Fundamentals of Linear Algebra KATSUMI NOMIZU Professor of Mathematics Brown University ‘McGRAW-HILL BOOK COMPANY New York — St, Lonis San Francisco — Toronto Lond — SydneyFUNDAMENTALS OF LINEAR ALGEBRA Copyright © 1966 by McGraw-Hill, Inc. ‘All Rights Reserved. Printed in the United States of America, This book, or * parts thereof, may not be reproduced in any form without permission of the publishers. Library of Congress Catalog Card Number 65-28732 46880 234567890 MP 7321069"Preface ‘The importance of Hinear algebra in the undergraduate mathematics curriculum is now so well recognized that it is hardly necessary to make any comment on why linear algebra should be taught. For the practical problem of how to teach it, one has to find an appropriate answer—depend- ing on the purpose of a given course and the ievel of the students in it, "The present book is primarily intended for use in a two-semester course on linear algebra combined with the elements of modern algebra and analy- tic geometry of n dimensions. It is designed, however, to be usable for a one-semester course on linear algebra alone or a one-semester course on linear algebra with analytic geometry. The presentation can be adjusted to an elementary level following a calculus course or to a more advanced level with a rigorous algebraic approach. Some specific recommendations are given in the Suggestions for Class Use. "The following is a brief description of the main content. After an intro~ ductory chapter (Chapter 1) explaining the motivations of the subject from various points of view, we develop the basic concepts and results on vector spaces, linear mappings, matrices, systems of linear equations, and bilinear functions in Chapters 2, 3, and 4. In Chapter 5 we introduce some basic algebraic concepts (fields, polynomials and their factorizations, rings, extensions of fields, modules) in order to make it possible to develop linear algebra in a general algebraic setting. In Chapter 6 the theory of determinants is treated for matrices over a commutative ring with identity by way of alternating n-linear functions. In Chapter 7 we discuss minimal polynomials and characteristic polynomials (including the Cayley-Hamilton theorem) and their applications (in particular, the Jordan forms); Schur’s lemma and complex structures are also treated. In Chapter 8 we deal with inner product spaces and prove the spectral decomposition theorems for normal transformations, in particular, hermitian, unitary, symmetric, and orthogonal transformations. As explained in the Suggestions for Class Use, we indicate various proofs for these theorems in the exercises, Chapters 9 and 10 provide a linear algebra approach to analytic geome- try of n dimensions, which is the most efficient way of introducing rigorously geometric concepts in affine and euclidean spaces. The introductory ma- terial in Sections 1.3 and 1.4 serves as a preview for the full geometricvi Preface development of linear algébra in these last two chapters. The knowledge of n-dimensional analytic geometry is basic for the study of topology, alge- braic geometry, and differential geometry; nevertheless, the author has often noticed remarkable lack of that basic knowledge among many students of mathematics. We have given many examples in order to illustrate important points of discussion and computational techniques or to introduce the standard models for the concepts at hand. Bxereises at the end of each section are of the following three kinds: 1. To test the understanding of basi¢ concepts and techniques given in the text 2, To offer more challenging problems of genuine interest based on the text material 3. To provide supplementary results and alternative proofs in order to amplify the understanding of the text material Problems of the second and third kinds are starred. After learning the material in this book, a reader will certainly be ready to proceed to a more advanced study of linear algebra in its most prolific sense; for example, through the theory of modules to homological algebra, through the theory of matrix groups to Lie groups and Lie algebras, through the theory of exterior algebras to differential and integral calculus on differ- entiable manifolds, through the theory of tensor algebras, projective and other geometries to differential geometry, through the theory of Banach and Hilbert spaces to functional analysis, and so on. We should have liked to include at least an elementary introduction to some of these sub- jeots, but they had to be left out entirely, 1t is hoped that the present book will give the reader a balanced background in linear algebra before he specializes in various directions. Tn concluding the preface I should like to acknowledge the invaluable help I have received from Mr. Carl Pomerance, a student at Brown Univer- sity, who has critically read the manuscript and suggested numerous im- provements in the presentation. My thanks go also to Mrs. Marina , Smyth for her expert help in proofreading. KATSUMI NOMIZUSuggestions for Class User 1. For a short course the following material may be used: Sections 1.1, 1.2 (or 1.3); Chapter 2; Chapter 8; Chapter 6; Section 7.1; Sections 8.1, 8.2, followed by one or two theorems selected out of Theorems 8.21, 8.22, 8.25, 8.80 (and their corollaries). For this selection observe the following: @. ‘Throughout the whole treatment, treat vector spaces and matrices over the real number field ® or the complex number field C. Thus the * notation & will always stand for R or C. In Chapter 6, a commutative ring with identity is to be replaced by Ror, InSection 7.1, the character- istic polynomial has to be defined less formally. * >. Instead of (a), one may insert Section 5.1 between! Sections 2.2 and 2.3 so that one can treat vector spaces and matrices over an arbitrary, field F. ¢. For the proofs of Theorems 8,21, 8.22, 8.25, 8.30 follow the sugges- tichs in suggestion 5. 2. Fora short course with emphasis on geometry the following material may be used: Sections 1.3, 1.4; Chapter 2; Sections 3.1 to 3.3; Sections 6.1 to 6.4; Sections 8.1, 8.2 followed by Theorem 8.30; Chapter 9; Chapter 10. 4, One may treat only vector spaces and matrices over R. b, For the proof of Theorem 8.80, follow the suggestions in 5. 8. A more satisfactory treatment of lineat algebra with the elements of modern algebra can be given by Chapters 1 to 8. 4, Sections 1.3, 1.4 may be omitted, although it is always recommended to illustrate various concepts on vector spaces and linear transformations by using geometric interpretations. b. For less emphasis on algebra, one may introduce only the material in Chapter 5 that is absolutely necessary for the development of Inear algebra as the need arises. F 4. ‘There are various ways of arriving at-the spectral theorems for normal transformations (matrices), in particular hermitian and unitary transforma- tions (matrices) in the complex case and symmetric and orthogonal trans formations (matrices) in the real case. They ate developed along the following lines in the main text.viii Suggestions for Class Use a. Theorem 8.19 (complex normal) and Theorem 8.20 (real normal) are based on Theorem 7.8 (on the minimal polynomial). Theorem 8.21 (hermitian) and Theorem 8.22 (unitary) follow immediately by using the results in Section 8.4. Bb. ‘Theorem 8.25 (symmetric) follows from Theorem 8.20 as soon as ‘Theorem 8.24 (that a symmetric transformation has the real characteristic roots) is proved, and this is proved in two ways. c. Theorem 8.30 (orthogonal) is proved first for the two-dimensional case and then by using the argument on the minimal polynomial (Theo- rem 7.5). &, For alternative proofs of the spectral theorems we suggest. the follow- ing: a. Theorem 8.21 (hermitian) can be proved as in Exercise 8.4, num- ber 15 (thus before introducing normal transformations and without refer- ence to the minimal polynomial). b, Theorem 8.22 (unitary) can be proved as in Exercise 8.4, number 16, in the same way as (a). ¢c. Theorem 8.19 (complex normal) can be proved as in Exercise 8.5, number 8, and Theorem 8.20 (real normal) as in Exercise 8.5, number 9. d, Theorem 8.25 (symmetric) can be proved as in Exercise 8.16, number 18, once the existence of an eigenvalue (a real characteristic root) is established (as in Exercise 8.6; number 16 or 17, without reference to hermitian transformations). e. Theorem 8.30 (orthogonal) can be proved as in Exercise 8.7, number 8 (by using Proposition 8.16) or as in Exercise 8,7, number 9 (by using ‘Theorem 8.25). f. There are other variations; see Exercises 8.5, numbers 7 and 15, » Exercise 8.6, number 19, and Exercise 8.7, numbers 12 and 13. 6. Finally, a word about the terminology used in the text. We assume familiarity with the notation concerning sets, mappings, and equivalence relations. We also assume that a reader is acquainted with the principle of mathematical induction. Sizice these ideas are now introduced at an early stage in many courses in calculus, we shall give only a very concise explanation of the terminology in the Appendix.Contents, Preface v Suggestions for Class Use vii Chapter 1 Introduction 1 Ll 1.2 Differential equations 7 13 14 Systems of linear equations 1 Vectors on the plane 10 Change of coordinates 22 Chapter 2 Vector Spaces 32 \ 21 22 23 24 2.5 Q and@r $2 5 Vector spaces 34 j Subspaces 40‘ Linear independence; basis 42 Dimensions of subspaces 43 Chapter 3 Linear Mappings and Matrices 52 3.1 3.2 33. 34 3.5 Linear-mappings 62 Matrices 60 Change of bases 67 Elementary operations “74 Systems of linear equations 81 Chapter 4 Construction of Vector Spaces 88 Ad 42 4.3 Dual space 88 Direct sum and quotient space 94 Bilinear functions 100 Chapter § Some Algebraic Concepts 112 5.1 52 5.3 5A 55 5.6 Fields 112 Polynomials 116 ‘Ideals and factorization in F [xz] 121 Extension of fields 127 Construction of extensions 183 Quotient fields 187x Contents 5.7 Extension of vector spaces 140 5.8 Modules 143 Chapter 6 Determinants 148 6.1 Permutations 148 6.2 Alternating n-linear functions 156 6.3 Determinants 161 6.4 Expansions of determinants 169 6.5 Determinant of 4 linear transformation 176 6.6 Cramer's formula 178 Chapter 7 Minimal Polynomials 181 7.1 Eigenvalues and characteristic polynomials 181 72 Minimal polynomials 187 7.3 Nilpotent transformations 197 74 Jordan canonical form 204 7.5 Schur's lemma and complex structures 209 Chapter 8 Inner Product 218 8.1 Inner product 218 82 Subspaces 227 88 Dual space 231 84 ‘Hermitian and unitary transformations 285 85 Normal transformations 239 8,6 Symmetric transformations 245 87 Orthogonal transformations 250 Chapter 9 Affine Spaces 259 259 9.2 Affine transformations 265 98 Affine subspaces 272 9.4 Representations of affine subspaces 276 Chapter 10 Euclidean Spaces 284 10.1 Euclidean spaces 284 10.2 Subspaces 288 10.3 Isometries 294 10.4 Classification of isometries $00 Appendix 309 Notation 313 Index 318So FUNDAMENTALS OF LINEAR ALGEBRAeet ee 1 Introduction ‘The purpose of this chapter is to motivate the study of linear algebra. We shall show how the concepts of vector spaces, linear mappings, and matrices arise naturally from various types of problems in mathematics. Discussions are rather informal and are not to be considered part of a systematic development (which starts in Chap. 2); Definition 1.3, on matrix multiplication, and Definition 1.5, on identity matrices, will be referred to later. . “" 11 SYSTEMS OF LINEAR EQUATIONS To start with an easy example; we recall how we can solve a aystem of two linear equations in two unknowns: ax + by = 4, on + dy =0, where a, 8, ¢,d and u,v are given (real) numbers and 2, y are unknowns. We assume that a,b are not both 0 and c, dare not both 0. Multiplying the first equation by d and subtracting from it b times the second equation, we obtain ay "(ad = be)z = du — bw, Similarly, by eliminating 2, we obtain (ad — bey = av — cu. Hf ad — be x 0, then we find the solution _dunb wan, ; ad—bo' =~ ad —be Inthe case where ad — be = 0, we proceed as follows. Iie x 0,letk = a/e, 50 that a = ck, Substituting this in ad = bc, we have cdk = cb. Since 0, wo get b = dk. Ic = 0, then d #0 and we let k = b/d and still obtain a = ch. If k = 0, we have a = b = 0, contrary to the assumption. Thus we see that there is k 7 0 such that a=ck and b=dk,2 Fundamentals of Linear Algebra Multiplying the second equation of (1.1) by k, we have che + dky = ke, that is, ax + by = kv. If u = ky, then this is the same as the first equation of (1.1). In this case the system (1.1) has infinitely many solutions (z,y), which are determined from the first: equation by giving arbitrary values to x (or y). On the other hand, if w >.kv, then (z,y) satisfying the second equation does not satisfy the first equation; that is, the system (1.1) has no solution at all. Summing up, we have: 1. If there is no number k #0 such that a = ck, b = dk, then (1.1) has a unique solution, : 2. If there is a number k #0 such that a = ck, b = dk, and u = vk, then (1,1) has infinitely many solutions, 3. If thore is a number & # 0 such that a = ck, b = dk but u = vk, then (1.1) has no solution at all. A geometric interpretation is the following. Hach of the equations in (1.1) represents a straight line on the plane with the usual ‘coordinate system. In case 1, two straight lines are distinct and meet at one point. In case 2, two lines coincide. In case 3, two lines are distinct but parallel to each other. We shall now consider a more general system of linear equations —a* system of m linear equations in n unknowns 1, .. . , t4— Guts + Gute bes + Aint, = Uy (1.2) Gai, Anata b+ Oana = Uy Fats +H Oma + + + + + Omnia = my where the essential data of the system are the mn coefficients a:;, 1 < i.< m, 1Sj Sn, and the numbers u;, 1 n-dimensional zero vector. 0. This solution is called the trivial solution of (1.5’); it may be the only solution, or there may be other solutions, At any rate, consider the set’ S of all solutions [namely, n-dimensional vectors satisfying (1.5’)). If x = [ai] and x’ = [2] are in S, then the vectors a + ah ony ta + ay ony and , | | |» carbitrary, tm + th, crs . which we denote by x + x’ and cx, respectively, are also solutions. Thus the set of solutions S has the property that x, x’ € S implies x +x’€ S and cx € Sfor any number c. Definition 1.4 Let x= [x] and x’ = [z{] be two n-dimensional vectors. The sum x +x’ is the n-dimensional vector a + xy tn + oy, and the scalar multiple cx, where c is an arbitrary number, is the n-dimen- sional vector cay Cn, In connection with the sums and scalar multiples of vectors, let us observe that matrix multiplication has the following two properties: (1.6) A@e+ x’) = Ax+ Ax’, (1.7) A(x) = c(Ax), t £i 4 6 Fundamentals of Linear Algebra where A is an m X n matrix and x, x’ are n-dimensional vectors. In fact, the ith entry of the vector A(x + x’) is equal to Dasles + 2)) = Vi aves + Di aves, fat jad ial which is equal to the ith entry of Ax + Ax’. Property (1.7) can be verified ina similar way. ‘The assertion on S which we made above can be considered as 8, conse- quence of (1.6) and (1,7);in fact, if Ax = Oand Ax’ = 0, then AQ +x!) = Ax+ Ax’ =0+0=0 and Alex) = chx = 0 = 0, showing that x ++ x’ and cx are solutions. Let us now consider the system (1.3), called an inhomogeneous system if u 0; we call (1.5) the homogeneous system associated with (1.3). If x and x’ are solutions of (1.3), that is, Ax=u and Ax’ =u, then, denoting x’ + (—1)x by x’ — x, we have AG! — x) = AG’ + (—1)x) = Ax’ + (—1)A’ = Ax —Ax=u—u=0, by virtue of (1,6) and (1.7). This shows that x’ — x = y is a solution of (1.5). Coriversely, suppose that x is a solution of (1.3) and y a solution of (1.5). Then x’ = x + y iso solution of (1.5), because Ax! = A(e+y) =Ax+Ay=u+0=u. We have seen that an arbitrary (or general) solution of system (1.3) is obtained from any particular’solution by adding an arbitrary (or general) solution of the associated homogeneous system (1.5). In order to find a solution of (1.3), one will, of course, try to reduce the system to a system of a simpler form which has the same solutions, One employs a number of elimination steps, the simplest form of which we reoalled.for system (1.1). One of the processes consists in multiplying one vquation by a certain number and adding it to another equation, For example, we replace the jth equation of (1.3) by (ca + ag)ar ++ + ++ (Cain + Oin)tn = Cus + Uy For the corresponding matrix A, this process will change the jth row lan a +++ ain] into [edu tan cda+ap +++ Con + an): with the accompanying change of u into uy Ua cus + u; > Ua Um It is obvious that this sort of operation does riot change the solutions. ‘The same thing is true of multiplying one equation by a nonzero number, which is the other kind of process one uses for solving a system of linear equations. This indicates that the practical method of solving a system of linear equations can be described neatly as a sequence of certain operations per- formed on the matrix A and the vector u. EXERCISE 11 | 1, Compute the following matrix produets: 2-1 1 0 8-1 1 ofl 2) © [ JL} | : 1 i][! o1 ‘] [: -1 @ |1. 3-1/8 2 5]; @ 2 3 A 5 oOjL. 3 4 1 o a @® @ -1 aft]: 3. 2 Aro the following products defined? 1 aye © [i ls): ) [FJ “rou, 8. Solve the following systems of linear equations: rs E =} @ | aty+e=1, @®) styteet, | 8a = 2y +22 = 8, Bx — Qy + 2z = 8, | am ymee 1 by tam 4, 12 DIFFERENTIAL EQUATIONS Tet us consider 2 differential i 18 (1.8) iB +. a= Introduction 78 Fundamentals of Linear Algebra where x = 2({) is a function of ¢ which we wish to find. It is known that a general solution is of the form (1.9) cr cos t+ cn sin t, where c;, @ are arbitrary constants. In order to verify this, we see that both functions cos ¢ and sin ¢ satisfy (1.8), and then we may proceed as follows. We consider a mapping (or operator) which associates to a func- tion x(t) (which is twice differentiable, to be precise) the function, ax ete which we denote by Dz. The operator D then has the property that Dex + cots) = cxDas + c2D2, where the sum of two functions 2; +2," and the scalar multiple cz are defined by Gta)® =a® +n, (cx) @) = x(t). ‘Thus it follows that Dz: = Oand Dzz = Oimply Deis + e102) = exDrr + Dra = 0; in particular, any function of the form (1.9) is a solution of (1.8). Jn order to prove that every solution of (1.8) is of the form (1.9), we use the result that (1.8) admits a unique solution for any initial conditions 2(0) and (dz/dt)(0). Let 2 = x(t) be an arbitrary solution and let (0) =r and (dz/dt)(0) = c,. Consider the function y(t) =e: cost + ea sin t, which is a solution. Moreover, y(0) = a: and (dy/dt)(0) = cz. By the uniqueness, we must have y(t) = x(t), as we wanted to show. Summing up, we see that the set of all solutions S of the differential equation (1.8) has the property that if x,y € S, then z + y and ez € 8 for every number c. There are two particular elements, namely, cost and sin t, in S, so that every element of S can be written in the form (1.9) and hence is determined uniquely by the pair (c,,c:). In this sense it is possible to consider S as the set of all two-dimensional vectors = : ‘The differential equation (1.8) is closely related to a system of differential equations: 2 -—y =0, (1.10) “ : we r+ ane‘Introduction 9 In fact, if we set y.— dz/dt [which is the first equation of (1.10)], then (1.8) may be written as the second equation of (1-10). In dealing with the system (1.10) let us write the pair of functions x and y in the form of a [il | ‘means ll , then (1.10) takes the form 1.10’) ali] a [+ J [3]: where on the right-hand side we multiply the matrix [3 4 with the vector a Tf we agree that a “vector” [7] formally in the samé way as before. We may consider 1.10") as a differential equation for the vector-valued function [zh namely, a function whose value for each t is the vector [26 : In érder to generallize the discussions above, let W, denote the set of all functions (on the real line) which are differentiable at loast n times. If wwe define the sum 2 -+ y and the scalar multiple ez for functions as before, it follows that 17, has the property that 2, y € Wa implies 2 +y € Wa and x € Wa for any number e. We consider (uit) p=> ind Log where a'(t) are certain functions for 0 0, thon — let C be the unique point on the line 4B such that the directed segment AC 3 — has the same sense of direction as AB and such that the length of AC is ¢ times the length of AB. We define ea to be the vector AC (cf. Fig. 8). ‘This vector is determined independently of the choice of the point A. Fig. 7 (em2) Pig. 8 If c = 0, then we define cw to be the zero vector. If c <0, then we define ca to be —((—c)d), where (~c)a has already been defined, since ~c> 0, and the first minus sign is that used in (d) of the properties of addition. ‘We may easily verify the following properties of scalar multiplication: e. (ab)a = a(ba) for any « € Vand any scalars a, b. fe = ax for any’ vector «x. g. a(a + 8) = aa + a6 for any scalar a and a, 6 € V. A. (@4+ b)a = aa + ba for any a € V and scalars a, b. For example, (g) is illustrated by Fig. 9. In this way the set of vectors V can be made into an algebraic system with addition and scalar multiplication which satisfy conditions a to h. catasua(a+s) Fig. 9| | Introduction 18 ‘These algebraic operations, together with the basic relationships 1 and 2 between vectors and points, are indeed sufficient to deduce all the geo- metric properties of the plane that do not depend on the notions of length and angle; these properties are called affine properties of the plane. Example 1.2, If AABC is a triangle, then yy ee AB + BC +CA =0. Example 1.3, For any points A, B (A # B), » point P is on the straight line de- termined by A and B if and only it AP = cAB for some scalar c. In particular, the — as point P such that AP = #4B, where 0 = = = OP = OP; + OP, OP; = OF, and OP; = 0B for some scalars c:andc:, We thus obtain = = = (1.13) OP = o0E, + ,0Ey. It is obvious that the pair of scalars (c1,c2) is determined uniquely by P and that, conversely, any pair of scalars (c,,¢2) determines a unique point P for which (1.13) is satisfied. We call (c,c2) the coordinates of P. Tt is con- venient to write P = (yt). Two points A = (aya) and B = (bbs) coincide if and only if a, = bs and az = by By P Ea, o zy A Fig. 10 ‘Starting from a basis {a:,a2} and an arbitrary point 0, we have thus obtained a rule which associates with each point P its coordinates (c1,¢2). "This rule is called the affine coordinate system determined by {ajax} and 0. ‘The point 0 is called the origin, and the points B, and E) are called the unit points. Wehave 0 = (0,0), E: = (1,0), and #: = (0,1). Let {e103} be a basis of V. If a is an arbitrary vector, then a can be represented in the form (1.14) @ = eye + Cre, where ,, cz are certain scalars which are uniquely determined by «. Equa- 7 = = = tion (1.14) is nothing bit (1.13) if we let a: = OF, a = OB;, and a = OP. Example 1.6. With respect to an affine coordinate system with origin O and cor- responding to a basis {a,a2}, let A = (ays) and B = (b,b). Then (1.18) AB = (hy — aidan + (bs — crdo-Introduction 17 ‘This follows from 2 SS AB = OB — OA = (bye + bron) — (aren + axes) = i= aon + Or = @r)ar. Example 1.7. With respect to an’ affine coordinate system, a straight line I ean be ‘expressed by a linear equation (1.16) erty + cry = d, where c; and cy are not both 0 [that is, 2 is the set of all points whose coordinates (21,22) satisfy (1.16)]. To prove this, let A = (a,,a,) and B = (bibs) be two distinet points — ont Apoint P = (cyt) is on Lif and only if AP = cAB for some sealar e. By (1:18) this can be expressed by mame — a) and mz a = Cb — a). Assuming by % ay, we have Gira) G— a), b= a which is a-linear equation of the form (1.16). If b, = ay then we get 1 = aj, which is a special case of (1.16). Example 1.8. In order to illustrate the usefulness of an affine coordinate system, we give another proof of the result in Example 1.5. Given a triangle AABC, we take the aifine coordinate system with origin B and the basis (BC,BA}. ‘Thus B = (0,0), C= (1,0), and A= (01). We then have L = (4,0), M = (4,4), and N = (0.4). ‘The equations for the medians AL, BM, and CN are, respectively, ww+y=1, z-y=0, and e+d=l1. We find that these three equations have s common solution = 2 = 4,y 4. This means that the three medians meet at the point (4). We also see that BG ~ 4Baf, ‘AG = FAL, ana CG = 40N. %— % We shalll now treat the metric properties of the plane, namely, properties concerning or involving the notions of length and angle. For a vector a, the length of a, denoted by |al|, is defined to be the length of a directed segment AB which represents a. (If AB = OD, then, of course, AB and (D have the same length.) Wehave (1.17) Ileal] =. |e} lle, (18), fee Al) < fell + lal, triangle inequality Infact, these follow from the definitions of ca and a + 8 (Fig. 4). Given two vectors a and 8, we define the inner product (2,8) as follows. Ifa = 0 or B = 0, we set (0,8) =0. If a0, 6 #0, then choose any point 0 and let a = Od and @ = OB. The two half-lines OA and OBi 4 18 Fundamentals of Linear Algebra determine two ‘angles; let @ be the smaller angle, so that 0<@<~= (Fig. 11). Then we set (19) (2,8) = |lal]| |16|) cos . Note that this definition is independent of the choice of 0. For a= 6 (and @ = 0), we have (1.20) (aa) = |lal|*. ‘Two nonzero vectors and i are said to be orthogonal (or perpendicu- lar or normal) if the angle @ equals +/2, and this condition is equivalent to (a8) = 0. A vector of length 1 is called a unit vector. B A ‘B A 8 8 ° o Example 19. Let a = OA be a unit veotor, For any nonzero vector 8 = OB, let 0 be the projection of B on thie line OA. (that ia, the foot of the perpendicular line to OA drawn from B; see Fig, 12). If OC = cOA and if o is the angle between OA and OB, then Fig. 11 c cos 6 = wi and Ga) = Il all cos @ = ¢ lal = 6, which implies c = (6,«). Thus = os 21) _ 06 = (6,2)04. ‘The inner product has the following properties: G22) Ga) = (a8). (1.28) (ca,8) = o(c8) = (08). B &B L\ dq cana 7 Fig. 2Introduction 19 (ca + a3, B) = (01,8) + (02,8), (@ Bu Ba) = (e8:) + (aus). Property (1.22) is obvious from the definition, and (1.28) is also obvious if c>0. Fore <0, we note that the angle between ca and 6 is x — 6, where fis the angle between a and 6. Thus (ca,8) = ||cal| [él] cos (r — 8) = =|c{ [lal] [\6l] cos @ = c(a,). Similarly, (a,¢8) = c(a,8), Ibis sufficient to prove (1.24) in the case where @ isa unit vector. For a nonzero vector B, let y = 8/|[A||- If (a: + ea, 7) = (ayy) + (any), then we may multiply both sides by ||é|| and find (a + a, 6) = (ax) + (a2,8) by using (1.23), Assume is a unit vector. Then the identity follows from the geometric interpretation of the inner product in Exam- ple 1.9 (see Fig. 13). Fig. 13 All the metric properties of the plane can be derived from the properties of the inner product. Let a and az be two unit vectors which are perpen dioular [that is, (cj) = (02) = Land (o102) = 0], Taking any point as the origin, we construct an affine coordinate system, ‘This coordinate system is a rectangular coordinate system, namely, an ordinary coordinate system based on two perpendicular lines with equal unit length on each line (coordinate axis) (see Fig. 14). 2a Fig, 14 |20 Fundamentals of Linear Algebra Let A’= (axa2) and B = (b,,bs) with respect to a rectangular coordinate system. Then we have | — AB = (b: — air + (br — ca)on and (AB,AB) = ((br — as)ou + (bs — c2)an, (bs — aiden + (bs — 02) 02) = (b1 — a) *(ayan) + (br — @)*(a2,02) = (br — a)? + (he =m), by using (aio) = Oand (a1,01) = (as2) = 1. ‘Thus the distance d(4,B) between A and B is given by (1.25) a(4,B) = VG — a) F Ge — wa). Example 1.10. Fix o rectangular coordinate system. with origin @, For a line which does not pass through O, let H = (hiha) be the foot of the perpendicular drawn from O — tol (Fig. 15). A point P = (2,2) is on the line Lif and only if (ia) =0. This condition may be expressed by (ea — hid + = adh = 0, hy he aa that is tb Se = VI t+ * Viet * Vista” ii We rewrite this in the form (1.26) yt, + am = d, where d > 0 and a,’ + a:* = 1. ‘This equation is called the normal form of the line 1, Note that d is the distance from Oto the line 1. The vector aa + dyay is a unit vector perpendicular to }, and is called a unit normal vector of I. Equation (1.26) is still valid if ! goes through 0, in which ease d=0. Pate, 2) Fig. 15 EXERCISE 13 3a am wm ODD 1. Verify (AB + BC) -+CD = AD = AB + (BC +CD). Draw » geometric illustration for these identities. a 2, If AABC is a triangle, show that AB = AC — BC.Introduction 94. 3. For any points Pi, <4 > = [Hint: Let @ = CA,y = AB. Let H be the intersection of BB’ and CC and set BH = &. 3 my = Write down (BH,CA) = (GH,AB) = 0 and derive (AH,EC) ~ 0.) 18, Let O be the intersection of the perpendicular bisectors of the three sides of a triangle AABC. Show that a a) OA + OB + 0C = 806, > 3 = where G is the center of gravity of AABC. [Hint: Let a = 04, 8 = OB, y = 00. — aay Express OL (L: midpoint of BC) and then OG in terms of a, 8, 7.] >, = 14. Let a= AB and 6 = AD. Show that the area of the parallelogram ABCD (with vertices in this order) is equal to V (aya) (8,8) — (a8), 15, Find the normal form of each of the straight lines: (a) 8a + 4x = 10; (@) 3% +42 +10 =0. 16, Find the distance from a point (yy) to the line given by the normal form (1.26). 17. Suppose that for an affine coordinate system the distance d(A,B) between any two points A = (a,ja2) and B = (by,b,) is given by the formula (1.25). Prove that the coordinate system is a rectangular coordinate system.sSereaN 22 Fundamentals of Linear Algebra 14. CHANGE OF COORDINATES ‘An affine coordinate system {21,22} is constructed from a point 0 (which will be the origin) and a basis {a:,a:} of V. For each point P we denote the coordinates of P by (m(P),m(P)). We shall study how two affine coordinate systems are related to each other. Let {21m} be an affine coordinate system constructed from O and {enn}. Let {2{,24} be an affine coordinate system constructed from 0’ and {aj,ai}. We first consider two special cases. Case 1. Assume a: = af and a; = af. Let O’ have coordinates (a,a2) with respect to {21,22}. ‘This means 00" = ase + aoa. For any point P we have OP = 2(P)au + 24(P)an — and OP = (Pa + 23(Par — Since OP = 00’ + O'P, we obtain (Por + 2x(P)os = (a + 24(P))oa + (de + 24(P)) on, that is, n(P)=2(P) +a, “m(P) = 24(P) +a. ‘This being the case for any point P, we may simply write (1.27) PAT Bat Oh 11 esaladion of coordi ase L2 = Ly + As, fa (27) = ory, _ th = ta — a The relationship is illustrated in Fig. 16. Case 2, Assume 0 = 0’. Since {asjaa] is a basis of V, every vector can be written uniquely in the form (1.14). Let 0.28) . et = anos + anes, of = duos + dena. For any point P we have (since 0 = 0") OP = al(Phaf + a(P)af = 2(P) (ancy + anos) + 23(P) (ane + O20) = (amy(P) + auey(P))ox + (aaa (P) + a2 (P)) an.Introduction 98 Comparing this with =. OP = (P)a + a2(P)as, we obtain - a(P) = auai(P) + auti(P), i na(P) = ant{(P) + an2{(P) | i I | i ' | i / | t I Fig. 16 f ; | Dropping P, we may write f (1.29) % = auti + ayry, E y= Oni ++ ati. See Fig. 17. Similarly, starting from a1 = dual + Buel, | (1.28) - a2 = dual + bua, Fig. 1724 ~~ Fundamentals of Linear Algebra we obtain af = buts + but, q.20) respectively, by a. elle , at] _ fbn cian [2]-h The maizices A=|% 2s and B Ox Cag, are related by (1.31) AB=BA= 2 = Gu(buta + bists) + aia(barts + basta) = (Gubu + aubu)er + (auby + aube)r2 and, sinilarly, 2 = (Onbs + Goabar)ai + (darbi2 + annbas) tae Puiting a = 1, a2 = 0, we get Qubu + Guba = 1, + abu + onba = 0. Putting 2: = 0, a = 1, we get aybu + dubs = 0, ll ry = bats + Dante. By using the matrix notation in Sec. 1.1, (1.29) and (1.29) can be expressed, allZ} 2 a], a, [i oh 01 In fact, substituting (1.20') in (1.29), we have abu + Gaba = 1. ‘These four conditions are equivalent to the matrix equation aL] Similarly, we get BA= 1 07. 0 1. Now the general case can be treated by combining the special cases 1 and 2. One passes from the coordinate system {2/,r{} to the coordinate system {ys,y2} constructed from O’ and {c,02} and then from {ys,y2} to ] [ * Lda Ga oSIntroduction 25 the system {21,29}. (See Fig. 18.) ‘The first passage is related as in case 2, so that = aut + anes, Ya = anal + Ons. ‘The second passage is related as in case 1, so that m= y+ ay T= Yat an fi oa ee ta Ca ‘a? i ay . a 0 oO Fig. 18 Therefore we obtain = Outi + awry + a, 1.32) ees ty = dat + ames + ae or fio af = buts + bum + by, ah = bats + dave + Ba. (1.82) can be expresced in matrix notation by Eat an ae a | fey (1.33) @2}= {an a de || me] 1 oo ali Similarly, (1.82') can be expressed by af] [ou be bi |[m (1.38") zh} =|bu be be || m|- 1 oo tli If we let Qn am ay bn ba bb dx an is B=|bx be be]? oon oo14 26 Fundamentals of Linear Algebra then we can verify 100 a4) AB=BA-|0 1 0}- 001 On the other hand, writing a-[3 }; where a = B= [3 aE where B = [“} [i] we see that AB Aa+6 ws-['0 i] 20 that Aa + B= 0, that is, (1.35) B=-Ae or a=—BB ‘Thus two affine coordinate systems {z1,r2} and {x{21} are related by (1.38) and (1.33'), where the matrices A, B and vectors a, B are related by (1.81) and (1.35). We shall make the following definitions. Definition 1.5 For any positive integer n, then X m matrix of the form 1.0--. 0 o1 0 0 0° 1 (that is, the entries on the diagonal are 1 and all the other entries are 0) is called the ‘dentity matrix (or unit matrix) of degree n. Definition 1.6 For an n-X n matrix A, an n Xn matrix B such that AB = BA = I, is called an cnverse of A. ‘The identity matrix I, has a remarkable property that (1.36) AL=LA=A for any n Xn matrix A. Hf we denote the (i,7) entry of I, by 8:;, then we have a - [1 foré=s, = 10 fori xj.Introduction 27 (6,; is called the Kronecker symbol.) . Condition (1.36) can be easily verified by going back to Definition 1.3 of the matrix product. Let us also observe (1.37) (AB)C = A(BC) for any n X m imatrices A, B, and C. In fact, if A = [ai], B = [by], and C = [cx] and if we denote by (AB): ((AB)C),j, ete., the (7,7) entry of AB, (AB)C, etc., then we have (AB)C)y = By (AB)acas = 3, abn and (ABC))a =F ain(BC)ay = - Be aidan so that (1.87) is valid. By using (1.37), we may show, that when an n Xn matrix A has an inverse B, that is, a matrix B such ‘that AB = BA = I,, such a matrix B is uniquely determined. In fact, if C is another n X m matrix such that AC = CA =I,, then (CA)B = C(AB) by (1.87). The left-hand side is equal to I,B = B, while the right-hand side is equal to CI, = C, thus proving B = C. The inverse of A is denoted by A~. Example 1.11, For 10 a i 0 -a =|0 1a). wehave At=/0 1 —m|- 00 1, Gane! This fact conforms with (1.27) and (1.27’). Example 1.12, In order to see whether acfon ae Lan an has an inverse, let X = [z 2] esate down ax = Ie. Then we obtain a system of four linear equations in four unknowns 2;;,1 < 4,7 <2. If this system has no solution, then A has no inverse. For example, : : has no inverse. We shall now consider how two rectangular coordinate systems {2,22} and {2{,23} are related to each other. Let 0; a, a: and 0’, af, af be as before, except that now we have (a;a,) = (af,a/) = 6; (using Kronecker's symbol). In case 1 (that is, a = aj and a = of), we have (1.27) and (1.27') just as before.28 Fundamentals of Linear Algebra | In case 2 (that is, 0 = 0’) we proceed as follows. Let us consider (xen) as illustrated in Fig. 19, To any vector B let us associate the angle (0 2m, 0 = (8 + 7/2) — 2x] as in Fig. 20.Introduction 99 b.@ =6431/2 [if 64 32/2>2n, w= 0+ 3n/2)—2n] as in Fig. 21. Fig. 21 In case a we have in (1.28) ® zi os = 00s (+2) = —sin 6, om = sin (6-4 2) = cose, Tn case b we have yz = cos 04 3) = sing ‘ 30 do = sin (6+ 5) = —cos 9. Thus the matrix A in (1.30) is equal to cos@ =—sin 6 + (1.38) Ro = [8 ee | in case a, sin 0. —cos 6 In particular, for @ = 0, we have co-[ I} We note (1.40) S(@) = R()SO). The matrix S(0) represents the change of axes illustrated in Fig. 22. Thus (1.80) takes the form a] _ [cose —siné] [ay (1.41) [2]- sin 0 [EI (1.89) so =(ome marl in case b.} i i } j 1 30 Fundamentals of Linear Algebra which is the rotation of the coordinate axes by angle é, or ‘t]__[eose —siné][1 0] fa | ae [2] = [ies @ cos8 Ilo -1fLen)’ | which is the change of direction of the second axis followed by the rotation (1.41). Finally, a general change of rectangular coordinate systems is given by (1.88), where the matrix A = [s: as] iseither R(6) or S(@). Gr, Aas o nay Fig. 22 Example 1.13. For any angles @ and w, we have ROR) = RE +e). This is obvious in view of the geometric interpretation of R(6) as the rotation by angle @ of the coordinate axes, ‘The identity is“ equivalent to the following formulas in trigonometry: . cos (9 + w) = cos 9 cose — sing sinw, sin (6 +a) = sin6 cos + cos6 sin w, Example 1.14. Fixing a rectangular coordinate system {z,%}, the equation : 1 a” i a3) [z] = R@) A | may be interpreted as a point transformation f of the plane which maps a point (21,22) into (y,,us) given by (1.43). ‘The distance between the points remains unchanged by this transformation, namely, d(f(A),f(B)) = d(4,B). (fis called the rotation of the plane by angle 6.) ‘This property can be verified as follows. IfA = (ajax), B = (bisbs), then (1.25) implies G(A,B)* = @, = a1)? + (ba — a1)*.Introduction 81 By the same formula, we have 4(f(A),fB))* = ((b, c08.4 — by sin 6) — (a; cos 6 — a» sin 6))* + (Qh sin 9 + bs cos 4) — (a; sin 6 + @ cos 9)? = (G1 = 41) cos@ — (a) sing)? + (i — a) sin @ + (; — as) cos 9)* = Gi — a1)? + br — a)3, so that d(f(4),f(B)) = a(4,B). EXERCISE 14 1. Find the inverse of @ IF a]: |G i]: @ [4 2 Does as | have an inverse? Ke 8. Show that A = [: § ] nas on nvense it ana only if ad — be 5£0, 4. Show that the inverse of R(@) is Rx — 6). 5. Write down R@) for @ = «/3, »/2, x. 6. With respect to a rectangular coordinate system, consider a point transforma- tion f which maps (x,,2) upon (1 ++ a1, 22 + a2), where a, a are given numbers. Shove that the distance between the points remains unchanged by f.(f'is called a translation.) 1. The equation z,2/9 + 23/4 = 1 represents an ellipse with respect to'a rec- tangular coordinate system. Let O = (0,0), Ai = (3,0), A; = (0,2). What is ‘the equation of the given ellipse with respect to, the affine coordinate system based on 0, area Ay, OA? 8. Let {z,m] be a rectangular coordinate system and consider the figure S' con- sisting of all points whose coordinates satisfy the equation 2% — 2a, + m4" — 4a, = 20. Taking a new rectangular coordinate system zi, 24 such that mamta m2, express S by a new equation. What is S geometrically?o 2 Vector Spaces Assuming familiarity with the real and complex number systems R and C, we first define veetor spaces R" and C*, which will be the models of n-dimen- sional vector spaces over R and ©, respectively. . We then proceed to an. axiomatic definition of vector space and introduce related basic concepts. With the notion of a field, which will be introduced in Chap. 3, discussion of vector spaces over sin arbitrary field will present no further difficulty. Those who wish to study vector spaces over an arbitrary field from the beginning should insert Sec. 5.1 perhaps between Secs. 2.2 and 2.3, In Definition 2.1, as elsewhere, F can be an arbitrary field. 421 Q@* AND @ Let R and @ denote the real and complex number systems, respectively. Let n be an arbitrary but fixed positive integer. We shall denote by R* the set of all ordered n-tuples (a, ... yx) of real numbers. Hach n-tuple is often denoted by a single Greek letter, say a, so that we write a= (@,.+. 0a) of, more briefly, a = (aj). Two ordered n-tuples a = (a;) and 6 = (b,) are regarded as equal to each other if and only if ay = by for every é,1 a+ € F*], and scalar multiplication, which associates to a scalar c and a vector # a second vector ca [or, what amounts to the same thing, a mapping (c,a) € F X F"—ca € F*). We shall denote by 0 the n-tuple (0, . . . ;0) and call it the zero vector. It is now quite straightforward to prove the following Proposition 2.1. Addition and ‘scalar multiplication in F" have the following properties: Lat8= B+ a (commutativity). 2 (a+ 8) += a+ (8+ 7) (associativity). 3. a +0 = a for any vector a. 4, For any a € F*, there is a vector denoted by —awith «+ (—a) = 0; in fact, if a = (a:), then —a = (—a;). 5. cla+§) =ca+ cB. 6 (c-+d)a = ca+da. % 7. (ed)a = e(da). 8 la =a. Inall the identities, a, 8,7 are arbitrary vectors and ¢, d are arbitrary scalars. Definition 2.1 ‘The set * with addition and scalar multiplication as defined above is called the n-dimensional standard vector space over F. Of course, according as F.= R or ©, we shall denote the corresponding vector space by F* (n-dimensional standard real vector space) or C* (n-dimensional standard complex vector space). In the vector space *, the sum of vectors can be defined for any number of vectors. For & vectors a1, ..., a» the suma+- +--+ ay is the vector whose, ith component is equal to the sum of the ith components of ax, +. ay We can easily verify that ates tae (ait: ++ ena) tai It is also clear that in taking the sum of any number of vectors the order of the vectors does not matter; for example, ea Fae + as = a + ay + a9 = os + ow + on, i f é| | 34 Fundamentals of Linear Algebra By a linear combination of c, .. +, a» we mean a vector of the form cart > ++ cx, where ci, .. +» Cr are & arbitrary scalars, called the coefficients of the linear combination. ‘We shall often ‘use the summation notation. For any umber of veetors a... , ay the sum a +++ ++ ais denoted by Y avand a linear combination cron + - - - + cram by xy Citi. In F* we define n special vectors a, . . . , é 88 follows. For each 7, eis the vector whose components are all 0 except for the ith component, which isequal to 1. The set of vectors {a, . . . ,és} is called the standard basis of F* for the following reason Proposition 2.2. Every vector a in F* can be expressed as a linear combi- nation of ey... , & ina unique manner (that is, the set of coefficients is uniquely determined by «). Proof. Given any vector a = (d102, . . - ,») in Fr, we have =) zo where the coefficient for ¢; is nothing but the ith component of a. The uniqueness of this representation follows from the definition of equality for vectors in Fr. EXERCISE 2.1 1. In @, calculate the following linear combinations: 2(1,0,4) + (—4)(2,1,5); 8(2,—1,—2) + 2(—3,2/3,3). 2 In C?, caloulate i(2, —1 +8) + @-1)0+4%4). 8 InQY, write (1,0,0) as a linear combination of (0,1,1), (1,0,1), and (1,1,0). & In 5*, prove that for any vectors a snd f there is one and only one vector £ such tndia-+£=6. (This vector is denoted by 8 — «.) Show that 6—« = 8 + (-a). 2.2 VECTOR SPACES Before proceeding to an axiomatic definition of vector space we shall define the concept of additive group. Definition 2.2 By an additive group we mean a set V together with a rule, called addition, which associates to any pair of elements a and b in V a unique element denoted by « + b in V so as to satisfy the following conditions: Latb=b+a.Vector Spaces 35 « 2 @+b)+e=at+G+e). 3. There exists a certain element, denoted by 0, such that ¢ + 0 = afor every EV. 4. For any a € V there exists an element a’ € V such that a + a’ = 0; Example 2.1. The set of all integers Z is an additive group where addition is the usual one, ‘The set of all rational numbers @ is also an additive group with respect to the usual addition, Similarly, @ and @ are additive groups. Example 2.2. The vector spe ve F* is an additive group (when we just consider vector addition and forget about sealar multiplication). A vector space, which we shall soon define in an abstract way, is an additive group with a further property, as illustrated by Example 2.2. For this reason we shall first study some of the basic properties of an arbi- trary additive group. . Proposition 2.3. Let V be an additive group. 1. There exists one and only one element 0 in V such that a +0 =a for everya€ V. (This unique element is called the zero element of V.) 2. For any a € V there exists one and only one element a’ € V such that a+a’=0. (This unique element is. called the inverse of a and is denoted by. —a.) 3, For any a,b € V there exists one and only one element x € V such that a-+2=b. (This element is called the difference of b and a and is denoted by b — a.) 4. Ifa+z =a for some a, then x = 0. Proof. 1. We know that there is at least one such element 0. Suppose there is another element 0’ with the same property. Then 0 + 0’ = 0. By condition 1 of Definition 2.2, we have 0 + 0’ = 0’ +0 = 0’, so that 0=0. 2. We know that there is at least one such element a’ (condi- tion 4 of Definition 2.2). Suppose that a”’ is another such element. Using condition 2 of Definition 2.2, we have a’.+ 0 = a! -+ (a+a") = (a’ +a) +a! =0+a"andhencea’ = a”. 3. Suppose there is an element 2 such that @-+2=b. By adding —a to both sides, we get (—a) + (a+z) = (-a)+b. By associativity (that is, condition 2 of Definition 2.2), we have (—a) + (a+ 2) = ((-a)+4)+2=0+2=2, 50 thatz =b+(-a). Con- versely, the element b + (—a) actually satisfies a + 2 = b, because a+ (b+ (-a)) =a+ ((-a) +b) = @+ (-a)) +d =O0+b=5. 4, By (8) of this proposition, there is only one element x such thata +2 =a,and, infact,2 =a + (—a) =0. [ i i86 Fundamentals of Linear Algebra Remark. For ony a € V, it is clear that —(—a) = a. In an additive group V we shall define the'sum of any number of ele- / mentsay,... , a,inductively} by (2.1) aye sae = (t+ + ++ ana) ta k We denote this sum by © ai. a Proposition 2.4. Let ay, . . . , Grim be arbitrary elements in an additive group V. Then oo (Be)+(¥.9)- Bs Proof. We use induction} on m. For m = 1, formula (2.2) is nothing but @.1) Assuming that (2.2) is valid for somo m, we shall prove (2.2) for m+ 1. We have (Bo)+ CE. «) = Bea) + (Ri) +s) = (Ba) +(Z4)) +e hem heme Dy ai) + Grime = DO. ae fot ae & We also observe that the sum 5 a; is independent of the order of the m1 elements a, . . . , a, as the following examples illustrate: a+ a+ ast as = (2 +41) + (a+ a) = (ata) $ (ta) = a+ a+ ast as aa + ay + as + a = ae + (1 + as + 2) = (a tart m) + =Htatata ‘Now we shall define the notion of a vector space. Let F be R or Cas in Sec. 2.1 (F can be an arbitrary field, which will be defined in Sec. 5.1). Definition 2.3 A veclor space over F is an additive group V for which there is, furthermore, a mapping (c,a) € F X V—ca€ V, called scalar multiplication, which has the following properties: 1. cla + 8) = cat cf. + For “induction” and “inductive definition” see Appendix., Vector Spaces 37 2 (c+d)a=ca+da. 3. (ed)a = c(da). 4, la =a Here a, # are arbitrary elements in V and c, d are arbitrary elements in F. ‘As in the case of F*, elements of F are called scalars and elements of V vectors. When F = &, V is called a real vector space; and when F = C, V is called a complex vector space. All the properties of an additive group are, of course, valid for a vector space with respect to addition. ‘The zero element 0 of V, called the zero vector, is a vector such that « +0 = aforeverya € V. Weare using the symbol 0 for this zero vector as well as for the scalar zero, but this should not cause any confusion. We now. prove some more basic rules concerning a vector space; for example, we have Oa = 0 for any « € V. In this equation, 0 on the left-hand gide is the scalar 0 and 0 on the right- hand side is the zero veetor. Proposition 2.5. Let V be a vector space over F. 1, Oa = 0 for every a € V. 2. cO =0 for every c € F. 3. If ca = 0, then c = Oora= 4, For any c € F anda € V, (—c)a = —(ca). Proof. 1. 0a = (0 + 0)a = 0a + Oa by condition 2 of Definition 2.3. By (4) of Proposition 2.3 we have Oz = 0. 2.0 =c(0 + 0) = c0 +0 by condition 1 of Definition 2.3. Again by (4) of Proposition 2.3 we have c0 = 0. .3. Suppose ca = 0 and ¢ 0. Multiplying ca =0 by we have : 0=e%(ca) = (ea =la=a by conditions 3 and 4 of Definition 2.3. 4. ca+ (—c)a = ((¢ + (—c))a = 0a = 0 by condition 2 of Definition 2.3 and (1) of this proposition. Recalling (2) of Proposition 2.3, we have (—e)a = —(ca). Definition’2.4 : k Given a finite number of vectors a:, . . . , oi, & vector of the form Zoi is called a linear combination of ox, . . . , ct with coeficients cr, ..., cx €.F. We now give some examples of vector spaces. Example 2.3. 5 is a vector space over F.88 Fundamentals of Linear Algebra Example 2.4. Let S = Gor © (or, more generally, an arbitrary field). Denote by F™ the set of all sequences a = (00)4:,0% - . - a, » - .), Where a's are elements of F ‘and are 0 except for a finite number of them. For a = (a;) and £ = (b,), we define +B = (dy + bot + Dita + bay + + + sn + Boy «= -) and Cox = (CaeyCtyCOr + + + s0ny + - -)y where ¢ € F. Then those sequences « + # and ca are again elements of $*. It is easy to verify that F° is a vector space over F with, respect to these operations. ‘The zero vector is (0,0,0,...,0,...). Example 2.5. For each a = (a:) in §*, we consider the expression of a usual poly- nomial Sa(@) = ao + ait + ass? + + +t + as", where dy is the last nonzero term of the sequence a = (a:). It is obvious that, con- versely, any polynomial expression with coefficients in S corresponds to en element of $* in this fashion. One can also easily verify that « + and ca correspond to the sum fa(z) +fa(z) of two polynomials and the sealar multiple ofa(z), respectively. ‘This means that the set of all polynomials with'coefficients in F is a vector space over &, which is essentially the same vector space as 5°. Example 2.6. Let A be an arbitrary set. Consider the set §4 of all functions on A with values in. For two such functions f and g, we define F +9) =f@) +92), (f)(2) = fle) where z € A and c € §. With respect to these operations 5 is a vector space over 5. ‘The zero vector is the function which is identically 0 on A. Example 2.7. By specifying the Set A and also by restricting the nature of functions in Example 2.6, we can obtain many examples of a vector space. (a) Let A = and. =@. ‘Then we have areal vector space consisting of all real- velued funétions on the real line. Let A be the unit interval (0,1) and ¢ = @. Then we have a real vector space consisting of all real-valued functions on the unit interval. (b) Let A =Q (or (0,1]) and = ® and consider only those funetions which ere continuous. Since the functions f(z) + g(z) and ¢f(e) are continuous together with JS) and g(z), we obtain a real vector space consisting of all real-valued continuous functions on @ (or [0,1]). (c) Let A be the set of natural numbers and consider the set of all functions on A with values in § each of which has the following property: the values f(n) are zero except for a finite number of natural numbers n. This set becomes a vector space over §. With any such function f, we associate (a;) € F*, where a: = f(i). It is obvious that our vector space is essentially the same as the vector space 5 in Example 2.4. Example 2.8. Another special case of Example 2.6 is the following. Let A be the set of all pairs (¢,j) where ¢ and j are integers such that 1 » or A= [ays]. mn °° Om, A display A of mn elements of F in this kind of form is called an m Xn matris over &. For each pair (i,j), 1 0}. @ la= @);m = 0}. (©) {a = (@);all 2/s are rational numbers). | a Lt We hese of a2 X Amaro ier] 2, j] mies ni snteae A arbitrary elements of §. Prove that W is » subspace of %. a! 3. When W; and W; are subspaces of a vector space V, is the union W; U Ws (set of all vectors which belong to W: or Ws) # subspace? 4, The set Gim can be considered as a subset of the complex veetor space @*. Is it a subspace of @*? What is Sp() in C*? ) 6. Let W bo a subspace of a vector space Vover §. For a, 6 € W, we shall write a= pmod Wif@—a€ W. Prove that this relation, called congruence mod W, has the following properties: H 3 (a) a = amod W for every a € Ve @) a= 6 mod W implies 6 = a mod W. (©) «= 6 mod W and 6 = y mod W imply'a = ymod W. (@) a = 8, mod W and a = f; mod W imply a + es = fi + A mod W. (©) a= mod W implies ca = of mod W, where c € F. 6. Lot a ~8 be e relation among elements in'a vector space V which has the properties a to ein number 5. Prove that W = a;0 0) is a subspace and that tho given relation a = f is equivalent to a = 6 mod W. 1: Let V be a vector space and let a, ... , an, 8 be elements in V. If 6 € Splay ... jan} and if 6 € Spla, ..- ,c], prove that ann € Spa, . . + ,am8}- 24 LINEAR INDEPENDENCE; BASIS | *Let V be a vector space over F. | Definition 2.7 | ‘A sot of cloments (ci «- « yu} in V is suid to bo Binearly indspondent i E, cias = 0 implies that cy = ---= c= 0. Otherwise, the sei is said to bbe linearly dependent.the bv rite has te be tot Vector Spaces 48 Jn other words, if ai,..., ax are linearly dependent, then there exist a» +» Gein F which are not all 0 such that E cia: = 0. a Example 2.11. In, «, é Of the standard basis are linearly independent, because ae E66 = uence) is 0 if and only fe, == c= 0. On the other hand, the vectors = (1,2,-3), a = (2,-1,1), as = (=1,8,—11) are linearly dependent, since 8a + (—2)ax +: (—1)ay = 0. For one vector a, a is linearly independent if and only if a #0. In fact, if a # 0 and ca = 0, thene = 0. On the other hand, if « = 0, then la = 0,80 that ais linearly dependent. It isalsoclear that if a, ... , a are linearly independent, then a, ..., om where 1 0, we have so Saw 2 Cy i a= —p— ao é ‘Thus any linear combination of ai, a2)... ; oy is 2 linear combination of 8, a -» + 1 dm if we replace a by the expression above. Thus Sp {oayas, . - + en} Sp (Bran, «+ + sen). Theorem 2.10. Let V beavector space over F. Assume that ds Py = 1 « giPa'E Splat ss saa} 2. 6.» yBm are linearly indepentient. : Thenm m, we shall derive a contradiction, The element A, is a linear combination of ai... ; ae: A= Ycxax. Since fi #0, at least one of the coeffciente cis not 0. By rearranging the in- dices, we may assume that c x0. By the lemma above, we have 8p (Bias, «+ + aa} = Sp {aan . . . aa}. Since f; belongs to this subspace, it isa linear combination of Br ay, aa! fr= 0 + ¥ sas, Not all cfs are 0, because otherwise we would have f:= af; contrary to the second assumption: Again by rearranging the indices, we may assume that ¢: x 0. Again by the lemma we have Sp {BiBrya, . . « on} = Sp fous, «+ on}. Now assume that we have proved for some k < m Gears ee ey ep ae ee The vector B41 belonging to this subspace is a linear combination of By... > Be cays, » . «yd bUt not a linear combination of 61, . . . , Bi since 61, . . » » Busi ate linearly independent. Thus, by rearranging the indices, we may assume that. - z Bur = Deadet Yo ova, where crys 7 0. & iFha By the lemma it follows that Sp (81,8) - - - Baines) «+ - jan} = Sp (BuBa + - + Piyoery + + + sn} = Sp {aon .- + son}.46 Fundamentals of Linear Algebra Since we are assuming m>n, we may finally arrive at the set| {BiB - - » Gn} which spans Sp{a, .. - ,0n}. But this is @ contra diction, since then 8,41 would be # linear combination of (1, - . + » Bu ‘We now prove the following important theorem. Theorem 2.11. Let V be a vector space over F of finite dimension and let {au .. . san} bea basis of V. Then every basis ts finite and has exactly n elements. Proof. Let S be any basis of V. Since S cannot contain more than n elements which are linearly independent by Theorem 2.10, it follows that S must be finite. Let S = {f:,... 8m}. Again by Theorem 2.10, we have'm cia: = O with o, #0. ‘Then we have cae es