1 Cans cei! mais As atx = [*] iss ' there be more than one such coefficient matrix? Explain, Show that ifw and v are n-veetors, then w+ Let A bean m xn matrix and B an n> p matrix. What, if anything, can you say about the matrix product AB when (@)_ A has a column consisting entirely of zeros? () has. row consisting entieely of zeros? MA = [ay] is ann nm matrix, then the trace of A, Tr(A), is defined asthe sum of al elements onthe main diagonal of A, THA) = Jay. Show each of the follow solution tothe system Ay by where b ing: (@) Tea) = eH), whete cis real number () A+B) = THA) + TB) (©) THAR) = TBA) @ TWAT) = Tray ( THATA)=0 Compute the tae (Se Exercise 4) ofeach ofthe fo lowing mate 22 9 of m2 4 4 : 3a -s Loo @ Joo dot Show tater ar no 2 x 2 maces A and B such hat ny assem) (@) Show that the jin col of he mati prot AB is equal othe marx product Ab, wher bys the Sunol of BI follows thatthe prodoct A can de writen in ems of columns 8 AB= [Ab Abs Ady (b) Show dt the th ow of he mai paduc AB is qual othe mix product where abe th tow of A I fll that the product AB can be ‘wien in tems of rows a ae ae ag B a. 48. . (Manufacturing Costs) 1.3 Matrix Moliplicaion 33 Show thatthe jh column of the matrix product AB is a Tinear combination ofthe columns of A with cocticients the entries in by, heh column of B. The veer 20 30 wo 1a ives the number of recsvers, CD players, speakers, and VD recorders that are on hand in an audi shop. The vestor 200 120 80 70. gives the price (in dollars) of each receiver, CD player, speaker, and DVD recorder, respectively. What does the dot product uy tel the shop owner? ‘A fumiture manufacturer makes chairs and tables, each of which must go through, fan assembly process and a finishing process. The times. required for these processes are given (in hours) by the ‘matrix Assembly Finishing oa et 22] char af 5d dan ‘Tho manufacturer has a plant in Salt Lake City and an- ‘other in Chieago. The hourly rates for each of the pro- ‘cesses are given (in dollars) by the matrix ‘Salt Lake Gig Chicago 9 10 » 2 What do the entries in the matrix product AB tell the manufacturer? (Medicine) _ dict research project includes adults and children of both sexes. ‘The composition of the partici- ‘pant inthe project is given by the matrix Assembly process Finishing process B Adults Children ac f 82 120 7 Mae 100 200 | Femate ° “The number of daily grams of protein, fat, and carboby- «date consumed by each child and adult is given by the matrix asbo- Protein Pat ydrate Adult 8 2 20 20 10 20 30 | ow34 Chapter 1 Linear Equotions and Matrices (a) How many grams of protein are consumed daily by (©). Show that (ka) -D ‘the males inthe project? + (kb) = kab), 53, Let A be an m x7 matrix whose entries are real numbers. (b) How many grams of fat are consumed daily by the Show that if AAT = O (the m xm matrix all of whose females in the project? SI. Let x be an n-vector. (@) Ist possible for xx (o be negative? Explain. (b) Ihx-x=0, whatis x? entries are zero), then A = O. JB 54, Use the matrices A and C in Exercise 11 and the matrix multiplication command in your software to compute AC and C/A, Discuss the resus 55, Using your software, compute BB and BB" for 52, Leta, band ¢ be n-vectors and let & be areal number, (a) Show that ab (b). Show that (a+b) + Theorem 1.1 gst} b pis al ct bee. Discuss the nature ofthe results. Algebraic Properties of Matrix Operations In this section we consider the algebraic properties of the matrix operations just defined. Many of these properties are similar to the familiar properties that hold for real numbers. However, there will be striking differences between the set of real numbers and the set of matrices in their algebraic behavior under certain ‘operations—for example, under multiptiation (as seen in Section 1.3). The prools ‘of most of the properties will be left as exercises. Properties of Matrix Addition Let A, B, and € be m x n matrices. (a) AY B=B+A. (b) A+ (BHC) = (A+B) +E, (@) There is a unique m x matrix’ O such that A+O=A ao for any m x n matrix A. The matrix Q is called the m x n zero matrix, (d) Foreach m x n matrix A, there is @ unique m x m1 matrix D such that AtD=0. Q We shall write D as —A, so (2) can be written as A+CA)=O. “The matrix —A is called the negative of A. We also note that —A is (—1)A. Proof (a) Let Asa}, #=[bs] A+B=C=[c;], and B+A= =[4] We must show that ¢jj = dy for all, j. Now ci) = aij + by and diy = by +43 for alli, j. Since ay and by; are real numbers, we have aij +biy = bij aij, which implies that ¢;; = dy for all i,jaan Bee) Theorem 1.2 1.4. Algebraic Properties of Matrix Operations 35 (©) Let U = [uy]. Then A+ U = A if and only iff ay + uy = ay, which holds iff and only if uj = 0. Thus U is the m xm matrix all of whose entries are zero: U is denoted by 0. . ‘The 2 x 2-zer0 matrix is 00 o=[5 6] It 4 -[! then 4-1) ,f0 0] _[4+0 -1+0]_y4 -1 2 a}tlo of=|2+0 340)> [2 3 . ‘The 2 > 3 zero matrix is o-fo 9 ° =lo 0 0 13 2 1-302 wan[t 3 thena-[-} 3 3) . Properties of Matrix Multiplication (a) IFA, B, and C are matrices of the appropriate sizes, then A(BC) = (ABYC. (b) IFA, B, and C are matrices of the appropriate sizes, then (A+ BNC = AC + BC. (©) IFA, B, and C are matrices of the appropriate sizes, then C(A+ B)=CA+CB. e Proof (a) Suppose that A is m xm, B is n x p. and C is p x q. We shall prove the result for the special case m = 2, n = 3, p = 4, and g = 3. The general proof is ‘completely analogous. = [bye Let A= [ai J (AByC [iy and Bey all, j, Now cy], AB = D=[dj], BC = J. We must show that fi, eu), iy for Tie connector “if and only iP" means thst both statements are tue or both statements are fal “Thus (Dif AYU = A, then gy my = ay and (i ify uy) =a, then A+ U = A. See Appendix "Intrdction o Proofs.”6 Chapter 1 Linear Equations and Matrices and ‘Then, by the properties satisfied by the summation notation, Dlaisdre + arabe + aisbs0euy fi ain J buen) + aia D> bxerj +033 Yds ‘ Ee (Yo) \ ‘he profs of (0) and (earl as Bers 4 . BoE “11 0 2 3 wf 2 3 1d), an 0 2 30 03 1 0. Ten < Oo 3 7 so 9 16 56 wo-[S 2 Je ef | 23 a)[f-$ *[-[p » %s and 102 19 -1 6 13]/2 3 0 4 16 56 is -§ -§ 6jjo 0 3)> [12 30 8] 2 1 0. Sine 1 0 223 o 01 afi 2 Ye eff $a} e-[} d 3-1Theorem 1.3 1.4. Algebraic Properties of Matrix Operations 37 1], [3 -1]_pis 0 scvoc=[ {}+[5 G]-[2 3] . Recall Example 9 in Section 1.3, which shows that AB need not always equal BA. This is the first significant difference between multiplication of matrices and multiplication of real numbers} and (verify) Properties of Scalar Multiplication If and s are real numbers and A and B are matrices of the appropriate sizes, then (a) nA) = (rs)A (b) (HAS rAtSA (© (A+B) =rAtrB (@) APB) = r(AB) = (rA)B : 1 A Hl Then Rn 6 of 12 roa=2[% § ]=[2 ts a] =e We also have 6-4 2 423 2-10 16 AB) [3 2 ] 402 -[ ]=280 2-3 af] 2 2)=[0 0 2% . ‘Scalar multiplication can be used to change the size of entries in a matrix to meet prescribed properties. Let 1 ‘Then for k = 4, the largest entry of kA is 1. Also if the entries of A represent the volume of products in gallons, for k = 4A gives the volume in quarts . ‘So far we have seen that multiplication and addition of matrices have much in common with multiplication and addition of real numbers. We now look at some ‘properties of the transpose,38 Chapter 1 Linear Equations and Matrices Theorem 1.4 Properties of ‘Iranspose Ifr isa scalar and A and B are matrices of the appropriate sizes, then @) (ANT =A (b) (A+B)! = AT + BP (© (AB)? = BTAT (@ (Ay =r aT Proof ‘We leave the proofs of (a), (b), and (A) as Bxercises 26 snL27.) (Let A = [ai] and B = [by]; let AB = C = [ey ]. We must prove that c? is the (i,j) entry in BTA. Now 7 pt aye bu = Ya Oh ft Tao, eh =e = Vbkal, =the Gj) entry in BTAT. i Let Bee 4 ot 2] 23 oi] a=([3 2 ‘Then 2 303 0 and B' =| 1 2 1 2-1, Also, “1 20. bo ow Eee ‘ Then1.4. Algebraic Properties of Matrix Operations 3 (On the other hand, 2 2 7) cape 2 T]euar. . ‘We also pote two other peculiar of mauix multiplication lita and bare real numbers, then ab = 0 can hold only if a or-b-is-zeras However, this is not us for maces a [J [3 4} If a, b, and ¢ are real numbers for which ab = ac and a 4 0, it follows that b = c. That is, we can cancel out the nonzero factor a. However, the ces, as the following example shows. then but BAC. ‘We summarize some of the differences between matrix multiplication and the multiplication of real numbers as follows: For matrices A, B, and C of the appropriate sizes, 1. AB need not equal BA 2. AB may be the zero matrix with A # O and B ZO. ~ 3. AB may equal AC with BA C. In this section we have developed a number of properties about matrices and their transposes. If a future problem involves these concepts, refer (0 these properties to help solve the problem. These results can be used to develop many more results40 Chapter 1_Linear Equations and Matrices Key Terms Properties of matrix addition Zero matrix Properties of matrix multiplication (RR exercises 1, Prove Theorem 1.1(b). 2, Prove Theorem 1.1(@). 3. Verify Theorom 1.2(a) forthe following matrices anf i} oo 3 i} o ef? ] 4, Prove Theorem 1.2(b) and (c). 5, Verity Theorem 1.2(¢) for the following matrices: ees a ee we-[s 4] 6 Let A = [ay] be the m x n matrix defined by ay = & and ay) = Gift j. Show that if B is any n> m matex, then AB = kB. 7. Let A be an m x n matrix and C = [er ex] a1 x memati, Prove that where A, is the jth row of A, cos sind sind cos 8 Leta (a) Determine a simple expression for A2. (b) Determire a simple expression for A? (©) Conjectute the form of a simple expression for A, a positive integer, (a) Prove or disprove your conjecture in pat (c). 9, Finda pair of unequal 2 x2 matrices A and 8, other than ‘those given in Example 9, such that AB = O. 10, Find two different 2x 2 matrices A such that 2 [lo ea[o i Properties of scalar multiplication Properties of transpose 11, Find two uncqual 2 % 2 matrices A and B sueh that "3 an-[! oh ind two different 2 x 2 matrices A such that A? = 0. 13, Prove Theorem 1.3(a). 14, Prove Theorem 1.300) 15, Verify Theorem 1.3(b) for r = 4, 8 (i a) 46, Prove Theorem 1.3(0) 47, Verily Theorem 1.3(6) for 42) 02 A={1 3], and B=] 43 32 21 Prove Theorem 1.3(0). \Nerify Theorem 1.3(4) for the following matrices: af) sf) ‘and 2, and A = 20, ‘Te mutts 4 contains the weight (n pounds) of abjcis packed on board spacecraft on earth. ‘The objess are to be wed on the moon whore things weigh about as ruc, Write an expression that calculates the weg ofthe objects onthe moon 21, (a) A is-a 360 x 2 matrix, The first column of is 080" cos 1", 608359 andthe second column is sinO® sin.» sn389°, The graph ofthe o- dered pais in A isa circle of radios | centered at the origin, Write an expression A for ordered pairs ‘whose praph isa cicle of radius 3 centered at the ogin () Explain how to prove the claims about the cirtes in pat, 22, Determine a salary such that Ax = rx, where oft] mf]a 14 12-4 boo 1] and a 44s Prove that if Ax = rx form x n matrix A,n x Imatix 38, S.and scalar r, then dy = ry, where y = sx for any Seale Determine a scalars such that A®x = sx when Ax = rx. WB 39, Prove Theorem 1 4a) Prove Theorem 1.4(8) and (). Verify Theorem 1.4(a, (band (@) for 132 42-1 ef tah [2 75] and rand B40. Verify Theorem L4(e) for 132 =f 1] “Uf (a) Compute (AB*)C. {) Compute BC and multiply the result by A on the sight. Hint: BYC is 1x D. (0) Exphin shy (ABC = (BCVA. Determine « constant k such that (kA) (eA) |. where =I he use? Find theee 2 x 2 matrices, A, B, and C swe that AB AC with B# Cand A #0. Let be an x matrix and ca real number. Show that cA = O,thene=00rA = 0. Determine all 2 x 2 matrices A such that AB = BA for any 2 x 2 matrix B Show that (A — B)? = AT ~ 87, Lot x; and x: be solutions tothe homogeneous linear sys- em AX = 0, Ja) Show that xy ++ x2 isa solution * Zan. 4b) Show that xj ~ x: isa solution, Algebraic Properties of Matrix Operations 41 (©) Forany scalar r, show that rxy isa solution, (@ For any scalars r and s, show that rx; + sx is @ solution Show that if Ax = b has more than one solution, then ‘thas infinitely many solutions, (Hine: fx and x: are solutions, consider xy = rx) + 3%, where rs = 1.) ‘Show that if x; and x3 are solutions to the linear system ‘AX = b, then x; — xis a solution to the associated ho- rmogencous system AX = 0. Let 6-1 1 10s A=|0 13-16 and x=] 210 o 8H 105 (a) Determine a sealar r such that A (b) Is it te that ATx = rx for the value r determined in pat (a)? Repeat Exercise 39 with 3.60 2.03 6.20 3.85) 335 120 360 2.40 125 and x=] 125 625 O10. 001 0.0001 isplay format to show as many decimal places as possi- bie, then compute 3.00 Lata Jt your soar, the B= 104A C=AFALALAFAFAFAFAEAT and D=B-C. adrian cea hs Seana o oem, ot caer aaa ht ons sna ek42 Chopter 1 Be Linear Equations and Matrices Special Types of Matrices and Partitioned Matrices ‘We have already introduced one special type of matrix O, the matrix all of whose. entries are zero, We now consider several other types of matrices whose structures are rather specialized and for which it will be convenient to have special names. Ann x nmatrix A = [qi] is called a diagonal matri fori 4 ‘Thus, for a diagonal matrix, the terms off the main diagonal are all zero. ‘Note that isa diagonal matrix. A sealar matrix is a diagonal matrix whose diagonal elements are equal. The scalar maisix Jy = [dy], where di; = 1 and diy = 0 for 1 # j, iscalled the n x n identity matris. 100 01 of. oot 10 0 2 o do], B=lo0 03 0 Then A, B, and J; are diagonal matrices; B and J; are scalar matrices; and J3 is the 3 x 3 identity mates, . Let Is easy to show (Exercise 1) that if A is any m x n matrix, then . Alp=A and [pA Also, if A isa scalar matrix, then A= rf, for some scalar r Suppose that 4 isa square.maifix, We now define the powers of a matrix, for a positive integer, by APS AAU OA for If A isn xn, we also define a Ash For nonnegative integers p and g, the familiar laws of exponents for the real num. bers can also be proved for matrix multiplication of a square matrix A (Exercise 8) APAT= APS and (AP) = AM, It should also be noted that the rule * (ABy? = APB? does not hold for square matrices unless A, BA (Exercise 9).DEFINITION 1.8 DEFINITION 1.9 Cares > N one 1.5. Special Types of Matrices and Partitioned Matrices 43 peciall Typ ‘Ann xn matrix A = [ay] is called upper triangular if a = fori > j It is called lower triangular if fori < j/A diagonal matrix is both upper ‘wiangular and lower triangular. ‘The matrix is upper triangular, and is lower triangular, . A matrix A with real ent 4c ¥ A matrix A with real entries is called skew symmet 123 A=|2 4 5 | isa symmetric matrix. . 35 6 0.2 3 B=|-2 0 ~4 | isa skew symmetric matrix, . aro) ‘We can make a few observations about symmetric and skew symmetric matrices; the proofs of most of these statements will be left as exercises, It follows from the preceding definitions that if A is symmetric or skew sym- ‘metric, then A is a square matrix If A is a symmetric matrix, then the entries of A are symmetric with respect to the main diagonal of AJ Also, A is symmetric if and only iff, = af and A is skew symmetric if and only ifa;j = —a Moreover, if ‘Ais skew symmetzic, then the entries on the main diagonal of A areal zero. An important property of symmetric and skew symmetric matrices is the following: If Aiisan n xn matrix, then we can show that A = $+ K, where S is symmetric and K is skew symmetric. Moreover, this decomposition is unique (Exercise 29). @ Partitioned Matrices If we start out with an m xn matsix A = [a;] and then cross out some, but not all, ofits tows of columns, we obtain a submatris of A. 4 5 -3 Let44° Chapter 1 Linear Equations and Matrices If we cross out the second row and third column, we get the submatrix 12 4 3 0-3) . ‘A matrix can be partitioned into submatrices by drawing horizontal fines between rows and vertical lines between columns. Of course, the partitioning can be carried out in many different ways. Eze | Te matrix an a2 asia as am aide as} TA Au Aa Ax @ ay an lan ws ay) day ag au as ccan be partitioned as indicated previously. We could also write ay) ay which gives another partitioning of A, We thus speak of partitioned matrices. i ESPV TEBE The augmented matrix (defined in Section 1.3) of a linear system is a partitioned matrix. ‘Thus, if Ax = b, we can write the augmented matrix of this system as [afb] : If A and B are both m x » matrices that are'partitioned in the same way; then A+ B is produced simply by adding the corresponding submatrices of A and B. Similarly if A is a partitioned matrix, then the scalar multiple cA is obtained by forming the scalar multiple of each submatrix. If A is partitioned as shown in (1) and bu AB= a (Aa Bu + ABo + AosBu) | (Ao.Bi2+ AmB + AB)1.5. Special Types of Matrices and Partitioned Matrices 48 Ecce and let Then AB=C 0 oO 0 7 where Cj, should be Aj, Bi, + Ar. Bai. We verify that Cy, is this expression as follows a anboraven=[5 5 9 Se fs ays 3 2 0 o},fi 3 0 =lo 2 2})*[6 0 -2 303 «0 [ea a)- ‘This method of muliplying parttoned matrices is also known asBlock mul- tiplieaton; Partitioned matices ean be used to pret advantage when matrices exceed the memory capacity ofa computer. Ths in mliplying two paitoned matrices, one can Keep the matrices on disk and bring into memory ony the submatrices required to form the subyatrx product. The products, ofcourse, can be downloaded 1s they are formed. he pationing must be done in sucha way that the products of coresponditg submatrices are define, Partitioning ofa matrix implies a subdivision ofthe information into blocks, or units The reverse process i o consider individual matrices as blocks and adjoin them to form a pattioned matrag/The only requirement is that afer the Blocks have been joined, all rows have the same number of entries and all columns have the same number of enifies.s 1 Es [1-1 0], ana D [e s 31 \ \ns we46 Chapter 1 Linear Equations and Matrices DEFINITION 1.10 Theorem 1.5 ‘Then we have woes 4} (2)-[23- 3} [fl el-[ Adjoining matrix blocks to expand information structures is done regularly in 8 variety of applications. I is common for a business to keep monthly sales data fora yearin a1 x 12 matrix and then adjoin such matrices o build a sales history matrix for a period of years. Similaely, results of new laboratory experiments are adjoined to existing data to update a database in a research facility ‘We have already noted in Example 7 that the augmented matrix ofthe linear system Ax = b is a partitioned matrix. Attimes we shall need to solve several linear systems in which the coefficient mautix A is the same, butte right sides of the systems are diferent, say, b, and d, In these cases we shal find i to consider the partitioned matrix [4 |b! ¢ !d]. (See Section 4.8.) and. MH Nonsingular Matrices ‘We now come to a special type of square matrix and formulate the notion corresponding to the reciprocal of a nonzero real number Ann x nm matrix A is called nonsingular, or invertible, if there exists ann x ni ‘matrix B such that AB = BA = Ip; such a Bis called an inverse of A. Otherwise, A is called singular, or noninvertible, Remark In Theorem 2,11, Section 2.3, we show that if AB = [,, then BA = ly ‘Thus, to verify that B is an inverse of A, we need verify only that AB = [y afoot [7] J scat a seco Pro Let B and C be inverses of A. Then AB=BA and AC=CA=1, We then have B = Bl, = B(AC) = (BA)C = I,C = C, which proves that the inverse of a matrix, if it exists, is unique. 71.5. Special Types of Matrices and Partitioned Matrices. 47 Because of this uniqueness, we write the inverse of a nonsingular matrix A as AW. Thus AA! =ATA= hy [s 3} ab cd ‘Then we must have wef IE d-o-[ 9) at+2e b+2d]_[1 0 Ba+de ab+4d}~ LO 1 Equating corresponding entries ofthese two matrices, we obtain the linear systems a+2=1 1 be ada 3a +4c=0 3b+4d=1 Let IFAM exists, let re so that The solutions are (verify) @ = —2,¢ 5. Moreover, since the matrix b= 1, and d Let If A exists, let ‘Then we must have web fe d-e-[ t)48 Chapter 1 Linear Equations and Matrices Theorem 1.6 Corollary 1.1 Theorem 1.7 Theorem 1.8 so that a+ b+2d]_f1 0 2a+4e 2b+4d]= lo 1 Equating corresponding entries of these two matrices, we obtain the linear systems at2 2a +4e b+2d and 0 2b + 4d ‘These linear systems have no solutions, so our assumption that A~! exists is in correct. Thus A is singular. . We next establish several properties of inverses of matrices. If A and B are both nonsingular n x n matrices, then AB is nonsingular and (By = Bat Proof We have (AB)(B~'A~!) = A(BB™!)A~! = (Al,)A~! = AA! = [,. Similarly, (B-!A-!)(AB) = J. Therefore AB is nonsingular. Since the inverse of a matrix is unique, we conclude that (AB)-! = B14! . Tf Ay, Aa... Ay are nx n nonsingular matrices, then AyA2--~ A, is nonsingular and (AyAg + Ag"! = AvlAg Ay! Proof Exercise 44 . If Ais a nonsingular matrix, then A~" is nonsingular and (A~!)~! = A Proof Exercise 45, . If A is nonsingular matrix, then AT is nonsingular and (A~ Proof ‘We have AA~! = Jy. Taking transposes of both sides, we get (YAP = IT = iy Taking transposes of both sides of the equation A~!A = [,, we find, similarly, that (aya), Wh caty . a-[} 3 ‘These equations imply that (A Ir1.5. Special Types of Matrices and Partitioned Matrices 49 ant wy e[? i] : 2 wf] ry eae . ‘Suppose that A is nonsingular. Then AB = AC implies that B = C (Bxer cise 50), and AB = O implies that B = O (Exercise 51). It follows from Theorem 1.8 that if A is a symmetric nonsingular matrix, then A“ is symmetric, (See Exercise 54.) then from Example 11 Also (verify), WE Lincer Systems and Inverses fA isan nxn mattis, then the linear system Ax = b isa system of n equations in ‘unknowns. Suppose that 4 is nonsingular. Then A~' exists, and we can multiply Ax = bby A~ on the left on both sides, yielding A (Ax) = Ab (tax = Ay x= A x=atb, @ Moreover, x = A~!b is clearly a solution to the given linear system. Thus, if A is nonsingular, we have a unique solution, We restate this result for emphasis: If A is ann x n matrix, then the linear system Ax = b has the unique solution Moreover, if b = 0, then the unique solution to the homogeneous system Ax =isx=0. If A is a nonsingular n x m matrix, Equation (2) implies that if the linear system Ax = b needs to be solved repeatedly for different b’s, we need compute A“ only once; then whenever we change b, we find the corresponding solution x by forming A~'b, Although this is certainly a valid approach, its value is of a ‘more theoretical rather than practical nature, since a more efficient procedure for solving such problems is presented in Section 2.5. ‘Suppose that A is the matrix of Example 11 so that Ir50 Chapter 1 Linear Equations and Matrices then the solution to the linear system Ax = bis eevef? ]f}-[-4) On the other hand, if _ fio = Loo}: [=l-[:] ' @ Application A: Recursion Relation; the Fibonacci Sequence In 1202, Leonardo of Pisa, also called Fibonacci,” wrote a book on mathematics in which he posed the following problem: A pair of newborn rabbits begins t breed at the age of 1 month, and thereafter produces one pair of offspring pe! ‘month. Suppose that we start with a pair of newly bom rabbits and that none o: the rabbits produced from this pair dies. How many pairs of rabbits will there bs at the beginning of each month? ‘At the beginning of month 0, we have the newly bor pair of rabbits Py. Att beginning of month 1 we still have only the original pair of rabbits P,, which have rot yet produced any offspring. At the beginning of month 2 we have the origina pair P; and its first pair of offspring, P:. At the beginning of month 3 we have the original pair P,, its first pair of offspring P: bor at the beginning of month 2, anc its second pair of offspring, Ps. At the beginning of month 4 we have Pi, P2, anc Py; Pa, the offspring of Pj; and Ps, the offspring of P2, Let up denote the numbe of paits of rabbits atthe beginning of month n. We see that then uy 3 ow ‘The sequence expands rapidly, and we get 1,1,2,3,5,8, 13, 21, 34, 55, 89, 144, “To obtain a formula for u,, we proceed as follows. The number of pairs 0 rabbits that are alive at the beginning of month n is uy—1, the number of pair ‘who were alive the previous month, plus the number of pairs newly bom at th beginning of month n. The latter number is u,_2, since a pair of rabbits produce a pair of offspring, starting with its second month of life. ‘Thus Uy ty Mya. GB “Tazonardo Fibonacci of Pisa (about 1170-1250) was bom and lived most of hs life ip Pisa, Ital: ‘When he was about 20, bis father was appointed director af Pisen commercial inerests in noes Aliica now a part of Algeria Leonardo aocompanied his ater to Africa and for several years travele extensively throughout the Mediterranean aea on behalf of his father. During sexe travels he lame the Hinds Arabic method of numeration snd ealeulstion and decied to promote is ws i aly. Th ‘was one purpose of his most famous book, Liber Abact, which appeared in 1202 and contained th ‘bie problem stated hee1.5. Special Types of Matrices and Partitioned Matrices SI That is, each number is the sum of its wo predecessors. The resulting sequence of numbers, called a Fibonacei sequence, occurs in a remarkable variety of applications, such as the distribution of leaves on certain trees, the arrangements of seeds on sunflowers, search techniques in numerical analysis, the generation of random numbers in statistics, and others ‘To compute 1 by the recursion relation (or difference equation) (3), we have to compute Wo, i.» +» p21. This can be rather tedious for large n. We now develop a formula that will enable us to calculate w, directly. In addition to Equation (3), we write Hyad = Minos so we now have ‘which can be written in matrix form as fe}-[ olf] « w=["] and A [i a sksn-1) EU ~-E)-E) ‘Then (4) can be written as Woot = AWy-2. Thus wy = Awe Wp = Aw) = A(Awo) = A?wo Ws = Av, = A(APW0) = ABW Woo = AM Wo, Hence, to find 1,, we merely have to calculate A"-!, which is still rather tedious if ‘is large. In Chapter 7 we develop a more eflicient way to compute the Fibonacci ‘numbers that involves powers of a diagonal matrix. (See the discussion exer in Chapter 7.)52 Chapter 1 Linear Equations and Matrices Key Terms Diagonal matrix Teeny attix Powers oft matix Upper wiangular matrix Lower triangular max ‘Symmettic matrix ‘Skew symmetric matrix ‘Submatrix Panitioning Partitioned matrix (EEE Exercises _ 1. (a) Show that if Ais ay m x n max, hen Lad = A and Al, = A (b) Show that if A is an n xm scalar matin, then ‘Arla for some el uber? 2. Prove that the sum, product, and scalar multiple of diag- ‘onal, scalar, and upper (lower) triangular matrices is di- ‘agonal, scalar, and upper (lower) triangular, respectively 3 Prove: If A and B are n x m diagonal matrices, then AB=BA. 302-1 6 32 a=|o -4 3] aa s=|o 2 4 9 0-0 0 0 Verify that A + B and AB are upper triangular. 8. Describe all matrices that are both upper and lower trian- ular éere[} 2]aar=[) conten 3-2 203 of the following @ a © 10 niaa=|2 1 31 ‘Compute each of the following: @ a () B 8. Let pandg be nonnegative integers and let A be asquare ‘matrix. Show chat fe) (AB) Anat AP and (AM) = AMP BA and p is a nonnegative integer, show that ABE, 10. If p is @ nonnogative integer and ¢ isa scalar, show that (At = cra. 11, Fora square matrix A and a nonnegative integer p, show that (AT)? = (4PY7 2 13. MM 15, 16. vw. 18. 19. a. 2. 23, Show that if A is a symmettic matrix, then A‘, & Nonsingular invertible) matrix Inverse Singular (ooninvertible) matrix Properties of nonsingular matrices Linear system with nonsingular coefficient matrix Fibonacei sequence For nonsingular matrix A and a nonnegative integer p, show that (4)! = (A)? For a nonsingular matix A and nonzero sealarf, show that (kay = A? {a) Show that every scalar matrix is symmetric (b) Is every scalar matix nonsingular? Explain (©) Isevery diagonal matrix a scalar matrix? Explain Find a 2% 2 matrix B_# O and B # Ip such that 1 B= tase [! ces B are there? Find a 2% 2 matrix 8 # O and B # fy such that AB = BA, where A 2) sms ts UF wm tm es B are there? Prove or disprove: For any nxn matrix A, ATA (a) Show that 4 is symmettic if and only if a forall i,j (b). Show that A is skew symmetric if and only if ay = ay forall, j (©). Show that if 4 is skew symmetric, then the elements, ‘on the main diagonal of A are all zero. ‘Show that if is a symmetric matrix, then AT is sym- Deseribe all skew symmetric scalar matrices. Show that if is any wr > matrix, then AAT and ATA are symmetric. Show hati A is any >< matrix, then (a) A+ AT is symmetic (bh) A— AT is skew symmetie, see is symmetsic, 1. Let A and B be syrumetric matrices. (a). Show that A + B is symmetric. (by Show that AB is syrumetric if and only if AB = BA.1.5. Special Types of Matrices ond Partitioned Matrices 53 2a) Show that if A is an upper triangular matrix, then "is lower triangular. 'b) Show thatif 4 isa lower wiangular matix, then A” is upper wiangular “isa skow symmetric matrix, what type of matin is 2° Justify your answer. ‘Sow that if is skew symmetric, then the elements on ain diagonal of A areal zero. Show that if Ais skew symmeti, then Ab is skew sym- src for any postive odd integer k ow that Ais an mn matrix, then A = $+K, where S'ssymmettic and K is skew symmetric, Also show that sis decomposition is unique. (Hint: Use Exercise 22.) 13-2 A=|4 6 2 S13 1 the matrices § and K described in Exercise 29. 2 Sorters mica [2 2] 40000 Eep=|0 -2 0),tintD-! 0 0 3 ind the inverse of each ofthe following matrices: wa[h 3] weft 3] 2 361s 4s ma whe meses 1 icensgstr mate whe nee [2 Sng A. KE nd (AB)! Fe. Suppose that Solve the lineue system Ax ‘mattices b: ff] [3] F. The linear system ACK nonsingular with vf 20 eelta D foreach ofthe following Dis such that A and C are 40, 1. a. 49. 50, . The linear system AT Find the solution x. ‘The linear system A” eB Yat ef] Find the solution x. ‘bis such that 4 is nonsingular a ‘bis such that 4 is nonsingular with 4 1 ate[f g] a 9=[ 3] Find the solution x “The liner system C7 Ax = Bis such that A and C are onsingul with 0 2) at & Find the solution x Consider the linear system Ax = b, where A is the matrix defined in Exercise 33(. , 6 fins sosin ith [1] : 0 tints t= [5 Find two 2 > 2 singular matrices whose sum is nonsingular Find two 2 x 2 nonsingular matrices whose sum is sin- sla Prove Corollary 1. Prove Theorem 17. Prove that if one row (column) of them x matrix A consists entirely of zeros, then A is singular. (Hint: Assume that Ais nonsingular: that i, there exists un m > n matrix B such that AB = BA = [,, Establish a contradiction.) Prove: If A is a diagonal matrix with nonzero diagonal entries @y1,.:2,--- ua, then A is nonsingular and A~' is a diagonal mateix with diagonal entties Yay Wags Wg 200 0 Let A=]0 -3 0 |. Compute a4 o 0 5 For an n x 1 diagonal matrix A whose diagonal entries Are dy, 22, --- dq, Compute AP for a nonnegative inte- er p. Show that if AB = AC and A is nonsingular, then B=C.51. 33. 55. 56. 57, Chapter 1 Linear Equations and Matrices Show thai is nonsingular and AB = O foran nxn matrix B, then B= 0. ab a only fad = be £0, Consider the homogeneous system AX =O, where Ais fen. If Ais nonsingula, show thatthe only soltion is the trivial one, x = 0 Prove that if A is symmetric and nonsingular, then AW! issymmetic Formulate the method for adding partitioned maces, and verify your method by patiioning the maioes 13-1 1 A=|2 1 0} and B 1 2-3 4 5 intwo different ways and finding their sum, Let A ‘Show that A is nonsingular if and Let A and B be the following matrices: 2 1 3 4 2 12 3-1 4 23 2 1 4 5-1 3 2 6 3 1 2 4 6 2-1 2 5 7. and 1203 4 1 2 13 2-1 1 os 4 203 2 13 5 7 302 4 6 1 Find AB by panitioning A and B in two different ways, ‘What type of matrix isa finear combination of symmetric ‘matrices? Justify your answer. ‘What type of matris is linear combination of scalar mae trices? Justify your answer “The matrix form of the recursion relation Ou gt — Gi is written as 1, ty where #62. a. 6. (a) Using ve= [it] = [0]: compat a Thon mk of Be wameehi nets (b) Express wy asa matrix times wo. ‘The matrix form ofthe recursion relation “ [in moe] mt a(S] a) op ~-[iJ-[i): compute mw Wy an We Thon make 9st of thtems of emcee ton aye » For he softwar you a sing. determine the com> and) pss equi edo cach te follow ie (Alin arowo column om exiting ma Express wor a8 matrix times wo (b) Construct the partitioned matrix [2 ¢] from existing matrices A and B, using appropriate size zero matrices, (©) Extract a submatrix from an existing mat ‘Most software for linear algebra has specific commands for extracting the diagonal, upper triangular par, and lower tiangular part of a matrix. Determine the core- sponding commands for the software that you ae using, and experiment with them, Determine the command for computing the inverse of a matrix in the software you use, Usually, if such a command is applied to a singular matrix, a warning message is displayed. Experiment with your inverse command to determine which of the following matrices are singular:1.6 Matrix Transformations 55 123 123 bag 2 w@ [456 wlase ts 780 789 @ amlo yg )p=|2 124 005 @ yr rel tb) A= [a], where 2-103 fer) Y sg, 16 B isthe inverse of nxn matrix A, then Definition 1.10 1 guarantees that AB = BA = I,. The unstated assump- a tion is that exact arithmetic is used. 1 computer arith= ‘metic is used to compute AB, then AB need not equal, $s 2sc00, 1Opand b ‘and, in fact, BA need not equal AB. However, both ABM. 66. For the software you ate using, determine the command and BA should be close t Is, In your software, use the for obiaining the powers A?, A... of a square matrix inverse command (see Exercise 63) and form the prox- A. Then, for ucts AB and BA for each of the following matrices: he fist column of io o 1 0 0 1s [ia 0 0 1 0 0 weft) oll | iis . ‘ o 9 0 0 1 144 o 0 0 0 0. @aslp us 3 ‘compute the matrix sequence A‘, k = 2, 3, 4, 5,6. De- pad serbe the Behavior of dt ask = 20 In Secon 1.1 we studied the method of elimination 67. Experiment with your softwar to determine the behav for solving linear systems AX = b. In Equation (2) ior of the matrix sequence A* as k —+ 00 for each of the of this section we showed thatthe solution i given by folowing matioes 1-1 0 command for automatically solving linear systems, and oie its inverse command, compare these two solution tech- ood niques on each of the following linear systems: Ha Matrix Transformations In Section 1.2 we introduced the notation ” for the set of all n-vectors with real entries. ‘Thus R® denotes the set of all 2-vectors and R* denotes the set of all 3-vectors. It is convenient to represent the clements of R? and R* geometcically a directed line segments in a rectangular coordinate system.* Our approach in this section is intuitive and will enable us to present some interesting geometric applications in the next section (at this early stage of the course). We return in Section 4.1 to a careful and precise study of 2- and 3-vectors. ‘The vector » in R? is represented by the directed line segment shown in Figure 1.6. The vector . [ | 7 You have undoubtedly sen rectangular coordinate systems in your precalculus oealnlus courses. ‘Ab, if A is nonsingular, Using your software's (a) A= [ a €56 Chapter 1 Linear Equations and Matrices FIGURE 1. Ca pais z wind FIGURE 1.7 in R® is represented by the directed line segment shown in Figure 1.7. Figure 1.8 shows geometric representations of the 2-vectors o-E} efi} El in a 2-dimensional rectangular coordinate system, Figure 1.9 shows geometric representations of the 3-vectors fj Eh-=8 in a 3-dimensional rectangular coordinate system. . FIGURE 1. FIGURE 1.91.6 Matrix Transformations 57 Functions occur in almost every application of mathematics. In this section wwe give a brief introduction from 4 geometric point of view to certain functions ‘mapping R" into R”. Since we wish to picture these functions, called matrix transformations, we limit most of our discussion in this section to the situation \here m and 1 have the values 2 or 3. In the next section we give an application of these functions to computer graphics in the plane, that is, for m and m equal to 2. In Chapter 6 we consider in greater detail a more genersl function, called a linear ‘tansformation mapping R” into R™. Since every matrix transformation is a linear transformation, we then learn more about the properties of matrix transformations. Linear transformations play an important role in many areas of mathematics as well as in numerous applied problems in the physical sciences, the social sci ences, and economics. If A is an m xn matrix and w is an n-vector, then the matrix product Au is fan m-vector. A function f mapping X” into R” is denoted by f: R° —» R™ A ‘matrix transformation is a function f: R" —> R” defined by f(w) = Au. The vector f(u) in R is called the image of u, and the set of all images of the vectors in R® is called the range of f. Although we ate limiting ourselves in this section to mattices and vectors with only real entries, an entirely similar discussion can be developed for matrices and vectors with complex entries. (See Appendix B.2.) STE (a) Let f be the matrix transformation defined by so=[ ile rieimgeare= [2] asso imesof[!] [9] ww tss=[! 2 “Jandcoiieremsictantomsin ess f(u) = Au. 1 ° “Then the image of | 0 is [} sie 1] is [3] sna ce imag ij P 3} 2 - *] [2] osm . : Observe that if A is an m xn matrix and f2 R" > R" is a matrix transformation mapping R” into R” that is defined by f(u) = Au, then a vector w in R” is in the range of ff only if we can find a vector v in R" such that f(¥) = w. “Appendix A, dealing with sets and functions, may be consulted! ws needed.58 Chapter 1 Linear Equations and Matrices fy») 7) ”» FIGURE 1.10 Reilection With respect to the x-axis, Ba Leta [ } 3 ]antconsider tne marx trnsfomaton ined by f(a) = Au Determine ite vector w = [isin the rnge of / Solution The questions equivalent asking whether theresa vector v=] uch that J (¥) = w. We have rt, ‘ 1 or udim= 4 20; +302 ==1 Solving this linear system of equations by the familiar method of elimination, we get uy = 2 and vp = 1 (verify). Thus w is in the range of f. In particular, i v= [Jean fo) =w. ' For matrix transformations where m and n are 2 or 3, we ean draw pictures showing the effect of the matrix transformation. This will be illustrated in the examples that follow, Let f + R® be the matrix transformation defined by ro=[! Je Thus, fu = [i} then vw-s(())- ‘The effect of the matrix transformation f, called reflection with respect to the avaxis in R?, is shown in Figure 1,10. In Exercise 2 we consider reflection with respect to the y-axis . Let f: R° — R® be the matrix transformation defined by o-a(()-C 3] ‘ThenFIGURE 1.11 Prajection FIGURE 1.12 1.6 Matrix Transformations 59 Figure 1.11 shows the effect of this matrix transformation, which is called projee- tion into the xy-plane. (Warning: Carefully note the axes in Figure 1.11.) Observe that if where s is any scalar, then fo= 3] Fea) Hence, infinitely many 3-vectors have the same image vector. See Figure 1.12, 4g: RY RY defined by 0. 1 ean) pps bebe ado suy wos) t : Let f: R® + R® be the matrix transformation defined by r 00 fuy=]0 r Olu oor where r is @ real number. It is easily seen that f(u) = ru. fr > 1, f is called dilation; if 0 R° by gin) = [5 aD We also have g(u) =u, so again if'r > 1, g is called dilation; if0 1 () Contaction:0- R? defined by i(l v)- [es =sing [x calls] ° ‘Then (5) can be written, using (3), as _ [cose — ysing] _[2" soon ees eee] [9 It then follows that the vector /'(u) is represented by the directed line segment from O to the point P’. Thus, rotation counterclockwise through an angle @ is @ . ‘matrix transformation. Key Terms ‘Matrix transformation Image “Mapping function) Reflection Range Projection ED brercises [In Exercises I through 8, sketch w and its image under each given matri transformation f. 2 3 1. fs RY > RP defined by 2. fs RY > R® (reflection with respect to the y-asis) de- (G)-f 1G) “-CoGh L fl Dilation Contraction Rotation Bf: R= RP defined by Q)-ft aIEh = E] 6. fs RY RP defined by CD-b ach E)os RY RD defined by CD-[; a F] “OES -E ln E-sveses 9 through 14, let F: R® > R be the matrix ro ~rmation defined by F(X) = AX, where 12 A=|o 1 1 ~ine whether each given vector w is in the range of f. yf ff al] el rvises 15 through 17, give a geometric description of wre ~srriv transformation f: R® > R® defined by fw) = Sa“ each given matrix A confi met a [i a [i ee 38. Some matrix transformations # have the property that “uy = fev), when w 2 ¥. Tha is, the smages of “ferent vectors can be the sume. For each of the fol- sing matet transformations f: R? — R? defined by ind two diferent vectors wand sch that = Ftv) = wor the given vector w [0 i t}e=[-] . A r A a 1.7 Computer Graphics (Optional) 63 wofe ttf) Let fs R? > R* be the linear transformation defined by ‘Au, where a= [8 sing fw) - sn] cos For g = 30°, f defines a counterclockwise rotation by an angle of 30° @ rT >) Ew (©) What is the smallest positive value of & for whieh Tw) =A'u=ul Let f: RY — R® be a matrix transformation defined by {Fu} = Au, where A is an m 2 mses, (a) Show that f(u++ ¥) = f(a) + f(v) for any wand v in R* (b) Show that feu) = ef'(u) for any win" and any real number ¢ fe), Show that f(cw + dy) = ef a) + df») for any w and vin R¥ and any real numbers and d. A°u, desetibe the action of 7; on w. At, describe the action of 73 on w, Let f: RY > R be a matrix transformation defined by (0) = Au, where A isan m % m matrix. Show that ifm ‘and Ware veetors in R® such that f(u) = Oand fv) = 0, where 0. then (cu + dy) = 0 for any real numbers © and d. (a) Lot O: RY > R® be the mattis transformation do- fined by O(u) = Ou, where O is the m xm ze10 matrix. Show that O(w) =O for all win R™ (by Let 1: RY RY be the matrix transformation de fined by 1(u) = Iau where [isthe identity matrix, (See Section 1.5.) Show that /(u) = u forall win Re ERA computer Graphics (Optional) We are all familiar with the astounding results being developed with computer ‘graphics in the areas of video games and special effects in the film industry. Com-64 Chapter 1 Linear Equations and Matrices FIGURE 1.16 puter graphics also play a major role in the manufacturing world, Computer-aided design (CAD) isused to create a computer model ofa product and then, by subject- ing the computer model toa variety of tests (carried out on the computer), changes to the current design can be implemented to obtain an improved design. One of the notable successes of this approach has been in the automobile industry, where the ‘computer model can be viewed from different angles to achieve & most pleasing and popular style and can be tested for strength of components, for roadability, for seating comfort, and for safety ina crash In this section we give illustrations of matrix transformations f: R? > R? that are useful in two-dimensional graphies. Let f: R? > R? be the matrix transformation that performs a reflection with respect to the x-axis. (See Example 4 in Section 1.6.) Then f is defined by eee a=[} 4 onl ELI To illustrate a reflection with respect to the x-axis in computer graphics, let the triangle T in Figure 1.16(a) have vertices Thus, we have (1.4), GD, and 2,6) ‘To reflect T with respect (o the x-axis, we let we) bh oldFIGURE 1.7 Computer Graphics (Optional) 65 w=[o “]E]-L) wel aM[e)=[4] ‘These three products can be written in terms of partitioned matrices as aise oleh 3 2] ‘Thus the image of T has vertices 1-4, GD, and Q, 8) and is displayed in Figure 1.16(b. . ‘The matrix transformation f: R® — R° that performs a reflection with respect to the line y = —x is defined by Jy) = BY, o-[2 4} To illustrate reflection with respect to the line y defined in Example 1 and compute the products a i d-[4 3 3] ‘Thus the image of T has vertices where x, we use the triangle T as Bln mow (4D, (1-3), and (6, -2) and is displayed in Figure 1.17, . To perform a reflection with respect to the x-axis on the triangle T of Example 1, followed by a reflection with respect to the line y = —x, we compute B(Ay;), B(Ay2), and B(AYs), Itis not difficult to show that reversing the order of these matrix transformations produces a different image (verify). ‘Thus the order in which graphics transformations are performed is important. This is not surprising, since matrix multiplication, unlike multiplication of reat numbers, does not satisfy the commutative propery.66 Chapter 1 Linear Equations and Matrices FIGURE 1.18 Rotations in a plane have been defined in Example 8 of Section 1.6. A plane figure is rotated counterclockwise through an angle @ by the matrix transformation f: R? > R? defined by f(v) = Av, where A [ons “ane . sing cos. Now suppose that we wish to rotate the parabola y = x? counterclockwise through 50°, We start by choosing a sample of points from the parabola, say, G [Sce Figure 1.18(a).] We then compute the images of these points. Thus, leting i eee ‘we compute the products (to four decimal places) (verify) 4.3498 14088 0 0.1299 4.9660 1.0391 -0.1233 00.5437 8.0832 4), (LD, 0, (4). and G,9). Aly v2 vs ve vs -[ The image points (4.3498, 1.0391), (-1.4088, -0.1233), (0,0), (0.1299, 0.5437), and (—4,9660, 8.0832) are plotted, as shown in Figure 1.18(b), and successive points are connected, showing the approximate image of the parabola. . © Rotations are particularly useful in achioving the sophisticated effects seen in areade games and animated computer demonstrations. For example, to show & Wheel spinning, we can rotate the spokes through an angle 6, followed by asecond1.7 Computer Graphics (Optional) 67 rotation through an angle @, and so on, Let the 2-vector w= al represent a spoke of the wheel; let f': R? —> R? be the matrix transformation defined by f() = Av, where a=[San onl) and let g: R? —+ R? be the matrix transformation defined by ¢(v) = By, where - [ne “ane sin@> cos }* 8(f(u)) = g(Au) = B(Au). a a pas ee ro Nea B(Au) = B(ajcol, (A) + a2colz(A)) = a; Bool; (A) + a2Beolz(A), and the final expression is a linear combination of column vectors Beoly(A) and Beol(A), which we can write as the product [Beols(4) aeoicy] |]. From the definition of matrix multiplication, [ Beol\(A) Beol;(A)] = BA, so wwe have B(Au) = (BAYu. ‘Thus, instead of applying the transformations in suecession, f followed by g, we can achieve the same result by forming the matrix product BA and using it to define a matrix transformation on the spokes of the wheel. ‘The matrix product B.A is given by [ae ral be “a siné> cost || sine cos = [ e08820086 — sin sine, —cos 6) sind} ~ sin cos = | sine: cose, + e050; sind; —siné sin@ + cosd sin és Since g(/(w) = BAU, it follows that this matrix transformation performs a rotation of u through the angle @; + 9. Thus we have pa a [oO +62) sin, +62) | sin: +62) cos(@, +6) J” Equating corresponding entries of the two matrix expressions for BA, we have the trigonometric identities for the sine and cosine of the sum of two angles: c0s(0) +3) = cos) cos 4y — sin sin bs sin(®, + #3) = c08 0 sin 2 + sind cos B See Exercises 16 and 17 for related results68 Chapter 1 Linear Equations and Matrices ‘A shear in the x-direction is the matrix transformation defined by Soy [3 ‘] ¥, where & is a scalar. A shear in the x-direction takes the point (x, y) to the point (-+ky, y). Thats, the point (x, ») is moved parallel to the x-axis by the amount ky. Co ler now the rectangle R, shown in Figure 1.19{a), with vertices 0,0), @,2), 4,0), and (4,2). If we apply the shear in the x-direction with k = 2, then the image of R is the parallelogram with vertices ©.0, (4,2), 4,0), and (8,2), shown in Figure 1.19(b). If we apply the shear in the x-direction with k then the image of & is the parallelogram with vertices 0,0), (6,2), 4,0), and (2,2), shown in Figure 1.19(). In Exercise 3 we consider shears in the y-direction, . y y y ‘ 4 Shear 4 2 2 2 of 2 6 Oo a a 6 8 2 2 4 6 o FIGURE 1.19 Other matrix transformations used in two-dimensional computer graphics are considered in the exercises at the end of this section. For a detailed discussion of ‘computer graphics, the reader is referred to the books listed in the Further Readings at the end of this section, In Examples 1 and 2, we applied « matrix transformation toa triangle, a figure that can be specified by its three vertices. In Example 3, the figure transformed ‘was a parabola, which cannot be specified by a finite number of points. In this case we chose a number of points on the parabola t0 approximate its shape and computed the images of these approximating points, which, when joined, gave an approximate shape of the parabola Let f: R? + R? be the matrix transformation called scaling defined by f(v) = A¥, where ho aotFIGURE 1.20 1.7 Computer Graphics (Optional) 69 with hand k both nonzero. Suppose that we now wish to apply this mates ans- formation 10 cece of radios I that is centered atthe origin (he uit etl, Unfortunately, a circle cannot be specified by a finite number of points. However, each pint onthe unit cicle is deseribed by an ordered pir (0s, sin0), where the aie @ takes on al values fom Oto 2 radians. ‘Thus we now represen an tritary pint onthe unit tle bythe ector a = [8 the anit citle that are obtained by applying the matix warsformaton fare given by h 0] [eos] _ [ease] _[x" romau=[5 f][Se]=[eon6]= [0] We recall that a circle of radius 1 centered at the origin is described by the equation Hence the images of ety By Pythagoras’s identity, sin’ @ + cos? @ = 1. Thus, the points (cos 6, sin #) lie on the circumference of the unit circle. We now want to write an equation describing the image of the unt circle, We have It then follows that which is the equation of an ellipse. Thus the image of the unit circle by the matrix transformation / is an ellipse centered at the origin, See Figure 1.20. . Oo Unit eile ipse FURTHER READINGS Cunningham, Steve. Computer Graphics: Programming, Problem Solving, and Visual Communication. New Jersey: Prentice Hall, 2007, Foley, James D., Andries van Dam, Steven K. Feiner, and John F. Hughes. Com- puter Graphics: Principles and Practice in C, 2d ed. Reading, Mass.: Addison Wesley, 1996,70 Chapter 1 Linear Equations and Matrices Rogers, D. R, and J. A. Adams. Mathematical Elements for Computer Graphics, 2d ed. New York: MeGraw-Hill, 1989. Shirley, Peter, Michael Ashikhmin, Michael Gleicher, Stephen Marschner, Erik Reinhard, Kelvin Sung, William ‘Thompson, and Peter Willemsen. Fiundamen- tals of Computer Graphics, 2d ed. Natick, Mass.: A. K, Peters, Ltd,, 2005 Key Terms Computer graphics Rotation Dilation Computer-aided design (CAD) Contraction Image Shear (ETE_Exercises 1. Let f1 R? > R¥ be the matsix wansformation defined by f(@) = Av, where 10 a=fy ih thats, eflston with espetio the y-axis. Find and Sheth ihe imge ofthe rectangle vith vetoes (1). 20,0. 3}and 2.3). 2 Let R be the rectangle with vertes (1,1), (1,4), B.D. zd Gud). Lat fe the sheat inthe a-ietion With i 5, Find and sete he image of 3. A shear in the y-direction is the matrix transformation F: RE R defined by f(¥) = Av, and of] where & is a scalar, Let R be the rectangle defined in Exercise 2 and let be the shear inthe y-direction with = 2, Find and sketch the image of R. 4. The matrix twansformation f: R? — R® defined by F(8) = Av, where ko orl and k is areal number, is called dilation if k > 1 and contraction if 0 < k < 1. Thus, dilation stretches a veo= tor, whereas contraction shrinks it, If is the rectangle defined in Exercise 2, find and sketch the image of R tor @) k=4 wb kaH 5. The matrix wansformation f: R® —> RP defined by f(y) = Ay, where bo ve[h oh A 10. tnd & is a real nomber, is called dilation in the x direction if k > 1 and contraction in the x-direction if 0 = k = L. If Ris the unit square and f is dilation in the x-direction with k = 2, find and sketch the image of R. “The matrix transformation fs R? > R® defined by f(s) = Av, where a) AS [3 Hy ‘and k is a real number, is called dilaton i ‘ireetion if k > 1 and contraction in the y- if0 < k < 1. I Ris the ent square and f isthe conrac: tion in the y-direction with X=}, find and sketch the image of R. . Let T be the tiangle with vertices (5,0), (0,3), ane (Q, —1) Find the coordinates ofthe vertices ofthe image ‘of T under the matrix transformation f defined by 21 roe 3 i) Lec Pe the triangle with vertices (1, 1), (-3, =3). ae 2,1), Find the coordinstes ofthe vertices of the imagi ‘of T under the matrix transformation defined by aos no=[4 Le beth counteract ato th 6017 ibe tangle died in Exerc an sketch ‘image of T under f. Lat fe refs wih espe he sian pecounesnekone ton oan. Sh thar the result of first performing fs and then fis not th same as first performing fj and then performing J,24 ‘wiangle defined in Exercise 8. Describe the image of T ‘under the matrix transformation f': A? — R® defined by P18) = Ay. (See also Exercise 21.) Let f be the matrix transformation defined in Example . Find and sketch the image of the rectangle with vertices 0.0), (1,0), (1, 1),and (0, 1) for = 2 and k = 3. Let 2? > R® be the matrix transformation defined by 79) = Av, where Le A be the singular matrix. [ 2] and tes 7 be the Find and sketch the image of the rectangle defined in Exercise 12 Exercises 14 and 15, let fr, fix fix and [i be the following transformations: Fit counterclockwise rotation through the angle @ {2 reflection with respeet to the x-axis reflection with respect tothe y-axis reflection with respect tothe line y = x Let $ denote the unit square. Determine two distinct ways to use the matrix transfor ‘mations defined on S to obtain the given image. You may pply more than one matrix transformation in succession. 0) y 1.7. Computer Graphics (Optional) 71 Oy y “1 15. Let S denote the triangle shown in the figure. y Determine two distinct ways to use the matrix transfor- ‘mations defined on $ to obtain the given image. You may apply more than one matrix transformation in succession, @ 2 2 (b) y 2 + 1, 4 372 Chapter 1 Linear Equations and Matrices 16, Refer to the discussion following Example 3 to develop the double angle identities for sine and cosine by using the matrix transformation f(F(w)) = A(AU), where [Sea “cea] #=[2] 1. tsa poset sine oh ne case ater ample 3 to develop sine and cosine expressions for the difference of two angles; 8) — @, Exercises 18 through 21 require the use of software that sup- ports computer graphics. 18, Define a triangle T by identifying its vertices and sketch iton paper. (@) Reflect T about the y-axis and record the resulting Figure on paper, as Figure 1. Rotate Figure | counterclockwise through 30° and record the resulting figure on paper, as Figure 2 Reflect T sbout the line y = x, dilate the resulting figure in the x-direction by a factor of 2, and record the new figure on puper, as Figure 3 Repeat the experiment in part (c), but interchange the order of the matrix transformations. Record the resulting figure on paper, as Figure 4 Compare Figures 3 and 4. ‘What does your answer in part (e) imply about the ‘order of the matrix transformations as applied to the wiangle? Consider the triangle T defined in Exercise 18. Record T on paper. (a) Reflect about the x-axis. Predict the result before execution of the command, Call this matrix transformation Ly Reflect the figure obtained in part (a) about the y= axis. Predict the result before execution ofthe com= ‘mand. Call dhs matix transformation L, o © @ ) 0 wv. o (©) Recoed on paper the figure that resulted from parts @) and (b. Examine the relationship between the figure obtained in part (b) and T.. What single matrix wansformation Ls will accomplish the same result?” Write @ formula involving Ly, Ly, and Ly that expresses the relationship you sav in part (). [Experiment with the formula in part (e) on several ‘other figures until you can determine whether this formula is correct, in general. Write a brief summary of your experiments, observations, and conclusions. (ay © o 20. Consider the unit square S and record S on paper. 2B. (a) Reflect § about the x-axis to obtain Figure 1. Now reflect Figure 1 about the y-axis to obtain Figure 2 Finally, reflect Figure 2 about the fine y = —x €© bain Figure 3. Record Figure 3 on paper. Compare $ with Figure 3. Denote the reflection about the x-axis as L, the reflection about the 9 axis as Lo, and the reflection about the Tine y as Ly. What formula does your comparison suggest ‘when L, is followed by 12, and then by L3 on $? I Mj, i = 1, 2, 3, denotes the matrix defining determine the entries of the matrix Ms M;My. Does this result agree with your conclusion in part (6)? o © (a) Experiment with the successive application ofthese three mateix transformations on other figures. . IF your computer graphics software allows you to select any 2 x 2 matrix fo use as a matrix transformation, perform the following experiment: Choose a singular matrix and apply it toa triangle, unit square, rectangle, and pentagon. Write a brief summary of your experiments, ‘observations, and conclusions, indicating the behavior of “singular” matrix transformations. . IF your software includes access to 2 computer algebra system (CAS), use i as follows: Let f(u) = Au be the matrix transformation defined by and let g(¥) = BY be the matix transformation defined by costs sind cn con) (a) Find the symbolic matrix BA. (b) Use CAS commands to simplify BA to obtain the c0s(8) +8) —Sin(6y +6) Sim(@; +62) c0s(Gh +8) IF your software includes aocess to 8 computer algebra system (CAS), use it as follows: Let f(w) = Au be the ‘matrix transformation defined by cos sind sin cos (a) Find the (symbole) matrix that defines the matrix transformation f( (0) (©) Use CAS commands to simply the matrix obtained in part (a) so that you obtain the double angle identities for sine and cosine.rtsix transformation defined aa [ooe sing FIGURE 1.21 Length of. e FIGURE 1.22 Length ofv | 24. Ir your software includes access to a computer algebra system (CAS), use it as follaws: Let f(a) = Au be the 1.8 Correlotion Coefficient (Optional) 73 (a) Find the (symbolic) matrix that defines the matrix transformation (FCF (ftw). by ° () Use CAS commands to simplify the matrix ob- 2] tained in part (a) so that you obtain the identities cos 8 for sin(49) and cos(48). EE] corretation Coefficient (Optional) As we noted in Section 1.2, we can use an n-vector to provide a listing of data. In this section we provide a statistical appl of the dot product to measure the strength of a linear relationship between two data vectors. Before presenting this application, we must note two additional properties that vectors possess: length (also known as magnitude) and direction. These notions will be carefully developed in Chapter 4; in this section we merely give the prop~ ‘without justification. ‘The length of the n-vector denoted as |v is defined as Wl = fot top te- teh tu a If n = 2, the definition given in Equation (1) can be established easily as follows: From Figure 1.21 we see by the Pythagorean theorem that the length of the directed tn) is y/v} + v3. Since this directed oh ‘we agree that |/v|), the length of the line segment from the origin to the point (0, line segment represents the vector veetor v, isthe length of the directed line segment. If = 3, a similar proof can be given by applying the Pythagorean theorem twice in Figure 1.22 lis easiest to determine the direction of an n-veetor by defining the angle between two vectors. In Sections 5.1 and 54, we define the angle @ between the nonzero vectors W and ¥ as the angle determined by the expression wy In those sections we show that -ls <1 uit ivi Hence, this quantity can be viewed as the cosine of an angle 0 < @ <7.4 Chapter 1 Linear Equations and Matrices FIGURE 1.23 We now tum to our statistical application, We compare two data n-vectors x and y by examining the angle @ between the vectors. The closeness of cos to —1 or I measures how near the two vectors are to being parallel, since the angle between parallel vectors is either 0 or = radians. Nearly parallel indicates a strong rolationship between the vectors. The smaller | cos | is, the less likely i is that the vectors are parallel, and hence the weaker the relationship is between the vectors. Table 1.1 contains data about the ten largest U.S. corporations, ranked by mat- ket value for 2004. In addition, we have included the corporate revenue for 2004, All figures are in billions of dollars and have been rounded to the nearest billion TABLE 1.1 Market Value Revenue Corporation (in Sbiltions) (in Sbillions) General Electric Corp, 329 152 Microsoft 287 Ea Pfizer 285 3 sxon Mobile 2m am Citigroup 255 108 ‘Wal-Mart Stores 244 288 Intel 197 34 ‘American International Group 195 99 IBM Comp, m 96 Jobason & Tohnson 161 a Source: Tine Aimanae 2006, Information Please, Peatson Edvcaton, Boston, Mis. 2008; and hpiwore-geohive-comhans ‘To display the data in Table 1.1 graphically, we form ordered pairs, (market value, revenue), for each of the corporations and plot this set of ordered pairs. The display in Figure 1.23 is called a seater plot. This display shows that the data are spread out more vertically than horizontally. Hence there is wider variability in the revenue than in the market value, 300) — 250 200 10 +4 10 6 : she : a | a a1.8 Correlation Coefficient (Optional) 75 If we are interested only in how individual values of market value and revenue go together, then we can rigidly translate (shift) the plot so that the pattern of points does not change. One translation that is commonly used is to move the center of the plot to the origin. (If we think of the dots representing the ordered pairs as ‘weights, we see that this amounts to shifting the center of mass to the origin.) To perform this translation, we compute the mean of the market value observations and subtract it from each market value; similarly, we compute the mean of the revenue observations and subtract it from each revenue value. We have (rounded toa whole number) Mean of market values 240, mean of revenues Subiracting the mean from each observation i called eentering the data, and the ‘corresponding centered data are displayed in Table 1.2. The corresponding scatter plot of the centered data is shown in Figure 1.24 TABLE 1 Q Centered Market Value Centered Revenue {in Sillons) (in S billions) 89 33 2m T T 4 -2 10h pe 45 66 took a sz 15 - sor : 4 169 ° 3 -85 ot . 45 20 or 2 ee ee es 3000 =» —n FIGURE 1.24 Note that the arrangement of dots in Figures 1.23 and 1.24 is the same; the scales of the respective axes have changed, A scatter plot places emphasis on the observed data, not on the variables involved as general entities. What we want is a new way to plot the information ‘that focuses on the variables. Here the variables involved are market value and revenue, so we want one axis for each corporation. This leads to a plot with ten ‘axes, which we are unable to draw on paper. However, we visualize this situation by considering 10-vectors, that is, vectors with ten components, one for each corporation. Thus we define a vector v as the vector of centered market values and a.76 Chapter 1 Linear Equations and Matrices aa FIGURE 1.25 vector w as the vector of centered revenues: 89 33 41 82 45 66 37 132 15 =i . 4]o "=| 169 43 85 45 20 68 23 19 “2 ‘The best we can do schematically is to imagine v and w as directed line segments ‘emanating from the origin, which is denoted by 0 (Figure 1.25). ‘The representation of the centered information by vectors, as in Figure 1.25. is called a vector plot. From statistics, we have the following conventions: © Ina vector plot, the length of a vector indicates the variability of the corre sponding variable, * Ina vector plot, the angle between veotors measures how similar the variables are to each other, ‘The statistical terminology for “how similar the variables are” is “how highly correlated the variables are.” Vectors that represent highly correlated variables have either a small angle or an angle close to 7 radians between them. Vectors that represent uncorrelated variables are nearly perpendicular; that is, the angle between them is near x /2. ‘The following chart summarizes the statistical terminology applied to the ge- cometric characteristics of vectors in a vector plot. Geometric Characteristics ‘Statistical Interpretation Length of a vector. ‘Variability ofthe variable represented. Angle between a pair of vectors The variables represented by the vectors is small tare highly positively correlated. Angle between a pir of vectors ‘The variables represented by the vectors is near x sre highly negatively correlated, Angle between a pair of vectors ‘The variables represented by the vectors is near x/2. are uncomrelated ot unrelated. The variables are sai to be perpendicular ‘or orthogonal From statistics we have the following measures of a sample of data (1,22. 2 Xnaty ak Sample size = n, the number of dataFIGURE 1.26 1.8 Correlation Coefficient (Optional) 77 the average of the data, Correlation coefficient: If the n-vectors x and y are data vectors where the ‘data have been centered, then the correlation coeficient, denoted by Cor(x,y), is computed by xy Corts, = itil Geometrically, Cor(x, y) is the cosine of the angle between vectors x and y. For the centered data in Table 1.2, the sample size is n = 10, the mean of, the market value variable is 240, and the mean of the revenue variable is 119. To determine the correlation coefficient for v and w, we compute ‘and thus 8 This result indicates that the variables market value and revenue are not highly correlated. This seems to be reasonable, given the physical meaning of the ables from a financial point of view. Including more than the ten top corporations ‘may provide a better measure of the correlation between market value and revenuc, “Another approach that can be investigated based on the scatter plots isto omit data that scem far from the grouping ofthe majority ofthe data. Such data are termed outliers, and this approach has validity for certain types of statistical studies. Figure 1.26 shows scatter plots that geometrically illustrate various cases for the value of the correlation coefficient. This emphasizes that the correlation coe! ficient is @ measure of linear relationship between a pair of data vectors x andy ‘The closer all the data points are to the line (in other words, the less scatter), the higher the correlation between the data Gy Pececrpostive —@)Pevlooneqave (©) Lese hamper) Less than pete, colton, comeltion: pasiivecoretation; nepative comelaion Cons 9)= Cory Oscorayel 1s Corleys 078 Chapter 1 Linear Equations and Matrices To compute the correlation coefficient of a set of ordered paits (x), yi), = 1,2... m where a y m ” x= and y= : Yat Pat wm Jn ‘we use the steps in Table 1.3. The computational procedure in Table 1.3 is called the Pearson product-moment correlation coefficient in statistics. TABLE 1.3 4. Compute the sample means for each data vector: yo 2, Determine the centered x-data and the centered y-data as the vectors x. and ¥., respectively, where eea[nok ao¥ + axl yea Du-F Foo FY 3. Compute the correlation coefficient as xerye Cony. M89 = eal ‘The correlation coefficient can be an informative statistic in a wide variety of applications. However, care must be exercised in interpreting this numerical est- mation of a relationship between data. As with many applied situations, the inter relationships can be much more complicated than they are perceived to be at first glance. Be warned that statistics can be misused by confusing relationship with cause and effect. Here we have provided a look at a computation commonly used in statistical studies that employ dot products and the length of vectors. A much ‘more detailed study is needed to use the correlation coefficient as part of a set of information for hypothesis testing. We emphasize such warnings with the following discussion, adapted from Misused Statistics, by A. J. Jafle and H. F. Spirer (Marcel Dekker, Inc., New York, 1987) Data involving divorce rate per 1000 of population versus death rate FIGURE 1.27 per 1000 of population were collected from cities in a certain region. Figure 1.27 shows a scatter plot of the data, This diagram suggests that1.8 Correlation Coefficient (Optional) 79 divorce rate is highly (negatively) comrelated with death rate, Based on this measure of relationship, should we infer that (divorces cause death? (Gi) reducing the divorce rate will reduce the death rate’) Certainly we have not proved any cause-and-effect relationship: hence ‘we must be careful to guard against statements based solely on a nu- ‘merical measure of relationship. Perpendicular vectors Sample mean Scatter plot Comelation coefficient Yector plot Outliers between vectors Correlated/uncorrelated variables “lel vectors Sample size E & Exercises 2. The data sets displayed in Figures A, B, C, and D have by the 8-veotors x and y, respectively, where ‘ne of the following correlation coefficients; 0.97, 0.93, 88, 0.76, Match the figure with its eonetation coeffi een, 12 24 33 39 52 61 69 1 andy ‘Compute the correlation coefficient between the input flows in x andthe resultant meter readings in y 3. Aneexperiment to measure the amplitude ofa shock wave resulting from the detonation of an explosive charge is Taw Faw D, ‘conducted by placing recorders at various distances from the charge. (Distances are 100s of feet.) A common ar ‘> 5 miter that measures flow rates is being calibrated. In rangement for the recorders is shown in the aecompany~ jal test, = 8 flows are sent to the meter and the ing figure. The distance of a recorder from the charge | Srresponding meter readings are recorded, Let the set ‘and the amplitude ofthe recorded shock wave are shown |“ Hlows and the corresponding meter readings be given inthe table, Compute the correlation coefficient between,80 Chapter 1 Linear Equations and Matrices the distance and amplitude data, TOL 2345678 (C—Explosive charge Recorder Distance Amplitude 200 126 200 199 400 93 400 98 500 19 500 78 500 80 700 60 700 64 li Supplementary Exercises 1, Determine the number of entries on or above the main di ‘agonal of ak x k matrix when @ ten WEAN OE=4 OL o 9 sina=[2 2] (a) Find a2 x k matrix B % O such that AB = O for kEL234 (b) Are your answers o part (a) unique? Explain, 3, Find all2 x 2 matrices with real entries ofthe form a-[s 2] such that A? = h, 4, An equal number of two-parent families, each with three children younger than ten years old were interviewed in cities of populations ranging from 25,000 t0 75,000. In terviewers collected data on (average) yearly living ex- ppenses for housing (rental/morigage payments), food, and clothing. The collected living expense data were rounded to the nearest 100 dollars. Compute the correlation coefficient between the population data and living expense data shown in the following table: ‘average Yearly City Population in 1000s) (in $ 1005) 30 65 35 8 40 70 50 nm oo 85 6 83 15 88 4: tic A (the le ean eldencnasna amateur are & mY @ Pintscurennst@=[) {] 100 (b) Finda square root of B=|0 0 0 000. fe), Finda square root of B = Ie. (@)_ Show that there is no square root of aL) 3]W. ML x, 16. rn 18 Lot A be an m x 1 mate, a) Deseribe the diagonal entries of A” A in terms of the columns of A. (b) Prove thatthe diagonal entries of A” are nonnegative. (0) Whenis A"A = 07 If Ais ann x matrix, show dhat (AY? = (AT) forany Positive integer k. Prove that every symmettic upper (or lower) triangular matrix is diagonal Lot A be an n x m skew symmetric matrix and x an n= sector. Show that x7 Ax = O forall xin Let A be an upper triangular matrix. Show that A is nonsingular if and only if all ube entries on the main diagonal of A are nonzero, ‘Show dha the produet of ewe 2 x 2 skew symmetric matrices is diagonal. Is this true for n x n skew symmetric matrices with n > 2? Prove that if TAT A) = 0, then A= O. Forn 1 matrices A and B, when does (A-+B)(A—B) = Be Develop a simple expression for the entries of A", where nnisa positive integer and If B= PAP™, express BP, BY,..., BY, where k is a positive integer, in terms of A. P, and P~' Prove that if is skew symmetric and nonsingular, then A“ is skew symmetric. Let A be ann x n matrix. Prove that if Ax = 0 forall nx 1 matrices x, then A= O. Let A be an m xn matrix, Prove that if Ax = x forall nT matrices x, then A Let A and B ben x matrices. Prove that if Ax = Bx for all x 1 matrices x, then. = B. TE A is an mx re matrix, then A is called idempotent if a (a) Verify that /, and O are idempotent. (b) Find an idempotent matrix that isnot J, or O. (6), Prove that the only nx nonsingular idempotent matrix is I, Let A and B ben xm idempotent matrices. (See Exercise 19) (@)_ Show that A is idempotent if AB = BA. A. a. Supplementary Exercises 81 (b) Show that if is idempotent, then AT is idempotent. (©) IA + B idempotent? Justify your answer. (@) Find all values of & for which kis also idempotent Let A be an idempotent matrix, (8) Show that A" (b) Show that f, ~ Ais also idempotent. If A is an nxn matrix, then A is called nilpotent if AY = 0 for some positive integer (2) Prove that every nilpotent matrix is singular. ord (@) Verity that A= |0-0- 1 | isnipotem. 000. (©) IFA is nilpotent, prove that J, — A is nonsingo- Tac, Hint: Find (Jy ~ A)! in the cases A = 0. = 1,2... and look for pattern] A forall integers n > 1 Lety =|? ]. Determine a vector w so that v+w atb-+e¢d.Itvisan n-vector, what is w? Use the result from Exercise 23 to develop a formula for the average of the entries in an vector v = | "> | in 1% terms of a ratio of dot products Foran mn matrix A, the main counter diagonal elements Me dip, dy—ts ++ dye (Noe that ais main counter diagonal element, provided that + j = n-+ 1.) The sum of the main counter diagonal elements is denoted Med(A), Yo. Med(4) meaning the sum of all the entries of A whose subscripts add ton +1 (a) Prove: Med(cA) ber, Prove: Med(A + B) = Med(A) + Med(R). Prove: Med(AT) = Med). Show by example that Med(A/B) need not be equal to Med(BA). An n xn matrix A is called block diagonal if it can be ppatttioned in such a way that all the nonzero entries are contained in square blocks A; ‘Med(4), where cis areal num- ) © @82 Chapter 1 Lineor Equations and Matrices (a) Panttion the following matrix into a block diagonal matrix: (b) IF isblock diagonal, then the linear system Ax ig said to be uncoupled, because it can be solved by considering the linear systems with coefficient matices A\, and right sides an appropriate portion of bb. Solve Ax = b by “uncoupling” the linear system when A isthe 4 x 4 matrix of part (a) and 27, Show that the produet of two 2 x 2 skew symmetric ma- twices is diagonal Is this true for n x m skew symmetric matrices with n > 2? _[4n @ “lo An jonal n xr matrix. Prove that if Aj, and ‘Az: are nonsingular, then A is nonsingular. [sx] bbe a partitioned matrix. If Ay, and Azz are nonsingular, show that A is nonsingular and find an expression for A 29, Let In Exercises 30 through 32, X and ¥ are n x 1 matrices whase entries are xy.t3, +5 Syd YY «ay Ys Respectively. The ‘outer product of X and ¥ isthe maerix product XY", which ives the m x m marév SO) ME Ye M2) YE Ye MeL Tay ade 30, (a) Form the outer product of X and Y, where “rll (b) Form the outer product of X and ¥, where x 1 2 7 | ana 1 0 3 2, 5 3A. Prove or disprove: The outer product of X and ¥ equals the outer produet of ¥ and X. 32, Prove that THXY?) = X"Y. ame a-[t 3] aw of J AB = Sots prot fc (A) with 0 (8), 34. Let W be ann x I matrix such that WTW = 1, The a x1 matrix = 1,-20WT is called a Householder* matrix. (Note that a House- ‘holder matrix is the identity matrix plus a scalar multiple ‘of an outer product.) (a) Show that His symmewic (b) Show that #1 = AT “Alston 8, Householder (1904-1993) was born in Rockfon, lino, sn died in Maliba, Califor ALSTON S, HOUSEHOLDER nia, He received his undergraduate degree Irom Northwestern University an his Master of Ars from Cornell University both in philosophy. He received his PhD. in mathematics in 1937 from the Univer- ‘i of Chicago. His early workin mathematics dealt with the applications of mathematics to biology In 1944, he began to work on problems dealing with World War I. In 1946, he Became a meriber ofthe Mathematics Division of Oak Ridge National Lahoratory and became is diector in 1948. At (Oak Ridge, his interests shifted from mathematical biology to numerical analysis. He is best known for his many important contributions tothe eof numerical linear algstr, in ation to his re search, Householder occupied a number of post in profesional organizations, served on a variety of ‘itor boards, and organized the important Gating conferences (now Known as the Hanseholder ‘Symponia}, which continue to this dy.35. A circulant of order n isthe n xm matix defined by © = eiteleys 695.469) ‘The elements of each row of C are the same as those in the previous rows, but shifted one position tothe right and Chapter Review True or False 1. A linear system of thre equations can have exactly three ditfecent solutions, 2. IA and B are n x n matrices with no zero entries, then. AB #0. 3. IFA isan x n matrix then A+ AT is symmetric. If A isanm xn matrix and xis m x 1 thea AN sa linear combination ofthe columns of A. §, Homogeneous linear systems ae always consistent 6 The sum of two n > 1 upper triangular matrices is upper triangular 7. The sum of two n x n symmetric matrices is symmetric. 8, 1 a linear system has @ nonsingular coefiient matt, then the system has a unique solution 9. ‘The product of two n x 1 nonsingular matrices is monsin- gala 10. 4 [2°] sets» mais sma a Quiz 1. Find all solutions to the following li method of elimination: 4 3. 38, 3 >. Determine all values ofr so that Chapter Review 83 ‘wrapped around. (a) Form the circulant C = cite(1, 2,3) (b) Form the circulant C = cino(1, 2, (@)_ Form the circulant © = cire(1, 0,0, 0,0). (@) Form the circulant C = cite(1, 2, 1,0, 0). Let C = cire(e, 2,3). Under what conditions is C sym- rmettie? Let € = cito(ey,c2, ++) and letx be them x 1 matrix ofall ones. Determine a simple expression for Cx. Verity that for C = cire(ey. 2.03), C7C = CCT, «a solution to the following linear system: x-dy $35 3 Determine all values of a and b so that (Cf i}-[ a] o 0 14 8 2 of mi v=lo -1 5 1 0 0 6 (a) Determine all values of « $0 thatthe (3,2) entry of LU is7, (b) Determine all values of band c so that the (2,3) en- uy of LU is. Let ube a vector in R* whose projection onto the 2-axis is shown in the figure, Determine the entries of the vector x84° Chopter 1 Linear Equations and Matrices Discussion Exercises L. In Section 1.5 we briefly introduced recursion relations. ‘We showed how the members of the Fibonacci sequence could be computed by using a mattix form. By sueces- ‘ive matrix multiplication we were able to produce 2 » 1 ‘matrices from which the members ofthe sequence could bbe extracted, An alternative approach is to derive a for- ‘mule that generates any member ofthe sequence without computing the preceding members. For two-term recursion relations of the form tig = dig. + b> We can derive such a formula by solving a 2 x 2 system of equa tions as follows: Associated with tis recusioa relation is 4 quadratic polynomial ofthe form r? = ar-+b, called the characteristic polynomial ofthe recursion relation. Label the roots of this polynomial as r) and rs, with ry r2 ‘Then it can be shown thatthe general term of the recur sion relation uy can be expressed inthe form ty = Cyr" + Cal)", “ ‘where the constants Cand C; are determined by solving, the system of equations generated by setting n = 0 and then = Tin (®) and using the initial values of up and 1 given to start the recursion relation. (a) Use this approach to find 2 formula forthe nth mem= ber of the Fibonacei sequence, (b) Since the members of the Fibonacci sequence ate i= tegers, explain why the formula from part (a)is rather amazing. (©), Given the coefficients a and b of a two-term recur sion relation of the form uy = ay: + Duy Con struct a general system of equations that can be used to determine the values of the constants Cy and Cs in ©, . We can use a 3” by 5" index eard to represent (a portion of) a plane in space. Take three such index cards and cut a slit in each one about halfway through the card along. the 5-inch edge. The slit wil let you model intersecting a pair of planes by passing the slit in one card through the slit in another card. Each card can be represented by an equation of the form ax + buy + ez = dj $0 that for 1, 2, 3.we have a system of three equations in three unknowns, Using the index cards, configure them so thet they represent four different consistent linear systems and ‘our different inconsistent systems. Sketch each config uration and provide a brief description of your diggram, (Hint: Two index cards placed exacily one on top of the ‘other represent the same plane.) 3. In Section 1.3 we defined matrix multiplication; if the matrix A ism x p and the matrix B is p xn, then the product, 4 ‘AB is an m x n matrix, Here we investigate the special cease where A ism x I and B is 1 I matrix A has a ero entry and matix B is teen (©) Suppose that the second ent of the ms x 1 mateix A is zero, Describe a rsuling pater of zeros that appearsin AA. (@ Lec C bea Sx $ matrix with identical columns. Ex- plain how to generate C by using a product of aL matrix A and a 1 $ mateix B. CExplicitly indicate the enties of A and 8.) In Sections 1.6 and 1.7 we introdueed matrix transfor ‘mations forthe manipulation of objects in the plane— for example, dilation, contraction, rotation, reflection, and shes. Another manipulation i called transtaton, whichis Tike sliding an object over a certain distance. To translate (2) For the case m = 3 and » tne vector = [*] inte plane, we ale vestor »-[2Je" RR given by action defines a function ¢ mapping ew sata ‘The function g is called a translation by the vector a, Ex- plain why translations are not mattix transformations for 2 [2) tn tte inSain) . In Chapter 1 we discussed systems of lincar equations in the plane and in space to help connect geometry tothe al- zzebraic expressions forthe systems. We observed that a pair of fines in the plane could have zero, one, or infinitely many intersection points; similarly, for thee planes in space. For nonlinear systems of equations there can be many other sets of intersection points. (a) The set of functions y = f(x) = 2°", n = 1,2,3, is the family of even power functions. Describe the set of intersection points for any pair of functions inthis family, (b) For the function y = fox termine another function intersect in exactly four points 2 where x = 0, de- (4) so that f and g6 People have long been intrigued by magie squares. In the past they were often associated withthe supernatural and hence considered to have magical properties. Today they are studied to illustrate mathematical properties and also as puzzles and games. We define a magic square as follows: A magie squate is an n x 1 matrix of positive inte- _zrs sch thatthe sum ofthe entries of each row, each col- ‘umn, and each main diagonal (see the diagram) is equal to the same (magic) constant K. agonal egoual 4 sun is 4 1 sum isk 4 | sini BK Meek OX AB 3 magic square Chapter Review 85 81 6 357 492 {is amagie square with (mage) constant K = 15, (Verify) For3 x 3 and other small square matrices A of positive integers, itis rather easy 10 check whether As & mag square by observation (@) Fora3 x3 matrix A, construct a method to check ‘whether A is a magic square by mukiplying A by particular matrices and by using an operation defined fn matrices. Specifically indicate your strategy and state how each ofthe portions ofthe preceding def- nition are checked ‘The matrix (b) Briefly discuss how to generalize the technique you ‘devise in part (a) ton xn matrices.86 CHAPTER Solving Linear Systems ea Echelon Form of a Matrix In Section 1.1 we discussed the method of elimination for solving linear systems, which you studied in high school, and in Section 1.3 we introduced the coefficient matrix and augmented matrix associated with a linear system. In this section we discuss operations on a matrix, which when applied to an augmented matrix can greatly simplify the steps needed to determine the solution of the associated linear system. The operations discussed in this section apply to any matrix, whether or not itis an augmented matrix. In Section 2.2 we apply the constructions developed in this section to the solution of linear systems. DEFINITION 2.1 ‘An m 3 n matrix A is said to be in reduced row echelon form if it satisfies the following properties: (@) All zero rows, if there are any, appear at the bottom of the matrix. (b) The first nonzero entry from the left of a nonzero row is a 1. This entry is called a leading one of its row. (©) For each nonzero row, the leading one appears to the right and below any Jeading ones in preceding rows. (@ Ifa column contains a leading one, then all other entries in that column are ‘A matrix in reduced row echelon form appears as a staircase (“echelon”) pat- tem of leading ones descending from the upper left comer of the matrix. An m xn matrix satisfying properties (a), (b), and (@) is said to be in row ‘echelon form. In Definition 2.1, there may be no zero rows. ‘A similar definition can be formulated in the obvious manner for reduced column echelon form and column echelon form.ee 21 Echelon Form of a Matrix 87 The following are matrices in reduced row echelon form, since they satisfy properties (a), (b), (and (4) cone and 20 or 00 on 23 00. ‘The matrices that follow are not in reduced row echelon form. (Why not?) . ‘The following are matrices in row echelon form: 0 0 0 0 0 > 2 4 7 0 0 5 1 0 0 0 ene 304 2 54. 1 2 304 25 12 0 0. . 1000 o100 oor.) 0001 9 3 2 1 0 . ‘A.useful property of matrices in reduced row echelon form (see Exercise 9) is that if A is an n x n matrix in reduced row echelon form # {y, then A has a row consisting entirely of zeros, We shall now show that every matrix ean be put into row (column) echelon form, or into reduced row (column) echelon form, by means of certain row (column) operations, An elementary row (column) operation on a matrix A is any one of the following operations88 Chapter 2. Solving Linear Systems Bae (@) ‘Type I: Interchange any two rows (columns). (b) ‘Type IL: Multiply a row (column) by a nonzero number. (©) Type Hil: Add a multiple of one row (column) to anoth We now introduce the following notation for elementary row and elementary column operations on matrices: ‘+ Interchange rows (columns) i and j, Type I: ner oe). * Replace row (column) i by & times row (column) i, Type TI: bry ny (ke >), * Replace row (column) j by & times row (column) i + row (column) j, Type Ill: key try ry (ke, $e) +6). Using this notation, its easy to keep track of the elementary row and column ‘operations performed on a matrix. For example, we indicate that we have interchanged the ith and jth rows of A as Ay... We proceed similarly for column operations. Observe that when a matrix is viewed as the augmented matrix of a linear system, the elementary row operations are equivalent, respectively, to interchanging two equations, multiplying an equation by a nonzero constant, and adding a ‘multiple of one equation (o another equation. 0 0 1 2 A=|2 3 0 -2 303 6 79 Interchanging rows 1 and 3 of A, we obtain wes Baron =]2 3 a0 Let Hoo I -9) d| Multiplying the third row of A by 4, we obtain 0 0 1 2 C=Ay,.,=/2 3 0 -2 ; rons Adding (—2) times row 2 of A t0 row 3 of A, we obtain oo 1 2 D=Aatmnon =| 2 3 0 2 =1 3516 5, Observe that in obtaining D from A, row 2 of A did not change. .Theorem 2.1 2.1 Echelon Form of a Matrix 89 An m x n matrix B is sid to be row (column) equivalent o an m xn matrix A if B can be produced by applying a finite sequence of elementary row (column) ‘operations to A If we add 2 times row 3 of A to its second row, we obtain 1 2 B= Arto =| 4 —3 1-2 so B is row equivalent to A. Interchanging rows 2 and 3 of B, we obtain 1 Ines =] = } = so C is row equivalent to B. Multiplying row I of C by 2, we obiain 80 D is row equivalent to C. It then follows that D is row equivalent to A, since we obtained D by applying three successive elementary row operations to A. Using the notation for elementary row operations, we have 4 “1 3 We adopt the convention that the row operations are applied in the order listed, i We can readily show (see Exercise 10) that (a) every matrix is row equivalent (0 itself; (b) if B is row equivalent to A, then A is row equivalent to B; and (c) if C is row equivalent to B and B is row equivalent to A, then C is row equivalent to A. In view of (b), both statements “B is row equivalent to A” and “A is row equivalent to B” can be replaced by “A and B are row equivalent.” A similar statement holds for column equivalence. Every nonzero m xn matrix A = row (column) echelon form. uj] is row (column) equivalent to a matrix in90 Chapter 2 Solving Linear Systems Bee Proof ‘We shall prove that A is row equivalent to a matrix in row echelon form. That is, by using only elementary row operations, we can transform A into a matrix in row ‘echelon form. A completely analogous proof by elementary column operations establishes the result for column equivalence. We start by looking in matrix A for the first column with a nonzero entry. This column is called the pivot column; the first nonzero entry in the pivot coluran is called the pivot. Suppose the pivot column is column j and the pivot occurs in row i. Now interchange, if necessary, rows 1 and i, getting matrix B = [ bi; ]. Thus the pivot by, is # 0. Multiply the first row of B by the reciprocal of the pivot, that is, by 1/bij, obtaining matrix C = [¢,; ]. Note that ¢1; = 1. Now if c,j,2 3° (yon a 0-2 -1 7 1 2 ° : Do (C)): obtain D, Dan says 10 Obi a=|o 4 nes 0 3 1o1-$ 1 2 or gas D= 203 4 O 2093 492 Chapter 2 Solving Linear Systems Deleting the first row of Dy yields the matrix A>, We repeat the procedure with Az instead of A. No rows of Az have to be interchanged. Pon ae eee 0 es 00 4 A= 2 = Ba. Do (Bs) tp, on, 0 0b1 0 0 304 aren 10 Tete ee ee Okay ee oul? 9 1 42 Finally, do (C3)-20, 4-07 =] 4 9 2 3 4 | Wobiain D: Pot spor 2 Oo Tae o 1 32 0 0 0 0 11-$ 12 n-|o ! g-2 4 oo 132 0 0 0 0 0 is in row echelon form and is tow equivalent to A. . When doing hand computations, itis sometimes possible to avoid fractions by suitably modifying the steps inthe procedure, Theorem 2.2 Every nonzero m x n matrix A — [ay] is row (column) equivalent to a unique ‘matrix in reduced row (column) echelon form. Proof We proceed! as in Theorem 2.1, obtaining matrix Hf in row echelon form that is row equivalent to A. Suppose that rows 1, 2,...,r of H are nonzero and that the leading ones in these rows occur in columns ¢), ¢2,..., ¢-. Then cy < ¢) < < ¢r Starting withthe lst nonzero row of H, we add suitable maltiples of this row to all rows above it to make all entries in column c, above the leading one in row r equal to zero, We repeat tis process with rows r ~ 1, r—2,...,and 2, making allEs 2.1 Echelon Form of a Matrix 93 entries above a leading one equal to zero, The result is & matrix K in reduced row echelon form that has been derived from H by elementary row operations and thus row equivalent to H. Since A is row equivalent to H., and H is row equivalent to K, then A is row equivalent to K. An analogous proof can be given to show that A is column equivalent to a matrix in reduced column echelon form. It can be shown, with some difficulty, that there is only one matrix in reduced row echelon {orm that is row equivalent to a given matrix. For a proof, see K. Hoffman and R. Kunze, Linear Algebra, 2d ed, (Englewood Cliffs, NJ: Prentice-Hall, 1971). = Remark i should be noted that a row echelon form of a matrix is not unique Find a matrix in reduced row echelon form that is row equivalent to the matrix A of Example 5. Solution ‘We start with the matrix #7 obtained in Example 5 in row echelon form that is row equivalent to A. We add suitable multiples of each nonzero row of H to zero out all entries above a leading 1. Thus, we start by adding (~3) times the third row of H Woits second row: roi-go4 an o 1 0-8 N= Homen=1q 9 1} er) 110 8 7 o 1 0-2 -§ = WM spanory = 2 fmon 1g 9 10 32 o 0 0 0 Oo Finally, we add (—1) times the second row of J to its first row: 100 9 8 o 1 0-8 3 oo 1 32 0 0 0 0 0 This is in reduced row echelon form and is row equivalent to A. Altematively, we ‘can express the reduced row echelon form of A as Mnenon iat eink .94 Chapter 2. Solving Linear Systems Key Terms Elimination method Reduced row echelon form Leading one EONS Exer 1. Find a row echelon form of each of the given matrices. ‘Record the row operations you perform, using the nota Remark The procedure given here for finding a matrix K in reduced row echelon form thet is row equivalent to a given matrix A is not the only one possible, For example, instead of first obtaining a matrix H in row echelon form that is row ‘equivalent to A and then transforming H to reduced row echelon form, we could proceed as follows, Hirst, zero out the entries below a leading 1 and then immy ately zero out the entries above the leading 1. This procedure is not as efficient as the procedure given in Example 6. Row echelon form Pivot columa Elementary row (column) operation Pivot Row (column) equivalent fermine its reduced row echelon form. Record the row ‘operations you perform, using the notation for elemen- tion for elementary row operations, tary row operations, 12-5 1 0-3 2 @ A=] 2-1 6 foo roid 22 7 MA=\o 9 1 2 toe 0 0 0 0 304-1 13 02 4 MAm) Ss 6 3 4=]0 1 0 1 0 2-2 2 oo 1-1 0 2. Find a row echelon form of each of the given matrices, __S. Find the reduced row echelon form of cach of the given Record the row operations you perform, using the nota- ‘ion for elementary row operations matrices. Record the row operations you perform, using the notation for elementary row operations. “11-1 0 3 1 0 2) @a=|3 4 1 1 wo @Aa={2 1 9 4-6 4 -2 14. 3024, bor nw 107 ) A=/2 -1 10 _Ja 22 43-2 Mam) 4 To 3. Bach of the given matrices isin row echelon form, De- “207-5 termine its reduced row echelon form. Record the row 6, Find the reduced row echelon form of each ofthe given ‘operations you perform, using the notation for elementary row operations 1 @ a=lo0 0 ) A= 1 0 0 0 4 2 1 3 1 0 0 5 -4 1 0, matrices, Revord the row operations you perform, using the notation for elementary row operations. m1 2-5 @a=| 2-1 6 2-2 7 1o-t 3 0441 M A=! 5 6 3 2 2 2. 4. Bach of the given matrices isin row echelon form, De- 7. Let.x,y,z, and w be nonzero real numbers. Label each of3 a 2.2 Solving Linear Systems 95 the following matices REF if itis in row echelon form, 10, Prove "REP in edad ow echelon form, or NF 6 (a) Every matix is ow equivalent toil NOC REF and not 1 . () IF Bis row equivalentio A, then A is row equivalent lx y ooo WB. @ foo loo io (© IC is row equivalent wo B and Bis row equivalent 00 w Coon to Athen C i row equivalent 10 A. 1 Let pies ae © wy -1 003 4 loo 0010 a=] 1 2 oo 0001 230 Ley, 5; and w be nonzero eal aumbers. Label each of snd a matrix a column echelon form thatis column the following mais REF iit isin row echelon form, (2) Bind mass column echelon form hat iseo IRREF if tis in educed row echslon form, or Nif iis om ul not BREF: (0) Finda matix in reduced column echelon form that lx oo iscolumn equivalent to A oe 0010 12, Repeat Exereise 11 forthe matix Ly a oot too 123 4 8 21 3-1 2 ayo 3 oro2 4 1 fr 10 oo4 13, Determine the reduced row echelon form of Let A be an m x m mattx in reduced ow echelon form. cos sind Prove that fs J, then A has a row consisting entirely A [ ¢ | of zeros sn cos EEE sotving Linear Systems In this section we {ise the echelon forms developed in Section 2.1 to more efficiently determine the solution of a linear system compared with the elimination method of Section 1.1. Using the augmented matrix of a linear system together ‘with an echelon form, we develop two methods for solving a system of mt linear equations in n unknowns, These methods take the augmented matrix of the linear system, perform elementary row operations on it, and obtain a new matrix that represents an equivalent linear system (i... a system that has the same solutions as the original linear system). The important point is thatthe later linear system can be solved more easily. ‘To sce how a linear system whose augmented matrix has a particular form can be readily solved, suppose that 12 oo4 o 0 represents the augmented matrix of a linear system. Then the solution is quickly96 Chapter 2 Solving Linear Systems Theorem 2.3 Corollary 2.1 Found from the corresponding equations xy =3-2n =3-6= ‘The task of this section isto manipulate the augmented matrix representing a given Jinear system into a form from which the solution can be found more eusily. ‘We now apply row operations tothe solution of linea systems. Lot Ax = b and Cx = d be two linear systems, each of m equations in n unknowns. I'he augmemed matrices [4 {1b} and [ae row equivalent, then the linear systems are equivalent; sl js, they have exacily the same solutions. Proof This follows from the definition of row equivalence and from the fat that the three clementary row operations on the augmented matrix are the three manipulations on linear systems, discussed in Section 1.1, which yield equivalent linear systems. We also note that if one system has no solution, then the other system has no solution, : Recall from Section 1.1 that the linear system of the form x1 Haig Hoe + Aint an1Xy + apts $06 + Aang =O a gtk + gnka Hy is called a homogenous system, We can also write (1) in matrix form as Ax=0 @ IFA and C are row equivalent m x m matrices, then the homogeneous systems Ax = Vand Cx = 0 ate equivalent Proof Exercise, . We observe that we have developed the essential features of two very straight. forward methods for solving linear systems. The idea consists of starting with the linear system AX = b, then obtaining a partitioned matrix [C | d] in either row ‘echelon form or reduced row echelon form that is row equivalent tothe augmented matrix [4 |b]. Now [C | dl] represents the linear system Cx = d, which is quiteCARL FRIEDRICH Vo poble ¥ heii wnat torpaN 2.2. Solving Linear Systems 97 simple to solve because of the structure of [ | @], and the set of solutions to this system gives precisely the set of solutions to Ax = bs; tha is, the linear systems Ax = band Cx = d ate equivalent. (See Section 1.1.) The method where [C | d] is in row echelon form is called Gaussian elimination; the method where [C d] is in reduced row echelon form is called Gauss*—Jordan’ reduction. Strictly speaking, the original Gauss-Jordan reduction was more along the lines described in the preceding Remark. The version presented in this book is more efficient. In actual practice, neither Gaussian elimination nor Gauss—Jordan reduction is used as much as the method involving the LU-factorization of A that is discussed in Section 2.5. However, Gaussian elimination and Gauss—lordan reduction are fine for small problems, and we use the latter heavily in this book. Gaussian elimination consists of two steps: Step 1. The transformation of the augmented matix [A | b] 10 the matrix [C | €] in row echelon form using elementary row operations Step 2. Solution of the linear system corresponding to the sugmented matrix [C | d] using back substitution For the case in which A is m x n, and the linear system Ax = b has @ unique solution, the matrix [C | dl] has the following form: Cal edict Gauss (1777-1855) was bom into a poor working-class fanily in Brunswick, Ger- ‘mang, and died in Gottingen, Germany the mest famous matbecatician inthe word. He was a child prodigy with a genius that di nocimpesshis father, wh elle him a"star-gazee" However, bis teach- fr were impressed enough to arrange forthe Duke of Brunsvick to provide a scholarship for Gauss a the local secondary school. Asa teenager thee, he made orginal discoveries in number theory and hogan to sgeeulate about non-Euclidean geometry. His seientite publications include impant contributions in number theory, mathematical astonoms. mathematical geography, satis, citferenial seometry, and magnetism. His diaries and private noes contain many otber discoveries that he never published "An austere, conservative man who had few eins and whose private life was generally unhapp he was very concemed that proper credit he given for scienlifc discoveries. When he relied oti results of eters, he was careful to acknowledge them: and when others independently discovered ‘esl in is private notes, he was quick t claim pro, Tn his esearch Gauss used a method of calculation that ater generations generalized to row re Auction of matrices and named i his honor, afhough the method was used in China alsost 2000 yeas catlice ‘Wilhelm Jordan (1842-1899) was born in soutien Germany. He atended college in Stutgat 1nd in 1868 became fll professor of geodesy atthe technical college in Karlsruhe, Geemany. He Participate in surveying several repions of Germany. Jordan was a proliic writer whose major work, Handbuch der Vermessingstunde (Handbook of Gender), Was ‘into French, Talan, ad Russian, He was considered a superb writer and an excellent teacher. Unfortunatly, the Gauss-Sordan reduction method has been widely atebuted to Carnill Jordan (1838-1922), a well-known French Iathematiclan, Moreover, i sem thatthe rethod was also discovered independently at the same time by B. . Clssny priest who lived in Luxembourg. This biographical sketch is based on 2h excelent article: S.C. Athoon and R. MeLaughlia, "Gauss-Jondan reduction: A brief istry” MAA Mowohly, 94 (1989), 130-14398 Chopter 2. Solving Linear Systems len cs + cn | dt 0 1 es cu | rae 00 0.0 1 id (The remaining cases are treated after Example 1.) This augmented matrix repre~ sents the linear system aitenxa tenmy to. + cin = dh Myton te Hany Bint + noite Xn ‘Back substitution proceeds from the nth equation upward, solving for one variable from each equation; dy dy ~Cntn%n dy — e235 — C2484 — + Conky dy — e192 — C1383 — + — Cah ‘The linear system x42y432=9 de yt 258 Br = S3 hhas the augmented matrix 2 359 [4 “1 1/8 0-113 ‘Transforming this matrix to row ec! 12319 {cjaj=|o 1 1/2]. 00 113 Using back substitution, we now have x=9-2y-32= 942-952, thus the solution is x = yy = 1.2 = 3, which is unique. .2.2 Solving Linear Systems 99 ‘The general case in which A is m x nm is handled ina similar fashion, but we need to elaborate upon several situations that can occur. We thus consider Cx = d, where C ism xn, and [Cj a] is in ow echelon form, Then, fr example, [C | a] mnight be ofthe following form: 1 ep es Cu 00 1 mw: cw oo O01 Cute 0 oo. ° 0 0 0 ‘This augmented matrix represents the linear system ai tener tests + + inte = dt wat eagty toot Conn = db St + ents on + Ory = dest Ox) + + + Ox, First if dior = 1 then Cx = d has no solution, since at least one equation is not satisticd. If disy = 0, which implies that dia — <> = dy = O (since [C |] was assumed to be in row echelon form), we then obtain xy — dey — het ~ C¢-1nkn = dt — C4-tndle and continue using back substitution to find the remaining unknowns corresponding to the Ieading entry in each row. Of course, in the solution some of the unknowns may be expressed in terms of others that can take on any values whatsoever. This merely indicates that Cx = d has infinitely ‘many solutions. On the other hand, every unknown may have a determined value, indicating that the solution is unique. Let 123 45 vqje{o 1 2 3-1 leidl=Fo 9 1 a3 oo 0 1 2100 Chapter 2. Solving Linear Systems ‘Then xy =9— Drs ty = 7 2xy — 3x5 = 7 — 2 — 2s) — 3x5 = = Hs p27 — 2x — By bats = 24 Sts 6 — 2x — Bxy — dy — Ss = =1 = 1085 any real number. ‘The system is consistent, and all solutions are of the form, Es " [efaj=]o 1 2 3/6]. ooo01 then Cx = d has no solution, since the last equation is Ox; + Oxy + Ors + Ox which can never be satisfied, . When using the Gauss-Jordan reduction procedure, we transform the augmented matrix [A | b] to [C | d]}, which is in reduced row echelon form. This means that we can sole the linear system Cx = d without back substitution, as the examples that follow show; but of course, it takes more effort to put a matrix in reduced row echelon form than to put it in row echelon form. It turns out that the techniques of Gaussian elimination and Gauss—Jordan reduction, as described in this book, require the same numberof operations. EE * Lo 0 01s sigje [0 190.6 (eial=|9 9 1 017 ooo1s then the unique solution is ne 4 =8 .2.2. Solving Linear Systems 101 cae 1120 then as fondu fas where x2, 15, and 2s can take on any real numbers, so the system has infinitely ‘many solutions. Thus a solution is of the form ati where r,s, and r are any real numbers. . ‘We now solve a linear system both by Gaussian elimination and by Gauss~ Jordan reduction, HRI Coser te tinea system x+2y+32= 6 x Sy $225 14 But yo r= 2 We form the augmented matrix 1203 [Alb]=]2 -3 2 Add (~2) times the first row to the second row: 1203 [ajo] -|3 “1-4 meen Lab at Add (—3) times the first row to the third row: 12 [4 | blonen-_ =| 0 -7 Saits=n [9 -3102 Chapter 2 Solving Linear Systems Multiply the third row by (4) and interchange the second and third rows: Add 7 times the second row to the third row: 12 [Afbloninse =]O 1 pis 0 0 Bikan Multiply the third row by 3; ‘This matrix is in row echelon form, This means that ¢ row, 3, and from the second pred so that y=4-2) From the first row, w+ 2y4+32=6, which implies that =6-2y-32 =6-2(-2) 33) = Thus x = 1, y = —2, and z = 3 is the solution. This gives the solution by Gaussian elimination. ‘To solve the given linear system by Gauss—Jordan reduction, we transform the last matrix to [C jd], which is in reduced row echelon form, by the following steps ‘Add (~2) times the third row to the second row: 1 [1 bontnons = [a 0 2 1 0 Now add (—3) times the third row to the first row: 2 1 0 :2.2. Solving Linear Systems 103 Finally, add (—2) times the second row to the first row: 0 oft 101-2 0 113 2, and z = 3, as before. . [CfA] on tes ee ‘The solution is x = Remarks. 1. As we perform elementary row operations, we may encounter a tow of the augmented matrix being transformed to reduced row echelon form whose first entries are zero and whose m + 1 entry isnot zero. In this case, we can stop ‘our computations and conclude that the given linear system is inconsistet. 2. In both Gaussian elimination and GaussJordan reduction, we ean use only row operations. Do not try to use any column operations. 1 Applications Linear systems arise in a great many applications. In this section we look at several of these. Quadratic Interpolation ‘Various approximation techniques in science and engineering use a parabola that passes through three given data points {(21, yi), (2, 92), (#3, Ys), where xy # xy for j. We call these distinet points, since the x-coordinates are all different. ‘The graph of a quadratic polynomial p(x) = ax?-+bx-+c isa parabola, and we use the given data points to determine the coefficients a, b, and cas follows. Requiring that p(x) = yi i = 1.2,3, gives us three linear equations with unknowns a, b, and c: ot axt tba + or adtbnbe=y e or ax? + bx + ys. Let apm tl A=|s}om 1 oa beth coin mt, ¥ = (‘| anty= [*} Then (2) ane ite ia ¢ > matrix equation form as Av = y whose augmented matrix on Lisle] ao xbox104 Chapier 2 Solving Linear Systems FIGURE 2.1 Bee We solve this linear system by Gaussian elimination or GaussJordan reduction, ‘obtaining values for a, b, and c. It can be shown that there is a unique solution to this linear system if and only ifthe points are distinct. The construction of the parabola that matches the points of the given data set is called quadratie interpolation, and the parabola is called the quadratic interpolant. This process can be generalized to distinct data sets of m+ I points and polynomials of degree n. We illustrate the construction of the quadratic in the following example: Find the quadratic interpolant forthe three distint points {(1, ~5). (1, 1), 2,7) Solution Setting up linear system (3), we find that its augmented matrix is (verify) Solving this linear system, we obtain (verify) a ‘Thus the quadratic interpolant is p(x) = Sx? — 3x ~ 7, and its graph is given in Figure 2.1. The asterisks represent the three data points. . 20, 0 ‘Temperature Distribution A simple model for estimating the temperature distribution on a square plate gives rise to linear system of equations. To construct the appropriate linear system, we use the following information: ‘The square plate is perfectly insulated on its top and bottom so that the only heat flow is through the plate itself. The four edges are held at various temperatures. To estimate the temperature at an interior point on the plate, we use the rule that it is the average of the temperatures at its Four compass-point neighbors, to the west, north, east, and south. Estimate the temperatures 7 oon the plate shown in Figure 2.2. 1,2,3,4, at the four equispaced interior pointsFIGURE 2.2 2.2. Solving Linear Systems 108 Solution ‘We now construct the linear system to estimate the temperatures. The points at which we need the temperatures of the plate for this model are indicated in Figure 2.2 by dots. Using our averaging rule, we oblain the equations 60 +1004 T+ Ts qh fie or -Ti +47 - T=140 The angone¢ mati is near ems) en ipie{oh 48 140 ibl=| oy 0 4 <1) oO -1 -1 40. Using Gaussian elimination or GaussJordan reduction, we obtain the unique solution (verify) 35° . 1 =65°, Ty 60°, Ty = 40°, and Global Positioning System A Global Positioning System (GPS) is a satellite-based global navigation system ‘enabling the user to determine his or her position in 3-limensional coordinates, without the need for further knowledge of navigational calculations. It was developed by the military as a locating utility, and the GPS system operated by the USS. Department of Defense became operational in 1995. GPS technology has proven to be a useful tool for & wide variety of civilian applications as well and is now available in low-cost units. These units have been incorporated into boats, au- tomobiles, airplanes, and handheld units available for general use such as hiking. GPS is based on satellite ranging, that is, calculating the distances between a receiver and the position of three or more satellites (four or more if elevation is desired) and then applying some mathematics. Assuming that the positions ofthe satellites are known, the location of the receiver can be calculated by determining the distance from each of the satellites to the receiver. GPS takes these three or more known references and measured distances and “trilaterates” the position of the receiver. Trilateration is a method of determining the relative position of an object, in this case, orbiting satellites. For GPS calculations, there are three position variables, x, y, and z, together with a fourth variable, 1, time. Time rust be considered, since the GPS receiver processes signals from the satellites to determine the distances involved. Even though the signals move at the speed of light, there are small time delays for transmission, together with other Factors like atmospheric conditions, that must be accounted for to ensure that accurate data are gathered. In this brief discussion of GPS we will use a simplified model to106 Chapter 2 Solving Linear Systems FIGURE 2.3 show how linear systems of equations enter into the mathematics that is part of the clever model involved in GPS. For “teal” GPS, we need to think in terms of three dimensions. In this con- text each satelite is represented by a sphere, and we need four spheres so that the location of the receiver can be determined by computing an intercept of the spheres; that is, a single point of intersection of the spheres. For our discussion, ‘We will think in two dimensions and consider thee satellites, each represented by circle for which the coordinates of the center are known. We will assume that cour GPS receiver can determine the distance between its location and the position of the satellite; thus, the radius of each circle is also known, Then, algebraically, wwe have the equations of three circles as shown in (4), where circle j has center (a,j) and radius r, for j = 1.2.3 (a+b am) +O - by = (x ay) + = bs? By construction, the location of the GPS receiver is on the circumference of each circle, so we are guaranteed that there is point (x,y) that satisfies each equation in (4). Itis this point that will provide the coordinates of the GPS receiver. In Figure 2.3 we illustrate the system ofthe (nonlinear) equations in (4) [Why is (4) rota system of linear equations? "A question that arises is, How do we solve the system of equations in (4), since they ate not linear equations? The answer is, We first expand each equation and then eliminate the terms that contain x* and y* by using algebra. In (5) we show the expansion of each of the equations in (4); note that a and y? appear in ceach equation 2 ® : 3 x2 Dae bal ty? — by + BR = x? = Daze +a} + y? = Way + BB x —lagx tal+y?—Why +h = ‘Now we rearrange each equation in (5) to obtain the expressions shown in (6). ©) xP 2a tal +? — Diy +87 Daye +43 +)? — Dry +83 x? —2asx tah ty? — sy +B} ‘Next, set the left side of the first equation in (6) equal to the left side of the second ‘equation in (6) and simplify. Do likewise forthe second and third equations in (6). This gives the linear system in x and y in (7). ~2ayx +a? = Ibyy +b} — 7}? = Dae + a — Poy +E = 7} © o vtast +02 —2byy +8 = Dax $d Day +6R— Finally, collect ike terms in and y to get the equations in 8) Hay = aa)x— 2b = body = FD + ah) + EO) og Das — an)x ~ 2(bs — bady = (FF — 73) + (a — a3) + — 6)2.2. Solving Linear Systems 107 ‘To simplify a bit further, we multiply each equation in (8) by —1 and show the ‘matrix formulation for the resulting 2 2 system in 9). ay — a2) 2b) ~ by) ] [x] _ [P+ Gad + G- ] 2as—a2) 2bs—br) || y. (rj — 72) + (a = a3) + (05 — by) So, given the coordinates of the centers and the radii of the three circles, we can ‘determine the location of the GPS receiver in the two-dimensional model ‘The coordinates of the centers and the rail of three circles are shown in Table 2.1 ‘The corresponding system of equations is given in (10) (verify), and its solution is x = 6 and y = 10 (verify). Thus, the coordinates of the GPS receiver in the two-dimensional system for these three circles is (6, 10). TABLE 2.1 Cirete 1 2 3 13,34) 2s 28 104] fx] _ [972 [7 )0)-[%] " [Next we present an approach for GPS in three dimensions. In this case each of the equations in the system that is analogous to those in (4) has the form (x aj)? + (y —by)* + ( — c))* = Gistance from receiver to satellite j ay for j = 1,2, 3,4, where (a;, b;,c3) isthe position of the satellite j.. The distance from the receiver to satellite j is computed by measuring the time it takes the signal from satellite j to reach the receiver. The satellite contains a clock mechanism that sends the time the signal was sent to the receiver, and we let r be the time the signal was received. Since the signal travels atthe speed of light, we can get @ ‘good approximation to the distance by using the basic formula distance = speed elapsed time. Thus there are now four unknowns: x, y, z, and r. We proceed algebraically as we did to get expressions that are analogous 0 those in (7) and (8). This wall yield a system of three equations in four unknowns analogous to the system in (9). We solve this system for r, y, and z in terms of time f. To determine the unknown f, we substitute these expressions for x,y, and z into any of the ‘equations in (11) and then solve the resulting quadratic polynomial for 1. Finally, ‘we use the resulting value of tin the expressions for x, y, and z to determine the location of the receiver in three dimensions. This approach uses a system that has infinitely many solutions and then cleverly uses the underlying physical situation to determine the “free” variable t108 Chapter 2. Solving Linear Systems Theorem 2.4 ‘The real-life situation is even more complicated than our approach outlined for the thtee-dimensional case. The satellites are continually moving, so their lo- ‘cations vary with time, inherent errors in time calculations creep in, and a number of other factors introduce more inaccuracies. Highly accurate estimation of the receiver’s position is beyond the scope of this course, but there are many books and discussions on the Internet that provide more detailed information, We have presented a basic component of GPS calculations, namely, that linear systems of ‘equations are involved. IM Homogeneous Systems Now we study a homogeneous system Ax = 0 of m linear equations inn unknowns, Consider the homogeneous system whose augmented matrix is, oe core ese eofeN Since the augmented matrix is in reduced row echelon form, the solution is seen tobe ay =2r mes where r and s are any real numbers. . In Example 10 we solved a homogeneous system of m (= 4) linear equations in m = 5) unknowns, where m <7 and the augmented matrix A was in reduced, row echelon form. We can ignore any row of the augmented matrix that consists entirely of zeros. Thus let rows 1, 2,.... of A be the nonzero rows, and let the 1 in row f occur in column ¢;. We are then solving @ homogeneous system of r equations in m unknowns, r n. Consider the homogeneous system xb ytotw ‘The augmented matrix is row equivalent to (verify) Hence the solution is any real number, . @ Application: Chemical Balance Equations ‘Chemical reactions can be described by equations. The expressions on the left side are called the reactants, and those on the right side are the products, which are produced from the reaction of chemicals on the left. Unlike mathematical equations, the two sides are separated by an arrow, either —>, which indicates that the reactants form the products, or <, which indicates a reversible equation; that is, once the products are formed, they begin to form reactants. A chemical equation is balanced, provided that the number of atoms of each type on the let isthe same as the number of atoms of the corresponding type on the right. In Example 12 ‘we illustrate how to construct « homogeneous system of equations whose solution provides appropriate values to balance the atoms in the reactants with those in the products Sodium hydroxide (NaOH) reacts with sulfuric acid (HS0,) to form sodium sul- fate (NaSO,) and water (HO). The chemical equation is, NaOH + H2S0, > NaSO, + H20.110 Chapter 2 Solving Linear Systems ‘To balance this equation, we insert unknowns, multiplying the chemicals on the left and right to get an equation of the form xNaOH + yHpSO, + zNaSO, + wH,O. Next, we compare the number of sodium (Na), oxygen (0), hydrogen (H), and sulfur (S) atoms on the left side with the numbers on the right. We obtain four linear equations: Nar x= 2e Orx bay Hix +2y Sty=z Observe that we made use of the subscripts because they count the number of atomts of a particular element. Rewriting these equations in standard form, we see that we have a homogeneous linear system in four unknowns: x = xtay—4e— w x2y 0 = 2w yor ‘Writing this system in matrix form, we have the augmented matrix 10-2 010 14-4 -1)0 12 0-240 0 1-1 00. ‘The reduced row echelon form is 10 “110 on tio o 0 1 bio} 0 0 0 0!0 and the solution is x = w, y = 4 w. Since w can be chosen arbitrarily and we are dealing with atoms, it is convenient to choose values so that all the unknowns are positive integers. One such choice is w = 2, which gives y= Land z= 1. In this case our balanced equation is 2NaOH + HSO, + NaySO, + 2H20. .2.2 Solving Linear Systems 111 Mf Relationship between Nonhomogeneous Linear Systems and Homogeneous Systems Let Ax = b, b 4 0, be a consistent linear system, If x» is a particular solution to the given nonhomogeneous system and xy, is a solution to the associated homogeneous system AX = 0, then x» + xj is a solution to the given system Ax = Moreover, every solution x to the nonhomogeneous linear system Ax = b can be written as x, -+x), where X, is a particular solution to the given nonhomogeneous system and x, is a solution to the associated homogeneous system Ax = 0. For a proof, see Exercise 29, Solving Linear Systems with Complex Entries Gaussian elimination and Gauss-Jordan reduction can both be used to solve linear systems that have complex entries. The examples that follow show how to solve linear system with complex entries by using these solution techniques. (For simplicity, we do not show the notation for row operations.) Further illustrations and exercises are given in Appendix B.2. Solve the linear system (-dx+ Q+dy=2421 Qe + (= 2y S143 by Gaussian elimination, Solution ‘The augmented matrix of the given linear system is i 2+i [2421 2 1am 143% To transform this matrix to row echelon form, we first interchange the two rows (Co avoid complicated fractions), obtaining 123i 2421 [iia [2.8] Multiply the first row by 3 Add —(1 ~ i) times the first row to the second row:112 Chapter 2 Solving Linear Systems Multiply the second row by = + 5 : ‘Then G2 Imi 0 2% 24: ow 4 by Gauss-Jordan reduction, Solution “Multiply the first row by AddiKey Terms ‘Gaussian elimination ‘Gauss-Jordan reduction Homogeneous system EER exercises aut ( 2+i third row to the first row: 22) sth ho sam ot 2 Multiply the third row by ( ie Hence the solution is ‘Back substitution (Quadratic interpolation (Quadratic interpotant 1. Each of the given linear systems is in row echelon form, Solve the system, (@ xt2y—2 yhras r=4 6b) x= 3y tact 2. Bach of the given linear systems i in row echelon form Solve the system. @ xty-zt2wa4 w= Wb) xyes ys 43. Each of the given linear systems is in reducod row echelon form, Solve the system. 2.2. Solving Linear Systems 113 2 1 0 times the . Global positioning system Chemical balance equations (@) xty Each of the given linear systems isin reduced row eche~ Jon form. Solve the system. @ x -2=5 tb) x yt r=? 3. Consider the linear system at yt214 Chapter 2. Solving Linear Systems (a) Find.all solutions, itany exist, by using the Gaussian climination method. (b) Find all solutions, if any exist, by using the Gauss— Jordan reduction method, Repeat Exercise $ for each of the following finear sys @) x4 yt2e +3023 xd ch w= 8 Bet yb rw o) © 1 2 -1 2 Sro yt2e—8w= 3 In Exercises 7 through 9, solve the linear system, with the given augmented mari, iit is consistent 10. au it 2340 @ jit 1100 on 12/0. 12 a fii pate 57 zn 12 @ | 13 Lo 1 w fo 2 1 oo b2it 12 110 eed 23 010 @ |i 0 21s) M5 7 ao 623i o 1 2}0 21 410 : Find a2 1 matix x with entries not ll zero such that a oi ss sto [f 1 Hint: Rewrite the matrix equation Ax = 4x as 4x — AX = (Alp ~ A) = O, and solve the homogeneous, linear system] Find a 2 x 1 matrix x with entries not all zero such that a a ss ote [2 4 2 13. my 15, 16. 19. Find a 3 x 1 matuix x with enties not all zero such that boa -t rood 4-45 Find a 3 x 1 matrix x with entries not all zero such that 12-1 101 4-45 In the following lineur system, determine all values of a for which the resulting linear system has Ix, where Al {@) no solution: (0) unique solution; (6) infinitely many solutions: xhye z=? abo 2=3 xt yt@-Seoa Repeat Exercise for the linear system xtoyt z ast Qe By @ = De Repeat Exercise 14 forthe linear system xboyt xblyt xt yt? ab wf] Show that the linear system AX solution if and only if ad — be # 0. has only the trivial Show that A = |@ only if ad ~ be #0, °OF (i) "ea munnef 22, Let fs RP > Ri be the matrix transformation defined *(G)-E4 Find an equation relating a, b, and c so that we can al- ‘ways compute values of x, y, and z for which Find x,y 23, Let f: R¥ > R® be the matrix transformation defined “Oe Find an equation relating a, b, and ¢ so that we can al- ‘ways compute values of x,y, and z for which Exercises 24 and 25 are optional 24, (a) Formulate the definitions of column echelon form ‘and reduced column echeton form of a matrix, (b) Prove that every m x 7 matrix is column equivalent ‘toa matrix in column echelon form, 2.2. Solving Linear Systems 115 25, Prove that every m x mattix is column equivalent to a “unique matrix in reduced column echelon form. 26, Find an equation relating a, b, and c so that the tinea system x+2y—32 De 439432 Sr 499-62 is consistent for any values of a,b, and that satisfy that ‘equation. 27, Find an equation relating a, b, and ¢ so that the linear system De Dy + Br- y 452 ay +22 is consistent for any values of a,b, and c that satisfy that ‘equation, 28, Show that the homogeneous system ant crt bn) thas a nontrivial solution if and only i satisfies the equa tion (a = r)( =r) = cd = 29, Let Ax =b,b 4 0, bea consistent linear system. (a) Show that if x, isa particular solution to the given nonhomogeneous system and xy is & solution to the associated homogeneous system Ax = @ then Xp +X; isa solution to the given system AX = b. (b) Show that every solution x to the nonhomogeneous linear system Ax = b can be written as x, +X, ‘where x, is a particular solution tothe given nonho- _mogeneous system and x; is solution tothe ass0~ ciated homogeneous system Ax = 0. [Hines Let x = Xp + (= Xp)d Determine the quadratic interpolant to each ofthe given data sets Follow the procedure in Example 7. fa) (0.2), 0,5),2,19) (b) (-1,2),.14). 0D) 31. (Calculus Required) Construct a linear system of equations to determine a quadratic polynomial p(x) = ax? b-e that satisfies the conditions p(0) = f(0), p'O) = F°@), wad p"(O) = f"(0), where f(x) = e* 3% (Caleulus Required) Construct a linear system of equations to determine a quadratic polynomial p(x) = ax? + bx be that satisfies the conditions p(1) = f(1), p’Q) = P(),and p"() = FC), where f(x) = xe116 Chapter 2. Solving Linear Systems 38, Determine the temperatures at the interior points 7; 1, 2, 3, 4 for the plate shown in the figure. (See Example 8) 30° cg so" 34, Determine the planar location (x, y) of a GPS receiver, using coordinates of the centers and radii forthe three circles given in the following tables (0 errr Cirle Center Radius 1 15,20) 5 2 6-12) B 3 (9,40) 41 00) rrnerenerrrenanat Cirele Center ——Radius 1 (10, 13) 25 2 «19, -19) B 3 14,3) 41 135. The location of a GPS receiver in a two-dimensional sys- emis (~4, 3). The data used in the calculation are given in the table, except thatthe radius of the first circle is missing. Determine the value of the missing pices of ata Circle Center Radius 1 (16,38) 2 2 37) o 3 G2, 80) 35 Girele Center ‘Radius 1 oy ” 2 (10,5) - 3 8,3) B 37. Suppose you have a “special edition” GPS receiver for ‘two-dimensional systems that contains three special but tons, labeled CI, C2, and C3. Each button when de~ prested draws a circle that corresponds to data received from one of three closest satellites. You depress button ‘Cl and then C2. The image on your handheld unit shows, «pair of cireles that are tangent to cach other. What is the location of the GPS receiver? Explain [38,_ Rusti formed when there isa chemical reaction between iron and oxygen, ‘The compound that is formed isthe red- dish brovin scales that cover the iron abject, Rust is iron ‘oxide whose chemical formula is Fe;0s, So a chemical equation for rusts, Fe +O: > Fe.0s, Balance this equation 39, Ethane i a gas similar to methane that burns in oxygen to tive carbon dioxide gas and steam, The steam condenses to form water droplets. The chemical equation for this, reaction is Cally + 03 + CO; +140. Balance this equation. In Exercises 40 and 41, solve each given linear system. 4. (-)x+ O+2y=1 (42 + (24 2D =i Morty 33 inty+ 2=3 yt In Bxercises 42 and 43, solve each linear system whose aug- ‘mented matrix is given. mere, efit) Sto Lod wef 42i aja i 2) 2 12 4) 2 36. The location of a GPS receiver in a two-dimensional sys- "E44, Determine whether the software you are using has & com tem is (6, 8). The data used in the calculation are given in the table, except that the radii of circles 1 and 2 are missing, Determine the values of missing pieces of data, ‘mand for compating the reduced rows echelon form of ‘a matrix, IF it does, experiment with that command on. ‘some ofthe previous exercises.4 46. Determine whether the software you ae using has a command for computing interpolation polynomials, given a set of ondered pairs. IF it does, use the command to de termine the quadratic interpolant for the datasets in Ex- excise 30. Determine whether the software you are using has a graphing option as part of a command for computing in- \erpolation polynomials, or if there isan easy way aval able to graph the interpolant. I it does, use its follows (a) Generate the graphs for the quadratic interpolants forthe data sots in Exercise 30. Prin out the graphs, 2.3 Elementary Matrices; Finding A™ o) 47 and then mark the dst poins on For the data st ((0, 0), (1, 1s (2), generat the quadratic interpolant and raph over the interval U0, Printout the graph and then mark the data ons onthe graph, ‘This data set isa sample of the function y = f(x) = V¥. Carefully sketch ‘he graph of on the printout making sure i goes through the datapoints Brie discuss the eror that soul he incurred i you were o evaluate the inte Polat at r= Dand «= 3 estimate V2 and V3, respectively raph, In this section we develop a method for finding the inverse of a mattix if it exists. To use this method we do not have to find out first whether A exists. We art to find A~ if in the course of the computation we hit a certain situation, then we know that A~' does not exist. Otherwise, we proceed to the end and obtain A-. This method requires that elementary row operations of types 1, Il, and IIL (see Section 2.1) be performed on A. We clarify these notions by starting with the following definition: Pein eed Oe Ann x n elementary matrix of type I, type II, or type TIT is a matrix obtained from the identity mairix 1, by performing a single elementary row or elementary column operation of type I, type Il, or type IIL, respectively. ‘The following are elementary matrices: 1 + and y=] 0 0 ‘Matrix £) is of type I—we interchanged the first and third rows of Is; E2 is of type I—we multiplied the second row of /; by (~2); Es is of type III—we added towice the second row of J; to the first row of Zs; and E, is of type IIl—we added three times the first column of Zs to the third column of J . Theorem 2.5 Lot A be an m > 1 matrix, and let an elementary row (column) operation of type L, type Hor type IIl be performed on A to yield matrix B. Let F be the elementary matrix obtained from In (Jp) by performing the same elementary row (column) operation as was performed on A. Then B Proof Exercise 1. EA(B = AB),118 Chapter 2 Solving Linear Systems Bie Theorem 2.6 Theorem 2.7 ‘Theorem 2.5 says that an elementary row operation on A can be achieved by premultiplying A (multiplying A on the left) by the corresponding elementary matrix E; an clementary column operation on A can be obtained by postmultiplying A (multiplying A on the right) by the corresponding elementary matrix. “Lt and let B= A_ory en) oeys then 5 3 0 B=|-1 2 a]. 300 2 Now let E = (J)-2e,)r)-on,3 then Let bee We can readily verify that B= BA. . If A and B are m xn matrices, then A is row (column) equivalent to B if and only there exist elementary matrices E;, E,..., By such that B = ExE, 1+ EEA (B= AE Ex- Ev E)- Proof We prove only the theorem for row equivalence. If A is row equivalent to B, then B results from A by a sequence of elementary row operations. This implies that there exist elementary matrices E}, B3,-.-, Bx such that B = EyEy1+-» E2EiA. Conversely, if B = Ex Ey-1 + E2EiA, where the E; are elementary matrices, then B results from A by a sequence of elementary row operations, which implies, that A is row equivalent to B. . An elementary matrix £ is nonsingular, and its inverse is an elementary matrix of the same type. Proof Exercise 6. . ‘Thus an elementary row operation can be “undone” by another elementary row operation of the same type. ‘We now obtain an algorithm for finding A~! if it exists; first, we prove the following lemma:Lemma 2.1° Theorem 2.8 Corollary 2.2 2.3. Elementary Matrices; Finding A’ 119 Let A be ann x n matrix and let the homogeneous system Ax = 0 have only the trivial solution x = 0. Then A is row equivalent to J,. (That is, the reduced row echelon form of is f.) Proof Let B be a matrix in reduced row echelon form that is row equivalent to A. Then the homogeneous systems Ax = 0 and Bx = 0 are equivalent, and thus Bx = 0 also has only the trivial solution. Its clear that if ris the number of nonzero rows of B, then the homogeneous system Bx = 0 is equivalent to the homogeneous system whose coefficient matrix consists of the nonzero rows of B and is therefore 1rx n. Since this last homogeneous system has only the trivial solution, we conclude from Theorem 2.4 that r 2 n. Since B ism x n,r , ..., Ex, then A = By B2-- By. Now each elementary matrix is nonsingular, and by Theorem 1.6, the product of nonsingular matrices is nonsingular; therefore, A is nonsingular. Conversely, if 4 is nonsingular, then Ax = Oimplies that A~!(Ax) = 0, s0 [4x = 0 or x = 0. Thus Ax = 0 has only the trivial solution. ena 2 then implies that 4 is row equivalent to Jy. This means that there exist elementary matrices E}, B2,..., Be such that Ey Ey E2BiA. Icthen follows that A = (Ey Bey +++ EaE,)~! = Ey 'Es--- By! Bp. Since the inverse of an elementary matrix is an elementary matrix, we have established the result, . I Ais nonsingular if and only if A is row equivalent to fy. (That is, the reduced row echelon form of A is y.) Proof If A is row equivalent to Jy, then In = ExEx-1--+ BxEvA, where Ei, E,.... Ex are elementary matrices. Therefore, it follows that A = E;'E3"--- Bj", Now the inverse of an elementary matrix is an elementary matrix, and so by Theorem 2.8, A is nonsingular. Conversely, if A is nonsingular, then A is a product of elementary matrices, A= EyEy-+ Fok}. Now A= Aly = EpBy-1~-> ExE yy, Which implies that A is row equivalent to Ip 7 ‘We can see that Lemma 2.1 and Corollary 2.2 imply that if the homogeneous system AX = 0, where A is m x n, has only the trivial solution x = 0, then A nonsingular. Conversely, consider Ax = 0, where A is n x m, and let A be nonsingular. Then A~' exists, and we have A~!(Ax) = A~"0 = 0. We also have A(x) = (A“!A)x = 1, = X, So X = O, which means that the homogeneous FA Tera isa theorem thas established forthe purpose of proving another theorem,120 Chapter 2 Solving Linear Systems Theorem 2.9 system has only the trivial solution. We have thus proved the following important theorem: ‘Tho homogeneous system of n linear equations in m unknowns AX = 0 has a nontrivial solution if and only if A is singular. (That is, the reduced row echelon form of A # In.) 7 [: j] Consider the homogeneous system Ax = 0; that is, b aIE]-b) ‘The reduced row echelon form of the augmented matrix is [0 10] Let A (verify), 50 a solution is xed y where r is any real number. Thus the homogeneous system has a nontrivial solution, and A is singular, . In Section 1.5 we have shown that if the n x n matrix A is nonsingular, then the system Ax = b has @ unique solution for every mx 1 matrix b. The converse Of this statement is also true. (See Exercise 30.) [Note that at this point we have shown thatthe following statements ate equivalent for an m xm matrix A: 1. Ais nonsingular, 2, Ax =0 has only the trivial solution, 3. Ais row (column) equivalent to J,. (The reduced row echelon form of A is In) 4. The linear system Ax = b has a unique solation for every m x 1 matrix b, 5, Aisa product of elementary matrices. ‘That is, any two of these five statements are pairwise equivalent. For exam ple, statements 1 and 2 are equivalent by Theorem 2.9, while statements 1 and 3 are equivalent by Corollary 2.2. ‘The importance of these five statements being equivalent is that we can always replace any one statement by any other one on the list. As you will see throughout this book, a given problem can often be solved in several alternative ways, and sometimes one procedure is easier to apply than another.Sue 2.3. Elementary Matrices; Finding A' 121 Finding A“! At the end of the proof of Theorem 2.8, A was nonsingular and EgBxy-- ExEy ‘This now provides an algorithm for finding A~!, Thus we perform clemer tary row operations on A until we get fy; the product of the elementary matic Ey Ej «+ EE) then gives A~!, A convenient way of organizing the computing process isto write down the partitioned matrix [A | J, }. Then (Be Exa +++ E2E,)[A | In] =[ExEx-1 + E2Bi A | ExEy1--- EE} =[b a7] That is, for A nonsingular, we transform the partitioned matrix [A { [to reduced row echelon form, obtaining [J, | A~*} Let Assuming that A is nonsingular, we Form tit 0 0 [atpj=jo 2.3 1 0 soso o1 ‘We now perform elementary row operations that transform [4 {J)] to [15 |A~!J; wwe consider [A | /;] as a3 x 6 matrix, and whatever we do to @ row of A we also do to the corresponding row of fs. In place of using elementary matrices directly, ‘we arrange our computations, using elementary row operations as follows:122 ‘Chapter 2 Solving Linear Systems. ‘Theorem 2.10 A b raid 0 0 Apply Sey brs ms 0203 10 sos ot oo 11 00 Apply try rs 0 23 10 0 0-4 ou rod 10) 0 Apply—trs ms oo en) oo 01 rot 10 0 Apply Sry +r) + mand oo ob oo STH 0 0 3 0-3 Por 0 fb OE Apply nin try re o 1 0 4 3 oo 1 $0 -! roo 44 Oo 1 0 i-Ro: G ou s 0 =} ‘We can readily ve ‘The question that arises at this point is how to tell when A is singular. ‘The answer is that A is singular if and only if A is row equivalent to matrix B, having at Jeast one row that consists entirely of zeros. We now prove this result. Ann xn matrix A is singular if and only if A is row equivalent to a matrix B that has a row of zeros. (That is, the reduced row echelon form of A has a row of e108.) Proof First, let A be row equivalent (0 a matrix B that has a row consisting entirely of zeros, From Exercise 46 of Section 1.5, it follows that B is singular, Now2.3. Elementary Matrices; Finding A” 123 B Ex Ey) +++ Ey A, where Ey, E2,..., Ey are elementary matrices. If A is nonsingular then B is nonsingular, a contradiction, Thus A is singular. Conversely, if is singular, then A is not row equivalent to /,, by Corollary 2.2. Thus A is row equivalent toa matrix B # 1,, which is in reduced row echelon form, From Exercise 9 of Section 2.1, it follows that B must have a row of zeros. . erence ‘This means that in order fo find A”', we do not have to determine, in advance, whether it exists. We merely start to calculate A~'; if at any point in the eom- ‘potation we find a matrix B that is row equivalent to A and has a row of zeros, then A“! does not exist. That is, we transform the partitioned matrix [| Jy] to reduced row echelon form, obtaining [C | D]. 16 = fp, then D = AIF C # 1, then C has a row of zeros and we conclude that A is singular. 7 1 3 Apply ~ny +12 + m, Let Apply ~5r +4) $n Apply ~3r: +85 > erjooruonuee oo -3) 4]. 0. the last matrix under A. Since B has a row of zeros, we stop and conclude that A i a singular matrix. : In Section 1.5 we defined ann x n matrix B to be the inverse of the mx 1 matrix A if AB = J, and BA = J, We now show that one of these equations follows from the other.124 Chopter 2. Solving Linear Systems Theorem 2.11 If A and B aren x n mattices such that AB = J, then BA = J}. Thus B = A~!. Proof We frst show that if AB = J, then A is nonsingular. Suppose that A is si gular, Then A is row equivalent to a matrix C with a row of zeros. Now C = E,By-1+-» EyA, where E;, E,..., Be are elementary matrices. Then CB Ey-1-+ E1AB, so AB is row equivalent to CB. Since CB has @ row of zeros, ‘we conclude from Theorem 2.10 that AB is singular. Then AB = /, is impossible, because J, is nonsingular. This contradiction shows that A is nonsingular, and so ‘A exists, Multiplying both sides of the equation AB = J, by A~! on the lef, we then obtain (verify) B= A . Remark Theorem 2.11 implies that if we want to check whether a given matrix B is A~!, we need merely check whether AB = J, or BA = I,. Thatis, we do not have to check both equalities Key Terms Elementary mati ‘Noosingular matrix Inverse matrix singular manic SEEM exercises a a _ 1. Prove Theorem 25 () How are A and B related? (Hint: Compute AB and 2. Let A bea 4 x 3 matrix. Find the elementary mats E BA) that, asa premultipier of A~that is, 8 EA~performs Set the following elementary row operations on A: too 1 (a) Mtipties the second rw of A by (2) ape (b) Adds 3 times the third row of A to the fourth row of - ! 4 (@) Find mats C in reduced row echelon form that e) inmrchanges the ist and third ows of A. so eet 64, Reso the operat Let A be a3 > 4 matrix. Find the elementary mates F ‘that, as a postmultiplier of A—that is, as "AF -performs (b) Apply the same operations to Zs that were used to the following elementary column operations on A: ‘obtain C. Denote the resulting matrix by B. (a) Adds (4) ines the fst column ofA tothe second (©) How are A and B related? (Hint: Compute AB and column of A BA) 6, Prove Theorem 2.7. Hint: Find the inverse of the ele erchanges the second and thin colunns of ())_Interchanges the second and third columns of 4 ‘mentary matrix of type I, type TI, and type IIL) (©) Multiples the thin column of by 4 ms 4 Lat 2. Finite imene of 4=[} 3] 100 a=|o10 123 201 8 Find theimerse of A=] 23 (a) Find a matrix C in reduced row echelon form that 124 is row equivalent to A. Record the row operations 9, Which ofthe given mates are singular? For the non- used. singular ones, find the inverse, (b) Apply the same operations to Zs that were used to 13 mit? obtain C. Denote the resulting matrix by B. |r 6 | 6123 123 @ ji 12} @lri2 012 ord 10, Invert each of the following matrices, if possible: 12-3 1 312 @) 3 (by }2 1 2 2015 33 3125 1 123 213 w fii 2) @lor2 110 103 11, Find the inverse, if t exists, of each ofthe following: w [pe] w fi 22 1 2 wf S23] 12, Find the inverse, if it exists, of each ofthe following: hora 0-2 0 0 malo 5s 0 302 1 trad 1301 2 ma=|p pt 5 9 1 6 [In Exercises 13 and 14, prove that each given matrix A is nonsingular and write it asa product of elementary matrices (Hint: irst, write the inverse a a product of elementary ma- ‘lees; then use Theorem 2.7.) wac[h3] macfoad 1S, 1A is «nosing nian whos mene is [7 T} find A hod we rat=|1 1 2] mda. be-nd 1. 18, Ww. a 2.3. Elementary Matrices; Finding A’ 125 Which of te following homogeneous systems have a sontival station? (a) r+2y+32=0 2y 4220 x4 2y432=0 @) xt ye so dyo ae ary 422 © drt y43e 2x + 2y = 42 2x —3y +5250 Which of the following homogeneous systems have a nontrivial solution? (@) ty +2 weyt z Br-yt 2 o) © exists, What is A™ For what values of @ does the homogeneous system @-Drt+ 2 a-Dy= have a nontrivial solution?” Prove that is nonsingular if and only it ad—be # 0. Ifthis condition holds, show that126 2 23, Chapter 2. Solving Linear Systems Lat 203-1 103 A=) oo 23 2 1B Find the elementary matrix that as @ postmultiplier of A performs the following elementary column operations on as (a) Multiples the third column of A by (~3) (b) Interchanges the second and third columns of A. {(e) Adds (~S) times the first column of 4 to the thind column of A. Prove that (wo m > matrices A and B are row equiv ‘lent if and only if there exists a nonsingular matcix P such that B= PA, (Hint: Use Theorems 2.6 and 2.8.) Let A and B be row equivalent m matrices, Show that if AB is non- ingular, then A and B must be nonsingular, (Hine: Use ‘Theorem 29.) 1 Let A and B be m xm matrices Let A bean m x1 matrix. Show that A is row equivalent 1 0 ifand only it A = 0. Show that A is row ‘equivalent to B if and only if is column equivalent to Br ‘Show hata square matrix which has a row or a coluran ‘consisting entirely of zeros must be singular. (a) Is(A+ By! =a + et? © wears tae ICA isan xm matsx, prove chat 4 is nonsingular ifand ‘only if the linear system Ax = b has a unique solution for every # x 1 matrix b. Prove thatthe inverse of a nonsingular upper (lower) t- angular matrix is upper (ower) triangular Ifthe software you use has a command for computing reduced row echelon fori, use it to determine whether the matrices Ain Exerises 9,10, and 11 have an inverse by Operating onthe matix [A | 7, (See Example 4) Repeat Exercise 32 onthe matrices given in Exercise 63 of Secon 15. EZ] equivalent Matrices ‘We have thus far considered A to be row (column) equivalent to B if B results from A by a finite sequence of elementary row (column) operations. A natural extension of this idea is that of considering B to arise from A by a finite sequence of elementary row or elementary column operations. This leads to the notion of equivalence of matrices. The material discussed in this section is used in Section 49. ee) If A and B are two m xn matrices, then A is equivalent to B if we obtain B from A by a finite sequence of elementary row or elementary column operations. AS we have seen in the case of row equivalence, we can show (see Exercise 1) that (a) every matrix is equivalent to itself; (b) if B is equivalent to A, then A is equivalent to B; (c) if C is equivalent to B, and B is equivalent to A, then C is equivalent to A. In view of (b), both statements “A is equivalent to B” and “ is equivalent to A” can be replaced by “A and B are equivalent.” We can also show that if two matrices are row equivalent, then they are equivalent. (See Exereise 4.) Theorem 2.12 the form If A is any nonzero m ¢; * Replace column j by & times column i + column j: ke; +.¢; > ¢ Let 11 2-1 12 1 0 -1 -4 1-2 1-205 4, To find a matrix of the form described in Theorem 2.12, which is equivalent to A, ‘we proceed as follows. Apply row operation —Lry +r —> rp to obtain 1 ot 2-1 o 1-4 1 Apply tei +45 = 3s. -1 -4 1 2 ply te + 1-205 -4 11 2-1 0 1-tt 0-33 3 Apply 0-33 3128 Chapter 2. Solving Linear Systems Theorem 2.13 Pot o2-n o1-att ‘ 03033 Apply hfs lo o 0 0, boi 2-1 ora ota Apply rs 0 0 0 0 tot 2-19 o 1rd o 0 0 0 Lo 0 0 0, 1 0 3-2 ee a ooo 0 Apply —3ey bes > es 0 0 0 0 1 0 0-2 o rat ar Apply 2¢s +e + es o 0 0 0, roo 0 o o 1-11 . 00 0 Apply les +e — €s 0 0 0 0 10 0 0 o 101 $0000 Apply —ley es + ey o 0 0 of 10 0 0 o 10 0 o 0 0 0 o 0 0 0, ‘This is the matrix desired. . ‘The following theorem gives another useful way to look at the equivalence of matrices: ‘Two m x n matrices A and B are equivalent if and only if B = PAQ for some nonsingular matrices P and Q.2.4 Equivalent Matrices 129 Proof Exercise 5. . ‘We next prove a theorem that is analogous to Corollary 2.2. Theorem 2.14 Ani x 1 matrix A is nonsingular if and only if A is equivalent to 1, Proof, IFA is equivalent to ,, then [, arises from A by a sequence of elementary row or clementary column operations. Thus there exist elementary matrices By, E>...» Ep, Fi, Pay -++ Fy such that Iy = Ey Ey- 0 EXER Pa Fy Let EB, ++ EpE, = P and Fi Fz-+- F, = Q. Then J, = PAQ. where P and Q are nonsingular. It then follows that A = P~'Q~!, and since P! and Q~! are nonsingular, A is nonsingular. Conversely, if A is nonsingular, then from Corollary 2.2 it follows that A is row equivalent to [,. Hence A is equivalent to Ip . Key Terms ‘Equivalent matrix Paritioned matix Exercises - 1. (4) Prove that every mates Ais equivalent wo itselt lo aa 34 122 (b) Prove that if B is equivalent to A then A i equiva @fi oo 2 3) mil 24 lento B. 3481 5 5 8-1 (©) Prove thatif is equivalent to B and B is quivalent ba 4 tA, then Cis equivalent to A. Pata 14 2, For each of the following matrices, find a matrix of the @ {2-2 1 ot @ {i -1 0 form described in Theorem 2.12 that is equivalent othe 4250 o 1 2 sven matrix 403 3-1 12-1 4 4, Show tha if A and B are row equivalent, then they are 512-3 124 equivalent @ |2 bt 4 3 wb y2 3 1 5. Prove Theorem 2.13. 20 12 201 3. 5 tot 6 Let [: 2 ;| 13-1 2 112 1-24 2-4-2 1 ot oft) offi 3. Repeat Exercise 2 for the following matices: Find a matrix B of the form described in Theorem 2.12 that is equivalent to A. Also, find nonsingular matrices P and Q such that B = PAQ. 73205130 Chapter 2 Solving Linear Systems 7. Repeat Exercise 6 for 40, For each of the following matriees A, find a matrix B # A that is equivalent co Az baa ae ae wafiod “pea tat waft fa) manly ott Las as fh taAtemm nna shovbstasemioen — @ Aap ota 02 Gift 0 3a 9, Let Aand B bem xn mattices. Show that A isequivalent 11, Let A and B be equivalent square matrices, Prove that A to B if and only if AT is equivalent to 8" i nonsingular if and only if B is nonsingular. E24 cv-Factorization (Optional) In this sation we discus variant of Gaussian elimination presented in Section 2.2) that decomposes a matrix as a product of a lower triangular matrix and an hppet angular mathx. This decomposition leads to an algo for solving fa linear system Ax = b that is the most widely used method on computers for ‘solving a linear system. A main reason for the popularity of this method is that it provides the cheapest way of solving a linear system for which we repeatedly have tochange the ight sds “This type of situation occurs often in aplied provlems, Ferexumpleanelectc lity company most etomine te inputs ie unknowns) needed o produce some required outputs ie right ies). The imps and outputs might be related by a linear system, whose coefficient matrix is fixed, while the right side changes from day to day, or even hour to hour, The decomposition discussed in this section is also useful in solving other problems in linear algebra. When 0 isan upper angular matrix all of whose diagonal ene are der: ‘ent from zero, then the linear system Ux = b can be solved without transforming, the augmented matrix [|b] to reduced row econ form orto row echelon form The augmented max of such a syst i given By mom ts 0 um ua 0 0 uss 0 0 0 ‘The solution is obtained by the following algorithm: be eae By = nnn2.5 lW-factorization (Optional) 131 This procedure is merely back substitution, which we used in conjunction with Gaussian elimination in Section 2.2, where it was additionally required that the diagonal entries be 1. In a similar manner, if L is @ lower triangular matrix all of whose diagonal entries are different from zero, then the linear system Lx = b can be solved by forward substitution, which consists of the following procedure: The augmented matrix has the form fy 0 0 fy 0 fa ba bs . far bon bs and the solution is given by as *= a be ae by - Vitae 42S vn ‘That is, we proceed from the first equation downward, solving for one unknown from cach equation. We illustrate forward substitution in the following example: ‘To solve the linear system Sx 4x; — 2x 2x; + 3x2 + Ay. ‘we use forward substitution, Hence we obtain from the previous algorithm 10 26 2x - 3x2 wy = POSE EL 3, 4 hich implies that the solution to the given lower triangular system of equations132 Chapter 2. Solving Linear Systems [4] . As illustrated in the discussion at the beginning of Section 2.2 and Example 1 of this section, the ease with which systems of equations with upper or lower triangular coefficient matrices can be solved is quite attractive. The forward substitution and back substitution algorithms are fast and simple to use. These are ‘used in another important numerical procedure for solving linear systems of equations, which we develop next. ‘Suppose that ann x n matrix A can be written as a product of a matrix L in Jower triangular form and a matrix U in upper triangular form; that is, A=LU. In this case we say that A has an LU-factorization or an LU-decomposition. The LU-factorization of a matrix A can be used to efficiently solve @ linear system Ax = b, Substituting LU for A, we have (Lu)x=b, or by (a) of Theorem 1.2 in Section 14, L(Ux) =b. Letting Ux = 2, this matrix equation becomes, La=b. Since 1. is in lower triangular form, we solve directly forz by forward substitution. Once we determine 2, since U is in upper triangular form, we solve Ux = 2 by back substitution, In summary, if an n x m matrix A has an LU -fectorization, then the solution of Ax = b can be determined by a forward substitution followed by a back substitution. We ilustrate this procedure in the next example. Consider the linear system 6m 2x — Atha = 2 Bu -3n- 6 = 4 Hx, + 8m +21 - Bus 8 6x 10x + Ixy = -43 ‘whose coeflicient matrix 6-2 -4 4 3-3-6 1 “12 8 2-8 6 0-0 72.5. lWFactorization (Optional) 133 hhas an LU-factorization, where 1 0 0 0 62-4 4 $10 0 02-4 } - 2-2 1 of ™ Flo 9 5 13 0 0 0 8 (verify). To solve the given system using this LU-factorization, we proceed as follows. Let Then we solve Ax = b by wi = b, First, let Ux = 2 and use forward substitution to solve La = b: 0 0 O}ry 2 10 olfa}_} 4 2 1 offjal=] 8 1-2 iflel L- We obtain + 2c +222 eS AB 2 2 $205 Next we solve Ux = 2, 6-2 -4 47 fa 2 0-2 -4 -1]]) x -5 0 0 5 2} Ja 2)" 0 0 o s}lxJ L-32 by back substitution. We obtain nam ‘ 8 242x4 x = -12 “shan oe S+ 415 + x4 69

Elementary Linear Algebra 9th Edition

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elementary Linear Algebra 9th Edition

Uploaded by

Copyright:

Available Formats

You might also like