.

PARTIAL STUDENT SOLUTION MANUAL

to accompany

LINEAR ALGEBRA with Applications

Seventh Edition

W. Keith Nicholson
December, 2011

TABLE OF CONTENTS
CHAPTER 1 SYSTEMS OF LINEAR EQUATIONS
1.1 Solutions and Elementary Operations
1.2 Gaussian Elimination
1.3 Homogeneous Equations
1.4 An Application to Network Flow
1.5 An Application to Electrical Networks
1.6 An Application to Chemical Reactions
Supplementary Exercises for Chapter 1

1
1
2
8
10
10
12
12

CHAPTER 2 MATRIX ALGEBRA

2.1 Matrix Addition, Scalar Multiplication, and Transposition
2.2 Equations, Matrices and Transformations
2.3 Matrix Multiplication
2.4 Matrix Inverses
2.5 Elementary Matrices
2.6 Matrix Transformations
2.7 LU-Factorization
2.8 An Application to Input-Output Economic Models
2.9 An Application to Markov Chains
Supplementary Exercises for Chapter 2

15
15
17
20
25
32
35
40
42
43
46

CHAPTER 3 DETERMINANTS AND DIAGONALIZATION

3.1 The Cofactor Expansion
3.2 Determinants and Matrix Inverses
3.3 Diagonalization and Eigenvalues
3.4 An Application to Linear Recurrences
3.5 An Application to Systems of Differential Equations
3.6 Proof of the Cofactor Expansion
Supplementary Exercises for Chapter 3

48
48
52
57
61
64
65
66

CHAPTER 4 VECTOR GEOMETRY

4.1 Vectors and Lines
4.2 Projections and Planes
4.3 More on the Cross Product
4.4 Linear Operators on R3
4.5 An Application to Computer Graphics
Supplementary Exercises for Chapter 4

68
68
73
81
83
84
85

CHAPTER 5 THE VECTOR SPACE Rn

5.1 Subspaces and Spanning
5.2 Independence and Dimension
5.3 Orthogonality
5.4 Rank of a Matrix
5.5 Similarity and Diagonalization
5.6 Best Approximation and Least Squares
5.7 An Application to Correlation and Variance
Supplementary Exercises for Chapter 5

86
86
87
89
91
93
95
98
99

CHAPTER 6 VECTOR SPACES

6.1 Examples and Basic Properties
6.2 Subspaces and Spanning Sets
6.3 Linear Independence and Dimension
6.4 Finite Dimensional Spaces
6.5 An Application to Polynomials
6.6 An Application to Differential equations
Supplementary Exercises for Chapter 6

101
101
104
106
110
112

CHAPTER 7 LINEAR TRANSFORMATIONS

7.1 Examples and Elementary Properties
7.2 Kernel and Image of a Linear Transformation
7.3 Isomorphisms and Composition
7.4 A Theorem about Differential Equations
7.5 More on Linear Recurrences

115
115
118
123
127
127

CHAPTER 8 ORTHOGONALITY
8.1 Orthogonal Complements and Projections
8.2 Orthogonal Diagonalization
8.3 Positive Definite Matrices
8.4 QR-Factorization
8.5 Computing Eigenvalues
8.6 Complex Matrices
8.7 An Application to Linear Codes over Finite Fields
8.8 An Application to Quadratic Forms
8.9 An Application to Constrained Optimization
8.10 An Application to Statistical Principal Component Analysis

131
131
133
138
139
140
141
144
146
150
150

CHAPTER 9 CHANGE OF BASIS

9.1 The Matrix of a Linear Transformation
9.2 Operators and Similarity
9.3 Invariant Subspaces and Direct Sums

151
151
157
160

CHAPTER 10 INNER PRODUCT SPACES

10.1 Inner Products and Norms
10.2 Orthogonal Sets of Vectors
10.3 Orthogonal Diagonalization
10.4 Isometries
10.5 An Application to Fourier Approximation

165
165
168
173
175
178

CHAPTER 11 CANONICAL FORMS

11.1 Block Triangular Form
11.2 Jordan Canonical Form

179
179
181

APPENDIX
A Complex Numbers
B Proofs
C Mathematical Induction

183
183
187
188

114

Section 1.1: Solutions and Elementary Operations

Chapter 1: Systems of Linear Equations

Solutions and Elementary Operations

Exercises 1.1

1(b) Substitute these values of x1 , x2 , x3 and x4 in the equation

2x1 + 5x2 + 9x3 + 3x4 = 2(2s + 12t + 13) + 5(s) + 9(s 3t 3) + 3(t) = 1
x1 + 2x2 + 4x3

(2s + 12t + 13) + 2(s) + 4(s 3t 3)

Hence this is a solution for every value of s and t.

2(b) The equation is 2x + 3y = 1. If x = s then y = 13 (1 2s) so this is one form of the general
solution. Also, if y = t then x = 21 (1 3t) gives another form.
4. Given the equation 4x 2y + 0z = 1, take y = s and z = t and solve for x: x = 41 (2s + 3).
This is the general solution.
5. (a) If a = 0, no solution if b = 0, infinitely many if b = 0. (b) If a = 0 unique solution x = b/a
for all b.



1 2 0
7(b) The augmented matrix is
0

(d) The augmented matrix is

8(b) A system with this augmented matrix is

2x

9(b)

Hence x = 3, y = 2.





1
4
3 4
(d)

5
4

Hence x = 17, y = 13.

10(b)

1
0

1
3
1
3

32
1
3
37

y + z =

= 1

3x + 2y + z =

1
9
10
9
73

4
1

13

. Hence x = 1 , y =
9

10
9 ,

z=

1
3

1
3

7
3 .

17

13

Section 1.2: Gaussian Elimination

11(b)

12

16

solution.


. The last equation is 0x + 0y = 36, which has no

5
36

14(b) False. The system x + y = 0, x y = 0 is consistent, but x = 0 = y is the only solution.

(d) True. If the original system was consistent the final system would also be consistent because
each row operation produces a system with the same set of solutions (by Theorem 1).
3(5x 2y ) + 2(7x + 3y  ) = 5

16 The substitution gives

; this simplifies to x = 5, y  = 1.

7(5x 2y ) + 5(7x + 3y  ) = 1

Hence x = 5x 2y  = 23 and y = 7x + 3y  = 32.

17 As in the Hint, multiplying by (x2 + 2)(2x 1) gives x2 x + 3 = (ax + b)(2x 1) + c(x2 + 2).
Equating coefficients of powers of x gives equations 2a + c = 1, a + 2b = 1, b + 2c = 3.
Solving this linear system we find a = 19 , b = 59 , c = 11
9 .
19 If John gets $x per hour and Joe gets $y per hour, the two situations give 2x + 3y = 24.6 and
3x + 2y = 23.9. Solving gives x = $4.50 and y = $5.20.

Exercises 1.2

Gaussian Elimination

1(b) No, No; no leading 1.

(d) No, Yes; not in reduced form because of the 3 and the top two 1s in the last column.
(f) No, No; the (reduced) row-echelon form would have two rows of zeros.

2(b)

5
4

2
1

3
0

1
1

11

42

13

13

11

0
1

0
0

0
0

0
0

11

13

11

0
1

3(b) The matrix is already in reduced row-echelon form. The nonleading variables are parameters;
x2 = r, x4 = s and x6 = t.
The first equation is x1 2x2 + 2x4 + x6 = 1, whence x1 = 1 + 2r 2s t.
The second equation is x3 + 5x4 3x6 = 1, whence x3 = 1 5s + 3t.
The third equation is x5 + 6x6 = 1, whence x5 = 1 6t.

Section 1.2: Gaussian Elimination

(d)
First carry the matrix to reduced

row-echelon form.
1

4(b)

The
The
The
The


3

2
1

nonleading variables are parameters; x3 = s, x5 = t.

first equation is x1 + 4x3 + 5x5 = 4, whence x1 = 4 4s 5t.
second equation is x2 + 2x3 x5 = 2, whence x2 = 2 2s + t.
third equation is x4 = 1.










1
1
1
1 0
1
2
1
2
1 2

3
3

17
37

. Hence x = 17 , y = 37 .

(d)

Note that the variables
order.


in the second equation


are in the wrong

2
2
3 1 2
1 31
3
1
3

. The nonleading variable y = t

6

is a parameter; then x =

0
2
3

0
+ 13 t

1
3 (t + 2).

2
(f) Again the order of the variables is reversed in the second equation.
2



2 3 5
. There is no solution as the second equation is 0x + 0y = 7.
0

5(b)

2
3

3
3

5
2

1
0

9
14
4

11

22

17
34

4
3

15

11

14

21

1
2
5

14

17
0

Take z = t (the nonleading variable). The equations give x = 21 15t, y = 17 11t.

1 2 1 2
1 2 1 2
1 2 1
2
0 1 1 3
0 1 1 3 .
(d) 2 5 3 1

0 0
0 7
1 4 3 3
0 2 2
1
There is no solution as the third equation is 0x + 0y + 0z = 7.

(f)

(h)

12

17

10

4
10
2

7
9
1

10
15
3

17

. Hence x = 7, y = 9, z = 1.

10

Section 1.2: Gaussian Elimination

0
0

. Hence z = t, x = 4, y = 3 + 2t.

6(b) Label the rows of the augmented matrix as R1 , R2 and R3 , and begin the gaussian algorithm
on the augmented matrix keeping track of the row operations:

35

R1

R2

R3

R2

32

R2 R1 .

R3 R1

At this point observe that R3 R1 = 4(R2 R1 ), that is R3 = 5R1 4R2 . This means
that equation 3 is 5 times equation 1 minus 4 times equation 2, as is readily verified. (The
solution is x1 = t 11, x2 = 2t + 8 and x3 = t.)

7(b)

1
1

x3 = 0.

(d)

0
3

14

14

14

2 ab

8(d)

14

14

14

8
4

0
25b
2ab ,

5+a
2ab

25b
2ab
5+a
2ab

5+a
y = 2ab
.

1
. Hence there is no solution if a = 5. If a = 5,
0 0 5+a



2
1
1
5
and the matrix is
. Then y = t, x = 1 + 25 t.

The unique solution is x =

1
b
Case 2 If ab = 2, it is
2
5

5+a

Case 1 If ab = 2, it continues

then b =

Hence x4 = t; x1 = 1, x2 = 1 t, x3 = 1 + t.






1
b
1
1
b
1
8(b)

.
a

. Hence x4 = t; x1 = 0, x2 = t,

1
2

b
2

1
2

b
2
a
2

ab
2

1
2

b
2

2a

2 ab

Section 1.2: Gaussian Elimination

Case 1 If a = 2 it continues:
The unique solution: x =

9(b)

b
c

1
2

2ab
2a .

12 t = 12 (1 t).

1
2

a 2c

b 2a + 4c

3
1

. Hence there is no solution if b = 1.

b1
2a
2ab
2a

2(1 b)

, so y = t, x =

b
2

b
2
2ab
2a

a 2c

0
1
2

1
2

1
2

y=

Case 2 If a = 2 the matrix is

If b = 1 the matrix is

1
2

b1
2a ,

a 2c
1

b 2a + 5c
3a b 6c

b 2a + 4c

Hence, for any values of a, b and c there is a unique solution x = 2a + b + 5c, y = 3a b 6c,
and z = 2a + b + 4c.

(d)

ac

ab

Case 1 If abc = 1, it continues:

ab

1 + abc

Hence we have the unique solution x = 0, y = 0, z = 0.

1 0 ab 0

, so z = t, x = abt, y = bt.
Case 2 If abc = 1, the matrix is
b
0
0 1

0 0
0
0
Note: It is impossible that there is no solution here: x = y = z = 0 always works.

(f)

a2

1
1

a2

1
1

2(a 1)

2(a 1)

1
0

Case 1 If a = 1 the matrix is

so y = t, x = t, z = 1.

a1

Case 2 If a = 0 the last equation is 0x + 0y + 0z = 1, so there is no solution.

Case
3 If a = 1 and a = 0,there
is a unique solution:

a1

1
0

1
0

Hence x = 1 a1 , y = 0, z = a1 .

a1

1
a

a1

Section 1.2: Gaussian Elimination

10(b)

1
2

12

3
2

; rank is 1.

(d) It is in row-echelon form; rank is 3.

(f)

11(b)

(d)

2
3

5
3
1

(f)

1
3
8

17
1

a2

1a

6a

2a

If a = 2 we get

If a = 0, a = 2, we get
12(b) False. A =

(d) False. A =

11

22

; rank = 3.

If a = 0 we get

1
4

1
2

; rank is 1.

a2

a2

4 2a2

2a

a2

a2

2a

a2

1
0

17

a2

a2

2a

4 a2

1
0
0

11

; rank is 2.

; rank = 2.

; rank = 2.

a2

2+a

; rank = 3.

(f) False. The system 2x y = 0, 4x + 2y = 0 is consistent, but the system 2x y = 1,

4x + 2y = 1 is not consistent

(h) True. A has 3 rows so there can be at most 3 leading 1s. Hence the rank of A is at most 3.

14(b) We begin the row reduction

b+c

c+a

a+b

b+c

ba

ab

ca

ac

. Now one of b a and

c a is nonzero (by hypothesis) so that row provides the second leading 1 (its row becomes

Section 1.2: Gaussian Elimination

[0 1

1]). Hence further row operations give

which has the given form.

b+c

1
0

b+c+a

16(b) Substituting the coordinates of the three points in the equation gives
1+1+a+b+c =0

a + b + c = 2

25 + 9 + 5a 3b + c = 0 5a 3b + c = 34
9 + 9 3a 3b + c = 0

1
2
1
2

34
18

5
6

24

24

2
6

3a + 3b c = 18

1
2

Hence a = 2, b = 6, c = 6, so the equation is x2 + y 2 2x + 6y 6 = 0.

18. Let a, b and c denote the fractions of the student popoulation in Clubs A, B and C respectively.
4
2
The new students in Club A arrived as follows: 10
of those in Club A stayed; 10
of those in
2
Club B go to A, and 10 of those in C go to A. Hence
2
2
4
10 a + 10 b + 10 c.

a=

Similarly, looking at students in Club B and C.

b=

1
7
10 a + 10 b

2
10 c

c=

5
1
10 a + 10 b

6
10 c.

Hence
6a + 2b + 2c = 0
a 3b + 2c = 0

6
1
5

5
8 t,

1
0
0

16

5a + b 4c = 0

14

16
14
7
= 8 t, c =

0
0
0

1
0
0

3
1
0

87

58

78

Thus the solution is a =

b
t. However a + b + c = 1 (because every student
5
7
8
belongs to exactly one club) which gives t = 52 . Hence a = 20
, b = 20
, c = 20
.

Section 1.3: Homogeneous Equations

Exercises 1.3
1(b) False. A =
(d) False. A =
(f) False. A =

Homogeneous Equations
1

(h) False. A =

2(b)

a2

a+3

Hence there is a nontrivial solution when a = 3 : x = 9t, y = 5t, z = t.

(d)

1
a

1a

1+a

a+1

Hence ifa = 1 and a = 1,

solution. The other cases are as follows:
there
is a unique, trivial

a=1:

a = 1 :

; x = t, y = t, z = 0.

; x = t, y = 0, z = t.

3(b) Not a linear combination. If ax + by + cz = v then comparing entries gives equations

2a + b + c = 4, a + c = 3 and a + b 2c = 4. Now carry the coefficient matrix to reduced
form:

Hence there is no solution.

1
0

(d) Here, if aa + by + cz = v then comparing entries gives equations 2a + b + c = 3, a + c = 0

and a + b 2c = 3. Carrying the coefficient matrix to reduced form gives

1
0

so the general solution is a = t, b = 3 + t and c = t. Taking t = 1 gives the linear

combination V = a + 2y z.
4(b) We must determine if x, y and z exist such that y = xa1 + ya2 + za3 . Equating entries here
gives equations x + 3y + z = 1, 3x + y + z = 9, 2y + z = 2 and x + z = 6. Carrying the
coefficient matrix to reduced form gives

Section 1.3: Homogeneous Equations

4
0

so the unique solution is x = 2, y = 1 and z = 4. Hence y = 2a1 a2 + 4a3 .

5(b) Carry the augmented matrix to reduced form:

1
2 1 1 1 0
1 2 0 2 3 0

1 2 2 0 1 0 0 0 1 1 2 0

1 2 3 1 3 0
0 0 0 0 0 0

Hence the general solution is x1 = 2r 2s 3t, x2 = r, x3 = s 2t, x4 = s and x5 = t.

In matrix form,
the general solution x = [x1 x2 x3 x4 x5 ]T takes the form

x=

2r 2s 3t

= r

s 2t
s
t

2
1
0
0
0

+ s

2
0

1
1
0

Hence X is a linear combination of the basic solutions.

5(d) Carry the augmented matrix to reduced form:

10

4
8

2s + 3t
s
t
0

2
1
0
0

2
0
1

= s

+t

Hence the general solution x = [x1 x2 x3 x4 x5 is

x=

]T

+ t

Hence X is a linear combination of the basic solutions.

3
0
1
0

x+y = 1
6(b) The system 2x + 2y = 2 has nontrivial solutions with fewer variables than equations.
x y = 1
7(b) There are n r = 6 1 = 5 parameters by Theorem 2 1.2.
(d) The row-echelon form has four rows and, as it has a row of zeros, has at most 3 leading
1s. Hence rank A = r = 1, 2 or 3 (r = 0 because A has nonzero entries). Thus there are
n r = 6 r = 5, 4 or 3 parameters.
9(b) Insisting that the graph of ax + by + cz + d = 0 (the plane) contains the three points leads
to three linear equations in the four variables a, b, c and d. There is a nontrivial solution by
Theorem 1.

10

Section 1.5: An Application to Electrical Networks

11. Since the system is consistent there are n r parameters by Theorem 2 Section 1.2. The
system has nontrivial solutions if and only if there is at least one parameter, that is if and
only if n > r.

Exercises 1.4

An Application to Network Flows

1(b) There are five flow equations, one for each junction:
f1 f2
f1

= 25
+ f3

+ f5

f2

= 50

+ f4

+ f7 = 60

f3 + f4

25

50

60

75

40

50

25

35

75

40

85

60

1
0

75

1
0

40
0

= 75

f5 + f6 f7 = 40
1

25

25

60

75

+ f6

40

85

60

35

40

40

If we use f4 , f6 , and f7 as parameters, the solution is

f1 = 85 f4 f7
f2 = 60 f4 f7

f3 = 75 + f4 + f6
f5 = 40 f6 + f7 .

2(b) The solution to (a) gives f1 = 55 f4 , f2 = 20 f4 + f5 , f3 = 15 f5 . Closing canal BC

means f3 = 0, so f5 = 15. Hence f2 = 35f4 , so f2 30 means f4 5. Similarly f1 = 55f4
so f1 30 implies f4 25. Hence the range on f4 is 25 f4 30.
3(b) The road CD.

Exercises 1.5

An Application to Electrical Networks

11

Section 1.5: An Application to Electrical Networks

2. The junction and circuit rules give:
Left junction

I1

I2

I3

Right junction

I1

I2

I3

Top circuit

5I1 + 10I2

Lower circuit

1
5
0

10

10

10

5
15 ,

0
0

Hence I1 =

3
5

I2 =

1
15
10

0
4
5.

4
5

10

and I3 =

10I2 + 5I3 = 10

1
0
0

1
3
2

2
15
3
5
4
5

4. The equations are:

Lower left junction

I1 I5 I6 = 0

Top junction

I2 I4 + I6 = 0

Middle junction

I2 + I3 I5 = 0

Lower right junction I1 I3 I4 = 0

Observe that the last of these follows from the others (so may be omitted).
Left circuit

10

10

10

1
0

1
1

Right circuit

10I3 + 10I4 = 10

Lower circuit

10I3 + 10I5 = 20

0
0

10

10

10

20

10

1
2

1
0

0
2

1
3

10I5 10I6 = 10

2
1

1
3

2
2

1
0
0

1
1

12

Supplementary Exercises - Chapter 1

1
2
3
2
3
2
1
2

Exercises 1.6

. Hence I1 = 2, I2 = 1, I3 = 1 , I4 = 3 , I5 = 3 , I6 = 1 .
2
2
2
2

An Application to Chemical Reactions

2. Suppose xN H3 + yCuO zN2 + wCu + vH2 O where x, y, z, w and v are positive integers.
Equating the number of each type of atom on each side gives
N : x = 2z

Cu : y = w

H : 3x = 2v

O:y=v

Taking v = t these give y = t, w = t, x = 23 t and z = 12 x = 13 t. The smallest value of t such

that there are all integers is t = 3, so x = 2, y = 3, z = 1 and v = 3. Hence the balanced
reaction is
2N H3 + 3CuO N2 + 3Cu + 3H2 O.
4. 15P b(N3 )2 + 44Cr(MnO4 )2 22Cr2 O3 + 88M nO2 + 5P b3 O4 + 90N O

Supplementary Exercises

Chapter 1

1(b) No. If the corresonding planes are parallel and distinct, there is no solution. Otherwise they
either coincide or have a whole common line of solutions.

2(b)

14

10

10

10

6
10
4
10

0
0

Hence x3 = s, x4 = t are parameters, and the equations give x1 =

1
x2 = 10
(1 + 4s t).

3(b)

If a = 1 the matrix is

1a

5 3a

4 a2

If a = 2 the matrix is

z = t.

a1

2
0
0

1
0
0

0
1
(16
10

1a

5 3a

4 a2

3(2 a)

6s 6t) and

16
10
1
10

6
10
1
10

4 a2

, so there is no solution.

, so x = 2 2t, y = t,

13

Supplementary Exercises - Chapter 1

If a = 1 and a = 2 there is a unique solution.

1a

5 3a

4 a2

3(2 a)

a2

5a+8
3(a1)
a2
3(a1)
a+2
3

3a5
a1

a2 4
a1
a+2
3

Hence x =

85a
3(a1) ,

y=

2
a1
3a5
a1

a2
3(a1) ,

z=

a+4
a1
a2 4
a1
a+2
3

a+2
3 .

4. If R1 and R2 denote the two rows, then the following indicate how they can be interchanged
using row operations of the other two types:

R
R2
R + R2
R + R2
R
2 .

1
1 1
R1
R1
R1
R2
R2
Note that only one row operation of Type II was used a multiplication by 1.

6. Substitute x = 3, y = 1 and z = 2 into the given equations. The result is

a

3 a + 2c = 0

2c = 3

that is

3b c 6 = 1

3b

= 9

3a + 2b

3a 2 + 2b = 5

= 7

This system of linear

for a, b and

equations

c has
unique solution:

8.

7
20

3
9

20

15

. Hence a = 1, b = 2, c = 1.

Hence the solution is x = 2, y = 3 t, z = t. Taking t = 3 i gives x = 2, y = i, z = 3 i,

as required.
If the real system has a unique solution, the solution is real because all the calculations in
the gaussian algorithm yield real numbers (all entries in the augmented matrix are real).

14
.

Supplementary Exercises - Chapter 1

15

Section 2.1: Matrix Addition, Scalar Multiplcation, and Tranposition

Chapter 2: Matrix Algebra

Exercises 2.1

Matrix Addition, Scalar Multiplication, and

Transposition

1(b) Equating entries gives four linear equations: a b = 2, b c = 2, c d = 6, d a = 2. The

solution is a = 2 + t, b = 4 + t, c = 6 + t, d = t.
(d) Equating coefficients gives: a = b, b = c, c = d, d = a. The solution is a = b = c = d = t, t
arbitrary.














3
6
1
9
30
7
2(b) 3
5
+7
=

+
2
1
3
10
7

1



9 30 + 7
14
=
=
3 10 7

20

(d) [3 1 2] 2 [9 3 4] + [3 11 6] = [3 1 2] [18 6 8] + [3 11
= [3 18 + 3 1 6 + 11 2 8 6] = [12 4 12]

(f)

(d) B + D is not defined as B is 2 3 while D is 3 2.

T



T

2+3
11
5
0
5
=
(f) (A + C)T =
=

3(b) 5C 5

T
3
2

2
1
0

0+2

15
10

1 + 0

=3
5
0

(h) 3

2
1

1
0

1
2

6
3

3
0



2
4

2
6

6]

4
1

1
6

(h) A D is not defined as A is 2 2 while D is 3 2.








2
3
2
3
.
= 5A 2
, subtract 3A from both sides to get
= 2A 2
4(b) Given 3A +
1
0
1
0







8
3
2
3
. Finally, multiply both sides
to both sides: 2A =
+2
=
Now add 2
0
1
0
1



8
4
by 21 : A = 12
= 1 .
1

5(b) Given 2A B = 5(A + 2B), add B to both sides to get

2A = 5(A + 2B) + B = 5A + 10B + B = 5A + 11B.
Now subtract 5A from both sides: 3A = 11B. Multiply by 31 to get A = 11
3 B.

16

Section 2.1: Matrix Addition, Scalar Multiplcation, and Tranposition

6(b) Given 4X + 3Y = A , subtract the first from the second to get X + Y = B A. Now
5X + 4Y = B
subtract 3 times this equation from the first equation: X = A 3(B A) = 4A 3B. Then
X + Y = B A gives Y = (B A) X = (B A) (4A 3B) = 4B 5A.
Note that this also follows from the Gaussian Algorithm (with matrix constants):


1
1
B
B

A
5
4
4 3
A

5 4 B
4 3 A
4 3
A






1
1
BA
1 0 4A 3B

.
0

5A 4B

4B 5A

7(b) Given 2X5Y = [1 2] let Y = T where T is an arbitrary 12 matrix.

Then 2X = 5T +[1 2]

so X = 52 T + 21 [1 2] , Y = T . If T = [s t] , this gives X = 52 s + 12 52 t + 1 , Y = [s t],

where s and t are arbitrary.
8(b) 5[3(A B + 2C) 2(3C B) A] + 2[3(3A B + C) + 2(B 2A) 2C]
= 5[3A 3B + 6C 6C + 2B A] + 2[9A 3B + 3C + 2B 4A 2C]
= 5 [2A B] + 2[5A B + C]
= 10A 5B + 10A 2B + 2C
= 20A 7B + 2C.



a b
9(b) Write A =
. We want p, q, r and s such that

=p

+q

+r

+s

p+q+r

q+s

r+s

Equating components give four linear equations in p, q, r and s:

p + q + r
q

= a
+ s = b

r + s = c
p

= d

The solution is p = d, q = 21 (a + b c d), r = 21 (a b + c d), s = 21 (a + b + c + d).

A + A = 0

11(b)

A + (A + A ) = A + 0 (add A to both sides)

(A + A) + A = A + 0 (associative law

0 + A = A + 0 (definition of A)

13(b) If A =

a1

A = A

0
..
.

a2
..
.

0
..
.

an

(property of 0)

and B =

b1

0
..
.

b2
..
.

0
..
.

bn

17

Section 2.2: Equations, Matrices and Transformations

a1 b1

then A B =

14(b)

st

0
..
.

a2 b2
..
.

0
..
.

an bn

so A B is also diagonal.

is symmetric if and only if t = st; that is t(s 1) = 0; that is s = 1 or t = 0.

(d) This matrix is symmetric if and only if 2s = s, 3 = t, 3 = s + t; that is s = 0 and t = 3.

 


 T


 T



8 0
1 0
1 0
2 0
T
T
T
15(b)
= 3A + 2
= (3A ) + 2
= 3A +
.
3 1
0 2
0 2
0 4












6
0
2
0
8 0
2 0
6
0
=
.
Hence 3A =

=
, so A = 13
3



 T


1
0
= 2A 5
(d) 4A 9
=
5
1 0
1 2







9
9
5 5
4
14
Hence 2A =

=
.
9 0
0 10
9 10





4
14
2
7
=
.
Finally A = 21
9
1

(2AT )T

10

1
0

1
2

16(b) We have AT = A as A is symmetric. Using Theorem 2: (kA)T = kAT = kA; so kA is

symmetric.
19(b) False. Take B = A for any A = 0.
(d) True. The entries on the main diagonal do not change when a matrix is transposed.
(f) True. Assume that A and B are symmetric, that is AT = A and B T = B. Then Theorem 2
gives
(kA + mB)T = (kA)T + (mB)T = kAT + mB T = kA + mB.
for any scalars k and m. This shows that the matrix kA + mB is symmetric.
20(c) If A = S +W as in (b), then AT = S T +W T = S W . Hence A+AT = 2S and AAT = 2W ,
so S = 21 (A + AT ) and W = 21 (A AT ).
22(b) If A = [aij ] then (kp)A = [(kp)aij ] = [k(paij )] = k[paij ] = k(pA).

Exercises 2.2

Equations, Matrices and Transformations

1(b) x1 3x2 3x3 + 3x4 = 5

8x2

+ 2x4 = 1

x1 + 2x2 + 2x3 +

= 2

x2 + 2x3 5x4 = 0

18

Section 2.2: Equations, Matrices and Transformations

2(b) x1

+ x2

1
2
3

2
4

By Theorem 4:


1
2
AX =
0

(d) By Definition 1:

AX =

AX =

5(b)

1
1
1

x3

x1
x2
x3

7
9

= x1

x1
x2
x3
x4

1
0

x2
x3
x4

+ x2

= x1

Observe that

2
2
0

8x1 + 7x2 3x3

(d)

2
7

4x2 + 5x3

x1 + 2x2 + 3x3
4x2 + 5x3

+ x3

8x1 + 7x2 3x3

4
6
4

x
y
z

1
3

2
3

0
0

2
1

+ x4

5
0

0 x1 + 2 x2 + 1 x3 + 5 x4

2x2 + x3 + 5x4

x1 + 2x2 + 3x3

2 + t
2 3t
t

2
2
0

+ t

is a solution to the given system of equations, and

(8) x1 + 7 x2 + (3) x3 + 0 x4

1
1

12

3 x1 + (4) x2 + 1 x3 + 6 x4

3x1 4x2 + x3 + 6x4

+ x2

+ x3

2x2 + x3 + 5x4

to the associated homogeneous system.

Hence x = t 2, y = 2 3t, z = t; that is

3x1 4x2 + x3 + 6x4

=
0

0 x1 + (4) x2 + 5 x3

1 x1 + 2 x2 + 3 x3

x1

+ x4

x2

By Theorem 4:

x1

+ x3

3(b) By Definition 1:


1
2
3

AX =
0

3
1

1
3
1

is a solution

19

Section 2.2: Equations, Matrices and Transformations

so

x1

x2
x3
x4

Here

9
2
0

. Hence x1 = 3 t, x2 = 4t 9, x3 = t 2, x4 = t,

3t

9 + 4t
2 + t
t

9
2
0

+ t

4
1
1

is a solution to the given equations, and

is a solution to the associated

4
1
1

homogeneous equations.

6 To say that x0 and x1 are solutions to the homogeneous system Ax = 0 of linear equations
means simply that Ax0 = 0 and Ax1 = 0. If sx0 + tx1 is any linear combination of x0 and
x1 , we compute:
A(sx0 + tx1 ) = A(sx0 ) + A(tx1 ) = s(Ax0 ) + t(Ax1 ) = s0 + t0 = 0
using Theorem 2. This shows that sx0 + tx1 is also a solution to Ax = 0.
8(b) The reduction of the augmented matrix is

2
6

11

11

7
3

is the general solution. Hence X =

10(b) False.



2
1

0
0

3
0

1
0
0

+ s

2
1
0
0
0

3
0

+ t

has a zero entry, but

so X =

5
0
2
0
1

3 + 2s 5t
s

1 + 2t

0
t

is the desired expression.

has no zero row.

(d) True. The linear combination x1 a1 + + xn an equals Ax where, by Theorem 1, A =

[a1 an ] is the matrix with these vectors ai as its columns.






2
1
1 1 1
, and this is not a linear
(f) False. If A =
and X = 0 then AX =
2

combination of

(h) False. If A =

1
2

1
1

and

1
1

1
2

1
1

because it is not a scalar multiple of


, there is a solution

1
2
1

for B =

1
2

0
0

. But there is no

20

Section 2.3: Matrix Multiplication

solution for B =




. Indeed, if

1
0

x
y
z

1
0

then x y + z = 1 and

x + y z = 0. This is impossible.




x
y
11(b) If
is reflected in the line y = x the result is
; see the diagram for Example 12,
y
x









x
y
0 1
x
0 1
2.4. In other words, T
=
=
. So T has matrix
.
y

(d) If

x
y

x
y
z

is rotated clockwise through the result is

x







0
1
x
0
1
=
so T has matrix
.
1

13(b) The reflection of

y
z

x
y
z

; see Example 14. Hence T

x
y

in the y-z plane keeps y and z the same and negates x. Hence

x
y
z

, so the matrix is

16 Write A = [a1 a2 an ] where ai is column i of A for each i. If B = x1 a1 + x2 a2 + + xn an

where the xi are scalars, then Ax = b by Theorem 1 where x = [x1 x2 xn ]T ; that is x is
a solution to the system Ax = b.
18(b) We are given that x1 and x2 are solutions to Ax = 0; that is Ax1 = 0 and Ax2 = 0. If t
is any scalar then, by Theorem 2, A(tx1 ) = t(Ax1 ) = t 0 = 0. That is, tx1 is a solution to
Ax = 0.
22 Let A = [a1 a2 an ] where ai is column i of A for each i, and write x = [x1 x2 xn ]T
and y = [y1 y2 yn ]T . Then
x + y = [x1 + y1 x2 + y2 xn + yn ]T

Hence we have
A(x + y) = (x1 + y1 )a1 + (x2 + y2 )a2 + + (xn + yn )an

Definition 1

= (x1 a1 + y1 a1 ) + (x2 a2 + y2 a2 ) + + (xn an + yn an )

Theorem 1 2.1

= (x1 a1 + x2 a2 + + xn an ) + (y1 a1 + y2 a2 + + yn an )

Theorem 1 2.1

= Ax + Ay.

Exercises 2.3
1(b)

1
2

1
0

2
4

Definition 1

Matrix Multiplication
2

212

4+04

39+0

6+0+0

17+4

2+0+8

1
0

6
6

2
10

21

Section 2.3: Matrix Multiplication

3]

(d) [1 3

1 3]

(f) [1

(h)

(j)



2
8

= [3 6 + 0 0 + 3 18] = [3 15]

a

b

c

= [2 1 24] = [23]

65

10 10

3 + 3

5 + 6

aa + 0 + 0

0+0+0

0+0+0

0+0+0

0 + bb + 0

0+0+0

0+0+0

0+0+0

0+

0 + cc

aa

bb

cc

2(b) A2 , AB, BC and C 2 are all undefined. The other productsare





2 12
1 4 10
7
6
2

2
6
, AC =
BA =
,B =
, CB =
1

CA =

2a + a1

2b + b1

a + 2a1

b + 2b1

Equating coefficients gives linear equations

5(b) A(BC) =

6(b) If A =

1
0

1
1



then A

2b + b1 = 2

a + 2a1 = 1

b + 2b1 = 4

2a + a1 = 7

The solution is: a = 3, a1 = 1; b = 0, b1 = 2.







8 2
2
2
6 0
2
4(b) A A 6I =

=

10

3(b) The given matrix equation becomes

16
1




14

and a = d. Hence A has the form A =

17

1

0
a
c

2
3

1
1

b
A becomes
d

0
, as required.

2
0

0
0

= (AB)C.

whence b = 0

7(b) If A is m n and B is p q then n = p because AB can be formed and q = m because BA

can be formed. So B is n m, A is m n.






1
0
1
0
1
1
1 0
1
0
1
1
,
,

8(b) (i)
,
,
(ii)
0 1
0 1
0 1
0 1
0 0
0 0

22

Section 2.3: Matrix Multiplication

12(b) Write A =
where P
,X=
, and Q =
. Then P X +
0 Q
0
1
0
0
0
1











2 1
2 1
P 2 P X + XQ
P2
0
XQ =
+
= 0, so A2 =
=
. Then A4 =
2
0
0
0
0
0
Q
0
Q2











P2
0
P2
0
P4
0
P6
0
6
4
2
=
, A = A A =
, . . . ; in general we claim
2
2
4
6
Q

that

A2k =

P 2k

Q2k

for k = 1, 2, . . .

(*)

This holds for k = 1; if it holds for some k 1 then

2(k+1)

2k

=A A =

P 2k

Q2k



P2

Q2

P 2(k+1)

Q2(k+1)

Hence (*) follows

byinduction



in k. 
1
2
1 3
Next P 2 =
, P3 =
, and we claim that
0

for m = 1, 2, . . .

(**)

It is true for m = 1; if it holds for some m 1, then

P

m+1

=P P =

1
0

m
1



1
0

(m + 1)
1

which proves (**) by induction.

As to Q, Q2 = I so Q2k = I for all k. Hence (*) and (**) gives

A2k =

Finally
A2k+1

13(b)

(d) I

A2k

(2k + 1)

A=



I
0

[X

X
I

P 2k

0 I
2

0

1

0

I]T = I

P 2k



1
1
1

I 2 + X0

IX + XI
0X + I 2

0I + I0

X T
I

2k

1
0

P 2k+1

P 2k X

for

= I2k

= IX T + X T I = Ok

k 1.

23

Section 2.3: Matrix Multiplication

0 X

(f)

0 X
0

0 X
0

0 X

0 X
0

0 X
0

0 X
I

X2

Continue. We claim that

0 X

2m

X2

X2

X2

Xm

=
for m 1. It is true if m = 1 and,
I 0
0 Xm
if it holds for some m, we have

2(m+1)
2m
2

0 X
0 X
0 X
Xm
0
X 0
X m+1
0

=
.
=
I 0
I 0
I 0
0 Xm
0 X
0
X m+1

Hence the result follows by induction on m. Now

2m+1
2m

m
m+1
0 X
0 X
0 X
X
0
0 X
0 X

=
m
I 0
I 0
I 0
0 X
I 0
Xm
0

for all m 1. It also holds for m = 0 if we take X 0 = I.

14(b) If Y A = 0 for all 1 m matrices Y , let Yi denote row i of Im . Then row i of Im A = A is

Yi A = 0. Thus each row of A is zero, so A = 0.
16(b) A(B + C D) + B(C A + D) (A + B)C + (A B)D
= AB + AC AD + BC BA + BD AC BC + AD BD
= AB BA.
(d) (A B)(C A) + (C B)(A C) + (C A)2 = [(A B) (C B) + (C A)] (C A) =
0(C A) = 0.
18(b) We are given that AC = CA, so (kA)C = k(AC) = k(CA) = C(kA), using Theorem 1.
Hence kA commutes with C.
20 Since A and B are symmetric, we have AT = A and B T = B. Then Theorem 2 gives
(AB)T = B T AT = BA. Hence (AB)T = AB if and only if BA = AB.

22(b) Let A =

. Then the entries on the main diagonal of A2 are a2 +x2 +y 2 , x2 +b2 +z 2 ,

y 2 + z 2 + c2 . These are all zero if and only if a = x = y = b = z = c = 0; that is if and only

if A = 0.

24 If AB = 0 where A = 0, suppose BC = I for some matrix C. Left multiply this equation by

A to get A = AI = A(BC) = (AB)C = 0C = 0, a contradiction. So no such matrix C exists.

24

Section 2.3: Matrix Multiplication

26. We have A =

, and hence A3 =

. Hence there are 3 paths

of length 3 from v1 to v4 because the (4,1)-entry of A3 is 3. Similarly, the fact that the
(3,2)-entry of A3 is 0 means that there are no paths of length 3 from v2 to v3 .



1 0
27(b) False. If A =
= J then AJ = A, but J = I.
0

(d) True. Since A is symmetric, we have AT = A. Hence Theorem 2 2.1 gives (I + A)T =
I T + AT = I + A. In other words, I + A is symmetric.



0 1
(f) False. If A =
then A =
 0 but A2 = 0.
0

(h) True. We are assuming that A commutes with A + B, that is A(A + B) = (A + B)A.
Multiplying out each side, this becomes A2 + AB = A2 + BA. Subtracting A2 from each side
gives AB = BA; that is A commutes with B.






2 4
1
2
(j) False. Let A =
and B =
. Then AB = 0 is the zero matrix so both
1

columns are zero. However B has no zero column.





1
2
2 4
(l) False. Let A =
and B =
as above. Again AB = 0 has both rows zero,
2

but A has no row of zeros.


28(b) If A = [aij ] the sum of the entries in row i is nj=1 aij = 1. Similarly for B = [bij ]. If
AB =C = [cij ] then cij is the dot product of row i of A with column j of B, that is
cij = nk=1 aik bkj . Hence the sum of the entries in row i of C is

n
n 
n
n
n
n





cij =
aik bkj =
aik
bkj =
aik 1 = 1.
j=1

j=1 k=1

k=1

j=1

k=1

Easier Proof: Let X be the n1 column matrix with every entry equal to 1. Then the entries
of AX are the row sums of A, so these all equal 1 if and only if AX = X. But if also BX = X
then (AB)X = A(BX) = AX = X, as required.

30(b) If A = [aij ] then the trace of A is the sum of the entries on the main diagonal, that is
tr A = a11 + a22 + + ann . Now the matrix kA is obtained by multiplying every entry of A
by k, that is kA = [kaij ]. Hence
tr (kA) = ka11 + ka22 + + kann = k(a11 + a22 + + ann ) = k tr A.
(e) If A = [aij ] the transpose AT is obtained by replacing each entry aij by the entry aji directly
across the main diagonal. Hence, write AT = [aij ] where aij = aji for all i and j. Let bi
T
denote the (i,
of AA
the dot product of row i of A and column i of AT ,
i)-entry
n. Then bi is 
n

that is bi = k=1 aik aki = k=1 aik aik = nk=1 a2ik . Hence we obtain
 n

n
n
n 
n




T
2
a2ik .
tr (AA ) =
bi =
aik =
i=1

This is what we wanted.

i=1

k=1

i=1 k=1

25

Section 2.4: Matrix Inverses

32(e) We have Q = P + AP P AP so, since P 2 = P,
P Q = P 2 + P AP P 2 AP = P + P AP P AP = P.
Hence Q2 = (P + AP P AP )Q = P Q + AP Q P AP Q = P + AP P AP = Q.
34(b) We always have
(A + B)(A B) = A2 + BA AB B 2 .

If AB = BA, this gives (A+B)(AB) = A2 B 2 . Conversely, suppose that (A+B)(AB) =

A2 B 2 . Then
A2 B 2 = A2 + BA AB B 2 .
Hence 0 = BA AB, whence AB = BA.
35(b) Denote B = [b1 b2 bn ] = [bj ] where bj is column j of B. Then Definition 1 asserts that
AB = [Ab1 Ab2 Abn ] = [Abj ],
that is column j of AB is Abj for each j. Note that multiplying a matrix by a scalar a is the
same as multiplying each column by a. This, with Definition 1 and Theorem 2 2.2, gives
Definition 1

a(AB) = a[Abj ]
= [a(Abj )]

Scalar Multiplication

= [A(abj )]

Theorem 2 2.2
Definition 1

= A(aB).
Similarly,

Definition 1

a(AB) = a[Abj ]
= [a(Abj )]

Scalar Multiplication
Theorem 22.2

= [(aA)bj )]
= (aA)B.

Definition 1

This proves that a(AB) = A(aB) = (aA)B, as required.

36 See the article in the mathematics journal Communications in Algebra, Volume 25, Number
7 (1997), pages 1767 to 1782.

Exercises 2.4

Matrix Inverses

2 In each case we need row operations that carry A to I; these same operations carry I to A1 .
In short [A I] [I A1 ]. This is called the matrix inversion algorithm.
(b)

We begin
row 2 from row
by subtracting



 1.


4 1 1 0
1 1 1 1
1 1
1
1

3 2 0 1
0
3 2
0 1
 5 3 4


2
1 1
1
1
1 0
15
5

. Hence the inverse is

3
4
3
4
0

1
5

2
3

1
4

26

Section 2.4: Matrix Inverses

(d)

1
5

(f)

1
3

11

1
0

1
10

Hence A1 =

10

14

. So A1 =

0
1

(j)

2
5

1 1

2
4

54

34

2
4
2
4

24

1
0

0
1

Hence A1 =

2
2

34

. Hence A1 =

5
4
14

1
10
2
10
9
10

(h)
We begin by subtracting
row 2from twice row 1:
3

4
10
2
10
14
10

1
10
2
10
1
10

2
1

14

2
4

14

0 2

2
10
1

2
10
2

5 2 5 .

3 2 1
5

1
4

4
1
2

2
2

27

Section 2.4: Matrix Inverses

(l)

Hence

A1

30

15

210

30

105
7

15

5
1
0

35

210
105
35

7
1

3(b) The equations are AX = B where A =

,X=
,B=
. We have (by the
1 4
y
1

4 3
. Left multiply AX = B by A1 to get
algorithm or Example 4) A1 = 51
1 2
1

X =A

AX = A

Hence x = 53 and y = 25 .

(d) Here A =

By the algorithm, A1 =

1
5

14

10

15

1
5

1
1 =
0

, x =

B=

y
z
9

4
10

4
15

1
5



0
1

1
5

3
2

The equations have the form Ax = b, so left multiplying by A1 gives

x = A1 (Ax) = A1 b =

Hence x =

23
5 ,

23
5
8
5

, b =
14

1
5

1
5

y = 85 , and z = 25
5 = 5.

14

10

4(b) We want B such that AB = P where P =

1
0
1

15

1
0

1
5

23
8
25

. Since A1 exists left multiply this

equation by A1 to get B = A1 (AB) = A1 P . [This B will satisfy our requirements because

AB = A(A1 P ) = IP = P ]. Explicitly

B = A1 P =

0
1

28

Section 2.4: Matrix Inverses

(2A)T

5(b) By Example 4, we have


3
1
1
T
2A = 5
so AT =
2

1
10

=

1
2
1

A = (A ) =

(d) We have (I

2AT )1

T T

Thus

=I

This gives AT =

1
2

1
1

1
2

(f) Given

Now

 1

A
=

1

(h) Given

(A1

2I)T

= 2

A
Hence

A1

= 2I 2

1 1

A = (A

= 2

6(b) Have A = (A1 )1 =

1
2

A=

T

1
10

. Since (2A)T = 2AT , we get

1

=


1
1

1
2

1
2

.


T

1

1
2

, so left multiply by this to obtain

, take inverses to get



1
2

1
1
1

1
2

, take transposes to get



2I = 2

, so

A=

so (because (U 1 )1 = U for any invertible matrix U)

A = (A ) =
1

1
10

T T

1
5


. Finally

(I 2AT ) =
2AT

1

 T


=

2
1
2

1
2

= 2

0
1

2
0
2

1
1

1
1

1
3
1

T

= 2

1
1

1
2

=2

0
1

by the algorithm.

1
1

1
2

. Finally

29

Section 2.4: Matrix Inverses

x

8(b) The equations are A

=
and
= B
where A =
y
1
y
y





5
4
. Thus B = A1 (by Example 4) so the substitution gives
4
3













x
x
x
x
7


AB
=I
=
. Thus x = 7, y = 1 so
=B
=



y
y
y
y
1



31
.

3
4
7
1

5
4

and B =


x
= A
=
y


4
7
=

25

9(b) False. A =

and B =

are both invertible, but A + B =

is not.

(d) True. If A4 = 3I then A( 13 A3 ) = I = ( 13 A3 )A, so A1 = 31 A3 .





1 0
1 0
(f) False. Take A =
and B =
. Then AB = B and B = 0, but A is not
0 0
0 0


0
invertible by Theorem 5 since Ax = 0 where x =
.
1

(h) True. Since A2 is invertible, let (A2 )B = I. Thus A(AB) = I, so AB is the inverse of A by
Theorem 5.
10(b) We are given C 1 = A, so C = (C 1 )1 = A1 . Hence C T = (A1 )T . This also has the form
C T = (AT )1 by Theorem 4. Hence (C T )1 = AT .
11(b) If a solution x to Ax = b exists, it can be found by left multiplication by C : CAx = Cb,
Ix = Cb, x = Cb.




3
3
(i) x = Cb =
here but x =
is not a solution. So no solution exists.
0
0



2
(ii) x = Cb =
in this case and this is indeed a solution.
1

1
so B 4 = (B 2 )2 =
1
0
1
0
0
1
0




1
0
Thus B B 3 = I = B 3 B, so B 1 = B 3 = B 2 B =

15(b) B 2 =



16 We use the algorithm:

c2

ues of c.

c2

c
1
c

0
1

. Hence

0
1

= I.


1
=

0
1

c2
3

c2

c2

for all val-

18(b) Suppose column j of A consists of zeros. Then Ay = 0 where y is the column with 1 in the
position j and zeros elsewhere. If A1 exists, left multiply by A1 to get A1 Ay = A1 0,
that is Iy = 0; a contradiction. So A1 does not exist.

30

Section 2.4: Matrix Inverses

(d) If each column of A sums to 0, then xA = 0 where x is the row of 1s. If A1 exists, right
multiply by A1 to get xAA1 = 0A1 , that is xI = 0, x = 0, a contradiction. So A1 does
not exist.

19(bii) Write A =

1
0

. Observe that row 1 minus row 2 minus row 3 is zero. If X =

[1 1 1] , this means XA = 0. If A1 exists, right multiply by A1 to get XAA1 =

0A1 , XI = 0, X = 0, a contradiction. So A1 does not exist.

20(b) If A is invertible then each power Ak is also invertible by Theorem 4. In particular, Ak = 0.

21(b) If A and B both have inverses, so also does AB (by Theorem 4). But AB = 0 has no inverse.







x
ax
a 0
x
22. If a > 1, the x-expansion T : R2 R2 is given by T
=
=
. We
y
y
0 1
y


1
1

a 0
0
a
=
have
, and this is an X-compression because a1 < 1.
0

24(b) The condition can be written as A(A3 + 2A2 I) = 4I, whence A[ 41 (A3 + 2A2 I)] = I.
By Corollary 1 of Theorem 5, A is invertible and A1 = 14 (A3 + 2A2 I). Alternatively, this
follows directly by verifying that also [ 14 (A3 + 2A2 I)]A = I.
25(b) If Bx = 0 then (AB)x = 0 so x = 0 because AB is invertible. Hence B is invertible by
Theorem 5. But then A = (AB)B 1 is invertible by Theorem 4 because both AB and B 1
are invertible.



2
1
1
1
1
26(b) As in Example 11, B Y A = (1) [1 3]
= [13 8], so
5
3

1

1
0
3 1
3 1 0

5 2 0 =
0
 5 2 
=


(1)1
1 3 1
13 8

2
1 0
5
3 0 .
13
8 1
(d) As in Example 11,

1
1

1
1

A1 XB 1
1

1
2

1
2

14

16

1

1
1

1
2



14
16

1
1



9
1
1
2

2
1

1
1

BB

16

, so

26(b) If A1 and B 1 exist, use block multiplication to compute







A X
A1 A1 XB 1
AA1 AA1 XB 1 + XB 1
=
=
1
1
0

14

= I2n

31

Section 2.4: Matrix Inverses

where A and B are n n. The product in the reverse order is also I2n so


1


A X
A1 A1 XB 1
=
.
1
0

28(d) If An = 0 write B = I + A + A2 + + An1 . Then

(I A)B = (I A)(I + A + A2 + + An1 )

= (I + A + A2 + + An1 ) A A2 A3 An
= I An
=I

Similarly B(I A) = I, so (I A)1 = B.

30(b) Assume that AB and BA are both invertible. Then
AB(AB)1 = I so AX = I where X = B(AB)1
(BA)1 BA = I so Y A = I where Y = (BA)1 B.
But then X = IX = (Y A)X = Y (AX) = Y I = Y , so X = Y is the inverse of A.
Different Proof. The fact that AB is invertible gives A[B(AB)1 ] = I. This shows that A is
invertible by the Corollary to Theorem 5. Similarly B is invertible.
31(b) If A = B then A1 B = A1 A = I. Conversely, if A1 B = I left multiply by A to get
AA1 B = AI, IB = A, B = A.
32(a) Since A commutes with C, we have AC = CA. Left-multiply by A1 to get C = A1 CA.
Then right-multiply by A1 to get CA1 = A1 C. Thus A1 commutes with C too.
33(b) The condition (AB)2 = A2 B 2 means ABAB = AABB. Left multiplication by A1 gives
BAB = ABB, and then right multiplication by B 1 yields BA = AB.
34 Assume that AB is invertible; we apply Part 2 of Theorem 5 to show that B is invertible. If
BX = 0 then left multiplication by A gives ABX = 0. Now left multiplication by (AB)1
yields X = (AB)1 0 = 0. Hence B is invertible by Theorem 5. But then we have A =
(AB)B 1 so A is invertible by Theorem 4 (B 1 and AB are both invertible).

35(b) By the hint, Bx = 0 where x =

1
3

so B is not invertible by Theorem 5.

36 Assume that A can be left cancelled. If Ax = 0 then Ax = A0 so x = 0 by left cancellation.

Thus A is invertible by Theorem 5. Conversely, if A is invertible, suppose that AB = AC.
Then left multiplication by A1 yields A1 AB = A1 AC, IB = IC, B = C.

38(b) Write U = In 2XX T . Then U is symmetric because

U T = InT 2(XX T )T = In 2X T T X T = In 2XX T = U.

32

Section 2.5: Elementary Matrices

Moreover U 1 = U because (since X T X = In )
U 2 = (In 2XX T )(I 2XX T )

= In 2XX T 2XX T + 4XX T XX T

= In 4XX T + 4XIm X T
= In .
39(b) If P 2 = P then I 2P is self-inverse because

(I 2P )(I 2P ) = I 2P 2P + 4P 2 = I.
Conversely, if I 2P is self-inverse then
I = (I 2P )2 = I 4P + 4P 2 .
Hence 4P = 4P 2 ; so P = P 2 .
41(b) If A and B are any invertible matrices (of the same size), we compute:
A1 (A + B)B 1 = A1 AB 1 + A1 BB 1 = B 1 + A1 = A1 + B 1 .
Hence A1 +B 1 is invertible by Theorem 4 because each of A1 , A+B, and B 1 is invertible.
Furthermore
(A1 + B 1 )1 = [A1 (A + B)B 1 ]1 = (B 1 )1 (A + B)1 (A1 )1 = B(A + B)1 A
gives the desired formula.

Exercises 2.5

Elementary Matrices

1(b) Interchange rows 1 and 3 of I, E 1 = E.

(d) Add (2) times row 1 of I to row 2. E 1 =

(f) Multiply row 3 of I by 5. E 1 =

1
5

2(b) A B is accomplished by negating row 1, so E =

1
0

0
0

(d) A B is accomplished by subtracting row 2 from row 1, so E =

(f) A B is acomplished by interchanging rows 1 and 2, so E =

33

Section 2.5: Elementary Matrices

3(b) The possibilities for E are

EA has a row different from C.



,



,



.

and

. In each case

4. If E is Type I, EA and A differ only in the interchanged rows.

If E is of Type II, EA and A differ only in the row multiplied by a nonzero constant.
If E is of Type II, EA and A differ only in the row to which a multiple of a row is added.

5(b) No. The zero matrix 0 is not invertible.

6(b)

12

1
0 2 6
5 1
0 1 3
52
2
0 1






1 0
7 12
1
1
0
7
1
2

so U A = R =
where U = 2
1
0 1 3 52
0 1 3
2
U is the product of the elementary matrices used at each stage:

12






U =

1
5

where E1 =

where E2 =

= E3 E2 E1 A where E3 =

= E1 A

= E2 E1 A

product of elementary matrices.

7(b)

2
1

1
1

0
1

Then E1 =




E1

and E2 =

1
0

12


. This matrix

=A

1
2

1
1

E2

, so U = E2 E1 .





6(d) Just as in (b), we get U A = R where R is reduced row-echelon, and

. So U =

is a

34
8(b)

Section 2.5: Elementary Matrices

v

1
0

1
1

0
1

= E1 A


= A

= E2 E1 A

= E3 E2 E1 A

= E4 E3 E2 E1 A

where E1 =

where E2 =

where E3 =

where E4 =

Thus E4 E3 E2 E1 A = I so

A = (E4 E3 E2 E1 )1
= E11 E22 E31 E41






0 1
1 0
=
1







Of course a different sequence of row operations yields a different factorization of A.

(d) Analogous to (b), A =

3
0
1

10. By Theorem 3, U A = R for some invertible matrix U. Hence A = U 1 R where U 1 is

invertible.












3 2 1 0
1 1 1 1
1
1 1
1
12(b) [A | I] =

2 1 0 1
0 1 2
3
2 1 0 1




1 2
1 0 1
2
. Hence, U A = R = I2 in this case so U = A1 .

so U =

0 1
2
3
2 3
Thus, r = rank A = 2 and, taking V = I2 , U AV = U A = I2 .

(d) [A | I] =

R=

1
0

1
1

2
3
2
3
4
0

. Hence, U A = R where U =

. Note that rank A = 2. Next,

2
3
2

and

35

Section 2.6: Matrix Transformations

[RT | I] =

Hence, (U AV )T = (RV )T = V T RT =

, so U AV =

so V T =

16. We need a sequence of elementary operations to carry [U A] to [I U 1 A]. By Lemma 1

these operations can be achieved by left multiplication by elementary matrices. Observe
[I

U 1 A] = [U 1 U

U 1 A] = U 1 [U

A].

(*)

Since U 1 is invertible, it is a product of elementary matrices (Theorem 2), say U 1 =

E1 E2 . . . Ek where the Ei are elementary. Hence (*) shows that [I U 1 A] = E1 E2 . . . Ek [U A],
so a sequence of k row operations carries [U A] to [I U 1 A]. Clearly [I U 1 A] is in reduced row-echelon form.
r

17(b) A A because A = IA. If A B, let A = U B, U invertible. Then B = U 1 A so B A.

r
r
Finally if A B and B C, let A = U B and B = V C where U and V are invertible. Hence
r
A = U(V C) = (U V )C so A C.



0 0 0
19(b) The matrices row-equivalent to A =
are the matrices U A whre U is invertible. If
0 0 1






a b
0 0 b
U =
then U A =
where b and d are not both zero (as U is invertible).
c d
0 0 d



a
b
Every such matrix arises use U =
it is invertible as a2 + b2 = 0 (Example 4
b

2.3).

22(b) By Lemma 1, B = EA where E is elementary, obtained from I by multiplying row i by k = 0.

Hence B 1 = A1 E 1 where E 1 is elementary, obtained from I by multiplying row i by k1 .
But then forming the product A1 E 1 is obtained by multiplying column i of A1 by k1 .

Exercises 2.6

1(b) Write A =

Matrix Transformations
3
2
1

,B =

2
0
5

and X =

5
6
13

. We are given T (A) and T (B), and are

asked to find T (X). Since T is linear it is enough (by Theorem 1) to express X as a linear
combination of A and B. If we set X = rA + sB, equating entries gives equations 3r + 2s = 5,
2r = 6 and r + 5s = 13. The (unique) solution is r = 3, s = 2, so X = 3A 2B. Since
T is linear we have







3
1
11
T (X) = 3T (A) 2T (B) = 3
2
=
.
5

11

36

Section 2.6: Matrix Transformations

2(b) LetA =

1
1
1
1

,B =

and X =

1
2

. We know T (A) and T (B); to find T (X) we

1
2
4

express X as a linear combination of A and B, and use the assumption that T is linear. If
we write X = rA + sB, equate entries, and solve the linear equations, we find that r = 2 and
s = 3. Hence X = 2A 3B so, since T is linear,

T (X) = 2T (A) 3T (B) = 2

3(b) In R2 , we have e1 =

1
0

and e2 =

0
1

0
1

. We are given that T (x) = x for each x in R2 .

In particular, T (e1 ) = e and T (e2 ) = e2 . Since T is linear, Theorem 2 gives

A = [T (e1 ) T (e2 )] = [e1 e2 ] =
Of course, T

x
y

x
y

x
y

1
1



0

x
y

1
0

0
1

for all

x
y

in R2 , so in this case

we can easily see directly that T has matrix

. However, sometimes Theorem 2 is
0
1
necessary.




1
0
(d) Let e1 =
and e2 =
. If these vectors are rotated counterclockwise through 4 , some
0
1

2 

2 
2
2

simple trigonometry shows that T (e1 ) =

and T (e2 ) =
. Since T is linear, the
2
2
2
2



2 2

matrix A of T is A = [T (e1 ) T (e2 )] = 21

.
2

4(b) Let e1 =
R3

0
0

, e2 =

0
1
0

and e3 =

0
0
1

is reflection in the u-z-plane, we have:

denote the standard basis of R3 . Since T : R3

T (e1 ) = e1 because e1 is perpendicular to the u-z-plane; while

T (e2 ) = e2 and T (e3 ) = e3 because e2 and e3 are in the u-z-plane.

So A = [T (e1 ) T (e2 ) T (e3 )] = [e1 e2 e3 ] =

5(b) Since y1 and y2 are both in the image of T, we have y1 = T (x1 ) for some x1 in Rn , and
y2 = T (x2 ) for some x2 in Rn . Since T is linear, we have
T (ax1 + bx2 ) = aT (x1 ) + bT (x2 ) = ay1 + by2 .
This shows that ay1 + by2 = T (ax1 + bx2 ) is also in the image of T.

37

Section 2.6: Matrix Transformations

7(b) It turns out that T2 fails for T : R2
R2 .T2 requires that T (ax) = aT (x) for all x in R2
0
and all scalars a. But if a = 2 and x =
then
1



T 2

0
1

=T

0
2

0
4

, while 2T

0
1

=2

0
1

0
2

Note that T1 also fails for this transformation T, as you can verify.











x
x+y
1
1
x
x
1
1
8(b) We are given T
= 2
= 2
for all
, so T is the matrix
y
x + y
1 1
y
y



1


1
1
1
2
2
transformation induced by the matrix A = 12
=
. By Theorem 4 we
1
1
1

recognize this as the matrix of the rotation R 4 . Hence T is rotation through = 4 .

(d) Here T

x
y

1
10

8x + 6y
6x 8y

1
10

formation induced by the matrix A =

6
1
10

6
8



x
y

6
8

for all

x
y

, so T is the matrix trans-

. Looking at Theorem 5, we see that A

is the matrix of Q3 . Hence T = Q3 is reflection in the line y = 3x.

x

10(b) Since T is linear, we have T

we have T

y
0

y
0

=T

y
z

because

y
0

+T

. Since T is rotation about the y axis,

0
z

is on the y axis. Now observe that T is rotation of

the x-z-plane through the angle from the x axis to the z axis. By Theorem 4 the effect of
T on the x-z-plane is given by








x
cos sin
x
x cos z sin

=
z

Hence T

x
0
z

sin

x cos z sin
0

x sin + z cos
x
y
z

= T

Hence the matrix of T is

cos

x sin + z cos

, and so
0
y
0

+T

x cos z sin
y

x sin + z cos

cos

sin

sin
0

cos

x
0
z

0
y
0

cos

+
0

sin

x cos z sin
0

x sin + z cos

sin
0

cos

x
y
z

12(b) Let Q0 denote

reflection
 in the x axis, and let R
denote rotation through . Then Q0 has
1
0
1
0
matrix A =
, and R has matrix B =
. Then R followed by Q0 is the
0

38

Section 2.6: Matrix Transformations

transformation Q0 R , and this has matrix AB =
matrix of reflection in the y axis.

1
0

0
1

by Theorem 3. This is the

(d) Let Q0 denote reflection in the x axis, and let R 2 denote rotation through 2 . Then Q0 has






1
0
0 1

matrix A =
, and R 2 has matrix B =
. Then Q0 followed by R 2 is the
0 1
1
0



0 1
by Theorem 3. This is the
transformation R 2 Q0 , and this has matrix BA =
1

matrix of reflection Q1 in the line with equation y = x.

(f) Let Q0 denote reflection

in the
in the line y = x. Then


 x axis, and let Q1 denote

reflection

1
0
0 1
Q0 has matrix A =
, and Q1 has matrix B =
. Then Q0 followed by Q1 is
0 1

1 0 
0 1
the transformation Q1 Q0 , and this has matrix BA =
by Theorem 3. This is the
1

matrix of rotation R 2 about the origin through the angle 2 .

13(b) Since R has matrix A, we have R(x) = Ax for all x in Rn . By the definition of T we have
T (x) = a R(x) = a(Ax) = (aA)x
for all x in Rn . This shows that the matrix of T is aA.
14(b) We use Axiom T2: T (x) = T [(1)x] = (1) T (x) = T (x).
17(b) The matrix of T is B, so T (x) = Bx for all x in Rn . Let B 2 = I. Then
T 2 (x) = T [T (x)] = B[Bx] = B 2 x = Ix = x = 1R2 (x) for all x in Rn .
Hence T 2 = 1Rn since they have the same effect on every column x.
Conversely, if T 2 = 1Rn then
B 2 x = B(Bx) = T (T (x)) = T 2 (x) = 1R2 (x) = x = Ix for all x in Rn .
This implies that B 2 = I by Theorem 5 2.2.


1

18 The matrices of Q0 , Q1 , Q1 and R 2 are



,



,

0
1

1
0

and

0
1

1
0

respectively. We use Theorem 3 repeatedly: If S has matrix A and T has matrix B then S T
has matrix AB.







0 1
1
0
0 1
=
, which is the matrix of R 2 .
(b) The matrix of Q1 Q0 is
1

(d) The matrix of Q0 R 2 is



0
1

1
0

0
1

1
0

which is the matrix of Q1 .

19(b) We have Pm [Qm (x)] = Pm (x) for all x in R2 because Qm (x) lies on the line y = mx. This
means Pm Qm = Pm .

39

Section 2.6: Matrix Transformations

20 To see that T is linear, write x = [x1 x2 xn ]T and y = [y1 y2 yn ]T . Then:


T (x + y) = T [x1 + y1 x2 + y2 xn + yn ]T
= (x1 + y1 ) + (x2 + y2 ) + + (xn + yn )

= (x1 + x2 + + xn ) + (y1 + y2 + + yn )
= T (x) + T (y),
T (ax)

= T ([ax1 ax2 axn ]T )

= ax1 + ax2 + + axn
= a(x1 + x2 + + xn )
= a T (x).

Hence T is linear, so its matrix is A = [T (e1 ) T (e2 ) T (en )] = [1 1 1] by Theorem 2.

Note that this can be seen directly because

x1

x2
..
.

xn

= x1 + + xn = [1 1 1]

x1
x2
..
.
xn

so we see immediately that T is the matrix transformation induced by [1 1 1]. Note that
this also shows that T is linear, and so avoids the tedious verification above.
22(b) Suppose that T : Rn R is linear. Let e1 , e2 , , en be the standard basis of Rn , and write
T (ej ) = wj for each j = 1, 2, , n. Note that each wj is in R. As T is linear, Theorem 2
asserts
T has matrix A = [T (e1 ) T (e2 ) T (en )] = [w1 w2 wn ]. Hence, given
that
x1

x=

x2
..
.

xn

in Rn , we have

T (x) = Ax = [w1 w2 wn ]

x1
x2
.
..
xn

= w1 x1 + w2 x2 + + wn xn = w x = Tw (x) for

all X in Rn where w = [w1 w2 wn ]T . This means that T = Tw . This can also be

seen without Theorem 2: We have x = x1 e1 + x2 e2 + + xn en so, since T is linear,
T (X) = T (x1 e1 + x2 e2 + + xn en )
= x1 T (e1 ) + x2 T (e2 ) + + xn T (en )
= x1 w1 + x2 w2 + + xn wn
= wx
= Tw (x).
for all x in Rn . Thus T = Tw .

40

Section 2.7: LU-factorization

T

24(b) Given linear transformations Rn Rm Rk we are to show that (S T )(ax) = a (S T )(x)

for all x in Rn and all scalars a. The proof is a straight forward computation:
(S T )(ax) = S[T (ax)]

Definition of S T

= S[a T (x)]

T is linear

= a [S[T (x)]]

S is linear

= a [(S T )(x)].

Exercises 2.7

1(b)

LU-factorization

(d)

A = LU where U is above and L =

(f)

3
1

1
3

1
1

Hence A = LU where U is above and L =

2
1
3

Hence A = LU where U is as above and L =

Definition of S T

23

12

= U.

2(b) The reduction to row-echelon form requires two row interchanges:

0
0
1

0
0
1

1
0
0

The elementary
are

matrices corresponding

(in order) to the interchanges

P1 =

and P2 =

= U. Hence

, so take P = P2 P1 =

= U.

41

Section 2.7: LU-factorization

We apply the LU-algorithm to P A:

PA =

0
0

1
0

Hence P A = LU where U is as above and L =

0
0

= U.

(d)
The reduction to row-echelon
form requires

two
row interchanges:

10

The elementary
matrices
corresponding
(in order)
to the interchanges
are

P1 =

and P2 =

We apply
A:
the LU-algorithmto P

PA =

10

so P = P2 P1 =

= U.

3(b) Write L =

tem LY = B is

2y1

, U =

y1 + 3y2

1
1
0

Hence P A = LU where U is as above and L =

1
2
2

1
0

, X =

x1
x2
x3
x4

, Y =

y1
y2
y3

. The sys-

= 2 and we solve this by forward substitu= 1

y1 + 2y2 + y3 = 1
tion: y1 = 1, y2 = 13 (1 y1 ) = 0, y3 = 1 + y1 2y2 = 0. The system U X = Y

42

Section 2.8: An Application to Input-Output Economic Models

is x1 + x2 x4 = 1 and we solve this by back substitution: x4 = t, x3 = 5,
x2 + x4 =

0 = 0
x2 = x4 = t, x1 = 1 + x4 x2 = 1 + 2t.

(d) Analogous to (b). The solution is: y =

8 2t

, x =

1
0

6t

1 t
t

, t arbitrary

5. If the rows in question are R1 and R2 , they can be interchanged thus:














R1
R1 + R2
R1 + R2
R1
R2

.
R2

R2

R1

R1

R1

6(b) Let A = LU = L1 U1 be LU-factorizations of the invertible matrix A. Then U and U1 have

no row of zeros so (being row-echelon) are upper triangular with 1s on the main diagonal.
1 is both lower triangular (L1 L) and upper triangular (U U 1 ) and so is
Thus L1
1
1 L = U1 U
1
diagonal. But it has 1s on the diagonal (U1 and U do) so it is I. Hence L1 = L and U1 = U.
7. We proceed by induction on n where
A and

B are n n. It is clear if n = 1. In general,




b 0
a
0
where A1 and B1 are lower triangular. Then
and B =
write A =
X A1
Y B1

ab
0
by Theorem 4 2.2, and A1 B1 is upper triangular by induction.
AB =
Xb + A1 Y A1 B1
Hence AB is upper triangular.
9(b) Let A = LU = L1 U1 be two such factorizations. Then U U11 = L1 L1 ; write this matrix as
D = U U11 = L1 L1 . Then D is lower triangular (apply Lemma 1 to D = L1 L1 ), and D is
also upper triangular (consider U U11 ). Hence D is diagonal, and so D = I because L1 and
L1 are unit triangular. Since A = LU, this completes the proof.

Exercises 2.8

1(b) I E =

An Application to Input-Output Economic

Models

.5

.1

.1

.4

.5

.2
.7

.1

3
3

1
3
0

. The equilibrium price struc-

ture P is the solution to (I E)P = 0; the general solution is P = [t 3t t]T .

(d) E =

.5

.2

.3

.1

.2

.2

.1

.1
0

.2

.1

.1

.1

.2
.4

Now add 3 times row 4 to row 2 to get:

10

2
9

11
3

43

Section 2.9: An Application to Markov Chains

0
0

7
3

13

14

28

13

46

94

23

47

14

28
13

47
23

equilibrium price structure p is the solution to (I E)p = 0. The solution is

14
23
17
23
47
23

. The

p = [14t 17t 47t 23t]T .

2. Here the input-output matrix is E =

I E =

1
0

1
1

so we get

1
1
0

. Thus the solution to (I E)p

is p1 = p2 = p3 = t. Thus all three industries produce the same output.





1a
b
1 a b
4. I E =

so the possible equilibrium price structures are

1 + a
b
0
0



bt
p=
, t arbitrary. This is nonzero for some t unless b = 0 and a = 1, and in that
(1 a)t





1
b
case p =
is a solution. If the entries of A are positive then p =
has positive
1a

entries.

7(b) One such example is E =

8. If E =

.4

.8

.7

.2

then I E =

, because (I E)1 = 54

1a
c

1d


. We have det(I E) = (1 a)(1 d) bc =

1 (a + d) + (ad bc)

= 1 tr E + det E. If det(I E) = 0 then Example 4 2.3 gives

(I E)1 =

1
det(IE)

1d
c

1a

. The entries 1 d, b, c, and 1 a are all between 0 and

1 so (I E)1 0 if det(I E) > 0, that is if tr E < 1 + det E.

3

9(b) If P =

(d) If p =

2
1

3
2
2

then P > EP so Theorem 2 applies.

then p > Ep so Theorem 2 applies.

Exercises 2.9

An Application to Markov Chains

1(b) Not regular. Every power of P has the (1, 2)- and (3, 2)-entries zero.

1







1
1 2
2t
2
2(b) I P =

so
(I

P
)s
=
0
has
solutions
s
=
. The entries
12
1
0
0
t

2 


1
1
3
is the steady state vector. Given s0 =
, we
of s sum to 1 if t = 3 , so s =
1
3

44

Section 2.9: An Application to Markov Chains

get s1 = P s0 =

1
2
1
2

, s2 = P s1

transitions with probability 38 .

.6

(d) I P =

s0 =

.4

s=

.2

t
1

.1

.5

.3

.7

.4

.2

0
0

3
4
1
4

, s3 = P s2 =

11

11

11

11
1
3

. The entries sum to 1 if t =

, s1 = P s0 =

0
0

.4
.2
.4

(f) I P =

s=

5t
7t
8t

s0 =

1
0
0

.9
.3
.6

.3

.3

.6

.9

.9

.6

, s2 = P s1 =

1
0
0

24

21

24

21

. The entries sum to 1 if t =

, s1 = P s0 =

.1
.3
.6

1
20

.34

4(b) The transition matrix is P =

.1

.1

.8

.3

.2

.1

.6

1
0

so (I P )s = 0 has solution

.28
.42
.30

state 2 after three transitions with probability .306.

.1

58

87
0

.350
.312
.338

1
3
1
3
1
3

. Given

. Hence it is in

, so (I P )s = 0 has solution

so the steady state vector is s =

.7

. So it is in state 2 after three

, s3 = P s2 =

.28

, s2 = P s1 =

.38

state 2 after three transitions with probability .312.

5
8
3
8

so the steady state vector is s =

, s3 = P s1 =

.244
.306
.450

5
20
7
20
8
20

. Given

. Hence it is in

where the columns (and rows) represent the

upper, middle and lower classes respectively and, for example, the last column asserts that,
for children of lower class people, 10%become upper class,
30% become middle

class and 60%

.3

remain lower class. Hence I P =

.1

.2

.1
.2

.1

.1

.3

.4

Thus the general solution to (I P )s = 0 is s =

2t
t

10

, so s =

10
1
4
1
2
1
4

1
2
0

is the steady state

solution. Eventually, upper, middle and lower classes will comprise 25%, 50% and 25% of this
society respectively.

6. Let States
1and 2 be late and on time respectively. Then the transition matrix in


1
1
3
2
P = 2 1 . Here column 1 describes what happens if he was late one day: the two entries
3

sum to 1 and the top entry is twice the bottom entry by the information we are given. Column

3 
2 is determined similarly. Now if Monday is the initial state, we are given that s0 = 41 .
4

3 

7 
8
16
Hence s1 = P s0 = v 5 and s2 = P s1 =
. Hence the probabilities that he is late
9
8

and on time Wednesdays are

7
16

and

9
16

16

respectively.

45

Section 2.9: An Application to Markov Chains

8. Let the states be the five compartments. Since each tunnel entry is equally likely,

1
1
1
0

P =

1
3
1
3

1
2

1
3

2
5
1
5
1
5

(a) Sincehe starts

in compartment

1,
1

s0 =

0
0
0
0

, s1 = P s0 =

0
1
3

1
2

2
5

, s2 = P s1 =

1
3
1
3

1
4
1
4
2
4

0
3
10
7
30
1
15

, s3 = P s2 =

7
75
23
120
69
200
53
300
29
150

Hence the probability that he is in compartment 1 after three moves is

(b) The steady state vector S satisfies (I P )s = 0. As

1
1
1
21

(I P ) =

31

31
0

31

3
5
15
51

41

in compartment 3 (in fact

12(a)

1p
p

q
1q

1
p+q

q
p

1pq
p+q

1
p+q

12

2
5
4
2

14
1
2

1
so the steady state is s = 16

5
16

23

25

2
0

7
75 .

. Hence, in the long run, he spends most of his time

of his time).
(1 p)q + qp

pq + (1 q)p

add to 1, it is the steady state vector.

1
p+q

q
p

. Since the entries of

(b) If m = 1
1
p+q

p
p

q
q

=
=

1
p+q
1
p+q

=P
In general, write X =

and Y =

P Y = (1 p q)Y . Hence if P m =

1
p+q X



q + p p2 pq
p p_p2 + pq

(p + q)(1 p)
(p + q)p

1
p+q X

1
p+q X

p + q pq q 2
(p + q)q
(p + q)(1 q)

. Then P X = X and

p
q
(1pq)m
+ p+q Y

P m+1 = P P m =

q q + pq + q 2

for some m 1, then

(1pq)m
1
PY
p+q P X +
p+q
(1pq)m
(1 p q)Y
p+q
(1pq)m+1
Y.
p+q

1
p+q

q
p

46

Supplementary Exercises - Chapter 2

Hence the formula holds for all m 1 by induction.

Now 0 < p < 1 and 0 < q < 1 imply 0 < p+q < 2, so that 1 < (p+q 1) < 1. Multiplying
through by 1 gives 1 > (1 p q) > 1, so (1 p q)m converges to zero as m increases.

Supplementary Exercises

Chapter 2

2(b) We have 0 = p(U) = U 3 5U 2 +11U 4I so that U(U 2 5U +11I) = 4I = (U 2 5U + 11I)U.

Hence U 1 = 14 (U 2 5U + 11I).
4(b) If xh = xm , then y = k(y z) = y + m(y z), whence (k m)(y z) = 0. But the matrix
y z = 0 (because y = z) so k m = 0 by Example 7 2.1.
 
6(d) Using (c), Ipq AIrs = ni=1 nj=1 aij Ipq Iij Irs . Now (b) shows that Ipq Iij Irs = 0 unless i = q
and j = r, when it equals Ips . Hence the double sum for Ipq AIrs has only one nonzero term
the one for which i = q, j = r. Hence Ipq AIrs = aqr Ips .
7(b) If n = 1 it is clear. If n > 1, Exercise 6(d) gives
aqr Ips = Ipq AIrs = Ipq Irs A
because AIrs = Irs A. Hence aqr = 0 if q = r by Exercise 6(b). If r = q then aqq Ips = Ips A is
the same for each value of q. Hence a11 = a22 = = ann , so A is a scalar matrix.

Supplementary Exercises - Chapter 2

.

47

48

Section 3.1: The Cofactor Expansion

Chapter 3: Determinants and Diagonalization

Exercises 3.1

The Cofactor Expansion

If A is a square matrix, we write det A = |A| for convenience.

1(b) Take 3

6 9

8 12

out of row 1, then subtract 4 times row 1 from row 2:

= 3 2 3 = 3 2 3 = 0

8 12
0 0

(d) Subtract row 2 from row 1:

a+1

a1

a1

= (a 1) a = 1

(f) Subtract 2 times row 2 from row 1, then expand along row 2:



2 0 3 0 4 13





4 13
1 2 5 = 1 2


= 39
5 =



3
0
0 3 0 0 3
0

(h) Expand along row 1:

(j) Expand along row 1:


0 a b




a 0 c = a a


b
b c 0

c
0

+ b

= a

= a(0) = 0

= a(bc) + b(ac) = 2abc

(l) Subtract multiples of row 1 from rows 2, 3 and

1 0 3 1 1 0 3 1



2 2 6 0 0 2 0 2 2 0 2

=

1 0 3 1 0 0 0 2 = 0 0 2


1 0 4
4 1 12 0 0 1 0 4

4, then expand along column 1:

=0

(n) Subtract multiplies of row 4 from rows 1 and 2, then expand along column 1:





4 1 3 1 0 9 7 5

9 7 5




3 1 0 2 0 5 3 1





0 1 2 2 = 0 1 2 2 = 5 3 1

1 2 2


1 2 1 1 1 2 1 1
Again, subtract multiplies of row 3 from rows


4 1 3 1




0 25 13
3 1 0 2

25 13




0 1 2 2 = 0 13 9 = 13 9


1 2 2
1 2 1 1
(p) Keep expanding along row 1:

1 and 2, then expand along column 1:

=

1
13

5
9

= (9 + 65) = 56

49

Section 3.1: The Cofactor Expansion

5(b)





(d)

2
5

= a





1

11

21

16

46




= a b

=




2

11

=






2

17

106

= ab(cd) = abcd.

17

= 106

a
b
c



6(b) Subtract row 1 from row2: a + b 2b c + b

2
2
2

a b c


= b b b


2 2 2

= (17) = 17


2

= 0 by Theorem 2(4).

7(b) Take 2 and 3 out of rows 1 and 2, then subtract row 3 from row 2, then take 2 out of row 2:

2a

2p + x
3x

a



2q + y 2r + z = 6 2p + x

x
3y
3z



a b c






= 6 2p 2q 2r = 12
x y z

2b

2c

8(b) First add rows 2 and 3 to row 1:

2a + p 2b + q 2c + r 3a + 3p + 3x


2p + x 2q + y 2r + z =
2p + x


2x + a 2y + b 2z + c 2x + a


a+p+x b+q+y c+r+z


2q + y
2r + z
= 3 2p + x
2x + a
2y + b
2z + c

2q + y

2r + z

= 12

3b + 3q + 3y

3c + 3r + 3z

2q + y

2r + z

2y + b

2z + c

Now subtract row 1 from rows 2 and 3, and then add row
2 plus twice row 3 to row 1, to get

a+p+x b+q+y c+r+z
3x
3y
3z




= 3 pa qb rc
qb
rc
= 3 p a




xp yq zr
xp
yq
zr
Next take 3 out of row 1, and
then add row
3 to row2, to get
x
x y z
y
z




= 9 p a q b r c = 9 a b c
p
p q r
q
r
Now use row interchanges and common row factors to get

50

Section 3.1: The Cofactor Expansion

= 9

= 9

9(b) False. The matrix A =

a
p
x

= 9

det R = 1.
(f) False. A =

(h) False. If A =

1
0

that det A = det

has zero determinant, but no two rows are equal.

(d) False. The reduced row-echelon form of A =

is R =

1
T
A

and B =

0
0

, but det A = 2 while

then det A = det B = 1. In fact, it is a theorem

holds for every square matrix A.

11(b) Use Theorem

5 twice:

0

A 0





0
= det
X B



C

Y Z
A

, det A = 1 = det AT .

10(b) Partition the

and use Theorem 5:
matrix as follows

0 3 0

1 2





1 4 0



1 3


1 2 2 1 1


2 1 1
=
1 0 2 = 5 1
0 0

1 3



1 0 2

3 0 1
0 0



3 0 1

0 0

(d)

= det

1
3

2
1

= 5(7) = 35.

det C = (det A det B) det C = 2(1)3 = 6

det C = (det A det B) det C = 2(1)3 = 6

14(b) Follow the Hint, take out the common factor in row 1, subtract multiples of column 1 from
columns 2 and 3, and expand along row 1:





1

x 1 3
x2 x2 x2
1
1
1










1
x 1
1
x1 =
2
1
x 1 = (x 2) 2
det 2
3 x + 2 2
3 x + 2 2 3 x + 2 2





1
0
0

3 x 3





= (x 2) 2
3
x 3 = (x 2)

x+5
1
3 x + 5

1
= (x 2)[x2 2x + 12] = (x 2)(x2 + 2x 12).

15(b) If we expand along column 2, the coefficient of z is

2
1

1
3

= (6 + 1) = 7. So c = 7.

16(b) Compute det A by adding multiples of row 1 to rows 2 and 3, and then expanding along
column 1:

51

Section 3.1: The Cofactor Expansion

det A =

1
x

x
2

x
x

x2 2

x2 + x

x2 x x2 3
2
(x + x)(x2 x) = (x4
3

= (x2 2)(x2 3)
Hence det A = 0 means x2 =

3
2

x2 2

x2 x

x2 + x
x2 3

5x2 + 6) x2 (x2 1) = 6 4x2

= 46 , so x = 26 .

(d) Expand along column 1, and use Theorem 4:

x y 0 0





x y 0
y
0 x y 0







det A =
= x 0 x y y x

0 0 x y
0 0 x
0
y 0 0 x

= x x3 y y 3

= x4 y4 = (x2 y 2 )(x2 + y 2 ) = (x y)(x + y)(x2 + y 2 ).

Hence det A = 0 means x = y or x = y (x2 + y2 = 0 only if x = y = 0).

x1
y1



x2
y2

21 Let x = . , y = .
,
and
A
=
where x + y is in
c

x
+
y

c
1
n

..
..

xn
yn
column j. Expanding det A along column j we obtain

T (x + y) = det A = ni=1 (xi + yi )cij (A)


= ni=1 xi cij (A) + ni=1 yi cij (A)
= T (x) + T (y).

where the determinant at the second step is expanded along column 1. Similarly, T (ax) =
aT (x) for any scalar a.
24. Suppose A is n n. B can be found from A by interchanging the following pairs of columns:
1 and n, 2 and n 1, . . . . There are two cases according as n is even or odd:
Case 1. n = 2k. Then we interchange columns 1 and n, 2 and n 1, . . . , k and k + 1, k
interchanges in all. Thus det B = (1)k det A in this case.

Case 2. n = 2k + 1. Now we interchange columns 1 and n, 2 and n 1, . . . , k and k + 2,

leaving column k fixed. Again k interchanges are used so det B = (1)k det A.
Thus in both cases: det B = (1)k det A where A is n n and n = 2k or n = 2k + 1.

Remark: Observe that, in each case, k and 12 n(n 1) are both even or both odd, so (1)k =
1
1
(1) 2 n(n1) . Hence, if A is n n, we have det B = (1) 2 n(n1) det A.

52

Section 3.2: Determinants and Matrix Inverses

Exercises 3.2

Determinants and Matrix Inverses

1(b) The cofactor matrix is


 1


 1


 1


 1

 1


 1


0 

1 


2 

1 


2 

0 


 3


 0

 1


 0


 1


 3


0 

1 


2 

1 


2 

0 


 3


 0


 1


 0

 1


 3





1 
1


1 

1 


1 

1 

1
9

 1

 2


 2

19 
 2
1
9


 2


 1

2 

1 

2 

1 

2 

2 

 2
19 
 2
1
9


 1


 2

91


 1


 2

2 

1 

2 

1 

2 

2 

1
9

 2

 2


 1

91 
 2
1
9


 1


 2

1 

2 


2 

2 





1 
2

1
9

3
6
6

The adjugate is the transpose of the cofactor matrix: adj A =

1
3

3M

The adjugate is the transpose of the cofactor matrix: adj A =

(d) In computing the cofactor matrix, we use the fact that det
matrix M. Thus the cofactor matrix is












1
2
2

1
9

1
3

det Mfor any 2 2

1
2
2

1
2

2
2
1

. Note that

the cofactor matrix is symmetric here. Note also that the adjugate actually equals the original
matrix in this case.
2(b) We compute the determinant by first adding column 3 to column 2:



0 c c 0 0 c







1 2 1 = 1 1 1 = (c) 1 1 = (c)(c) = c2 .



c 0
c c c c 0 c

This is zero if and only if c = 0, so the matrix is invertible if and only if c = 0.

(d) Begin by subtracting row 1 from row 3, and then subtracting column 1 from column 3:






4 c 3 4 c 3 4 c 1






c 2 c = c 2 c = c 2 0 = 1 c 1 = 2.

2 0



5 c 4 1 0 1 1 0 0
This is nonzero for all values of c, so the matrix is invertible for all c.

(f) Begin by subtracting c times row 1 from row 2:

1 c 1 1

c
1
2



c 1 1 = 0 1 c2 1 + c = 1 c 1 + c


1
c
0 1 c 0

1
c

(1 + c)(1 c)
1

1+c
c

53

Section 3.2: Determinants and Matrix Inverses

Now take the common factor (1 + c) out of row 1:

1 c 1




1c 1
c 1 1 = (1 + c)
= (1 + c)[c(1 c) 1] = (1 + c)(c2 c + 1) = (c3 + 1).


1

c
0 1 c

This is zero if and only if c = 1 (the roots of c2 c + 1 are not real). Hence the matrix is
invertible if and only if c = 1.

3(b) det(B 2 C 1 AB 1 C T ) = det B 2 det C 1 det A det B 1 det C T

= (det B)2 det1 C det A det1 B det C
= det B det A
= 2
4(b) det(A1 B 1 AB) = det A1 det B 1 det A det B =

1
1
det A det B

det A det B = 1.

Note that the following proof is wrong:

det(A1 B 1 AB) = det(A1 AB 1 B) = det(I I) = det I = 1.

The reason is that A1 B 1 AB may not equal A1 AB 1 B because B 1 A need not equal
AB 1 .
6(b) Since C is 3 3, the same is true for C 1 , so det(2C 1 ) = 23 det C 1 = det8 C . Now we
compute det C by taking 2 and 3 out of columns 2 and 3, subtracting column 3 from column
2:






2p a + u 3u
p a + u u
p a u






det C = 2q b + v 3v = 6 q b + v v = 6 q b v
2r c + w 3w
r c + w w
r c w
Now take 1 from column 2, interchange columns 1 and





p a u
a p u
a b c










q
b
v
b
q
v
det C = 6
= 6
= 6 p q r
r c w
c r w
u v w

Finally det 2C 1 =

8
det C

8
18

= 49 .

2, and apply Theorem 3:

= 6 3 = 18.

7(b) Begin by subtracting row 2 from row 3, and then expand along column 2:

2b

a+1

4d
2

2(c 1)

2b

4d

= 2

2c

2b

4d

2c

= 4

Interchange rows and use Theorem 3, to get

a c
a b
= 8(2) = 16.



= 8
= 8
c d
b d
8(b) x =



 9
4 



 1 1 


 3
4 



 2 1 

5
11

5
11 ,

y=












3
2
3
2





1 

4 

1 
9

21
11

21
11 .

2d

2c

= 8

54

Section 3.2: Determinants and Matrix Inverses

(d) The coefficient matrix has determinant:

4 1 3 0 1 0





6 2 1 = 14 2 5 = (1)




3 3 2 15 3 11

14

15

11

= 79

Hence Cramers rule gives

1 1 3
1 1 3






1
1 2
1


x = 79 0 2 1 = 79 0 2 1 = 79
2
1 3
0 2
2
5




4 1 3
4 1 3






1
1
1 6


y = 79 6 0 1 = 79 6 0 1 = 79
7
3 1 2
7 0 5




4 1 1
4 1 1






1
1
1 6 2


z = 79 6 2 0 = 79 6 2 0 = 79
7 2
3 3 1
7 2 0
[Cij ]T where

1 1
1
is det A C32 . Now C32 =

9(b) A1 =
A1

5
4

1
det A adj

A=

1
det A

1
5

1
5

12
79

= 37
79

=2.
79

[Cij ] is the cofactor matrix.

Hence the

(2, 3)-entry of

1 2 1 1 2 1




= 4. Since det A = 3 1 1 = 0 5 4 =




0 4 7 0 4 7

= 51, the (2, 3) entry of A1 is

4
51

4
51 .

10(b) If A2 = I then det A2 = det I = 1, that is (det A)2 = 1. Hence det A = 1 or det A = 1.
(d) If P A = P , P invertible, then det P A = det P, that is det P det A = det P . Since det P = 0
(as P is invertible), this gives det A = 1.
(f) If A = AT , A is n n, then AT is also n n so, using Theorem 3 3.1 and Theorem 3,
det A = det(AT ) = det[(1)AT ] = (1)n det AT = (1)n det A.
If n is even this is det A = det A and so gives no information about det A. But if n is odd it
reads det A = det A, so det A = 0 in this case.
15. Write d = det A, and let C denote the cofactor matrix of A. Here
AT = A1 = d1 adj A = 1d C T .
Take transposes to get A = d1 C, whence C = dA.

19(b) Write A =

[Cij ] =

c
c


 2
1 



 c
c 


 c


 c

 c


 2


c 

c 


c 

1 

1
c

. Then det A = c2 (Exercise 2) and the cofactor matrix is


 1


 c

 0


 c


 0


 1


1 

c 


c 

c 


c 

1 


 1


 c


 0


 c


 0


 1


2 

c 


c 

c 

c 

2 

c2

c
c2
c

55

Section 3.2: Determinants and Matrix Inverses

1
det A adj A

Hence A1 =

(d) Write A =

[Cij ] =


 2


 c


 c


 2

1
det A adj

Hence A1 =

(f) Write A =


 4


 c


3 

c 

1
1
c


c 

4 


3 

4 

A=


3 

c 


 c


 5


 4


 5

 4


 c

[Cij ] =


 c


 1

 c


 1

Hence A1 =

1 

c 

1
c3 +1


1 

c 


1 

1 

 c

 0

 1


 0


 1


 c

c2

c2

1
c

for any c = 0.


c 

2 
1
2

8 c2
c

c2

8 c2

1
c

10

c2

c
c

c2 6
8

c2

c2 10
c

c2

. Then det A = (c3 + 1) (Exercise 2) so det A = 0 means c = 1 (c

c2 + 1

c2


c 

c 

1
det A adj A

1c


2 

c 

[Cij ]T =

1
2

is real).The cofactor
matrix
is 






 1

 1

. Then det A = 2 (Exercise 2) and the cofactor matrix is


 4


 5


3 

4 


 c


 c

1
c2

c 4


 c
c 



 5

4

[Cij ]T =

1
c2

1 

c 


1 

c 


1 

1 


 c


 0


 1


 0

 1


 c

1
[C ]T
c3 +1 ij

c 1
c+1

c2


1 

1 


c 

1 


c 

1 

1
c3 +1

c1
c2
c

c1

(c2 + 1)
c+1

(c2 + 1)
c

c2

(1 + c)

c+1
(c + 1)
1

c2

c2

, where c = 1.

20(b) True. Write d = det A, so that d A1 = adj A. Since adj A = A1 by hypothesis, this gives
d A1 = A1 , that is (d 1)A1 = 0. It follows that d = 1 because A1 = 0 (see Example 7
2.1).
(d) True. Since AB = AC we get A(B C) = 0. As A is invertible, this means B = C. More
precisely, left multiply by A1 to get A1 A(B C) = A1 0 = 0; that is I(B C) = 0; that
is B C = 0, so B = C.

(f) False. If A =

then adj A = 0. However A = 0.

56

Section 3.2: Determinants and Matrix Inverses

(h) False. If A =
(j) False. If A =
(l) False. If A =


1

then adj A =


0
0

1
1

, and this has no row of zeros.

then det(I + A) = 1 but 1 + det A = 1.

then det A = 1, but adj A =

1
0

1
1

= A.

22(b) If p(x) = r0 + r1 x + r2 x2 , the conditions give linear equations for r0 , r1 and r2 :

r0

= p(0) = 5

r0 +

r1

r2

= p(1) = 3

r0 + 2r1 + 4r2 = p(2) = 5

The solution is r0 = 5, r1 = 4, r2 = 2, so p(x) = 5 4x + 2x2 .
23(b) If p(x) = r0 + r1 x + r2 x2 + r3 x3 , the conditions give linear equations for r0 , r1 , r2 and r3 :
r0

p(0)

p(1)

r0 +

r1

r2

r3

r0

r1

r2

r3

= p(1) =

r0 2r1 + 4r2 8r3

The solution is r0 = 1, r1 =

5
3 ,

p(2) = 3

r2 = 21 , r3 = 76 , so p(x) = 1 53 x + 12 x2 + 76 x3 .

24(b) If p(x) = r0 + r1 x + r2 x2 + r3 x3 , the conditions give linear equations for r0 , r1 , r2 and r3 :

r0
r0 +

r1

r2

= p(0) =

r3

= p(1) =

1.49

r0 + 2r1 + 4r2 +

8r3

= p(2) =

0.42

r0 + 3r1 + 9r2 + 27r3 = p(3) = 11.33

The solution is r0 = 1, r1 = 0.51, r2 = 2.1, r3 = 1.1, so

p(x) = 1 0.51x + 2.1x2 1.1x3 .

The estimate for the value of y corresponding to x = 1.5 is

y = p(1.5) = 1 0.51(1.5) + 2.1(1.5)2 1.1(1.5)3 = 1.25
to two decimals.
26(b) Let A be an upper triangular, invertible, n n matrix. We use induction on
n=1
n. If
a X
and
it is clear (every 1 1 matrix is upper triangular). If n > 1 write A =
0 B

b
Y
in block form. Then
A1 =
Z C

1 0
ab
+
XZ
aY
+
XC

= AA1 =
.
0 I
BZ
BC

57

Section 3.3: Diagonalization and Eigenvalues

So BC = I, BZ = 0. Thus C = B 1 is upper triangular by induction (B is upper

triangular

b Y
is
because A is) and BZ = 0 gives Z = 0 because B is invertible. Hence A1 =
0 C
upper triangular.

28. Write d = det A. Then

1
d

= det(A1 ) = det

= 21. Hence d =

4, we have A adj A = dI, so adj A = A1 (dI) = dA1 =

1
21

34(b) Write d = det A so det A1 = d1 . Now the adjugate for A1 gives

1
21 .

By Theorem

A1 (adj A1 ) = d1 I.
Take inverses to get (adj A1 )1 A = dI. But dI = (adj A)A by the adjugate formula for A.
Hence
(adj A1 )1 A = (adj A)A.

1
Since A is invertible, we get adj A1
= adj A, and the result follows by taking inverses
again.
(d) The adjugate formula gives
AB adj (AB) = det AB I = det A det B I.
On the other hand
AB adj B adj A = A[(det B)I]adj A
= A adj A (det B)I
= (det A)I (det B)I
= det A det B I.

Thus AB adj (AB) = AB adj B adj A, and the result follows because AB is invertible.

Exercises 3.3

Diagonalization and Eigenvalues

x2

4

= x2 x 6 = (x 3)(x + 2); hence the eigenvalues are 1 = 3, and
1(b) cA (x) =
1
x+1
2 = 2. Take these values for x in the matrix
xI A for cA (x):






4
1 4
1 4
.
; x1 =
1 = 3 :

0 0
1 4
1







1
4
4
1 1
2 = 2 :

; x2 = .
1
1
0
0
1


3
0
4
1
.
So P = [x1 x2 ] =
has P 1 AP =
1 1
0 2

58

Section 3.3: Diagonalization and Eigenvalues

(d) To compute cA (x) we first add row 1 to row 3:

x 1 1
x 1 1
x 1 1
3
3






x
6 = 2
x
6 = 2
x
cA (x) = 2
1
x2 0 x2 x2 0
1
x5


1 x + 4
= (x 2)[x2 4x + 4] = (x 2)3 .
= (x 2)

x

x + 4
4
0

So the eigenvalue is 1 = 2 of multiplicity 3. Taking x = 3 in the matrix xI A for cA (x) :

s 3t

, x =

; x1 =

s
t

1
0

, x2 =

3
0
1

Hence there are not n = 3 basic eigenvectors, so A is not diagonalizable.

x 1 0


(f) Here cA (x) = 3 x 1 = x3 3x 2. Note that 1 is a root of cA (x) so x + 1 is a
2 0 x
factor. Long division gives cA (x) = (x + 1)(x2 x 2). But x2 x 2 = (x + 1)(x 2), so
cA (x) = (x + 1)2 (x 2).

Hence, the eigenvalues are 1 = 1 and 2 = 2. Substitute 1 = 1 in the matrix

xI cA (x) gives

2
1
2

so the solution involves only 1 parameter. As the multiplicity of 1 is 2, A is not diagonalizable

by Theorem 5. Note that this matrix and the matrix in Example 9 have the same characteristic
polynomial but the matrix in Example 9 is diagonalizable, while this one is not.


x 2 1


1


x 2 1
2
= (x 1)

(h) cA (x) = 0
x1
0

1 x 2 = (x 1) (x 3). Hence the eigenvalues
1
1
x2
are 1 = 1,2 = 3. Take
thesevalues forx in the matrixxI A
for cA(x):
1 1 :

2 = 3 :

0
1

0
0

; x2 =

1
0
1

; x1 =

1
0
1

Since n = 3 and there are only two basic eigenvectors, A is not diagonalizable.
2(b) As in Exercise 1, we find 1 = 2 and 2 = 1; with corresponding eigenvectors x1 =








1
2 1
2
0
1
, so P =
satisfies P AP = D =
. Next compute
and x2 =
2

b=
Hence b1 =

7
3

b1
b2

P01 v

so, as 1 is dominant,

1
3

2
1

k
xk
= b1 1 x1

1
2



7 k
32

3
1
2
1

1
3

7
5

2
1

59

Section 3.3: Diagonalization and Eigenvalues

(d) Here 1 = 3, 2 = 2 and 3 = 1; x1 =

P =

2
3

Hence vk
= 32 3k

. Now P 1 =
1
0
1

1
6

1
0
1

, x2 =

1
1
3

, so P01 v0 =

and x3 =
1
6

9
2
1

1
2
3

, and

and hence b1 = 3 .
2

4. If is an eigenvalue for A, let Ax = x, x = 0. Then

A1 x = (A I)x = Ax x = x x = ( )x.
So is an eigenvalue of A1 = A I (with the same eigenvector). Conversely, if
is an eigenvalue of A1 , then A1 y = ( )y for some y = 5. Thus, (A I)y = ( )y,
whence Ay y = y y. Thus Ay = y so is an eigenvalue of A.








n
1
0
1 0
1 0
1
=
, the hint gives
8(b) Direct computation gives P AP =
, Since
0 2n
0 2
0 2






1
0
9 8 2n 12(1 2n )
1 =
An = P
P
.
n
n
n
0

9(b) A =

6(2 1)

92 8

. We have cA (x) = x(x 2) so A has eigenvalues 1 = 0 and 2 = 2 with







1
1
1 1
basic eigenvectors x1 =
and x2 =
. Since [x1 x2 ] =
is invertible, it is a
0
2


0 2
1 1
diagonalizing matrix for A. On the other hand, D + A =
is not diagonalizable by
0

Example 10.

11(b) Since A is diagonalizable, let P 1 AP = D be diagonal. Then P 1 (kA)P = k(P 1 AP ) = dD

is also diagonal, so kA is diagonalizable too.
(d) Again let P 1 AP = D be diagonal. The matrix Q = U 1 P is invertible and
Q1 (U 1 AU)Q = P 1 U(U 1 AU)U 1 P = P 1 AP = D is diagonal.
This shows that U 1 AU is diagonalizable with diagonalizing matrix Q = U 1 P.














1 0
2
1
1 0
2
1
1 1
+
and both
and
are diagonalizable.
12.
=
0 1
0
2
0 1
0 1
0
2



1 1
However,
is not diagonalizable by Example 10.
0

14. If A is n n, let 1 , 2 , , n be the eigenvalues, all either 0 or 1. Since A is diagonalizable

(by hypothesis), we have P 1 AP = D where D = diag(1 , . . . , n ) is the diagonal matrix
with 1 , . . . , n down the main diagonal. Since each i = 0, 1 it follows that 2i = i for
each i. Thus D2 = diag(21 , . . . , 2n ) = diag(1 , . . . , n ) = D. Since P 1 AP = D, we have
A = P DP 1 . Hence
A2 = (P DP 1 )(P DP 1 ) = P D2 P 1 = P DP 1 = A.

60

Section 3.3: Diagonalization and Eigenvalues

18(b) Since r = 0 and A is n n, we have


x

I A ] = rn det I A .
r
r


As cA (x) = det[xI A], this shows that crA (x) = rn cA xr .
crA (x) = det[xI rA] = det[r

x

20(b) If is an eigenvalue of A1 then A1 x = x for some column x = 0. Note that = 0 because

A1 is invertible and x = 0. Left multiplication by A gives x = Ax, whence Ax = 1 X.
Thus, 1 is an eigenvalue of A; call it = 1 . Hence, = 1 as required. Conversely, if is
any eigenvalue of A then = 0 by (a) and we claim that 1 is an eigenvalue of A1 . We have
Ax = x for some column x = 0. Multiply on the left by A1 to get x = A1 X; whence
A1 x = 1 x. Thus 1 is indeed an eigenvalue of A1 .
21(b) We have Ax = x for some column x = 0. Hence, A2 x = Ax = 2 x, A3 = 2 Ax = 3 x, so
(A3 2A + 3I)x = A3 x 2Ax + 3x = 3 x 2x + 3x = (3 2 + 3)x.
23(b) If is an eigenvalue of A, let Ax = x for some x = 0. Then A2 x = Ax = 2 x, A3 X =
2 AX = 3 X, . . . . We claim that Ak x = k x holds for every k 1. We have already checked
this for k = 1. If it holds for some k 1, then Ak x = k x, so
Ak+1 x = A(Ak x) = A(k x) = k Ax = k (x) = k+1 x.
Hence, it also holds for k + 1, and so Ak x = k x for all k 1 by induction. In particular, if
Am = 0, m 1, then m x = Am x = 0x = 0. As x = 0, this implies that m = 0, so = 0.
24(a) Let A be diagonalizable with Am = I. If is any eigenvalue of A, say Ax = x for some
column x = 0, then (see the solution to 23(b) above) Ak x = k x for all k 1. Taking k = m
we have x = Am x = m x, whence m = 1. Thus is a complex mth root of unity and so lies
on the unit circle by Theorem 3, Appendix A. But we are assuming that is a real number
so = 1, so 2 = 1. Also A is diagonalizable, say P 1 AP = D = diag(1 , , n ) where
the i are the eigenvalues of A. Hence D2 = diag(21 , , 2n ) = I because 2i = 1 for each i.
Finally, since A = P DP 1 we obtain A2 = P D2 P 1 = P IP 1 = I.
27 (a) If A is diagonalizable and has only one eigenvalue , then the diagonalization algorithm
asserts that P 1 AP = I. But then A = P (I)P 1 = I, as required.
(b) Here the characteristic polynomial is cA (x) = (x 1)2 , so the only eigenvalue is = 1.
Hence A is not diagonalizable by (a).






1
1
1
1
2
4
4
4
. In this case A =
so
31(b) The matrix in Example 1 is
2

cA (x) = det

1
4

14
x

= x2 14 x

3
4

= (x 1)(x + 34 )

Hence the dominant eigenvalue is = 1, and the population stabilizes.

3 1 



x 53 15
5
5
(d) In this case A =
so cA (x) = det
= x2 35 x 35 . By the quadratic
3 0
3
x

1
1
formula, the roots are 10
[3 69], so the dominant eigenvalue is 10
[3 + 69] 1.13 > 1, so
the population diverges.

61

Section 3.4: An Application to Linear Recurrences

2
5

34 Here the matrix A in Example 1 is A =

where is the adult reproduction rate.
2 0






x 25
Hence cA (x) = det
= x2 x 45 , and the roots are 12 2 + 16
5 . Thus
2
x



1
the dominant eigenvalue is 1 = 21 + 2 + 16
5 , and this equals 1 if and only if = 5 .

So the population stabilizes if = 15 . In fact it is easy to see that the population becomes
extinct (1 < 1) if and only if < 15 , and the population diverges (1 > 1) if and only if
> 15 .

Exercises 3.4

An Application to Linear Recurrences

1(b) In this case xk+2 = 2xk xk+1 , so vk+1 =




1
1
Diagonalizing A gives P =
and D =
1

b=

xk
xk+1

xk =

b1
b2

4 k
31

4
3

P01 v0


1
1

1
k
3 (2)

1
k
3 (2)

1
3

2
2



1
2

1
k
3 [4 (2) ]

1
2

xk+1
2xk xk+1
1

4
3
13



xk
xk+1

= Avk .

. Hence

. Thus

for each k. Comparing top entries gives

(2)k
3

 k 
k
1
1 4 2
(2)
3

for large k.

4
1
k
Here 2 is the dominant eigenvalue, so xk = 13 (2)k [ (2)
k 1] 3 (2) if k is large.

xk+1

(d) Here xk+2 = 6xk xk+1 , so vk+1 =




1
1
2
A gives P =
and D =
2

b=
Now

xk
xk+1

b1
b2

P01

v=

4 k
52

6xk xk+1

1
2

1
5

1
k
5 (3)



1
3



xk
xk+1

= Avk . Diagonalizing

. Hence

1
1

4
5
1
5

, and so, looking at the top entries we get

xk = 45 2k + 15 (3)k = 51 [2k+2 + (3)k ].

 k 
k
(3)k
2
Here xk = 5
1 + r 3
(3)
for large k, so 3 is dominant.
5

2(b) Let vk =

and D =

xk
xk+1
xk+2
1
0
0

. Then A =

. Then

giving the general formula

2
b1
b2
b3

, and diagonalization gives P =

= P01 v0 =

13
1
3

21
0
1
2

12
1
3
1
6

1
0
1

1
2

0
1
2

62

Section 3.4: An Application to Linear Recurrences

vk = 12 (1)1k

1
1
1

Thus equating first entries give

+ (0)(2)k

1
2
4

+ 1 1k
2

1
1
1

xk = 21 (1)k + 12 1k = 12 [(1)k + 1].

Note that the sequence xk here is 0, 1, 0, 1, 0, 1, . . . which does not converge to any fixed value
for large k.
3(b) If a bus is parked at one end of the row, the remaining spaces can be filled in xk ways to
fill it in; if a truck is at the end, there are xk+2 ways; and if a car is at the end, there
are xk+3 ways. Since one (and only one) of these three possibilities must occur, we have
xk+4 = xk + xk+2 + xk+3 must hold for all k 1. Since x1 = 1, x2 = 2 (cc or t), x3 = 3
(ccc, ct or tc) and x4 = 6 (cccc, cct, ctc, tcc, tt,b), we get successively, x5 = 10, x6 = 18,
x7 = 31, x8 = 55, x9 = 96, x10 = 169.
5. Let xk denote the number of ways to form words of k letters. A word of k + 2 letters
must end in either a or b. The number of words that end in b is xk+1 just add a b to a
(k + 1)-letter word. But the number ending in a is xk since the second-last letter must be
a b (no adjacent as) so we simply add ba to any k-letter word. This gives the recurrence
xk+2 = xk+1 + xk which is the same as in Example 2, but with different initial conditions:
x0 = 1 (since the empty word is the only one formed with no letters) and x1 = 2. The
eigenvalues, eigenvectors, and diagonalization remain the same, and so






1
1
+ b2 k2
vk = b1 k1
2

where 1 = 21 (1 +

5) and 2 = 21 (1 5). Comparing top entires gives

xk = b1 k1 + b2 k2 .

By Theorem 1 2.4, the constants b1 and b2 come from

b1
b2

= P01 v0 . However, we vary

the method and use the initial conditions to determine the values of b1 and b2 directly. More
precisely, x0 = 1 means 1
= b1 + b2 whilex1 = 2 means 2 = b1 1 + b2 2 . These equations

. It follows that
have unique solution b1 = 253
and b2 = 253
5
5
xk =

2 5


 k
 k
(3 + 5) 1+2 5 + (3 + 5) 12 5

for each k 0.

7. In a stack of k + 2 chips, if the last chip is gold then (to avoid having two gold chips together)
the second last chip must be either red or blue. This can happen in 2xk ways. But there are
xk+1 ways that the last chip is red (or blue) so there
are 2xk+1
 ways these possibilities can

0 1
occur. Hence xk+2 = 2xk + 2xk+1 . The matrix is A =
with eigenvalues 1 = 1 + 3
2 2







1
1
and 2 = 1 3 and corresponding eigenvectors x1 =
and x2 =
. Given the
2

initial conditions x0 = 1 and x1 = 3, we get






b1
2
1
1
1
1
= P0 v0 = 3
=
b2

1
2 3

2 3

2 3

2 3

2 + 3
2+

63

Section 3.4: An Application to Linear Recurrences

Since Theorem 1 2.4 gives
vk =

b1 k1

1
1

+ b2 k2

1
2

comparing top entries gives

xk = b1 k1 + b2 k2 =

2 3


(2 + 3)(1 + 3)k + (2 + 3)(1 3)k .
yk +yk+1
2

= 21 yk
+ 12yk+1 .
1
The eigenvalues are 1 = 1 and 2 = 12 , with corresponding eigenvectors x1 =
and
1



2
x2 =
. Given that k = 0 for the year 1990, we have the initial conditions y0 = 10 and

9. Let yk be the yield for year k. Then the yield for year k + 2 is yk+2 =

y1 = 12. Thus

Since

b1
b2

P01 v0

Vk =

1
3

k
34
3 (1)

then
yk =
For large k, yk

11(b) We have A =

write x =

34
3

k
34
3 (1)

1
1

2
3



4
3

12

 1 k
2

 k
+ 23 (2) 12 =

10

34
3

1
3

34
2

 1 k
2 .

so the long term yield is 11 31 million tons of wheat.

0
1
c

so cA (x) = x3 (a + bx + cx2 ). If is any eigenvalue of A, and we

, we have

Ax =

2
a + b

+ c2

because cA () = 0. Hence x is a -eigenvector.

2
3

= x

5
5
11
12(b) We have

p = 6 from (a), so yk = xk + 6 satisfies yk+2 = yk+1 + 6yk with

y0 = y1 = 6 . Here
0 1
1 1
with eigenvalues 3 and 2, and diagonalizing matrix P =
. This gives
A=
6 1
3
2


k+1 (2)k+1 , so x = 11 3k+1 (2)k+1 5 .
yk = 11
k
30 3
30
6

13(a) If pk is a solution of (*) and qk is a solution of (**) then

qk+2 = aqk+1 + bqk

pk+2 = apk+1 + bpk + c(k).
for all k. Adding these equations we obtain
pk+2 + qk+2 = a(pk+1 + qk+1 ) + b(pk + qk ) + c(k)
that is pk + qk is also a solution of (*).

64

Section 3.5: Systems of Differential Equations

(b) If rk is any solution of (*) then rk+2 = ark+1 + brk + c(k). Define qk = rk pk for each k.
Then it suffices to show that qk is a solution of (**). But
qk+2 = rk+2 pk+2 = (ark+1 + brk + c(k)) (apk+1 + bpk + c(k)) = aqk+1 + bqk
which is what we wanted.

Exercises 3.5

An Application to Systems of Differential

Equations

so cA (x) =

x+1

= (x 4)(x + 2).

1(b) The matrix of the system is A =

1
3
1
x3








5
5
1 1
1
1 = 4 :

; an eigenvector is X1 =
.

1 1


0 0 

1 
1 5
1 5
5
2 = 2 :

; an eigenvector is X2 =
.
1 5
0 0
1






4
0
1
5
Thus P 1 AP =
where P =
. The general solution is
0

1 x

f = c1 X1 e

2 x

+ c2 X2 e

= c1

1
1

4x

+ c2

5
1

e2x .

Hence, f1 (x) = c1 e4x + 5c2 e2x , f2 (x) = c1 e4x c2 e2x . The boundary condition is f1 (0) = 1,
f2 (0) = 1; that is








1
1
5
= f (0) = c1
+ c2
.
1

Thus c1 + 5c2 = 1, c1 c2 = 1; the solution is c1 = 32 , c2 = 13 , so the specific solution is

f1 (x) = 31 (5e2x 2e4x ),

(d) Now A =

. To evaluate cA (x), first subtract row 1 from row 3:

cA (x) =

f2 (x) = 31 (2e4x + e2x ).

x2

x1

x2

x2
2

x 1

x+1

x2

= (x + 1)(x 2)(x 4).

1 = 1 :

2 = 2 :

2
3

2
2

2
0

1
2

8
7
10
7

1
2
0

x4
0

; X1 =

; X2 =

x+1

x2

8
10
7
1

2
1

65

Section 3.6: Proof of the Cofactor Expansion

3 = 4 :

2
2
3

Thus P 1 AP =

2
0
3

where P =

8
10
7

f = c1 X1 ex + c2 X 2 e2x + c3 X3 e4x = c1

That is

8
10
7

1
0

; X3 =

1
0
1

. The general solution is

ex + c2

1
2
1

e2x + c3

1
0
1

e4x .

f1 (x) = 8c1 ex + c2 e2x + c3 e4x

f2 (x) = 10c1 ex 2c2 e2x

f3 (x) = 7c1 ex + c2 e2x + c3 e4x .

If we insist on the boundary conditions f1 (0) = f2 (0) = f3 (0) = 1, we get

8c1 +

c2

+ c3 =

10c1

2c2

2c1

c2

+ c3 = 1.

The coefficient matrix is P is invertible, so the solution is unique: c1 = 0, c2 = 21 , c3 = 32 .

Hence
f1 (x) = 21 (3e4x e2x )
f2 (x) = e2x

f3 (x) = 12 (3e4x e2x ).

Note that f1 (x) = f3 (x) happens to hold.
3(b) Have m (t) = k m(t), so m(t) = cekt by Theorem 1. Then the requirement that m(0) = 10
gives c = 10. Also we ask that m(3) = 8, whence 10e3k = 8, e3k = 54 . Hence (ek )3 = 45 , so
(ek ) = ( 54 )1/3 . Thus m(t) = 10( 45 )t/3 .
Now, we want the half-life t0 satisfying m(t0 ) = 21 m(0), that is 10( 45 )t0 /3 = 5 so t0 =
3 ln(1/2)
ln(4/5) = 9.32 hours.
5(a) Assume that a g = Ag where A is n n. Put f = g A1 b where b is a column of constant
functions. Then f  = g = Ag = A(f + A1 b) = Af + b, as required.
6(b) Assume that f1 = a1 f1 + f2 and f2 = a2 f1 . Differentiating gives f1 = a1 f1 + f2 = a1 f1 + a2 f1 .
This shows that f1 satisfies (*).

Exercises 3.6

Proof of the Cofactor Expansion Theorem

66

Supplementary Exercises - Chapter 3

2. Consider the rows Rp , Rp+1 , . . . , Rq1 , Rq . Using

Rp
R

p+1

..
Rp+1
.

..

.
Rq1

qp

Rq1
Rq

interchanges
Rq
Rp
Hence 2(q p) 1 interchanges are used in all.

Supplementary Exercises

adjacent interchanges we have

Rq

Rp+1

..

. .

qp1

interchanges q1
Rp

Chapter 3

2(b) Proceed by induction on n where A is n n. If n = 1, AT = A. In general, induction and (a)

give
det[Aij ] = det[(Aij )T ] = det[(AT )ij ].
Write AT = [aij ] where aij = aji , and expand det(AT ) along column 1:
det(AT ) =

n

j=1

aj1 (1)j+1 det[(AT )j1 ] =

n


a1j (1)1+j det[A1j ] = det A

j=1

where the last equality is the expansion of det A along row 1.

.

Supplementary Exercises - Chapter 3

.

67

68

Section 4.1: Vectors and Lines

Chapter 4: Vector Geometry

Exercises 4.1




1(b) 







(d) 



1
1
2

1
0
2





(f) 
3





 =






 =



1
1
2

Vectors and Lines

12 + (1)2 + 22 =

(1)2 + 02 + 22 =





 = |3| 12 + 12 + 22 = 3 6



2(b) A vector u in the direction of

2
1
2

must have the form u = t

2
1
2

for a scalar t > 0.


Since 8u is a unit vector, we want u = 1; that is 1 = |t| (2)2 + (1)2 + 22 = 3t, which

gives t =

1
3.

Hence u =

4(b) Write u =

2
1
2

1
3

2
1
2

and v =





difference: u v = 



1
1





(d) As in (b), the distance is 




2
0

1



 =


4
0
2

. The distance between u and v is the length of their

02 + (1)2 + 12 =

3
2
0

 
 
 
 = 
 
 

1
2
2

2.




 =



12 + (2)2 + (2)2 = 3.

6(b) In the diagram, let E and F be the midpoints of sides BC and AC respectively. Then
1

F C = 2 AC and CE = 21 CB. Hence

1 1 1

F E = F C + CE = 2 AC + 2 CB = 2 (AC + CB) = 12 AB
7 Two nonzero vectors are parallel if and only if one is a scalar multiple of the other.
(b) Yes, they are parallel: u = (3)v.
(d) Yes, they are parallel: v = (4)u.

8(b) QR = p because OP QR is a parallelogram (where O is the origin).

69

Section 4.1: Vectors and Lines

(d) RO = (p + q) because OR = p + q.

9(b) P Q =

1
1
6

2
1
0

1
1
5



, so 
P Q =

(1)2 + (1)2 + 52 =



P Q = 0.

(d) Here P = Q are equal points, so P Q = 0. Hence

(f) P Q =

2 3.

1
1
4

3
1
6

10(b) Given Q(x, y, z) let q =

Let v =

1
3

x
y
z

2
2

= 2

and p =

1
1

3
0
1



. Hence 
P Q = |2|

be the vectors of Q and P. Then

P Q = q p.

(i) If P Q = v then q p = v, so q = p + v =

5
1
2

(ii) If P Q = v then q p = v, so q = p v =

. Thus Q = Q(5, 1, 2)

. Thus Q = Q(1, 1, 4).

1
4

11(b) If 2(3v x) = 5w + u 3x then 6v 2x = 5w + u 3x, so

x = 5w + u 6v =

12(b) We have au + bv + cw =

au + bv + cw = x =

1
3
0

a
a
2a

25

0
b
2b

24

1
0

c
0
c

0
6

a +

a+b

= 3

2a + 2b c = 0.
The solution is a = 5, b = 8, c = 6.

2a + 2b c

+ c = 1
b

a+c

gives equations
a

(1)2 + 12 + (1)2 =

27 = 3 3.

26
4

19

. Hence setting

70

Section 4.1: Vectors and Lines

13(b) Suppose
for a, b, c:

5
6
1

3a + 4b + c

= au+bv +cw =

. Equating coefficients gives linear equations

a + c
b+c

3a + 4b + c =

+ c =

b + c = 1
This system has no solution, so no such a, b, c exist.

14(b) Write P = P (x, y, z) and let p =

y
z

, p1 =

and p2 =

1
2
0

be the vectors of P ,

P1 and P2 respectively. Then

p p2 = P2 P = p2 + 14 (P2 P1 ) = p2 + 14 (p1 p2 ) = 41 p1 + 34 p2 .
Since p1 and p2 are known, this gives

p=

Hence P = P

5


1

5
4, 4, 2

1
4

+ 3
4

1
2
0

1
4

5
2

17(b) Let p = OP and

Q respectively.

Then

q =
OQ denote
the vectors of the points P and
0
1
1
1
1

q p = P Q = 4 and p = 3 , so q = (q p) + p = 4 + 3 = 7 .
3

Hence Q = Q(0, 7, 3).

18(b) We have u2 = 20, so the given equation is 3u + 7v = 20(2x + v). Solving for x gives

26

40x = 3u 13v =

12

13

26

20
13
14

20(b) Let S denote the fourth point. We have RS = P Q, so

OS = OR + RS = OR + P Q = 1 +
0

Hence x =

4
4
2

Hence S = S(1, 3, 2).

1
3
2

1
40

20
13

21(b) True. If v w = 0 then v w = 0 by Theorem 1, so v = w.

(d) False. v = v for all v but v = v only holds if v = 0.
(f) False. If t < 0 they have opposite directions.
(h) False. By Theorem 1, 5v = |5| v = 5 v so it fails if v = 0.
(j) False. If w = v where v = 0, then v + w = 0 but v + w = 2 v =
 0.

14

71

Section 4.1: Vectors and Lines

22(b) One direction vector is d = QP =

2
1
5

vector equation of the line is

p = p0 + td =

3
1
4

. Let p0 =

+ t

2
1
5

3
1
4

be the vector of P . Then the

when p =

y
z

is the vector of an arbitrary point on the line. Equating coefficients gives the parametric
equations of the line
x = 3 + 2t
y = 1 t
z = 4 + 5t.

because P1 (1, 1, 1) is on the line, and take d =

to be parallel to d. Hence the vector equation is p = p0 + td =

(d) Now p0 =

p=

x
y
z

1
1

x=1+t
, the scalar equations are y = 1 + t .

1
1
1

because the line is

+ t

1
1
1

. Taking

z =1+t

(f) The line with parametric equations

x = 2t
y=1
z=
t

has direction vector d =

1
0
1

the components are the coefficients of t. Since our line is

parallel to this one, d will do as direction vector. We are given the vector p0 =

point on the line, so the vector equation is

The scalar equations are

p = p0 + td =

2
1
1

+ t

x = 2t

y = 1
z = 1 + t.

1
0
1

1
1

of a

72

Section 4.1: Vectors and Lines

23(b) P (2, 3, 3) lies on the line

y
z

4t
3

1 2t

since it corresponds to t = 2. Similarly

Q(1, 3, 9) corresponds to t = 5, so Q lies on the line too.

24(b) If P = P (x, y, z) is a point on both lines then

x=1t

for some t because P lies on the first line.

y = 2 + 2t
z = 1 + 3t
x = 2s

for some s because P lies on the second line.

y =1+s
z=3

If we eliminate x, y, and z we get three equations for s and t:

1 t = 2s
2 + 2t = 1 + s
1 + 3t = 3.
The last two equations require t = 43 and s = 11
3 , but these values do not satisfy the first
equation. Hence no such s and t exist, so the lines do not intersect.

x

(d) If

y
z

is the vector of a point on both lines, then

Eliminating

x
y
z

x
y
z
x
y
z

gives

three equations for s and t:

2
12

4
1
5

+ t

+ s

+ t

1
0
1

1
0

for some t (first line)

for some s (second line).

2
7
12

+ s

0
2
3

.Equating coefficients gives

4+t=2
1 = 7 2s
5 + t = 12 + 3s.
This has a (unique) solution t = 2, s = 3 so the lines do intersect. The point of intersection
has vector

1
5

+t

0
1

1
5

0
1

1
3

73

Section 4.2: Projections and Planes

(equivalently

29. Let a =

1
2

2
7
12

+ s

and b =

0
2
3

2
0
1

2
7

12

0
2
3

2
1
3

).

be the vectors of A and B. Then d = b a =

1
1
1

is a

direction vector for the line through A and B, so the vector c of C is given by c = a + td for
some t. Then






AC  = c a = td = |t| d and BC  = c b = (t 1)d = |t 1| d .






Hence AC  = 2 BC  means |t| = 2 |t 1| , so t2 = 4(t 1)2 , whence 0 = 3t2 8t + 4 =

5
3

(t 2)(3t 2). Thus t = 2 or t = 23 . Since c = a + td, this means c =

1
0

or c =

3
1
3
4
3

31(b) If there are 2n points, then Pk and Pn+k are opposite ends of a diameter of the circle for each

k = 1, 2, . . . . Hence CP k = CP n+k so these terms cancel in the sum CP 1 +CP 2 + +CP 2n .

Thus all terms cancel and the sum is 0.


33. We have 2EA = DA because E is the midpoint of side AD, and 2AF = F C because F is 31

the way from A to C. Finally DA = CB because ABCD is a parallelogram. Thus

2EF = 2(EA + AF ) = 2EA + 2AF = DA + F C = CB + F C = F B.

Hence EF = 21 F B so F is in the line segment EB, 31 the way from E to B. Hence F is the
trisection point of boty AC and EB.

Exercises 4.2

Projections and Planes

1(b) u v = u u = 12 + 22 + (1)2 = 6
(d) u v = 3 6 + (1)(7) + 5(5) = 18 + 7 25 = 0
(f) v = 0 so u v = a 0 + b 0 + c 0 = 0
2(b) cos =

,
u,v
,
u ,
v

182+0

10 40

(d) cos =

,
u,v
,
u ,v 

6+63

6(3 6)

= 12 . Hence = 3 .

(f) cos =

,
u,v
,
u ,v 

0214

25 100

= 12 . Hence =

3(b) Writing u =

2
1
1

20
20

= 1. Hence = .

2
3 .

and v =

1
2

x
2

, the requirement is

= cos 3 =

uv
u v

2x+2
.
6 x2 +5

Hence 6(x2 + 5) = 4(4 x)2 , whence x2 + 16x 17 = 0. The roots are x = 17 and x = 1.

74

Section 4.2: Projections and Planes

4(b) The conditions are u1 v = 0 and u2 v = 0, yielding equations

3x y + 2z = 0
2x

+ z

The solutions are x = t, y = t, z = 2t, so v = t

= 0
1
1
2

(d) The conditions are u1 v = 0 and u2 v = 0, yielding equations

2x y + 3z = 0
0 = 0.

The solutions are x = s, y = 2s + 3t, z = t, so v = s

1
2
0

+ t

0
3
1


2

3
2 






6(b) P Q =  2 
 = 9 + 4 + 16 = 29
 4 
 2


2 
 2 




QR =  7 
 = 4 + 49 + 4 = 57
 2 
 2

2 
 5 



 = 25 + 25 + 36 = 86.

5
P
R
=




 6 

2 





 2
Hence P R = P Q + QR . Note that this implies that the triangle is right angled,
that P R is the hypotenuse, and hence that the angle at Q is a right angle. Of course, we can

confirm this latter fact by computing P Q QR = 6 14 + 8 = 0.

8(b) We have AB =

Hence =

2
1
1

and
AC =

1
2
1

so the angle at A is given by

2+21
AB AC


= 12 .
cos = 


  =
6 6
AB  AC 

or 60 . Next BA =

2
1
1

and
BC =

1
1

so the angle at B is given by

BA BC
21+2



cos = 
= 12 .


  =
6
6
BA BC 

75

Section 4.2: Projections and Planes

Hence =

. Since
add to , the angle at C is
the
triangle
angles in any

1
1

However, CA = 2 and CB = 1 , this can also be seen directly from

1

= 3 .

CA CB
1 + 2 + 2
 


= 21 .
cos = 
  =
6
6
CA CB 

10(b) proj,v (u) =

uv
v
v2

122+1
16+1+1

uv
(d) proj,v (u) = v
2v =

5
21

2
1
4

1
21

53
26
20

u u1 =

3
2
1

orthogonal to v.

4
2

1 .

1

11
18

= 1
2

6+1+0
4+1+16

4
2

5
21

2
1

. Then u2 = u u1 =

1
4

3
1
0

. As a check, verify that u2 v = 0, that is u2 is orthogonal to v.

27
53

uv
v
v2
6
4
1

12(b) Write p0 =

uv
v
v2

(d) Take u1 = projv (u) =

1 =

1

1882
36+16+4

11(b) Take u1 = proj,v (u) =

, d =

= 1881
36+16+1

=
3
1
4

1
53

27
53

6
4
1

26

1
1
3

and write u =
P0 P = p p0 =
01+16
9+1+16

3
1
4





45





1

from P to the line is QP  = u u1  =  26 41 
=


44
its vector. Then

71 15 34
Hence Q = Q( 26
, 26 , 26 ).

. Then u2 is given by u2 =

u1 = P0 Q and compute it as u1 = projd P0 P =

q = p0 + u1 =

. As a check, verify that u2 v = 0, that is u2 is

, p =

=
1
26

15
26

1
4

3
1
4

1
4

. Write

. Then the distance

5642. To compute Q let q be

15
26

1
26

71
15
34

76

Section 4.2: Projections and Planes

,i
,j

,i
,j

,k

13(b) u v = det
(d) u v = det

,k

= 0i 0j + 0k =

14(b) A normal is n = AB AC =

1
1

15
8

= 0.

= 4i 15j + 8k =

= det

8
17

,i
,j

,k

17

Since the plane passes through B(0, 0, 1) the equation is

23
32
11

23(x 0) 32(y 0) 11(z 1) = 0, that is 23x + 32y + 11z = 11

(d) The plane with equation 2x y + z = 3 has normal n =

2
1
1

. Since our plane is parallel

to this one, 8n will serve as normal. The point P (3, 0, 1) lies on our plane, the equation is
2(x 3) (y 0) + (z (1) = 0, that is 2x y + z = 5.

1

(f) The plane contains P (2, 1, 0) and P0 (3, 1, 2), so the vector u = P P0 = 2 is parallel to

the plane. Also the direction vector d =

n = u d = det

2
3
2

1
0

of the line is parallel to the plane. Hence

is perpendicular to the plane and so serves as a

normal. As P (2, 1, 0) is in the plane, the equation is

2(x 2) + 3(y 1) + 2(z 0) = 0,

(h) The two direction vectors d1 =

n = d1 d2 = det

P (3, 1, 0) so the equation is

1
1
3

and d2 =

2
7
3

x
a
z

2
1
1

are parallel to the plane, so

will serve as normal. The plane contains

2(x 3) + 7(y 1) + 3(z 0) = 0,

Note that this plane contains the line

that is, 2x + 3y + 2z = 7.

3
1
0

that is 2x + 7y + 3z = 1.

+t

1
3

by construction; it contains

the other line because it contains P (0, 2, 5) and is parallel to d2 . This implies that the lines
intersect (both are in the same plane). In fact the point of intersection is P (4, 0, 3) [t = 1 on
the first line and t = 2 on the second line].

77

Section 4.2: Projections and Planes

(j) The set of all points R(x, y, z) equidistant from both P (0, 1, 1) and Q(2, 1, 3) is deter 

2 


  


 2
mined as follows: The condition is P R = QR , that is P R = QR , that is
x2 + (y 1)2 + (z + 1)2 = (x 2)2 + (y + 1)2 + (z + 3)2 .

This simplifies to x2 + y 2 + z 2 2y + 2z + 2 = x2 + y2 + z 2 4x + 2y + 6z + 14; that is

4x 4y 4z = 12; that is x y z = 3.

15(b) The normal n =

2
1
0

to the given plane will serve as direction vector for the line. Since the

line passes through P (2 1 3), the vector equation is

(d) The given lines have direction vectors d1 =

d = d1 d2 = det

1
1
2

1
1
1

x
y
z

and d2 =

1
2
3

1
3

+ t

2
1
0

, so

is perpendicular to both lines.

Hence d is a direction vector for the line we seek. As P (1, 1, 1) is on the line, the vector
equation is

y
z

+ t

1
1

(f) Each point on the given line has the form Q(2 + t, 1 + t, t) for some t. So P Q =

This is perpendicular to the given line if P Q d = 0 (where d =

1
1
1

1+t
t
t2

is the direction vector

1
of the given line). This condition
4 is(1 + t) + t + (t 2) = 0, that is
t = 3 . Hence the line

we want has direction vector

3
1
3
5
3

. For convenience we use d =

want passes through P (1, 1, 2), the vector equation is

7

4 1
3, 3, 3

is the point of intersection of the two lines.]

x
y
z

1
1
2

+t

16(b) Choose a point P0 in the plane, say P0 (0, 6, 0), and write u = P0 P =

. As the line we
4
1
5

3
5
1

. [Note that

. Now write

78

Section 4.2: Projections and Planes

n=

2
1
1

for the normal to the plane. Compute

u1 = projn (u) =

un
n=
n2

2
6

1
1

The distance from P to the plane is u1  = 13 6.

0

Since p0 = 6
and q are the vectors of P0 and Q, we get
0

Hence Q = Q

q = p0 + (8u 8u1 ) =

7

2 2
3, 3, 3

0
6
0

17(b) A normal to the plane is given by

2

n = PQ PR =
2
4

3
1
3

3
5
1

= det

Thus, as P (4, 0, 5) is in the plane, the equation is

10(x 4) + 6(y 0) + 8(z 5) = 0;

1
3

2
2

10
6
8

that is 5x 3y 4z = 0.

The plane contains the origin P (0, 0, 0).

19(b) The coordinates of points of intersection satisfy both equations:
3x + y 2z = 1
x + y + z = 5.
Solve

2
1

1
5

5
14

23
5
2

2
7

Take z = 2t, to eliminate fractions, whence x = 2 + 3t and y = 7 5t. Thus

is the line of intersection.

y
z

2 + 3t
7 5t
2t

2
7
0

= +t

5
2

20(b) If P (x, y, z) is an intersection point, then x = 1 + 2t, y = 2 + 5t, z = 3 t since P is on the

line. Substitution in the equation of the plane gives 2(1 + 2t) (2 + 5t) (3 t) = 5, that
is 1 = 5. Thus there is no such t, so the line does not intersect the plane.

79

Section 4.2: Projections and Planes

(d) If P (x, y, z) is an intersection point, then x = 1 + 2t, y = 2 + 5t and z = 3 t since P is on

the line. Substitution
in the
of

equation
the plane gives 1(1 + 2t) 4(2 + 5t) 3(3 t) = 6,
x

whence t =

8
19 .

Thus

y
z

3
19
78
19
65
19

21(b) The line has direction vector d =

so P

3
0
2

3

78 65
19 , 19 , 19

is the point of intersection.

which is a normal to all such planes. If P0 (x0 , y0 , z0 )

is any point, the plane 3(x x0 ) = 0(y y0 ) + 2(z z0 ) = 0 is perpendicular to the line. This
can be written 3x + 2z = 3x0 + 2z0 , so 3x + 2z = d, d arbitrary.

a

(d) If the normal is n =

and c are not all zero.

= 0, the plane is a(x 3) + b(y 2) + c(z + 4) = 0, where a, b

(f) The vector u = P Q =

1
1

is parallel to these planes so the normal n = b is

c

orthogonal to u. Thus 0 = u n = a + b c. Hence c = b a and n =

ba

. The plane

passes through Q(1, 0, 0) so the equation is a(x 1) + b(y 0) + (b a)(z 0) = 0, that is

ax + by + (b a)z = a. Here a and b are not both zero (as n = 0). As a check, observe that
this plane contains P (2, 1, 1) and Q(1, 0, 0).

a

(h) Such a plane contains P0 (3, 0, 2) and its normal n =

vector d =

2
1

must be orthogonal to the direction

of the line. Thus 0 = dn = a2bc, whence c = a2b and n =

(where a and b are not both zero as n = 0). Thus the equation is
a(x 3) + b(y 0) + (a 2b)(z 2) = 0,

a
b

a 2b

that is ax + by + (a 2b)z = 5a 4b

where a and b are not both zero. As a check, observe that the plane contains every point
P (3 + t, 2t, 2 t) on the line.
23(b) Choose P1 (3, 0, 2) on the first line. The distance in question is the distance

from P1 to
the

4
3

second line. Choose P2 (1, 2, 2) on the second line and let u = P2 P1 = 2 . If d = 1

0

is the direction vector for the line, compute

u1 = projd (8u) =

ud
d
d2

10
10 [3

1 0]T = [3 1 0]T .



then the required distance is u u1  = [1 3 0]T  = 10.

80

Section 4.2: Projections and Planes

24(b) The cross product n =

1
1
1

3
1
0

of the two direction vectors serves as a

normal to the plane. Given P1 (1, 1, 0) and P2 (2, 1, 3) on the lines, let u = P1 P2 =
Compute

u1 = projn (u) =

7
14

1
3

1
2

The required distance is u1  = 21 1 + 9 + 4 = 21 14.

1
3
2

1
0
3

Now let A = A(1 + s, 1 + s, s) and B =

B(2 + 3t, 1
+ t, 3) be the points on the two
1 + 3t s

lines that are closest together. Then AB = t s is orthogonal to both direction

vectors d1 =

1
1
1

and d2 =

3
1
0

3s

. By Theorem 3 this means d1

AB = 0 = d2 AB,

giving equations 4t 3s =4, 10t  4s = 3. The solution is t = 12 , s = 2, so the points are

A = A(3, 1, 2) and B = B 27 , 21 , 3 .

 37 13 
1
24(d) Analogous to (b). The distance is 66 , and the points are A( 19
3 , 2, 3 ) and B = B 6 , 6 , 0 .

26(b) Position the cube with and vertex at the origin and sides along the
axes. Assume

positive
a

each side has length a and consider the diagonal with direction d =

that do not meet d are:

(the dot product is 0).

a
0

a
0

and

a
a

. The face diagonals

, and all are orthogonal to d

a
a

28. Position the solid with one vertex at the origin and
x,
oflengths
a, b,
c, along
the positive

sides,
a

y and z axes respectively. The diagonals are

a
b

b
c

, and

The possible dot products are (a2 + b2 + c2 ), (a2 b2 + c2 ), (a2 + b2 c2 ), and one of
these is zero if and only if the sum of two of a2 , b2 and c2 equals the third.
34(b) The sum of the squares of the lengths of the diagonals equals the sum of the squares of the
lengths of the four sides.
38(b) The angle between u and u + 8v + w
8 is given by
cos =

u (u + v + w)
uu+uv+uw
u2 + 0 + 0
u
=
=
=
.
u u + v + w
u u + v + w
u u + v + w
u + v + w

Similarly the angles , between v and w and u + v + w are given by

cos =

v
u + v + w

and

cos =

w
.
u + v + w

81

Section 4.3: More on the Cross Product

Since u = v = w we

= = .
 get cos = cos = cos , whence

NOTE: u + v + w = u2 + v2 + w2 = u 3 by part (a), so cos = cos =

cos = 13 . Thus, in fact = = = .955 radians, (54.7 ).




x0 x1
39(b) If P1 (x, y) is on the line then ax + by + c = 0. Hence u = P1 P0 =
so the distance is
y0 y1



 u n  |u n|
|a(x0 x) + b(y0 y)|
|ax0 + by0 + c|



projn u = 
=
.
2 n = n =
2
2
a +b
a2 + b2
n

41(b) This follows from (a) because v2 = a2 + b2 + c2 .

44(d) Take x1 = z2 = x, y1 = x2 = y and z1 = y2 = z in (c).

Exercises 4.3

More on the Cross Product

3(b) One vector orthogonal to u and v is uv = det





5



1
1

5
5

. We have u v =

5
k 1 2





1
 = 5 3. Hence the unit vectors parallel to uv are

5 3


4(b) The area of the triangle is

1
2

5
5

33

the area of the parallelogram ABCD. By Theorem 4,


 


 2

 0 
4







 1 
1
1 



Area of triangle = 2 AB AC  = 2  1

2
 = 2  0  = 0.
 1

 0 
2

1
1

Hence AB and AC are parallel.

(d) Analogous to (b). Area = 5.

5(b) We have u v =
Theorem 5.

4
5
1

so w (u v) = 7. The volume is |w (u v)| = |7| = 7 by

6(b) The line through P0 perpendicular to the plane has direction vector n, and so has vector
equation p8 = p80 + t8n where p8 = [x, y, z]T . If P (x, y, z) also lies in the plane, then n p =
ax + by + cz = d. Using p = p0 + tn we find
d = n p = n p0 + t(n n) = n p0 + t n2


d n p0
d n p0
8n. Finally, the distance from P0 to the plane
Hence t =
,
so
p
=
p
+
0
n2
n2
is



 d n p0
 |d n p0 |



P
P
=
p

p

=
8n
.
 0
0

=
2
n
n

82

Section 4.3: More on the Cross Product

10. The points A, B and C are all on one line if and only if the parallelogram they determine has

area zero. Since this area is AB AC, this happens if and only if AB AC = 80.
12. If u and v are perpendicular, Theorem 4 shows that u v = u v . Moreover, if w is
perpendicular to both u and v, it is parallel to u v so w (u v) = w u v because
the angle between them is either 0 or . Finally, the rectangular parallepiped has volume
|w (u v)| = w u v = w (u v)
using Theorem 5.

15(b) If u =
y , v = q and w = m then, by the row version of Exercise 19 3.1, we
z
r
n
get

l+p

y m+q

k z n+r

i x l
i x p

+ det j y m = u v + u w.
= det
j
y
q

k z n
k z r

u (v + w) = det
j

v1

w1

u1

16(b) Let v =
v2 , w = w2 and u = u2 . Compute
u3
w3
v3
v [(u v) + (v w) + (w u)] = v (u v) + v (v w) + v (w u)

v1 w1 u1

= 0 + 0 + det
v2 w2 u2
v3 w3 u3
by Theorem 1. Similarly

w1 u1 v1

w [[(u 8v ) + (8v w) + (w u)]] = w (u 8v ) = det

w2 u2 v2 .
w3 u3 v3
These determinants are equal because each can be obtained from the other by two column
interchanges. The result follows because (v w) x = v x w x for any vector x.

83

Section 4.4: Linear Operators on R3

22. If v1 and v2 are vectors of points in the planes (so v1 n = d1 and v2 n = d2 ), the distance
is the length of the projection of v2 v1 along n
; that is



 (v2 v1 ) n
 |(v2 v1 ) n|
|d2 d1 |
=
projn (v2 v1 ) = 
8
n
=
.


n2
n
n

Linear Operators on R3

Exercises 4.4
1(b) By inspection, A =

1
2

1
1

of projection on y = x.


3
(d) By inspection, A = 51
of reflection in y = 2x.


1
(f) By inspection, A = 2

; by the formulas precedinging Theorem 2, this is the matrix

; by the formulas preceding Theorem 2, this is the matrix

3.

; by Example 5 2.5 this is the matrix of rotation through

2(b) For any slope m, projection on the line y = mx has matrix

1
1+m2

m2

(see the discussion

preceding


 Theorem

2). Hence
 the projections on the lines y = x and y = x have matrices
1
1
1
1
1
and 12
, respectively, so the first followed by the second has matrix (note
2
1

the order)

1
2



1
2

1
4

= 0.

It follows that projection on y = x followed by projection on y = x is the zero transformation.

Note that this conclusion can also be reached geometrically. Given any vector 8v, its
projection p8 on the line y = x points along that line. But the line y = x is perpendicular
to the line y = x, so the projection of p
8 along y = x will be the zero vector. Since 8v was
arbitrary, this shows again that projection on y = x followed by projection on y = x is the
zero transformation.

17

3(b) By Theorem 3:

(d) By Theorem 3:

(f) By Theorem 2:

1
21

1
30

1
25

20

22

8
4

28

20

10

12

12

16

20

10
20

1
7

26

0
1

1
21

=
1
25

11

1
15

93
0

124

32
1
35

84

Section 4.5: An Application to Computer Graphics

(h) By Theorem 2:

1
11

2
5
0

4(b) This is Example 1 with = 6 . Since cos

1
2

3
1

=
=

1
11

28

49
18

3
2

and sin

. Hence the rotation of 8v =

1
2

is

3
2
1
2

= 12 , the matrix is

1
0
3

3
2

1

3

1
2

12

1
6

6. Denote the rotation by RL, . Here the rotation takes place about the y axis, so RL, (8j) = 8j.
In the x-z

plane the effect of RL, is to rotate counterclockwise through

 , and

this
 has
cos

matrix

RL,

sin

sin

0
1

cos

sin
cos

by Example 4 2.6. So, in the x-z plane, RL,

cos

. Hence RL, (8i) =

and RL, (8k) =

sin

matrix of RL, is [RL, (8i) RL, (8j) RL, (8k)] =

cos

sin

sin

cos

1
0

sin
0

cos

cos
sin

and

. Finally, the

9(a) Write 8v = [x y]T . Then PL (8v ) = projd,(8v ) =











a 

a2 x + aby
a2 ab
x
ax+by
,
v d,
1
1
8
= a2 +b2
= a2 +b2
.
d = a2 +b2
2
2
2
,
b
abx + b y
ab b
y
d



a2 ab
1
Hence the matrix of PL is a2 +b2
. Note that if the line L has slope m this retrieves
ab b2



1
m
1
the formula 1+m
preceding Theorem 2. However the present matrix works for
2
m m2


1
vertical lines, where d8 =
.
0

Exercises 4.5

An Application to Computer Graphics

1(b) Translate to the origin, rotate and then translate back. As in Example 1, we compute

2
1

1
2

0
1
0

1
2
1

2+2

3 2 + 4

2
2
2

2
2
2

0
0

0
0

7 2+2

3 2+4

3 2+2

5 2+4

2+2

2+4

5(b) The line has a point w

8=

5 2 + 2

9 2+4

, so we translate by w,
8 then reflect in y = 2x, and then trans


3 4
1
late back by w.
8 The line y = 2x has matrix 5
. Thus the matrix (for homogeneous
1

85

Supplementary Exercises - Chapter 4

coordinates) is

Hence for w
8 =

1
4
1

1
5

we get 1
5

Supplementary Exercises

2
5

1
4
1

1
5

9
18
5

1
5

4
2
5

. Hence the point is P

9

18
5, 5

Chapter 4

4. Let p and w be the velocities of the airplane and the wind. Then p = 100 knots and
w = 75 knots and the resulting actual velocity of the airplane is v = w + p. Since w and p
are orthogonal. Pythagoras theorem gives v2 = w2 +p2 = 752 +1002 = 252 (32 +42 ) =
75
252 52 . Hence v = 25 5 = 125 knots. The angle satisfies cos = w
v = 125 = 0.6 so

= 0.93 radians or 53 .
6. Let v = [x y]T denote the velocity of the boat in the water. If c is the current velocity then
c = (0, 5) because it flows south at 5 knots. We want to
choose
 v so that the resulting actual
z

velocity w of the boat has easterly direction. Thus w =

for some z. Now w = v + c so
0







z
x
0
x
=
+
=
. Hence z = x and y = 5. Finally, 13 = v = x2 + y 2 =
0
y
5
y5

x2 + 25 gives x2 = 144, x = 12. But x > 0 as w heads east, so x = 12. Thus he steers
v = [12 5]T , and the resulting actual speed is w = z = 12 knots.

86

Section 5.1: Subspaces and Spannng

Chapter 5: The Vector Space Rn

Exercises 5.1

Subspaces and Spanning

1(b) Yes. In fact, U = span

(d) No.

(f) No.

2
0
0

is in U but 2

0
1
0

0
1
0
2
0
0

0
0
1

is in U but (1)

0
1
0

so Theorem 1 applies.

4
0
0

is not in U.

0
1
0

is not in U.

2(b) No. If x = ay + bz equating first and third components gives 1 = 2a + b, 15 = 3b; whence
a = 3, b = 5. This does not satisfy the second component which requires that 2 = a b.
(d) Yes. x = 3y + 4z.
3(b) No. Write these vectors as a1 , a2 , a3 and a4 , and let A = [a1 a2 a3 a4 ] be the matrix with
these vectors as columns. Then det A = 0, so A is not invertible. By Theorem 5 2.4, this
means that the system Ax = b has no solution for some column b. But this says that bis not
a linear combination of the ai by Definition 1 2.2. That is, the ai do not span R4 .

1

For a more direct proof,

0
0
0

is not a linear combination of a1 , a2 , a3 and a4 .

10. Since ai xi is in span {xi } for each i, Theorem 1 shows that span {ai xi } span {xi } . Since
xi = a1
i (ai xi ) is in span {ai xi } , we get span {xi } span {ai xi } , again by Theorem 1.
12. We have U = span{x1 , , xk } so, if y is in U, write y= t1 x1 + + tk xk where the ti are in
R. Then Ay = t1 Ax1 + + t1 Axk = t1 0 + + tk 0 = 0.
15(b) x= (x + y)y is in U because x+y and y = (1)y are both in U and U is a subspace.
16(b) True. If we take r = 1 we see that x= 1x is in U.
(d) True. We have span {y, z} span {x, y, z} by Theorem 1 because both y and z are in
span {x, y, z} . In other words, U span {x, y, z} .

For the other inclusion, it is clear that y and z are both in U = span {y, z} , and we are
given that x is in U. Hence span {x, y, z} U by Theorem 1.
'

(
1
2
(f) False. Every vector in span
,
has second component zero.
0

Section 5.2: Independence and Dimension

87

20. If U is a subspace then S2 and S3 certainly hold. Conversely, suppose that S2 and S3 hold.
It is here that we need the condition that U is nonempty. Because we can then choose some
x in U, and so 0= 0x is in U by S3. So U is a subspace.
22(b) First, 0 is in U + W because 0= 0+0 (and 0 is in both U and W ). Now suppose that P and
Q are both in U + W, say p= x1 + y1 and q= x2 + y2 where x1 and x2 are in U, and y1 and
y2 are in W. Hence
p + q = (x1 + y1 ) + (x2 + y2 ) = (x1 + x2 ) + (y1 + y2 )
so p + q is in U + W because x1 + x2 is in U (both x1 and x2 are in U), and y1 + y2 is in
W. Similarly
aP = a(x1 + y1 ) = ax1 + ay1
is in p+q because ax1 is in p and ay1 is in Q. Hence U + W is a subspace.

Exercises 5.2

Independence and Dimension

1(b) Yes. The matrix with these vectors as columns has determinant 2 = 0, so Theorem 3
applies.
1(d) No. (1, 1, 0, 0) (1, 0, 1, 0) + (0, 0, 1, 1) (0, 1, 0, 1) = (0, 0, 0, 0) is a nontrivial linear combination that vanishes.
2(b) Yes. If a(x + y) + b(y + z) + c(z + x) = 0 then (a + c)x + (a + b)y + (b + c)z = 0. Since we
are assuming that {x, y, z} is independent, this means a + c = 0, a + b = 0, b + c = 0. The
only solution is a = b = c = 0.
(d) No. (x+ y) (y +z) + (z +w) (w +x) = 0 is a nontrivial linear combination that vanishes.
3(b) Write x1 = (2, 1, 0, 1), x2 = (1, 1, 1, 1), x3 = (2, 7, 4, 1), and write U = span{x1 , x2 , x3 }.
Observe that x3 = 3x1 +4x2 so U = {x1 , x2 } . This is a basis because {x1 , x2 } is independent,
so the dimension is 2.
(d) Write x1 = (2, 0, 3, 1), x2 = (1, 2, 1, 0), x3 = (2, 8, 5, 3), x4 = (1, 2, 2, 1) and write
U = span{x1 , x2 , x3 , x4 }. Then x3 = 3x1 + 4x2 and x4 = x1 + x2 so the space is span{x1 , x2 }.
As this is independent, it is a basis so the dimension is 2.
4(b) (a + b, a b, b, a)= a(1, 1, 0, 1) + b(1, 1, 1, 0) so

U = span {(1, 1, 0, 1), (1, 1, 1, 0)} . This is a basis so dim U = 2.

(d) (ab, b+c, a, b+c)= a(1, 0, 1, 0)+b(1, 1, 0, 1)+c(0, 1, 0, 1). Hence U = span {(1, 0, 1, 0), (1, 1, 0, 1), (0, 1, 0, 1
This is a basis so dim U = 3.
(f) If a + b = c + d then a = b + c + d. Hence U = {(b + c + d, b, c, d) | b, c, d in R} so
U = span {(1, 1, 0, 0), (1, 0, 1, 0), (1, 0, 0, 1)} . This is a basis so dim U = 3.
5(b) Let a(x + w) + b(y + w) + c(z + w) + dw = 0, that is ax + by + cz + (a + b + c + d)w = 0. As
{x, y, z, w} is independent, this implies that a = 0, b = 0, c = 0 and a + b + c + d = 0. Hence
d = 0 too, proving that {x + w, y + w, z + w, w} is independent. It is a basis by Theorem 7
because dim R4 = 4.

88

Section 5.2: Independence and Dimension

6(b) Yes. They are independent (the matrix with them as columns has determinant 2) and so
are a basis of R3 by Theorem 7 (since dim R3 = 3).
(d) Yes. They are independent (the matrix with them as columns has determinant 6) and so
are a basis of R3 by Theorem 7 (since dim R3 = 3).
(f) No. The determinant of the matrix with these vectors as its columns is zero, so they are not
independent (by Theorem 3). Hence they are not a basis of R4 because dim R4 = 4.
7(b) True. If sy + tz = 0 then 0x + sy + tz = 0, so s = t = 0 by the independence of {x, y, z}.
(d) False. If x =0 let k = 2, x1 =x and x2 = x. Then each xi = 0 but {x1 , x2 } is not
independent.
(f) False. If y= x and z= 0 then 1x + 1y + 1z = 0, but {x, y, z} is certainly not independent.
(h) True. The xi are not independent so, by definition, some nontrivial linear combination vanishes.
10. If rx2 + sx3 + tx5 = 0 then 0x1 + rx2 + sx3 + 0x4 + tx5 + 0x6 = 0. Since the larger set is
independent, this implies r = s = t = 0.
12. If t1 x1 + t2 (x1 + x2 ) + + tk (x1 + x2 + + xk ) = 0 then, collecting terms in x1 , x2 , . . . ,
(t1 + t2 + + tk )x1 + (t2 + + tk )x2 + + (tk1 + tk )xk1 + tk xk = 0.
Since {x1 , x2 , . . . , xk } is independent we get
t1 + t2 + + tk = 0
t2 + + tk = 0
..
.
tk1 + tk = 0
tk = 0.
The solution (from the bottom up) is tk = 0, tk1 = 0, . . . , t2 = 0, t1 = 0.
T
2
16(b) We show that AT is invertible.

 Suppose A x = 0, x in R . By Theorem 5 2.4, we must
s
show that x= 0. If x=
then AT x = 0 gives as + ct = 0, bs + dt = 0. But then
t

s(ax+by)+t(cx+dy) = (sa+tc)x+(sb+td)y = 0. Hence s = t = 0 because {ax+by, cx+dy}

is independent.

17(b) Note first that each V 1 xi is in null(AV ) because (AV )(V 1 xi ) = Axi = 0. If t1 V 1 x1 +
+ tk V 1 xk = 0 then V 1 (t1 x1 + + tk xk ) = 0 so t1 x1 + + tk xk =)0 (by multiplication
*
by V ). Thus t1 = = tk = 0 because {x1 , . . . , xk } is independent. So V 1 x1 , . . . , V 1 xk
is independent. To see that it spans null(AV ), let y be in null(AV ), so that AV y = 0.
Then V y is in null A so V y = s1 x1 + + sn xn because {x1 , . . . , xn } spans null A. Hence
y = s1 V 1 x1 + +sk V 1 xk , as required.

89

Section 5.3: Orthogonality

20. We have {0} U W where dim{0} = 0 and dim W = 1. Hence dim U is an integer between
0 and 1 (by Theorem 8), so dim U = 0 or dim U = 1. If dim U = 0 then U = {0} by Theorem
8 (because {0} U and both spaces have dimension 0); if dim U = 1 then U = W again by
Theorem 8 (because U W and both spaces have dimension 1).

Exercises 5.3
1(b)

Orthogonality
,

1 (1, 1, 1), 1 (4, 1, 5), 1 (3, 3, 1)

3
42
14

vector.

where in each case we divide by the norm of the

3(b) Write e1 = (1, 0, 1), e2 = (1, 4, 1), e3 = (2, 1, 2). Then

e1 e2 = 1 + 0 1 = 0, e1 e3 = 2 + 0 2 = 0, e2 e3 = 2 4 + 2 = 0,

so {e1 , e2 , e3 } is orthogonal and hence a basis of R3 . If x = (a, b, c), Theorem 6 gives

x e1
x e2
x e3
a+4b+c
2ab+2c
x=
e3 = ac
e3 .
2 e1 +
18 e2 +
9
2 e1 +
2 e2 +
e1 
e2 
e3 2
3(d) Write e1 = (1, 1, 1), e2 = (1, 1, 0), e3 = (1, 1, 2). Then

e1 e2 = 1 1 + 0 = 0, e1 e3 = 1 + 1 2 = 0, and e2 e3 = 1 1 + 0 = 0.

Hence {e1 , e2 , e3 } is orthogonal and hence is a basis of R3 . If x = (a, b, c), Theorem 6 gives
x e2
x e3
x e1
ab
a+b2c
e3 = a+b+c
e3 .
x=
3 e1 + 2 e2 +
6
2 e1 +
2 e2 +
e1 
e2 
e3 2
4(b) If e1 = (2, 1, 0, 3) and e2 = (2, 1, 2, 1) then {e1 , e2 } is orthogonal because e1 e2 = 4
1+03 = 0. Hence {e1 , e2 } is an orthogonal basis of the space U it spans. If x = (14, 1, 8, 5)
is in U, Theorem 6 gives
x e1
x e2
40
x=
e2 = 42
14 e1 + 10 e2 = 3e1 + 4e2 .
2 e1 +
e1 
e2 2
We check that these are indeed equal. [We shall see in Section 8.1 that in any case,


x e1
x e2
x
e
+
e
is orthogonal to every vector in U.]
1
2
e1 2
e2 2

5(b) The condition that (a, b, c, d) is orthogonal to each of the other three vectors gives the following
equations for a, b, c, and d.
a
2a +
a

Solving we get:

c + d = 0
b

3b

1 0 1 1 0

2 1
1 1 0

1 3 1
0 0

1 0 1
1

0 1 3 3
0 0 11 10

+ c d = 0
+ c

0
0
0

= 0
1

The solution is (a, b, c, d) = t(1, 3, 10, 11),t in R.

1
3

1
3

1
11
3
11
10
11

0
0

0
0
0

90

Section 5.3. Orthogonality

6(b) 2x + 7y2 = (2x + 7y) (2x + 7y)

= 4(x x) + 14(x y) + 14(y x) + 49(y y)

= 4 x2 + 28(x y) + 49 y2

= 36 56 + 49
= 29.

(d) (x 2y) (3x + 5y) = 3(x x) + 5(x y) 6(y x) 10(y y)

= 3 x2 (x y) 10 y2
= 27 + 2 10
= 19.
7(b) False. For example, if x = (1, 0) and y = (0, 1)in R2 ,then {x, y}is orthogonal but x+y = (1, 1)
is not orthogonal to x.
(d) True. Let x and y be distinct vectors in the larger set. Then either both are xi s, both are
yi s, or one is an xi and one is a yi . In the first two cases xy = 0 because {xi } and {yj } are
orthogonal sets; in the last case xy = 0 by the given condition.
(f) True. Every pair of distinct vectors in {x} are orthogonal (there are no such pairs). As x= 0,
this shows that {x} is an orthogonal set.
9 Row i of AT is cTi so the (i, j) entry of AT A is cTi cj = ci cj . This is 0 if i = j, and 1 if i = j.
That is AT A = I.
11(b)
Take x= (1, 1, 1) and y= (r1 , r2 , r3 ). Then |x y| x y byTheorem 2; that is |r1 + r2 + r3 |
3 r12 + r22 + r32 . Squaring both sides gives
r21 + r22 + r32 + 2(r1 r2 + r1 r3 + r2 r3 ) 3(r12 + r22 + r32 )
Simplifying we obtain r1 r2 + r1 r3 + r2 r3 r12 + r22 + r32 , as required.
12(b) Observe first that
(x + y) (x y) = x2 y2

holds for all vectors x and yin

(*)

Rn .

If x+y and xy are orthogonal then (x + y) (x y) = 0, so x2 = y2 by (*). Taking

positive square roots gives x = y.
Conversely, if x = y then certainly x2 = y2 , so (*) gives (x + y) (x y) = 0.
This means that x+y and xy are orthogonal.

15. If is an eigenvalue of AT A, let (AT A)x = x where x = 0 in Rn . Then:

Ax2 = (Ax)T (Ax) = (xT AT )Ax = xT (AT Ax) = xT (x) = x2 .
Since x =
 0 (because x = 0), this gives =

Ax2
x2

0.

91

Section 5.4: Rank of a Matrix

Exercises 5.4

1(b)

2
2
4

Rank of a Matrix

2
0

1
0

1
2

21
0

0
*
)
Hence, rank A = 2 and [1 12 12 ]T , [0 0 1]T is a basis of row A. Thus {[2 1 1]T , [0 0 1]T }
is also a basis of row A. Since the leading 1s are in columns 1 and 3, columns 1 and 3 of A
are a basis of col A.









1
2
1
3
1 2 1 3
1 2 1 3
(d)

)
*
Hence, rank A = 2 and [1 2 1 3]T , [0 0 0 1]T is a basis of row A. Since the leading 1s
are in columns 1 and 4, columns 1 and 4 of A are a basis of col A.

2(b) Apply the gaussian algorithm to the matrix with these vectors as rows:

1
1

0
0

25

12

32

Hence, {[1 1 0 0 0]T , [0 2 2 5 1]T , [0 0 2 3 6]T } is a basis of U (where we

have cleared fractions using scalar multiples).
(d) Write these columns as the rows of the following matrix:

Hence,

6
8

10
12

1
5
6

0
1

0
0

12

0
0
1

6
4
8

36

is a basis of U.

6
1
24
0

6
1
1
0

3(b) No. If the 3 columns were independent, the rank would be 3.

No. If the 4 rows were independent, the rank would be 4, a contradiction here as the rank
cannot exceed the number of columns.
(d) No. Suppose that A is m n. If the rows are independent then rank A = dim(row A) = m
(the number of rows). Similarly if the columns are independent then rank A = n (the number
of columns). So if both the rows and columns are indeependent then m = rank A = n, that
is A is square.
(f) No. Then dim(null A) = n r = 4 2 = 2, contrary to null(A) = Rx where x = 0.
4. Let cj denote column j of A. If x = [x1 xn ]T Rn then Ax = x1 c1 + + xn cn by
Definition 1 2.2. Hence
col A = span{c1 , . . . , cn } = {x1 c1 + + xn cn | xj R} = {Ax | x Rn }.

92

Section 5.4: Rank of a Matrix

7(b) The null space of A is the set of columns X such that AX = 0. Applying gaussian elimination
to the augmented matrix gives:

3
5
5
2
0 0
1 0
2
2
1 0


1 0 2 2

1 0

0 5 1 4 3 0
1 1 1 2 2 0
0 1 1 4 3 0


2 0 4 4 2 0
0 0
0
0
0 0

1 0
2
2
1 0
1 0 0 6 5 0


0 1 1 4 3 0
0 1 0 0
0 0

0 0
4
16
12 0
0 0 1
4
3 0
0 0
0
0
0 0
0 0 0
0 0
0

6s + 5t

4s

3t
Hence, the set of solutions is null A =
| s, t in R = span B where

6
5

0 0

B = 4 , 3 . Since B is independent, it is the required basis of null A. We have

1 0

r = rank A = 3 by the above reduction, so n r = 5 3 = 2. This is the dimension of null

A, as Theorem 3 asserts.

8(b) Since A is m n, dim(null A) = n rank A. To compute rank A, let R = [r1 r2 rn ].

Then A = CR = [r1 C r2 C rn C] by block multiplication, so
col A = span{r1 C, r2 C, . . . , rn C} = span{C}
because some ri = 0. Hence rank A = 1, so dim(null A) = n rank A = n 1.
9(b) Let A = [c1 ... cn ] where cj is the jth column of A; we must show that {c1 , ..., cn } is
independent. Suppose that x1 c1 + + xn cn = 0, xi in R. If we write x = [x1 xn ]T , this
reads Ax = 0 by Definition 1 2.2. But then x is in null A, and null A = 0 by hypothesis. So
x = 0, that is each xi = 0. This shows that {c1 , ..., cn } is independent.
10(b) If A2 = 0 then A(Ax) = 0 for all xin Rn , that is {Ax | X in Rn } null A. But col A =
{Ax |xin Rn }, so this shows that col A null A. If we write r = rank A, taking dimensions
gives r = dim(col A) dim(null A) = n r by Theorem 3. It follows that 2r n; that is
r n2 .
12 We have rank(A) = dim[col(A)] and rank(AT ) = dim[row(AT )]. Let {c1 , c2 , . . . , ck } be a
basis of col(A); it suffices to show that {cT1 , cT2 , . . . , cTk } is a basis of row(AT ). But if t1 cT1 +
t2 cT2 + . . . + tk cTk = 0, tj in R, then (taking transposes) t1 c1 + t2 c2 + . . . + tk ck = 0 so each
tj = 0. Hence {cT1 , cT2 , . . . , cTk } is independent. Given v in row(AT ) then vT is in col(A), say
vT = s1 c1 + s2 c2 + . . . + sk ck , sj in R. Hence v = s1 cT1 + s2 cT2 + . . . + sk cTk so {cT1 , cT2 , . . . , cTk }
spans row(AT ), as required.
15(b) Let {c1 , . . . , cr } be a basis of col A where r = rank A. Since Ax = b has no solution, bis not
in col A = span{c1 , , cr } by Exercise 12. It follows that {c1 , . . . , cr , b} is independent [If
a1 c1 + + ar cr + ab = 0 then a = 0 (since bis not in col A), whence each ai = 0 by the

93

Section 5.5: Similarity and Diagonalization

independence of the ci ]. Hence, it suffices to show that col[A, B] = span{c1 , , cr , b} . It is

clear that bis in col[A, b] , and each cj is in col[A, b] because it is a linear combination of
columns of A (and so those of [A, b]). Hence
span {c1 , . . . , cr , b} col[A, b].
On the other hand, each column xin col[A, b] is a linear combination of band the columns of A.
Since these columns are themselves linear combinations of the cj , so xis a linear combination
of band the cj . That is, xis in span{c1 , . . . , cr , b} .

Exercises 5.5

Similarity and Diagonalization

1(b) det A = 5, det B = 1 (so A and B are not similar). However, tr A = 2 = tr B, and rank
A = 2 = rank B (both are invertible).
(d) tr A = 5, tr B = 4 (so A and B are not similar). However, det A = 7 = det B, so
rank A = 2 = rank B (both are invertible).
(f) tr A = 5 = tr B; det A = 0 = det B; however rank A = 2, rank B = 1 (so A and B are not
similar).
3(b) We have A B, say B = P 1 AP. Hence B 1 = (P 1 AP )1 = P 1 A1 (P 1 )1 , so

A1 B 1 because P 1 is invertible.


x3
0
6


= (x + 3)(x2 5x 24) = (x + 3)2 (x 8). So the eigenvalues
x+3
0
4(b) cA (x) = 0

5
0
x2
are 1 = 3,
eigenvectors:

2 = 8. To findthe associated
1 = 3:

2 = 8:

Since

P =

11

1
0
1
0

1
1

6
0

x4

(d) cA (x) = 0
2

E1 =

6
0
1
0

6
0
5

65
0
0

; basic eigenvectors

; basic eigenvector

x2
3

0
2

x1

0
1

6
0
5

1
0

is a basis of eigenvectors, A is diagonalizable and

will satisfy P 1 AP =
0

3
0
0

= (x 4)2 (x + 1). For = 4,

2
3

1
0

. Hence A is not diagonalizable by Theorem 6 because the dimension of E4 (A) = 1

while the eigenvalue 4 has multiplicity 2.

94

Section 5.5: Similarity and Diagonalization

8(b) If B = P 1 AP and Ak = 0, then B k = (P 1 AP )k = P 1 Ak P = P 1 0P = 0.

9(b) Let the diagonal entries of A all equal . If A is diagonalizable then P 1 AP = I by Theorem
3 for some invertible matrix P. Hence A = P (I)P 1 = (P IP 1 ) = I.
10(b) Let P 1 AP = D = diag{1 , 2 , . . . , n } . Since A and D are similar matrices, they have the
same trace by Theorem 1. That is
tr A = tr (P 1 AP ) = tr D = 1 + 2 + + n .

12(b) TP (A)TP (B) = (P 1 AP )(P 1 BP ) = P 1 AIBP = P 1 ABP = TP (AB).

13(b) Assume that A is diagonalizable, say A D where D is diagonal, say D = diag(1 , . . . , n )
where 1 , , n are the eigenvalues of A. But A and AT have the same eigenvalues (Example
5 3.3) so AT D also. Hence A D AT , so A AT as required.
17(b) We use Theorem 7. The characteristic polynomial of B is computed by first adding rows 2
and 3 to row 1. For convenience, write s = a + b + c, k = a2 + b2 + c2 (ab + ac + bc).



cB (x) =

xc
a
b

xa

xb

xs

xs

xs

xa

xb


= (x s) x2 (a b)2 (a c)(b c)
= (x s)(x2 k).

Hence, the eigenvalues of B are s,

Thus a2 + b2 + c2 ab + ac + bc.

xs
a
b

0
x + (a b)
bc

0
ac

x (a b)

k and k. These must be real by Theorem 7, so k 0.

20(b) To compute cA (x) = det(xI A), add x times column 2 to column 1, and expand along row
1:


x

1
0
0

0
0


0

x
1
0

0
0


0

0
x
1
0
0


..

..
..
..
..
..
cA (x) = .

.
.
.
.
.


0

0
0
0

1
0


0

0
0
0

x
1


r0 r1 r2 r3 rk2 x rk1




0
1
0
0

0
0




x2
x
1
0

0
0




0
0
x
1
0
0




..
..
..
..
..
..
=

.
.
.
.
.
.




0
0
0
0

1
0




0
0
0
0
x
1


r0 r1 x r1 r2 r3
rk2 x rk1
Now expand along row 1 to get

95

Section 5.6: Best Approximation and Least Squares

cA (x) =

x2

0
0
..
.

1
..
.

0
..
.

r0 r1 x

r2

r3

rk2

0
..
.

x rk1

This matrix has the same form as xI A, so repeat this procedure. It leads to the given
expression for det(xI A).

Exercises 5.6

1(b) Here A =

Best Approximation and Least Squares

, B =

1
0
8

This is invertible and the inverse is

, X =

96

(AT A)1 =

1
144

120

x
y
z

. Hence, AT A =

120

216

288

516

168

216

288

1
36

24
30
54

26

20

12

12

30

54

72

129

42

72

12
12
12

Here the (unique) best approximation is

Z = (AT A)1 AT B =

1
36

24
30

30

54

72

129

42

54

72

44
15
29

1
36

60
138
285

1
12

20
46
95

Of course this can be found more efficiently using gaussian elimination on the normal equations
for Z.

 4


 1 2





1 4
3
4
21
1 1 1 1
1 1 1 1
TY =
=
=

,
M
2(b) Here M T M =

10
42

21

133

. We solve the normal equation (M T M)A = M T Y by inverting M T M:

A = (M T M )1 M T Y =

1
91

133
21

21
4

Hence the best fitting line has equation y =



.

64
13

10
42




1
. =

91

6
13 x.

4
(d) Analogous to (b). The best fitting line is y = 10

17
10 x.

448
42




1
. =
13

64
6

96

Section 5.6: Best Approximation and Least Squares

3(b) Now M T M =

MT Y =

16

16

1
0
2
3

16

.
=

.
=

29

29

83

29

83

353

. .

16
70

We use (MM T )1 to solve the normal equations even though it is more efficient to solve them
by gaussian elimination.

3348
642
426
540
.127
6
1
1
642
571
187 16 =
102 = .024 .
A = (M T M)1 (M T Y ) =
4248
4248
426

187

91

Hence the best fitting quadratic has equation y = .127 .024x + .194x2 .

4(b) In the notation of Theorem 3: Y =

MT M =

14

36

34

36

98

90

34

90

85

, M =

1
5

10

, and (M T M )1 =

1
92

02

20

12

21

22

22

32

23

230

92

34

92

36

36
76

Thus, the (unique) solution to the normal equation is

Z = (M T M)1 M T Y =

The best fitting function is thus

1

5(b) Here Y =

1
5
9

MT M =

(1)2

02

22

32

14

14

98

10

10
2

17

46

18

1
2
46 [23x + 33x

, M =

1
46

115



sin 2
sin(0)

sin()
 3 

sin

46
18
38

Hence, the best fitting functions

2
14

2
1
20 [18 + 21x

103

1
2

1
0

Thus, the (unique) solution to the normal equations is

1
40

111

and (M T M)1 =

Z = (M T M)1 M T Y =

41

1
46

1
46

14

12

49

+ 28 sin

. Hence

24
7

35

31
2
203
2
19
2

 x 
2 ].

35

10

10

49

1
20

17

46

18

33
30

21
28

. Hence,

23

18

115

+ 30(2)x ].

24

.194

822

70

46

18
38

97

Section 5.6: Best Approximation and Least Squares

7. To fit s = a + bx where x = t2 , we have
T

M M=

M Y =

1
98

Hence A = (M T M)1 M T Y =

98




14

14

95
80
56



=
231
919

decimal places. Hence the best fitting equation is

14

14

98

231
919

1
98

9772
477

99.71
4.87


to two

y = 99.71 4.87x = 99.71 4.87t2 .

Hence the estimate for g comes from 12 g = 4.87, g = 9.74 (the true value of g is 9.81).
Now fit s = a + bt + ct2 . In this case

MT M =

Hence

MT Y =

A = (M T M)1 (M T Y ) =

1
4

76

95
80
56

84

84

98

20

24

=
20
24
6

231
423
919

14

14

36

14

36

98

231
423
919

1
4

404
6

18

101
32
92

so the best quadratic is y = 101 23 t 92 t2 . This gives 92 = 12 g so the estimate for g is

g = 9 in this case.
9 We want r0 , r1 , r2 , and r3 to satisfy
r0 + 50r1 + 18r2 + 10r3 = 28
r0 + 40r1 + 20r2 + 16r3 = 30
r0 + 35r1 + 14r2 + 10r3 = 21
r0 + 40r1 + 12r2 + 12r3 = 23
r0 + 30r1 + 16r2 + 14r3 = 23.
We settle for a best approximation. Here

A=

50

18

10

40

20

16

35

14

10

40

12

12

30

16

14

B=

28
30
21
23
23

98

Section 5.7: An Application to Correlation and Variance

AT A =

(AT A)1 =

1
50160

The best fitting function is

195

80

62

195

7825

3150

2390

80

3150

1320

1008

2390

1008

796

62

1
50160

So the best approximation

Z = (AT A)1 (AT B) =

1035720

16032

16032
10080

16032
10080

45300

10080

416

632

632

45300

1035720

2600

800

16032
416

632
800

45300

2180

10080
632
2600

2180

800

2180

45300
800

2180
3950

3950

125
4925
2042
1568

5.19
0.34

0.51
0.71

y = 5.19 + 0.34x1 + 0.51x2 + 0.71x3 .

10(b) f (x) = a0 here so the sum of squares is
s = (y1 a0 )2 + (y2 a0 )2 + + (yn a0 )2
n

=
(yi a0 )2
i=1

n

=
(a20 2a0 yi + yi2 )
i=1

  
 
= na20 2
yi a0 +
yi2 .

a quadratic in a0 . Completing the square gives







1  2
1 2
2
yi
yi
yi
.
s = n a0
n
n

This is minimal when a0 = n1
yi .

13(b) It suffices to show that the columns of M =

r0

1
..
.
1

+ r1

ex1
..
.
ex2

0
..
.
0

1
..
.

ex1
..
.

exn

are independent. If

, then r0 + r1 exi = 0 for each i. Thus, r1 (exi exj ) = 0 for all

i and j, so r1 = 0 because two xi are distinct. Then r0 = r1 ex1 = 0 too.

Exercises 5.7

An Application to Correlation and Variance

99

Supplementary Exercises - Chapter 5

2. Let X = [x1 x2 x10 ] = [12 16 13 14] denote the number of years of education. Then
1
1
x
= 10
xi = 15.3, and s2x = n1
(xi x
)2 = 9.12 (so sx = 3.02).
Let Y = [y1 y2 y10 ] = [31 48 35 35] denote the number of dollars (in thousands) of
1
1
yearly income. Then y = 10
ti = 40.3, and s2y = n1
(yi y)2 = 114.23 (so sy = 10.69).
The correlation is r =

X Y 10
xy
= 0.599.
9sx sy

4(b) We have zi = a + bxi for each i, so

Hence

 
 
1
1
1
z =
(a + bxi ) =
na + b
xi = a + b
xi = a + b
x.
n
n
n

1 
1 
1  2
(zi z)2 =
[(a + bxi ) (a + b
x)]2 =
b (xi x
)2 = b2 s2x .
n1
n1
n1

The result follows because b2 = |b| .

s2z =

Supplementary Exercises for Chapter 5

1(b) False. If r = 0 then rx is in U for any x.
1(d) True. If x is in U then x = (1)x is also in U by axiom S3 in Section 5.1.
1(f) True. If rx + sy = 0 then rx + sy + 0z = 0 so r = s = 0 because {x, y, z} is independent.
False. Take n = 2, x1 = [

1
1

] and x2 = [

is not independent.

1
1

]. Then both x1 and x2 are nonzero, but {x1 , x2 }

1(j) False. If a = b = c = 0 then ax + by + cz = 0 for any x, y and z.

1(l) True. If t1 x1 + t2 x2 + + tn xn = 0 implies that each ti = 0, then {x1 , x2 , .xn } is
independent, contrary to assumption.

1
0
0
1

0
0
0
0
1(n) False. [ ], [
], [ ], [ ] is not independent.
0
0
0
1

1(p) False. {x, x + y, y} is never independent because 1x + (1)(x + y) + 1y = 0 is a nontrivial

vanishing linear combination.
1(r) False. Every basis of R3 must contain exactly 3 vectors (by Theorem 5 5.2). Of course a
nonempty subset of a basis will be independent, but it will not span R3 if it contains fewer
than 3 vectors.

100
.

Supplementary Exercises - Chapter 5

101

Section 6.1: Examples and Basic Properties

Chapter 6: Vector Spaces

Exercises 6.1

Examples and Basic Properties

1(b) No: S5 fails 1(x, y, z) = (1x, 0, 1z) = (x, 0, z) = (x, y, z) for all (x, y, z) in V. Note that the
other nine axioms do hold.
(d) No: S4 and S5 fail: S5 fails because 1(x, y, z) = (2x, 2y, 2z) = (x, y, z); and S4 fails because
a[b(x, y, z)] = a(2bx, 2by, 2bz) = (4abx, 4aby, 4abz) = (2abx, 2aby, 2abz) = ab(x, y, z). Note
that the eight other axioms hold.
2(b) No: A1 fails for example (x3 + x + 1) + (x3 + x + 1) = 2x + 2 is not in the set.
(d) No: A1 and S1 both fail. For example x + x2 and 2x are not in the set. Hence none of the
other axioms make sense.






a b
x y
(f) Yes. First verify A1 and S1. Suppose A =
and B =
are in V, so a+c = b+d
c d
z w



a+x b+y
and x + z = y + w. Then A + B =
is in V because
c+z

d+w

(a + x) + (c + z) = (a + c) + (x + z) = (b + d) + (y + w) = (b + y) + (d + w).



ra rb
Also rA =
is in V for all r in R because ra + rc = r(a + c) = r(b + d) = rb + rd.
rc

rd

A2, A3,

S2,S3, S4, S5. These hold for matrices in general.
0 0
A4.
is in V and so serves as the zero of V .
0 0





a b
A5. Given A =
with a + c = b + d, then A =
c

a
c

is also in V because

a c = (a + c) = (b + d) = b d. So A is the negative of A in V .
(h) Yes. The vector space axioms are the basic laws of arithmetic.
(j) No. S4 and S5 fail. For S4, a(b(x, y)) = a(bx, by) = (abx, aby), and this need not equal
ab(x, y) = (abx, aby); as to S5, 1(x, y) = (x, y) = (x, y) if y = 0.
Note that the other axioms do hold here:
A1, A2, A3, A4 and A5 hold because they hold in R2 .
S1 is clear; S2 and S3 hold because they hold in R2 .
(l) No. S3 fails: Given f : R R and a, b in R, we have
[(a + b)f](x) = f ((a + b)x) = f (ax + bx)
(af + bf)(x) = (af)(x) + (bf )(x) = f(ax) + f(bx).
These need not be equal: for example, if f is the function defined by f (x) = x2 ;
Then f(ax + bx) = (ax + bx)2 need not equal (ax)2 + (bx)2 = f(ax) + f(bx).

102

Section 6.1: Examples and Basic Properties

Note that the other axioms hold. A1-A4 hold by Example 7 as we are using pointwise addition.
S2. a(f + g)(x) = (f + g)(ax)
definition of scalar multiplication in V
= f(ax) + g(ax)

definition of pointwise addition

= (af)(x) + (ag)(x) definition of scalar multiplication in V

= (af + ag)(x)
definition of pointwise addition
As this is true for all x, a(f + g) = af + ag.
S4. [a(bf)](x) = (bf)(ax) = f [b(ax)] = f[(ba)x] = [(ba)f](x) = [abf](x) for all x,
so a(bf ) = (ab)f.
S5. (1f)(x) = f(1x) = f(x) for all x, so 1f = f.
(n) No. S4, S5 fail: a (b X) = a (bX T ) = a(bX T )T = abX T T = abX, while (ab) X = abX T .
These need not be equal. Similarly: 1 X = 1X T = X T need not equal X.
Note that the other axioms do hold:
A1-A5. These hold for matrix addition generally.
S1. a X = aX T is in V.
S2. a (X + Y ) = a(X + Y )T = a(X T + Y T ) = aX T + aY T = a X + a Y.
S3 (a + b) X = (a + b)X T = aX T + bX T = a X + b X.
4. A1. (x, y) + (x1 , y1 ) = (x + x1 , y + y1 + 1) is in V for all (x, y) and (x1 , y1 ) in V.
A2. (x, y) + (x1 , y1 ) = (x + x1 , y + y1 + 1) = (x1 + x, y1 + y + 1) = (x1 , y1 ) + (x1 , y).
A3. (x, y) + ((x1 , y1 ) + (x2 , y2 )) = (x, y) + (x1 + x2 , y1 + y2 + 1)
= (x + (x1 + x2 ), y + (y1 + y2 + 1) + 1)
= (x + x1 + x2 , y + y1 + y2 + 2)
((x, y) + (x1 , y1 )) + (x2 , y2 ) = (x + x1 , y + y1 + 1) + (x2 , y2 )
= ((x + x1 ) + x2 , (y + y1 + 1) + y2 + 1)
= (x + x1 + x2 , y + y1 + y2 + 2).
These are equal for all (x, y), (x1 , y1 ) and (x2 , y2 ) in V .
A4. (x, y) + (0, 1) = (x + 0, y + (1) + 1) = (x, y) for all (x, y), so (0, 1) is the zero of V .
A5. (x, y) + (x, y 2) = (x + (x), y + (y 2) + 1) = (0, 1) is the zero of V (from A4)
so the negative of (x, y) is (x, y 2).
S1. a(x, y) = (ax, ay + a 1) is in V for all (x, y) in V and a in R.
S2. a[(x, y) + (x1 , y1 )] =

a(x + x1 , y + y1 + 1) = (a(x + x1 ), a(y + y1 + 1) + a 1)

= (ax + ax1 , ay + ay1 + 2a 1)

a(x, y) + a(x1 , y1 ) = (ax, ay + a 1) + (ax1 , ay1 a 1)

= ((ax + ax1 ), (ay + a 1) + (ay1 + a 1) + 1)
= (ax + ax1 , ay + ay1 + 2a 1).

These are equal.

S4. a[b(x, y)] = a(bx, by+b1) = (a(bx), a(by+b1)+a1) = (abx, aby+ab1) = (ab)(x, y).
S5. 1(x, y) = (1x, 1y + 1 1) = (x, y) for all (x, y) in V .
5(b) Subtract the first equation from the second to get x 3y = v u, whence x = 3y + v u.

103

Section 6.1: Examples and Basic Properties

Substitute in the first equation to get
3(3y + v u) 2y = u
7y = 4u 3v
y = 47 u 37 v.
Substitute this in the first equation to get x = 75 u 27 v.

It is worth noting that these equations can also be solved by gaussian elimination using u
and v as the constants.









a 0
0 b
c
c
0 0
6(b) au + bv + cw = 0 becomes
+
+
=
.
0

Equating corresponding entries gives equations for a and b.

a + c = 0,

b + c = 0,

b + c = 0,

a c = 0.

The only solution is a = b = c = 0.

(d) au + bv + cw = 0 means a sin x + b cos x + c1 = 0 for all choices of x. If x = 0, 2 , ,
we get,respectively, equations b + c = 0, a + c = 0, and b + c = 0. The only solution is
a = b = c = 0.
7(b) 4(3u v + w) 2[(3u 2v) 3(v w)] + 6(w u v)
= (12u 4v + 4w) 2[3u 2v 3v + 3w] + (6w 6u 6v)
= (12u 4v + 4w) (6u 10v + 6w) + (6w 6u 6v)
= 4w.
10. Suppose that a vector z has the property that z + v = v for all v in V. Since 0 + v = v also
holds for all v, we obtain z + v = 0 + v, so z = 0 by cancellation.
12(b) (a)v+av = (a+a)v = 0v = 0. Since also (av)+av = 0 we get (a)v+av = (av)+av.
Thus (a)v = (av) by cancellation.
Alternatively: (a)v = [(1)a]v = (1)(av) = av using part 4 of Theorem 3.
13(b) We proceed by induction on n (see Appendix A). The case n = 1 is clear. If the equation
holds for some n 1, we have
(a1 + a2 + + an + an+1 )v = [(a1 + a2 + + an ) + an+1 ]v
= (a1 + a2 + + an )v + an+1 v

by S3

= (a1 v + a2 v + + an v) + an+1 v

by induction

= a1 v + a2 v + + an v + an+1 v
Hence it holds for n + 1, and the induction is complete.
15(c) Since a = 0, a1 exists in R. Hence av = aw gives a1 av = a1 aw; that is 1v = 1w, that is
v = w.
Alternatively: av = aw gives av aw = 0, so a(v w) = 0. As a = 0, it follows that
v w = 0 by Theorem 3, that is v = w.

104

Section 6.2: Subspaces and Spanning Sets

Exercises 6.2

Subspaces and Spanning Sets

1(b) Yes. U is a subset of P3 because xg(x) has degree one more than the degree of g(x). Clearly
0 = x 0 is in U. Given u = xg(x) and v = xh(x) in U (where g(x) and b(x) are in P2 ) we
have
u + v = x(g(x) + h(x)) is in U because g(x) + h(x) is in P2
ku = x(kg(x)) is in U for all k in R because kg(x) is in P2

(d) Yes. As in (b), U is a subset of P3 . Clearly 0 = x0+(1x)0 is in U. If u = xg(x)+(1x)h(x)

and v = xg1 (x) + (1 x)h1 (x) are in U then
u + v = x[g(x) + g1 (x)] + (1 x) [h(x) + h1 (x)]
ku = x[kg(x)] + (1 x)[kh(x)]
both lie in U because g(x) + g1 (x) and h(x) + h1 (x) are in P2 .
(f) No. U is not closed under addition (for example u = 1 + x3 and v = x x3 are in U but
u + v = 1 + x is not in U). Also, the zero polynomial is not in U.
2(b) Yes. Clearly 0 =


a + a1
u + u1 =
c + c1

ku =

ka

kb

kc

kd

b + b1
d + d1

is in U. If u =

and u1 =

a1

b1

c1

d1

are in U then

is in U because (a + a1 ) + (b + b1 ) = (a + b) + (a1 + b1 )
= (c + d) + (c1 + d1)
= (c + c1 ) + (d + d1 ).

is in U because ka + kb = k(a + b) = k(c + d) = kc + kd.

(d) Yes. Here 0 is in U as 0B = 0. If A and A1 are in U then AB = 0 and A1 B = 0, so

(A + A1 )B = AB + A1 B = 0 + 0 = 0 and (kA)B = k(AB) = k0 = 0 for all k in R. This
shows that A + A1 and kA are also in U.






1 0
0 0
and A1 =
are both in U,
(f) No. U is not closed under addition. In fact, A =
0 0
0 1



1 0
but A + A1 =
is not in U.
0

3(b) No. U is not closed under addition. For example if f and g are defined by f(x) = x + 1 and
g(x) = x2 + 1, then f and g are in U but f + g is not in U because (f + g)(0) = f(0) + g(0) =
1 + 1 = 2.
(d) No. U is not closed under scalar multiplication. For example, if f is defined by f (x) = x,
then f is in U but (1)f is not in U (for example [(1)f ]( 21 ) = 12 so is not in U).

105

Section 6.2: Subspaces and Spanning Sets

(f) Yes. 0 is in U because 0(x + y) = 0 = 0 + 0 = 0(x) + 0(y) for all x and y in [0, 1]. If f and g
are in U then, for all k in R:
(f + g)(x + y) = f(x + y) + g(x + y)
= (f (x) + f (y)) + (g(x) + g(y))
= (f (x) + g(x)) + (f (y) + g(y))
= (f + g)(x) + (f + g)(y)

(kf)(x + y) = k[f (x + y)] = k[f (x) + f (y)] = k[f(x)] + k[f(y)]

= (kf)(x) + (kf)(y)
Hence f + g and kf are in U.

5(b) Suppose X =

x1
..
.

xn

= 0, say xk = 0. Given Y =

y1
..
.
yn

let A be the m n matrix with kth

column x1
k Y and the other columns zero. Then Y = AX by matrix multiplication, so Y is
in U. Since Y was an arbitrary column in Rn , this shows that U = Rm .

6(b) We want r, s and t such that 2x2 3x + 1 = r(x + 1) + s(x2 + x) + t(x2 + 2). Equating
coefficients of x2 , x and 1 gives s + t = 2, r + s = 3, r + 2t = 1. The unique solution is
r = 3, s = 0, t = 2.
(d) As in (b), x = 23 (x + 1) + 13 (x2 + x) 31 (x2 + 2).
7(b) If v = su + tw then x = s(x2 + 1) + t(x + 2). Equating coefficients gives 0 = s, 1 = t and
0 = s + 2t. Since there is no solution to these equations, v does not lie in span{u, w} .









1 4
1 1
2 1
(d) If v = su + tw, then
=s
+t
. Equating corresponding entries
5

gives s + 2t = 1, s + t = 4, 2s + t = 5 and s = 3. These equations have the unique solution

t = 1, s = 3, so v is in span{u, w} ; in fact v = 3u w.
)
*
8(b) Yes. The trigonometry identity 1 = sin2 x + cos2 x for all x means that 1 is in span sin2 x, cos2 x .
(d) Suppose 1 + x2 = s sin2 x + t cos2 x for some s and t. This must hold for all x. Taking x = 0
gives 1 = t; taking x = gives 1 + 2 =)t. Thus 2 +*2 = 0, a contradiction. So no such s
and t exist, that is 1 + x2 is not in span sin2 x, cos2 x .
)
*
9(b) Write U = span 1 + 2x2 , 3x, 1 + x , then successively
x = 31 (3x) is in U
1 = (1 + x) x is in U

x2 = 21 [(1 + 2x2 ) 1] is in U.

)
*
Since P2 = span 1, x, x2 , this shows that P2 U. Clearly U P2 , so U = P2 .

106

Section 6.3: Linear Independence and Dimension

11(b) The vectors uv = 1u+(1)v, u+v, and w are all in span {u, v, w} so span{u v, u + w, w}
span{u, v, w} by Theorem 2. The other inclusion also follows from Theorem 2 because
u = (u + w) w
v = (u v) + (u + w) w
w=w
show that u, v and w are all in span{u v, u + v, w} .
14. No. For example (1, 1, 0) is not even in span{(1, 2, 0), (1, 1, 1)} . Indeed (1, 1, 0) = s(1, 2, 0) +
t(1, 1, 1) requires that s + t = 1, 2s + t = 1, t = 0, and this has no solution.
18. Write W = span{u, v2 , . . . , vn } . Since u is in V we have W V. But the fact that a1 = 0
means
v1 = a11 u aa12 v2 aan1 vn
so v1 is in W. Since v2 , . . . , vn are all in W, this shows that V = span{v1 , v2 , . . . , vn } W.
Hence V = W.

21(b) If u and u + v are in U then v = (u + v) u = (u + v) + (1)u is in U because U is closed

under addition and scalar multiplication.
22. If U is a subspace then, u1 + au2 is in U for any ui in U and a in R by the subspace test.
Conversely, assume that this condition holds for U. Then, in the subspace test, conditions (2)
and (3) hold for U (because 1v = v for all v in V ), so it remains to show that 0 is in U. This
is where we use the assumption that U is nonempty because, if u is any vector in U then
u + (1)u is in U by assumption, that is 0 U.

Exercises 6.3

Linear Independence and Dimension

1(b) Independent. If rx2 + s(x + 1) + t(1 x x2 ) = 0 then, equating coefficients of x2 , x and 1,

we get r t = 0, s t = 0, s + t = 0. The only solution is r = s = t = 0.














1 1
0 1
1 0
1 1
0 0
(d) Independent. If r
+s
+t
+u
=
, then
1

r + t + u = 0, r + s + u = 0, r + s + t = 0, s + t + u = 0. The only solution is r = s = t = u = 0.

2(b) Dependent. 3(x2 x + 3) 2(2x2 + x + 5) + (x2 + 5x + 1) = 0












1 1
0
1
1
0
1
1
+
+
+0
=
(d) Dependent. 2
0

(f) Dependent.

5
x2 +x6

1
x2 5x+6

6
x2 9

= 0.

3(b) Dependent. 1 sin2 x cos2 x = 0 for all x.

4(b) If r(2, x, 1) + s(1, 0, 1) + t(0, 1, 3) = (0, 0, 0) then, equating components:
2r + s
xr
r

+ s + 3t = 0.

107

Section 6.3: Linear Independence and Dimension

Gaussian
elimination

gives
2 1 0 0
1 1

x 0 1 0 2 1

1 1 3 0
x 0

3
0
1

0
0
0

1 3x

0
0
0

1 + 3x

0
0
0

This has only the trivial solution r = s = t = 0 if and only if x = 1

3 . Alternatively, the
coefficient matrix has determinant


2 1 0
2
1 0
x
1
det x 0 1 = det x 0 1 = det
= (1 + 3x).
1

This is nonzero if and only if x = 13 .

5(b) Independence: If r(1, 1, 1) + s(1, 1, 1) + t(1, 1, 1) = (0, 0, 0) then r + s + t = 0,

r s + t = 0, r + s t = 0. The only solution is r = s = t = 0.

Spanning: Write U = span{(1, 1, 1), (1, 1, 1), (1, 1, 1)} . Then (1, 0, 0) = 12 [(1, 1, 1) +
(1, 1, 1)] is in U; similarly (0, 1, 0) and (0, 0, 1) are in U. As R3 = span{(1, 0, 0), (0, 1, 0), (0, 0, 1)} ,
we have R3 U.Clearly U R3 , so we have R3 = U.

(d) Independence: If r(1 + x) + s(x + x2 ) + t(x2 + x3 ) + ux3 = 0 then

r + (r + s)x + (s + t)x2 + (t + u)x3 = 0,
so r = 0, r + s = 0, s + t = 0, t + u = 0. The only solution is r = s = t = u = 0.
)
*
Spanning: Write U = span 1 + x, x + x2 , x2 + x3 , x3 . Then x3 is in U; whence
x2 = (x2 + x3 ) x3 is
x = (x + x2 ) x2 is in U; whence 1 = (1 + x) x is in
) in U;2whence
*
U. Hence P3 = span 1, x, x , x3 is contained in U. As U P3 , we have U = P3 .
)
*
)
*
6(b) Write U = a + b(x + x2 ) | a, b in R = span B where B = 1, x + x2 . But B is independent
because s + t(x + x2 ) = 0 implies s = t = 0. Hence B is a basis of U, so dim U = 2.
(d) Write U = {p(x) | p(x) = p(x)} . As U P2 , write p(x) = a + bx + cx2 be any member
of U. )The condition p(x)* = p(x)) becomes
a )+ bx +
cx2 = a bx + cx2 , so b = 0. Thus
*
*
U = a + bx2 | a, b in R = span 1, x2 . As 1, x2 is independent (s + tx2 = 0 implies
s = 0 = t), it is a basis of U, so dim U = 2.
'




 (



1
1
1
1
x y
7(b) Write U = A | A
=
A . If A =
, A is in U if and only if





, that is

xy

zw

x
z

This holds if and only if x = y + w and z = y, that is







y+w y
1
1
=y
+w
A=
y

'



,

(

x+z

y+w

Hence U = span B where B =

. But B is independent here because
1 0
0 1








1
1
1 0
0 0
s
+t
=
means s + t = 0, s = 0, s = 0, t = 0, so s = t = 0. Thus
1

B is a basis of U, so dim U = 2.

108

Section 6.3: Linear Independence and Dimension

'




 (



1
1
0
1
x y
(d) Write U = A | A
=
A . If A =
then A is in U if and only
1 0
1 1
z w














x y
1
1
0
1
x y
xy x
z
w
if
=
; that is
=
.
z

zw

This holds if and only if z = x y and x = w; that is







x
y
1 0
A=
=x
+y
xy

zx

wy

'



(


1 0
0
1
Thus U = span B where B =
,
. But B is independent because s
1 1
1 0





0
1
0 0
t
=
implies s = t = 0. Hence B is a basis of U, so dim U = 2.
1

8(b) If X =


+

x+z

y+w

the condition AX = X is
=
and this holds if and
0
0
z w









x y
1 0
0 1
only if z = w = 0. Hence X =
=x
+y
. So U = span B where
0 0
0 0
0 0
'



(
1 0
0 1
B=
,
. As B is independent, it is a basis of U, so dim U = 2.
0

10(b) If the common column sum is m, V has the form

a
q
r

| a, b, p, q, r, s, m in R = span B where
b
p
s
V =

mab mpq mrs

0
1
0
0
0 0
1
0 0
0 0 0
B = 0 0 0 , 0 0 0 , 1 0 0 , 0 0 0 ,

0 1 0
1 0 0
1 0 0
1 1 1

0 0
1
0 0
0
0
0
0

0 1 0 , 0 0 0 , 0 0 1 .

The set B is independent (a linear combination using coefficients a, b, p, q, r, s, and m yields

the matrix in V, and this is 0 if and only if a = b = p = q = r = s = m = 0.) Hence B is a
basis of B, so dim V = 7.

11(b) A general polynomial in P3 has the form p(x) = a + bx + cx2 + dx3 , so

)
*
V = (x2 x)(a + bx + cx2 + dx3 ) | a, b, c, d in R
)
*
= a(x2 x) + bx(x2 x) + cx2 (x2 x) + dx3 (x2 x) | a, b, c, d in R
= span B
)
*
where B = (x2 x), x(x2 x), x2 (x2 x), x3 (x2 x) . We claim that B is independent.
For if a(x2 x)+bx(x2 x)+cx2 (x2 x)+dx3 (x2 x) = 0 then (a+bx+cx2 +dx3 )(x2 x) = 0,
whence a + bx + cx2 + dx3 = 0 by the hint in (a). Thus a = b = c = d = 0. [This also follows
by comparing coefficients.] Thus B is a basis of V, so dim V = 4.
12(b) No. If P3 = span{f1 (x), f2 (x), f3 (x), f4 (x)} where fi (0) = 0 for each i, then each polynomial
p(x) in P3 is a linear combination
p(x) = a1 f1 (x) + a2 f2 (x) + a3 f3 (x) + a4 f4 (x)

109

Section 6.3: Linear Independence and Dimension

when the ai are in R. But then
p(0) = a1 f1 (0) + a2 f2 (0) + a3 f3 (0) + a4 f4 (0) = 0

for every p(x) in P3 . This is not the case, so no such basis of P3 can exist. [Indeed, no such
spanning set of P3 can exist.]
'







(
1 0
1 1
1 0
0 1
(d) No. B =
,
,
,
is a basis of invertible matrices.
0 1
0 1
1 1
1 1














1 0
1 1
1 0
0 1
0 0
Independent: r
+s
+t
+u
=
gives
0

r + s + t = 0,
s + u = 0,

t + u =0, r
+ s + t+ u = 0. The only solution is r = s = t = u = 0.
0 1
1 1
1 0
Spanning:
=

is in span B
0 0
0 1





0 1 
0 0
1 0
1 0
=

is in span B

1 0 
1 1 
0 1 


0 0
0 1
0 1
0 0
=

is in span B
0 1
1 1
0 0
1 0







1 0
1 0
0 0
=

is in span B
0 0
0 1
0 1
'







(
0 1
0 0
0 0
1 0
Hence M22 = span
,
,
,
span B. Clearly span B
0

M22 .

(f) Yes. Indeed, 0u + 0v + 0w = 0 for any u, v, w, independent or not!

(h) Yes. If su + t(u + v) = 0 then (s + t)u + tv = 0, so s + t = 0 and t = 0 (because {u, v} is
independent). Thus s = t = 0.
(j) Yes. If su + tv = 0 then su + tv + 0w = 0, so s = t = 0 (because {u, v, w} is independent).
This shows that {u, v} is independent.
(l) Yes. Since {u, v, w} is independent, the vector u + v + w is not zero. Hence {u + v + w} is
independent (see Example 5 5.2).
(n) Yes. If Iis a set of independent vectors, then |I| nby the fundamental theorem because
V contains a spanning set of nvectors (any basis).
15. If a linear combination of the vectors in the subset vanishes, it is a linear combination of the
vectors in the larger set (take the coefficients outside the subset to be zero). Since it still
vanishes, all the coefficients are zero because the larger set is independent.
19. We have su + tv = s(au + bv) + t(cu + dv) = (sa + tc)u + (sb + td)v. Since {u, v} is
independent, we have
su + tv = 0 if and only if sa + tc = 0 and sb + td = 0





a c
s
0
if and only if
=
.
b

Hence

{u , v }

is independent if and only if



s
t

By Theorem 5 2.4, this is equivalent to A being invertible.

0
0

implies

s
t

0
0

110

Section 6.4: Finite Dimensional Spaces

23(b) Independent: If r(u+v)+s(v+w)+t(w+u) = 0 then (r+t)u+(r+s)v+(s+t)w+0z = 0.

Thus r+t = 0, r+s = 0, s+t = 0 (because {u, v, w, z} is independent). Hence r = s = t = 0.
(d) Dependent: (u + v) (v + w) + (w + z) (z + u) = 0 is a nontrivial linear combination
that vanishes.
26. If rz + sz 2 = 0, r, s in R, then z(r + sz) = 0. If z is not real then z = 0 so r + sz = 0. Thus
s = 0 (otherwise z = r
whence r = 0. Conversely, if z is real then rz + sz 2 = 0
s) is real),
*
when r = z, s = 1, so z, z 2 is not independent.

29(b) If U is not invertible, let Ux = 0 where x= 0 in Rn (Theorem 5, 2.3). We claim that no set
{A1 U, A2 U, . . .} can span Mmn (let alone be a basis). For if it did, we could write any matrix
B in Mmn as a linear combination
B = a1 A1 U + a2 A2 U +
Then Bx = a1 AUx + a2 A2 Ux + = 0 + 0 + = 0, a contradiction. In fact, if entry k of
xis nonzero, then Bx = 0 where all entries of B are zero except column k, which consists of
1s.
33(b) Suppose U W = 0. If su + tw = 0 with u and w nonzero in U and W, then su = tw
is in U W = {0}. Hence su = 0 = tw. So s = 0 = t (as u = 0 and w = 0). Thus {u, v}
is independent. Conversely, assume that the condition holds. If v = 0 lies in U W, then
{v, v} is independent by the hypothesis, a contradiction because 1v + 1(v) = 0.
36(b) If p(x) = a0 + a1 x + + an xn is in On , then p(x) = p(x), so
a0 a1 x + a2 x2 + a3 x3 + a4 x4 = a0 a1 x a2 x2 a3 x3 a4 x4 .
)
*
3 , x5 , . . .
Hence a0 = a2 = a4 = = 0 and p(x) = a1 x+a3 x3 +a5 x5 + ). Thus On = span
x,
x
*
is spanned) by the odd powers
of x in Pn . The set B = x, x3 , x5 , . . . is independent
*
(because 1, x, x2 , x*3 , x4 , . . . is independent) so it is a basis of On . If)n is even, B =
)
*
x, x3 , x5 , . . . , xn1 has n2 members, so dim On = n2 . If n is odd, B = x, x3 , x5 , . . . , xn
n+1
has n+1
2 members, so dim On = 2 .

Exercises 6.4

Finite Dimensional Spaces

1(b) B = {(1, 0, 0), (0, 1, 0), (0, 1, 1)} is independent as r(1, 0, 0) + s(0, 1, 0) + t(0, 1, 1) = (0, 0, 0)
implies r = 0, s + t = 0, t = 0, whence r = s = t = 0. Hence B is a basis by Theorem 3
because dim R3 = 3.
)
*
(d) B = 1, x, x2 x + 1 is independent because r1 + sx + t(x2 x 1) = 0 implies r t = 0,
s t = 0, and t = 0; whence r = s = t = 0. Hence B is a basis by Theorem 3 because
dim P2 = 3.
2(b) As dim P2 = 3, any independent set of
vectors is a basis by
) three
* Theorem 3. But we have
2
2
2
2
(x +3)+2(x+2)+(x 2x1) = 0, x )+ 3, x + 2, x 2x 1 , so is dependent.
However
*
any other subset of three vectors from x2 + 3, x + 2, x2 2x 1, x2 + x is independent
(verify).

111

Section 6.4: Finite Dimensional Spaces

3(b) B = {(0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 1, 1), (1, 1, 1, 1)} spans R4 because
(1, 0, 0, 0) = (1, 1, 1, 1) (0, 1, 0, 0) (0, 0, 1, 1)

is in span B

(0, 0, 0, 1) = (0, 0, 1, 1) (0, 0, 1, 0)

is in span B

and, of course, (0, 1, 0, 0) and (0, 0, 1, 0) are in span B. Hence B is a basis of R4 by Theorem
3 because dim R4 = 4.
)
*
(d) B = 1, x2 + x, x2 + 1, x3 spans P3 because x2 = (x2 + 1) 1 and x = (x2 + x) x2 are in
span B (together with 1 and x3 ). So B is a basis of P3 by Theorem 3 because dim P3 = 4.
4(b) Let z = a + bi; a, b in R. Then b = 0 as z is not real and a = 0 as z is not pure imaginary.
Since dim C = 2, it suffices (by Theorem 3) to show that {z, z} is independent. If rz + s
z=0
then 0 = r(a + bi) + s(a bi) = (r + s)a + (r s)bi. Hence (r + s)a = 0 = (r s)b so (because
a = 0 = b) r + s = 0 = r s. Thus r = s = 0.
5(b) The four polynomials in S have distinct degrees. Use Example 4 6.3.
6(b) {4, 4x, 4x2 , 4x3 } is such a basis. There is no basis of P3 consisting of polynomials have the
property that their coefficients sum to zero. For if it did then every polynomial in P3 would
have this property (since sums and scalar multiples of such polynomials have the same property).
7(b) Not a basis because (2u + v + 3w) (3u + v w) + (u 4w) = 0.
(d) Not a basis because 2u (u + w) (u w) + 0(v + w) = 0.
8(b) Yes, four vectors can span R3 say any basis together with any other vector.
No, four vectors in R3 cannot be independent by the fundamental theorem (Theorem 2 6.3)
because R3 is spanned by 3 vectors (dim R3 = 3).
10. We have det A = 0 if and only if A is not invertible. This holds if and only if the rows of
A are dependent by Theorem 3 5.2. This in turn holds if and only if some row is a linear
combination of the rest by the dependent lemma (Lemma 3).
11(b) No. Take X = {(0, 1), (1, 0)} and D = {(0, 1), (1, 0), (1, 1)}. Then D is dependent, but its
subset X is independent.
(d) Yes. This is follows from Exercise 15 6.3 (solution above).
15. Let {u1 , ..., um }, m k, be a basis of U so dim U = m. If v U then W = U by Theorem
2 6.2, so certainly dim W = dim U. On the other hand, if v
/ U then {u1 , ..., um , v} is
independent by the independent lemma (Lemma 1). Since W = span{u1 , ..., um , v}, again
by Theorem 2 6.2, it is a basis of W and so dim W = 1 + dim U.
18(b) The two-dimensional subspaces of R3 are the planes through the origin, and the one-dimensional
subspaces are the lines through the origin. Hence part (a) asserts that if U and W are distinct
planes through the origin, then U W is a line through the origin.
23(b) Let vn denote the sequence with 1 in the nth coordinate and zeros elsewhere. Thus v0 =
(1, 0, 0, . . .), v1 = (0, 1, 0, . . .) etc. Then a0 v0 + a1 v1 + + an vn = (a0 , a1 , . . . , an , 0, 0, . . .)
so a0 v0 + a1 v1 + + an vn = 0 implies a0 = a1 = = an = 0. Thus {v0 , v1 , . . . , vn }
is an independent set of n + 1 vectors. Since n is arbitrary, dim V cannot be finite by the
fundamental theorem.

112

Section 6.5: An Application to Polynomials

25(b) Observe that Ru = {su | s in R}. Hence Ru + Rv = {su + tv | s in R, t in R} is the set of

all linear combinations of u and v. But this is the definition of span{u, v} .

Exercises 6.5

An Application to Polynomials

2(b) f (0) (x) = f (x) = x3 + x + 1, so f (1) (x) = 3x2 + 1, f (2) (x) = 6x, f (3) (x) = 6. Hence, Taylors
theorem gives
f (3) (1)
f (2) (1)
(x 1)2 +
(x 1)3
2!
3!
= 3 + 4(x 1) + 3(x 1)2 + (x 1)3 .

f(x) = f (0) (1) + f (1) (1)(x 1) +

(d) f (0) (x) = f(x) = x3 3x2 + 3x, f (1) (x) = 3x2 6x + 3, f (2) (x) = 6x 6, f (3) (x) = 6. Hence,
Taylors theorem gives
f(x) = f (0) (1) + f (1) (1)(x 1) +
= 1 + 0(x 1) +

f (3) (1)
f (2) (1)
(x 1)2 +
(x 1)3
2!
3!

0
(x 1)2 + 1(x 1)3
2!

= 1 + (x 1)3 .
6(b) The three polynomials are x2 3x + 2 = (x 1)(x 2), x2 4x + 3 = (x 1)(x 3) and
x2 5x + 6 = (x 2)(x 3), so use a0 = 3, a1 = 2, a2 = 1, in Theorem 2.
7(b) The Lagrange polynomials for a0 = 1, a1 = 2, a2 = 3, are
(x 2)(x 3)
= 21 (x 2)(x 3)
(1 2)(1 3)
(x 1)(x 3)
1 (x) =
= (x 1)(x 3)
(2 1)(2 3)
(x 1)(x 2)
2 (x) =
= 21 (x 1)(x 2).
(3 1)(3 2)

0 (x) =

Given f (x) = x2 + x + 1:
f(x) = f(1)0 (x) + f(2)1 (x) + f(3)2 (x)
= 32 (x 2)(x 3) 7(x 1)(x 3) +

13
2 (x

1)(x 2).

10(b) If r(x a)2 + s(x a)(x b) + t(x b)2 = 0, then taking x = a gives t(a b)2 = 0,
so t = 0 because a = b; and taking x = b gives r(b a)2 = 0, so r = 0. Thus, we are
left with
) s(x a)(x b) = 0. If x is any
* number except, a, b, this implies s = 0. Thus
B = (x a)2 , (x a)(x b), (x b)2 is independent in P2 ; since dim P2 = 3, B is a
basis.

113

Section 6.6: An Application to Differential Equations

11(b) Have Un = {f(x) in Pn | f(a) = 0 = f(b)} . Let {p1 (x), . . . , pn1 (x)} be a basis of Pn2 ; it
suffices to show that
B = {(x a)(x b)p1 (x), . . . , (x a)(x b)pn1 (x)}
is a basis of Un . Clearly B Un .

Independent: Let s1 (x a)(x b)p1 (x) + + sn1 (x a)(x b)pn1 (x) = 0. Then
(xa)(xb)[s1 p1 (x)+ +sn1 pn1 (x)] = 0, so (by the hint) s1 p1 (x)+ +sn1 pn1 (x) = 0.
Thus s1 = s2 = = sn1 = 0.

Spanning: Given f(x) in Pn with f(a) = 0, we have f (x) = (x a)g(x) for some polynomial
g(x) in Pn1 by the factor theorem. But 0 = f(b) = (b a)g(b) so (as b = a) g(b) = 0. Then
g(x) = (x b)h(x) with h(x) = r1 p1 (x) + + rn1 pn1 (x), ri in R, whence
f(x) = (x a)g(x)
= (x a)(x b)g(x)
= (x a)(x b)[r1 p1 (x) + + rn1 pn1 (x)]
= r1 (x a)(x b)p1 (x) + + rn1 (x a)(x b)pn1 (x).

Exercises 6.6

An Application to Differential Equations

1(b) By Theorem 1, f(x) = cex for some constant c. We have 1 = f(1) = ce1 , so c = e. Thus
f (x) = e1x .
(d) The characateristic polynomial is x2 + x 6 = (x 2)(x + 3). Hence f(x) = ce2x + de3x
for some c, d. We have 0 = f(0) = c + d and 1 = f(1) = ce2 + de3 . Hence, d = c and
1
e2x e3x
c = e2 e
3 so f(x) = e2 e3 .
(f) The characteristic polynomial is x2 4x+4 = (x2)2 . Hence, f(x) = ce2x +dxe2x = (c+dx)e2x
for some c, d. We have 2 = f (0) = c and 0 = f (1) = (c d)e2 . Thus c = d = 2 and
f (x) = 2(1 + x)e2x .
(h) The characteristic polynomial is x2 a2 = (x a)(x + a), so (as a = a) f (x) = ceax + deax
for some c, d. We have 1 = f(0) = c + d and 0 = f (1) = cea + dea . Thus d = 1 c and
c = 1e1 2a whence
f(x) = cax + (1 c)eax =

eax ea(2x)
.
1 e2a

(j) The characteristic polynomial is x2 + 4x + 5. The roots are = 2 i, so

f(x) = e2x (c sin x + d cos x) for some real c and d.
 
We have 0 = f(0) = d and 1 = f 2 = e (c). Hence f(x) = e2x sin x.

4(b) If f(x) = g(x) + 2 then f  + f = 2 becomes g  + g = 0, whence g(x) = cex for some c. Thus
f (x) = cex + 2 for some constant c.

114

Supplementary Exercises Chapter 6

3

5(b) If f(x) = x3 then f  (x) = x2 and f  (x) = 2x, so

f  (x) + f  (x) 6f (x) = 2x x2 + 2x3 .
3

is a particular solution.
Hence, f (x) = x
3
3
g(x) = h(x) f(x) = h(x) + x3 . Then

Now, if h = h(x) is any solution, write

g  + g 6g = (h + h 6h) (f  + f  6f ) = 0.
So, to find g, the characteristic polynomial is x2 + x 6 = (x 2)(x + 3). Hence we have
g(x) = ce3x + de2x , where c and d are constants, so
h(x) = ce3x + de2x

x3
3 .

6(b) The general solution is m(t) = 10( 45 )1/3 . Hence 10( 45 )t/3 = 5 so t =

3 ln(1/2)
ln(4/5)

= 9.32 hours.

7(b) If m = m(t) is the mass at time t, then the rate m (t) of decay is proportional to m(t), that
is m (t) = km(t) for some k. Thus, m km = 0 so m = cekt for some constant c. Since
m(0) = 10, we obtain c = 10, whence m(t) = 10ekt . Also, 8 = m(3) = 10e3k so e3k = 45 ,
 t/3
 1/3
, m(t) = 10(ek )t = 10 45
.
ek = 45
9. In Example 4, we found that the period of oscillation is
 2
= 0.044.
k = 15

Supplementary Exercises

.
k

Hence

= 30 so we obtain

Chapter 6

2(b) Suppose {Ax1 , . . . , Axn } is a basis of Rn . To show that A is invertible, we show that Y A = 0
implies Y = 0. (This shows AT is invertible by Theorem 5 2.3, so A is invertible). So assume
that Y A = 0. Let c1 , . . . , cm denote the columns of Im , so Im = [C1 , C2 , . . . , Cm ] . Then
Y = Y Im = Y [c1 c2 . . . cm ] = [Y c1 Y c2 . . . Y cm ] , so it suffices to show that
Y cj = 0 for each j. But cj is in Rn so our hypothesis shows that cj = r1 Av1 + + rn Avn
for some rj in R. Hence,
cj = A(r1 v1 + + rn vn )
so Y cj = Y A(r1 v1 + + rn vn ) = 0, as required.
4. Assume that A is m n. If x is in null A, then Ax = 0 so (AT A)x = AT 0 = 0. Thus x is in
null AT A, so null A null AT A. Conversely, let x be in nullAT A; that is AT Ax = 0. Write

Ax = y =

y1
..
.

ym

2 = yT y = (Ax)T (Ax) = xT AT Ax = xT 0 = 0. Since the y are real

Then y12 + y22 + + ym
i
numbers, this implies that y1 = y2 = = ym = 0; that is y = 0, that is Ax = 0, that is x is
in null A.

115

Section 7.1: Examples and Elementary Properties

Chapter 7: Linear Transformations

Exercises 7.1

Examples and Elementary Properties

1(b) T (X) = XA where A =

X = (x, y, z) is thought of as a row matrix. Hence, ma-

trix algebra gives T (X + Y ) = A(X + Y ) = AX + AY

T (r) = A(rX) = rA(X) = rT (X).

= T (X) + T (Y ) and

(d) T (A+B) = P (A+B)Q = P AQ+P BQ = T (A)+T (B); T (rA) = P (rA)Q = rP AQ = rT (A).

(f) Here T [p(x)] = p(0) for all polynomials p(x) in Pn . Thus
T [(p + q)(x)] = T [p(x) + q(x)] = p(0) + q(0) = T [p(x)] + T [q(x)]
T [rp(x)] = rp(0) = r[T p(x)].

(h) Here Z is fixed in Rn and T (X) = X Z for all X in Rn . We use Theorem 1, Section 5.3:
T (X + Y ) = X + Y ) Z = X Z + Y Z = T (X) + T (Y )
T (rX) = (rX) Z = r(X Z) = rT (X).

(j) If v = (r1 rn ) and w = (s1 sn ) then, v + w = (r1 + s1 rn + sn ). Hence:

T (v + w) = (r1 + s1 )e1 + + (rn + sn )en
= (r1 e1 + + rn en ) + (s1 e1 + + sn en ) = T (v) + T (w).
Similarly, for a in R, we have av = (ar1 arn ) so
T (av) = (ar1 )e1 + + (arn )en = a(r1 e1 + + rn en ) = aT (v).

2(b) Let A =

...

...

0
..
.

0
..
.

0
..
.

...
..
.

0
..
.

...

, B =

...

...

0
..
.

0
..
.

...
..
.

0
..
.

...

0
..
.
0

, then A + B =

...

...

0
..
.

0
..
.

0
..
.

...
..
.

0
..
.

...

Thus, T (A) = rank A = 2, T (B) = rank B = 2 and T (A + B) = rank(A + B) = 1. Thus

T (A + B) = T (A) + T (B).

(d) Here T (v) = v+u, T (w) = w+u, and T (v+w) = v+w+u. Thus if T (v+w) = T (v)+T (w)
then v + w + u = (v + u) + (w + u), so u = 2u, u = 0. This is contrary to assumption.
Alternatively, T (0) = 0 + u = 0, so T cannot be linear by Theorem 1.
3(b) Because T is linear, T (3v1 + 2v2 ) = 3T (v1 ) + 2T (v2 ) = 3(2) + 2(3) = 0.

116

Section 7.1: Examples and Elementary Properties

(d) Since we know the action of T on

combination of these vectors.

1
7

1
1

and

=r

1

1
1

, it suffices to express

+s

1
1

1
7

as a linear

Comparing components gives 1 = r + s and 7 = r + s. The solution is r = 4, s = 3, so


















1
1
1
1
1
0
1
3
T
=T 4
3
= 4T
3T
=4
3
=
.
7

(f) We know T (1), T (x + 2) and T (x2 + x), so we express 2 x + 3x2 as a linear combination of
these vectors:
2 x + 3x2 = r 1 + s(x + 2) + t(x2 + x).
Equating coefficients gives 2 = r + 2s, 1 = s + t and 3 = t. The solution is r = 10, s = 4
and t = 3, so
T (2 x + 3x2 ) = T [r 1 + s(x + 2) + t(x2 + x)]

= rT (1) + sT (x + 2) + tT (x2 + x)
= 5r + s + 0
= 46.

In fact, we can find the action of T on any vector a + bx + cx2 in the same way. Observe that
a + bx + cx2 = (a 2b + 2c) 1 + (b c)(x + 2) + c(x2 + x)
for any a, b and c, so
T (a + bx + cx2 ) = (a 2b + 2c)T (1) + (b c)T (x + 2) + cT (x2 + x)
= (a 2b + 2c) 5 + (b c) 1 + c 0
= 5a 9b + 9c.

This retrieves the above result when a = 2, b = 1 and c = 3.

4(b) Since B = {(2, 1), (1, 1)} is a basis of R2 , any vector (x, y) in R2 is a linear combination
(x, y) = r(2, 1) + s(1, 1). Indeed, equating components gives x = 2r + s and y = r + s so
r = 13 (x y), s = 13 (x + 2y). Hence,
T (x, y) = T [r(2, 1) + s(1, 1)]
= rT (2, 1) + sT (1, 1)
= 13 (x y)(1, 1, 1) + 13 (x + 2y)(0, 1, 0)


= 13 (x y), y, 13 (x y)
= 13 (x y, 3y, x y).

117

Section 7.1: Examples and Elementary Properties

In particular, T (v) = T (1, 2) = 31 (3, 6, 3) = (1, 2, 1).

x+2y
This works in general. Observe that (x, y) = xy
3 (2, 1) + 3 (1, 1) for any x and y, so
since T is linear,
x+2y
T (x, y) = xy
3 T (2, 1) + 3 T (1, 1)

for any choice of T (2, 1) and T (1, 1).

'







1 0
0 1
1 0
(d) Since B =
,
,
,
0

is a linear combination



a

=r

+s

(


is a basis of M22 , every vector

+t

+u

Indeed, equating components and solving for r, s, t and u gives r = a c + b, s = b, t = c b,

u = d. Thus,















a b
1 0
0 1
1 0
0 0
T
= rT
+ sT
+ tT
+ uT
c

= (a c + b) 3 + b (1) + (c b) 0 + d 0
= 3a + 2b 3c.

5(b) Since T is linear, the given conditions read

T (v) + 2T (w) = 3v w
T (v) T (w) = 2v 4w.
Add twice the second equation to the first to get 3T (v) = 7v9w, T (v) = 73 v3w. Similarly,
subtracting the second from the first gives 3T (w) = v + 3w, T (w) = 31 v + w. [Alternatively,
we can use gaussian elimination with constants 3v w and 2v 4w.]
8(b) Since {v1 , . . . , vn } is a basis of V, every vector v in V is a unique linear combination
v = r1 v1 + + rn vn , ri in R. Hence, as T is linear,

T (v) = r1 T (v1 ) + + rn T (vn ) = r1 (v1 ) + + rn (vn ) = v = (1)v.

Since this holds for every v in V, it shows that T = 1, the scalar operator.
12. {1} is a basis of the vector space R. If T : R V is a linear transformation, write T (1) = v.
Then, for all r in R :
T (r) = T (r 1) = rT (1) = rv.
Since T (r) = rv is linear for each v in V, this shows that every linear transformation
T : R V arises in this way.
15(b) Write U = {v V | T (v) P }. If v and v1 are in U, then T (v) and T (v1 ) are in P . As P
is a subspace, it follows that T (v + v1 ) = T (v) + T (v1 ) and T (rv) = rT (v) are both in P ;
that is v + v1 and rv are in U. Since 0 is in Ubecause T (0) = 0 is in P it follows that U
is a subspace.

118

Section 7.2: Kernel and Image of a Linear Transformation

18. Assume that {v, T (v)} is independent Then T (v) = v (or else 1v + (1)T (v) = 0) and
similarly T (v) = v.
Conversely, assume that T (v) = v and T (v) = v. To verify that {v, T (v)} is independent,
let rv + sT (v) = 0; we must show that r = s = 0. If s = 0, then T (v) = av where a = rs .
Hence v = T [T (v)] = T (av) = aT (v) = a2 v. Since v = 0, this gives a = 1, contrary to
hypothesis. So s = 0, whence rv = 0 and r = 0.
21(b) Suppose that T : Pn R is a linear transformation such that T (xk ) = T (x)k holds for all
k 0 (where x0 = 1). Write T (x) = a. We have T (xk ) =) T (x)k = ak = Ea (xk )*for each k by
assumption. This gives T = Ea by Theorem 2 because 1, x, x2 , . . . , xi , . . . , xn is a basis of
Pn .

Exercises 7.2

Kernel and Image of a Linear Transformation

1(b) We have ker TA = {X | AX = 0} ; to determine this space, we use gaussian elimination:

2 1 1 3 0
1 0
3
1 0
1 0
3
1 0
1 0 3 1 0 0 1 7 1 0 0 1 7 1 0 .



1 1 4 2 0
0 1 7 1 0
0 0
0
0 0

Hence ker TA =

3s t
7s t
s
t

| s, t in R = span

3
7
1
0

dependent so nullity of TA = dim(ker TA ) = 2. Next

1
1
0

. These vectors are in

)
*
im TA = AX | X in R4

2 1 1 3
s

1 0
3
1
=
|
r,
s,
t,
u
in R
t

1 1 4 2
y

2
1
1
3

= r 1 + s 0 + t 3 + u 1 | r, s, t, u in R .

Thus im TA = col A as is true in general. Hence dim(im TA ) = dim(col A) = rank A, and

we can compute this by carrying A to row-echelon form:

7
0

Thus dim(im TA ) = rank A = 2. However, we want a basis of col A, and we obtain this by
writing the columns of A as rows and carrying the resulting matrix (it is AT ) to row-echelon
form:

119

Section 7.2: Kernel and Image of a Linear Transformation

Hence, by Lemma 2 5.4,

0
1

0
1

3
1

0
0

is a basis of im TA = col A. Of course this once

again shows that rank TA = dim(col A) = 2.

(d) ker TA = {X | AX = 0} so, as in (b), we use gaussian elimination:

Hence, ker TA =

t
2t
t

| t in R = span

1
2
1

0
0

. Thus the nullity of TA is

dim (ker TA ) = 1. As in (b), im TA = col A and we find a basis by doing gaussian elimination
on AT :

Hence, im TA = col A = span

1
0
1
1

1
3

0
1
1
2

, so rank TA = dim(im TA ) = 2.

2(b) Here T = P2 R2 given by T [p(x)] = [p(0) p(1)] . Hence

ker T = {p(x) | p(0) = p(1) = 0} .

If p(x) = a + bx + cx2 is in ker T, then 0 )= p(0) =
* a and
that p(x) = bx bx2 , and so ker T = span x x2 . Thus
im T is a subspace of R2 . We have (1, 0) = T (1 x) and
im T = R2 . Thus {(1, 0), (0, 1)} is a basis of im T .

0 = p(1)* = a + b + c. This means

)
x x2 is a basis of ker T. Next,
(0, 1) = T [x] are both in im T, so

(d) Here T : R3 R4 given by T (x, y, z) = (x, x, y, y). Thus,

ker T = {(x, y, z) | (x, x, y, y) = (0, 0, 0, 0)} = {(0, 0, z) | z in R} = span {(0, 0, 1)} .
Hence, {(0, 0, 1)} is a basis of ker T. On the other hand,
im T = {(x, x, y, y) | x, y in R} = span {(1, 1, 0, 0) , (0, 0, 1, 1)} .
Then {(1, 1, 0, 0) , (0, 0, 1, 1)} is a basis of im T.

120

Section 7.2: Kernel and Image of a Linear Transformation

(f) Here T : M22 R is given by T

'

Hence,
hand,

= a + d. Hence

( '

a
ker T =
|a+d=0 =
c d
c
'





1
0
0 1
0
= span
,
,
a

'



,

im T =
So {1} is a basis of im T.

(
| a, b, c in R
a
(
0
.
b

is a basis of ker T (being independent). On the other

'


a+d |

in M22

= R.

(h) T : Rn R, T (r1 , r2 , . . . , rn ) = r1 + r2 + + rn . Hence,

ker T = {(r1 , r2 , . . . , rn ) | r1 + r2 + + rn = 0}
= {(r1 , r2 , . . . , rn1 , r1 rn1 ) | ri in R}
= span {(1, 0, 0, . . . , 1) , (0, 1, 0, . . . , 1) , . . . , (0, 0, 1, . . . , 1)} .
This is a basis of ker T . On the other hand,
im T = {r1 + + rn | (r1 , r2 , . . . , rn ) is in Rn } = R.
Thus {1} is a basis of im T .
(j) T : M22 M22 is given by T (X) = XA where A =
'

ker T = {X | XA = 0} =
'



0 1
= span
,
0

Thus,

'



,

(

Thus,



,

(

(



. Writing X =

(

'

(
| y, w in R

is a basis of ker T (being independent). On the other hand,

im T = {XA | X in M22 } =
'

'

(
'

| x, z in R = span



,

(

is a basis of im T .

3(b) We have T : V R2 given by T (v) = (P (v), Q(v)) where P : V R and Q : V R are

linear transformations. T is linear by (a). Now
ker T = {v | T (v) = (0, 0)}
= {v | P (v) = 0 and Q(v) = 0}

= {v | P (v) = 0} {v | Q(v) = 0}
= ker P ker Q.

Section 7.2: Kernel and Image of a Linear Transformation

121

4(b) ker T = {(x, y, z) | x + y + z = 0, 2x y + 3z = 0, z 3y = 0, 3x + 4z = 0} . Solving:

4
0
1
1
1 0
1
1
1 0
1 0
3



1
2 1 3 0

0 3 1 0 0 1 3 0 .
0 3 1 0
0 3 1 0
0 0 0 0



0 0
0 0
3
0
4 0
0 3 1 0

Hence, ker T = {(4t, t, 3t) | t in R} = span{(4, 1, 3)} . Hence,{(1, 0, 0), (0, 1, 0), (4, 1, 3)}
is one basis of R3 containing a basis of ker T . Thus
{T (1, 0, 0), T (0, 1, 0)} = {(1, 2, 0, 3), (1, 1, 3, 0)}

is a basis of im T by Theorem 5.
6(b) Yes. dim(im T ) = dim V dim(ker T ) = 52 = 3. As dim W = 3 and im T is a 3-dimensional
subspace, im T = W. Thus, T is onto.
(d) No. If ker T = V then T (v) = 0 for all v in V, so T = 0 is the zero transformation. But
W need not be the zero space. For example, T : R2 R2 defined by T (x, y) = (0, 0) for all
(x, y) in R2 .
(f) No. Let T : R2 R2 be defined by T (x y) = (y 0) for all (x, y) R2 . Then ker T = {(x, 0) |
x R} = im T.
(h) Yes. We always have dim(im T ) dim W (because im T is a subspace of W ). Since
dim(ker T ) dim W also holds in this case:
dim V = dim(ker T ) + dim(im T ) dim W + dim W = 2 dim W.
Hence dim W

1
2

dim V.

(j) No. T : R2 R2 given by T (x, y) = (x, 0) is not one-to-one (because ker T = {(0, y) | y R}
is not 0).
(l) No. T : R2 R2 given by T (x, y) = (x, 0) is not onto.
(n) No. Define T : R2 R2 by T (x, y) = (x, 0), and let v1 = (1, 0) and v2 = (0, 1). Then {v1 , v2 }
spans R2 , but {T (v1 ), T (v2 )} = {v1 , 0} does not span R2 .
7(b) Given w in W, we must show that it is a linear combination of T (v1 ), . . . , T (vn ). As T is onto,
w = T (v) for some v in V. Since V = span{v1 , . . . , vn } we can write v = r1 v1 + + rn vn
where each ri is in R. Hence
w = T (v) = T (r1 v1 + + rn vn ) = r1 T (v1 ) + + rn T (vn ).
8(b) If T is onto, let v be any vector in V . Then v = T (r1 , . . . , rn ) for some (r1 , . . . , rn ) in Rn ;
that is v = r1 v1 + + rn vn is in span{v1 , . . . , vn } . Thus V = span{v1 , . . . , vn } . Conversely,
if V = span{v1 , . . . , vn }, let v be any vector in V. Then v is in span{v1 , . . . , vn } so r1 , . . . , rn
exist in R such that
v = r1 v1 + + rn vn = T (r1 , . . . , rn ).
Thus T is onto.

122

Section 7.2: Kernel and Image of a Linear Transformation

10. The trace map T : M22 R is linear (Example 2, Section 7.1) and it is onto (for example,
r = tr[diag (r, 0, . . . , 0)] = T [diag (r, 0, . . . , 0)] for any r in R). Hence the dimension theorem
gives dim(ker T ) = dim Mnn dim(im T ) = n2 dim(R) = n2 1.
12. Define TA : Rn Rm and TB : Rn Rk by TA (x) = Axand TB (x) = Bxfor all xin Rn .
Then the given condition means ker TA ker TB , so dim(ker TA ) dim(ker TB ). Hence
rank A = dim(im TA ) = n dim(ker TA ) n dim(ker TB ) = dim(im TB ) = rank B.
)
*
15(b) Write B = x 1, x2 1, . . . , xn 1 . Then B ker T because T (xk 1) = 1 1 = 0 for
all k. Hence span B ker T. Moreover, the polynomials in B are independent (they have
distinct degrees), so dim(span B) = n. Hence, by Theorem 2 6.4, it suffices to show that
dim(ker T ) = n. But T : Pn R is onto, so the dimension theorem gives dim(ker T ) =
dim(Pn ) dim(R) = (n + 1) 1 = n, as required.
20. If we can find an onto linear transformation T : Mnn Mnn with ker T = U and im T = V,
then we are done by the dimension theorem. The condition ker T = U suggests that we define
T by T (A) = A AT for all A in Mnn . By Example 3, T is linear, ker T = U, and im T = V.
This is what we wanted.
22. Fix a column y = 0 in Rn , and define T : Mmn Rm by T (A) = Ay for all A in Mmn . This
is linear and ker T = U, so the dimension theorem gives
mn = dim(Mmn ) = dim(ker T ) + dim(im T ) = dim U + dim(im T ).
Hence, it suffices to show that dim(im T ) = m, equivalently (since im T Rm ) that T is
onto. So let xbe a column in Rm , we must find a matrix A in Mmn such that Ay = x. Write
A in terms of its columns as A = [C1 C2 . . . Cn ] and write y = [y1 y2 . . . yn ]T .
Then the requirement that Ay = x becomes

x = [C1

C2

...

y1

y2

Cn ]
..
.

yn

= y1 C1 + y2 C2 + + yn Cn .

(*)

Since y = 0, let yk = 0. Then Ay = x if we choose Ck = yk1 x and Cj = 0 if j = k. Hence T

is onto as required.
29(b) Choose a basis{u1 , . . . , um } of U and (by Theorem 1 6.4) let {u1 , . . . , um , . . . , un } be a basis
of V. By Theorem 3 7.1, there is a linear transformation S : V V such that
S(ui ) = ui if 1 i m
S(ui ) = 0

if i > m.

Hence, ui is in im S for 1 i m, whence U im S. On the other hand, if w is in im S,

write w = S(v), v in V. Then ri exist in R such that
v = r1 u1 + + rm um + + rn un

123

Section 7.3: Isomorphisms and Composition

so
w = r1 S(u1 ) + + rm S(um ) + + rn S(un )
= r1 u1 + + rm um + 0.
It follows that w is in U, so im S U. Then U = im S as required.

Exercises 7.3

Isomorphisms and Composition

1(b) T is one-to-one because T (x, y, z) = (0, 0, 0) means x = 0, x + y = 0 and x + y + z = 0, whence

x = y = z = 0. Now T is onto by Theorem 3.
Alternatively: {T (1, 0, 0), T (0, 1, 0), T (0, 0, 1)} = {(1, 1, 1), (0, 1, 1), (0, 0, 1)} is independent, so T is an isomorphism by Theorem 1.
(d) T is one-to-one because T (X) = 0 implies U XV = 0, whence X = 0 (as U and V are
invertible). Now Theorem 3 implies that T is onto and so is an isomorphism.
(f) T is one-to-one because T (v) = 0 implies kv = 0, so v = 0 because k = 0. Hence, T is onto
if dim V is finite (by Theorem 3) and so is an isomorphism. Alternatively, T is onto because
T (k 1 v) = k(k1 v) = v holds for all v in V.
(h) T is onto because T (AT ) = (AT )T = A for every n m matrix A (note that AT is in Mmn
so T (AT ) makes sense). Since dim Mmn = mn = dim Mnm , it follows that T is one-to-one
by Theorem 3, and so is an isomorphism. (A direct proof that T is one-to-one: T (A) = 0
implies AT = 0, whence A = 0.)
4(b) ST (x, y, z) = S(x + y, 0, y + z) = (x + y, 0, y + z); T S(x, y, z) = T (x, 0, z) = (x, 0, z). These
are not equal (if y = 0) so ST = T S.
















a b
c a
c 0
a b
a 0
0 a
(d) ST
=S
=
; TS
=T
=
. These are not
c

equal for some values of a, b, c and d (nearly all) so ST = T S.

5(b) T 2 (x, y) = T [T (x, y)] = T (x + y, 0) = [x + y + 0, 0] = (x + y, 0) = T (x, y). This holds for all
(x, y), whence T 2 = T .












a b
a b
a+c b+d
1 a+c b+d
1
2
=T 2
(d) T
=T T
= 2T
c

1
4

a+c

(a + c) + (a + c)

(b + d) + (b + d)

(a + c) + (a + c)

(b + d) + (b + d)

This holds for all

b+d

1
2

a+c

a+c

b+d

a+c

b+d

b+d

=T

, so T 2 = T.

6(b) No inverse. For example T (1, 1, 1, 1) = (0, 0, 0, 0) so (1, 1, 1, 1) is a nonzero vector in

ker T . Hence T is not one-to-one, and so has no inverse.

124

Section 7.3: Isomorphisms and Composition

(d) T is one-to-one because T

implies a + 2c = 0 = 3c a and

b + 2d = 0 = 3d b, whence a = b = c = d = 0. Thus T is an isomorphism by Theorem 3. If












a b
x y
a b
x y
x + 2z y + 2w
1
T V
=
, then
=T
=
. Thus
c

3z x

3w y

x + 2z = a
x + 3z = c
y + 2w = b
y + 3w = d.
The solution is x = 15 (3a 2c), z = 15 (a + c), y = 15 (3b 2d), w = 51 (b + d). Hence






3a 2c 3b 2d
a b
1
1
.
T
=5
c

A better way to find

T 1

a+c

b+d

is to observe that T (X) = AX where A =

(*)

. This matrix

is invertible which easily implies that T is one-to-one (and onto), and if S : M22 M22 is
1 by Theorem 5.
defined by S(X) =
A1 X then
 ST = 1M22 and T S = 1M22 . Hence S = T
Note that A1 =

1
5

3
1

2
1

which gives (*).

(f) T is one-to-one because, if p in P2 satisfies T (p) = 0, then p(0) = p(1) = p(1) = 0. If

p = a + bx + cx2 , this means a = 0, a + b + c = 0 and a b + c = 0, whence a = b = c = 0,
and p = 0. Hence, T 1 exists by Theorem 3. If T 1 (a, b, c) = r + sx + tx2 , then
(a, b, c) = T (r + sx + tx2 ) = (r, r + s + t, r s + t).
Then r = a, r + s + t = b, r s + t = c, whence r = a, s = 12 (b c), t = 12 (b + c 2a). Finally
T 1 (a, b, c) = a + 12 (b c)x + 12 (b + c 2a)x2 .
7(b) T 2 (x, y) = T [T (x, y)] = T (ky x, y) = [ky (ky x), y] = (x, y) = 1R2 (x, y). Since this holds
for all (x, y) in R2 , it shows that T 2 = 1R2 . This means that T 1 = T.
(d) It is a routine verification that A2 = I. Hence
T 2 (X) = T [T (X)] = A[AX] = A2 X = IX = X = 1M22 (X)
holds for all X in M22 . This means that T 2 = 1M22 , and hence that T 1 = T.
8(b) T 2 (x, y, z, w) = T [T [x, y, z, w]] = T (y, x y, z, w) = ((x y), y (x y), z, (w))
= (y
x, x, z, w).
2 (x, y, z, w) = T (y x, x, z, w) = (x, y, z, w).
T 3 (x, y, z, w) = T T


T 6 (x, y, z, w) = T 3 T 3 (x, y, z, w) = T 3 [x, y, z, w] = (x, y, z, w) = 1R4 (x, y, z, w).

Hence, T 6 = 1R4 so T 1 = T 5 . Explicitly:


T 1 (x, y, z, w) = T 2 T 3 (x, y, z, w) = T 2 (x, y, z, w) = (y x, x, z, w).

125

Section 7.3: Isomorphisms and Composition

9(b) Define S : Mnn Mnn by S(A) = U 1 A. Then
ST (A) = S(T (A)) = U 1 (U A) = A = 1Mnn (A)

so ST = 1Mnn

T S(A) = T (S(A)) = U(U 1 A) = A = 1Mnn (A)

so T S = 1Mnn .

Hence, T is invertible and T 1 = S.

T

10(b) Given V W U with T and S both onto, we are to show that ST : V U is onto. Given
u in U, we have u = S(w) for some w in W because S is onto; then w = T (v) for some v in
V because T is onto. Hence,
ST (v) = S[T (v)] = S[w] = u.
This shows that ST is onto.
12(b) If u lies in im RT write u = RT (v), v in V . Thus u = R[T (v)] where T (v) in W, so u is in
im R.
T

13(b) Given V U W with ST onto, let w be a vector in W. Then w = ST (v) for some v in
V because ST is onto, whence w = S[T (v)] where T (v) is in U. This shows that S is onto.
Now the dimension theorem applied to S gives
dim U = dim(ker S) + dim(im S) = dim(ker S) + dim W
because im S = W (S is onto). As dim(ker S) 0, this gives dim U dim W.
14. If T 2 = 1V then T T = 1V so T is invertible and T 1 = T by the definition of the inverse of a
transformation. Conversely, if T 1 = T then T 2 = T T 1 = 1V .
16. Theorem 5, Section 7.2 shows that {T (e1 ), T (e2 ), . . . , T (er )} is a basis of im T. Write

U = span{e1 , . . . , er } . Then B = {e1 , . . . , er } is a basis of U, and T : U im T carries B

to the basis {T (e1 ), . . . , T (er )} . Thus T : U im T is itself an isomorphism. Note that
T : V W may not be an isomorphism, but restricting T to the subspace U of V does result
in an isomorphism in this case.

19(b) We have V = {(x, y) | x, y in R} with a new addition and scalar multiplication:

(x, y) (x1 , y1 ) = (x + x1 , y + y1 + 1)
a  (x, y) = (ax, ay + a 1).
We use the notation and  for clarity. Define
T : V R2 by T (x, y) = (x, y + 1).
Then T is a linear transformation because:
T [(x, y) (x1 , y1 )] = T (x + x1 , y + y1 + 1)
= (x + x1 , (y + y1 + 1) + 1)
= (x, y + 1) + (x1 , y1 + 1)
= T (x, y) + T (x1 , y1 )

126

Section 7.3: Isomorphisms and Composition

T (a  (x, y)] = T (ax, ay + a 1)
= (ax, ay + a)
= a(x, y + 1)
= aT (x, y).
Moreover T is one-to-one because T (x, y) = (0, 0) means x = 0 = y + 1, so (x, y) = (0, 1), the
zero vector of V. (Alternatively, T (x, y) = T (x1 , y1 ) implies (x, y + 1) = (x1 , y1 + 1), whence
x = x1 , y = y1 .) As T is clearly onto R2 , it is an isomorphism.

24(b) T S[x0 , x1 ...) = T [0, x0 , x1 , ...) = [x0 , x1 , ...) so T S = 1V . Hence T S is both onto and oneto-one, so T is onto and S is one-to-one by Exercise 113. But [1, 0, 0, ...) is in ker T while
[1, 0, 0, ...) is not in im S.
26(b) If p(x) is in ker T, then p(x) = xp (x). If we write p(x) = a0 + a1 x + + an xn , this becomes
a0 + a1 x + + an1 xn1 + an xn = a1 x 2a2 x2 nan xn .
Equating coefficients gives a0 = 0, a1 = a1 , a2 = 2a2 , . . . , an = nan . Hence we have,
a0 = a1 = = an = 0, so p(x) = 0. Thus ker T = {0}, so T is one-to-one. As T : Pn Pn
and dim Pn is finite, this implies that T is also onto, and so is an isomorphism.
27(b) If T S = 1W then, given w in W, T [S(w)] = w, so T is onto. Conversely, if T is onto, choose a
basis {e1 , . . . , er , er+1 , . . . , en } of V such that {er+1 , . . . , en } is a basis of ker T. By Theorem 5,
7.2, {T (e1 ), . . . , T (en )} is a basis of im T = W (as T is onto). Hence, a linear transformation
S : W V exists such that S[T (ei )] = ei for i = 1, 2, . . . , r. We claim that T S = 1W , and
we show this by verifying that these transformations agree on the basis {T (e1 ), . . . , T (er )} of
W. Indeed
T S[T (ei )] = T {S[T (ei )]} = T (ei ) = 1W [T (ei )]
for i = 1, 2, . . . , n.

28(b) If T = SR, then every vector T (v) in im T has the form T (v) = S[R(v)], whence im T im
S. Since R is invertible, S = T R1 implies im S im T, so im S = im T.
Conversely, assume that im S = im T. The dimension theorem gives

dim(ker S) = n dim(im S) = n dim(im T ) = dim(ker T ).

Hence, let {e1 , . . . , er , . . . , en } and {f1 , . . . , fr , . . . , fn } be bases of V such that {er+1 , . . . , en }
and {fr+1 , . . . , fn } are bases of ker S and ker T, respectively. By Theorem 5, 7.2, {S(e1 ), . . . , S(er )}
and {T (f1 ), . . . , T (fr )} are both bases of im S = im T. So let g1 , . . . , gr in V be such that
S(ei ) = T (gi )
for each i = 1, 2, . . . , r.
Claim: B = {g1 , . . . , gr , fr+1 , . . . , fn } is a basis of V .

Proof. It suffices (by Theorem 4, 6.4) to show that B is independent. If

a1 g1 + + ar gr + br+1 fr+1 + + bn fn = 0,
apply T to get
0 = a1 T (g1 ) + + ar T (gr ) + br+1 T (fr+1 ) + + bn T (fn )
= a1 T (g1 ) + + ar T (gr ) + 0

127

Section 7.4: A Theorem about Differential Equations

because T (fj ) = 0 if j > r. Hence a1 = = ar = 0; whence 0 = br+1 fr+1 + + bn fn . This

gives br+1 = = bn = 0 and so proves the claim.
By the claim, we can define R : V V by

R(gi ) = ei for i = 1, 2, . . . , r
R(fj ) = ej

for j = r + 1, . . . , n.

Then R is an isomorphism by Theorem 1, 7.3, and we claim that SR = T. We show this by

verifying that SR and T have the same effect on the basis B in the claim. The definition of
R gives
SR(gi ) = S[R(gi )] = S(ei ) = T (gi ) for i = 1, 2, . . . , r
SR(fj ) = S[ej ] = 0 = T (fj )

for j = r + 1, . . . , n.

Hence SR = T .
29. As in the hint, let {e1 , e2 , . . . , er , . . . , en } be a basis of V where {er+1 , . . . , en } is a basis of
ker T. Then {T (e1 ), . . . , T (er )} is linearly independent by Theorem 5, 7.2, so extend it to a
basis {T (e1 ), . . . , T (er ), wr+1 , . . . , wn } of V. Then define S : V V by
S[T (ei )] = ei for 1 i r
S(wj ) = ej

for r < j n.

Then, S is an isomorphism (by Theorem 1) and we claim that T ST = T. We verify this by

showing that T ST and T agree on the basis {e1 , . . . , er , . . . , en } of V (and invoking Theorem
2, 7.1).
If 1 i r: T ST (ei ) = T {S[T (ei )]} = T (ei )
If r + 1 j n: T ST (ej ) = T S[T (ej )] = T S[0] = 0 = T (ej )
where, at the end, we use the fact that ej is in ker T for r + 1 j n.

Exercises 7.4

A Theorem about Differential Equations

Exercises 7.5

More on Linear Recurrences

1(b) The associated polynomial is

p(x) = x3 7x + 6 = (x 1)(x 2)(x + 3).
Hence, {[1), [2n ), [(3)n )} is a basis of the space of all solutions to the recurrence. The general
solution is thus,
[xn ) = a[1) + b[2n ) + c[(3)n )
where a, b and c are constants. The requirement that x0 = 1, x1 = 2, x2 = 1 determines a,
b, and c. We have
xn = a + b2n + c(3)n

128

Section 7.5: More on Linear Recurrences

for all n 0. So taking n = 0, 1, 2 gives
a + b + c = x0 = 1
a + 2b 3c = x1 = 2
a + 4b + 9c = x2 = 1.
The solution is a =

15
20 ,

b=

8
20 ,

3
c = 20
, so

xn =

1
20 (15

+ 2n+3 + (3)n+1 ) n 0.

2(b) The associated polynomial is

p(x) = x3 3x + 2 = (x 1)2 (x + 2).

As 1 is a double root of p(x), [1n ) = [1) and [n1n ) = [n) are solutions to the recurrence by
Theorem 3. Similarly, [(2)n ) is a solution, so {[1), [n), [(2)n )} is a basis for the space of
solutions by Theorem 4. The required sequence has the form
[xn ) = a[1) + b[n) + c[(2)n )
for constants a, b, c. Thus, xn = a + bn + c(2)n for n 0, so taking n = 0, 1, 2, we get
a +
a +

c = x0 =

b 2c = x1 = 1

a + 2b + 4c = x2 =
The solution is a = 59 , b = 69 , c = 49 , so


xn = 19 5 6n + (2)n+2

1.

n 0.

(d) The associated polynomial is

p(x) = x3 3x2 + 3x 1 = (x 1)3 .

)
*
Hence, [1n ) = [1), [n1n ) = [n) and [n2 1n ) = [n2 ) are solutions and so [1), [n), [n2 ) is a basis
for the space of solutions. Thus
xn = a 1 + bn + cn2 ,

a, b, c constants. As x0 = 1, x1 = 1, x2 = 1, we obtain
a

= x0 =

a +

b +

c = x1 = 1

a + 2b + 4c = x2 =
The solution is a = 1, b = 4, c = 2, so

xn = 1 4n + 2n2

n 0.

This can be written

xn = 2(n 1)2 1.

1.

129

Section 7.5: More on Linear Recurrences

3(b) The associated polynomial is
p(x) = x2 (a + b)x + ab = (x a)(x b).
Hence, as a = b, {[an ), [bn )} is a basis for the space of solutions.

4(b) The recurrence xn+4 = xn+2 +2xn+3 has r0 = 0 as there is no term xn . If we write yn = xn+2 ,
the recurrence becomes
yn+2 = yn + 2yn+1 .
Now the associated polynomial is x2 2x + 1 = (x 1)2 so basis sequences for the solution
space for yn are [1n ) = [1, 1, 1, 1, . . .) and [n1n ) = [0, 1, 2, 3, . . .). As yn = xn+2 , corresponding
basis sequences for xn are [0, 0, 1, 1, 1, 1, . . .) and [0, 0, 0, 1, 2, 3, . . .). Also, [1, 0, 0, 0, 0, 0, . . .) and
[0, 1, 0, 0, 0, 0, . . .) are solutions for xn , so these four sequences form a basis for the solution
space for xn .
7. The sequence has length 2 and associated polynomial x2 + 1. The roots are nonreal: 1 = i
and 2 = i. Hence, by Remark 2,
[in + (i)n ) = [2, 0, 2, 0, 2, 0, 2, 0, . . .) and [i(in (i)n )) = [0, 2, 0, 2, 0, 2, 0, 2, . . .)
are solutions. They are independent as is easily verified, so they are a basis for the space of
solutions.

130
.

Section 7.5: More on Linear Recurrences

Section 8.1: Orthogonal Complements and Projections

131

Chapter 8 Orthogonality
Exercises 8.1

Orthogonal Complements and Projections

1(b) Write x1 = (2, 1) and x2 = (1, 2). The Gram-Schmidt algorithm gives
e1 = x1 = (2, 1)
x2 e1
e2 = x2
e1
e1 2
= (1, 2) 45 (2, 1)
= 51 {(5, 10) (8, 4)}
= 53 (1, 2).
In hand calculations, {(2, 1), (1, 2)} may be a more convenient orthogonal basis.
(d) If x1 = (0, 1, 1), x2 = (1, 1, 1), x3 = (1, 2, 2) then
e1 = x1 = (0, 1, 1)
x2 e1
e2 = x2
e1 = (1, 1, 1) 22 (0, 1, 1) = (1, 0, 0)
e1 2
x3 e1
x3 e2
e3 = x3
e2 = (1, 2, 2) 20 (0, 1, 1) 11 (1, 0, 0) = (0, 2, 2).
2 e1
e1 
e2 2
2(b) Write e1 = (3, 1, 2) and e2 = (2, 0, 3). Then {e1 , e2 } is orthogonal and so is an orthogonal
basis of U = span{e1 , e2 } . Now x = (2, 1, 6) so take
x e1
x e2
x1 = projU (x) =
e2
2 e1 +
e1 
e2 2
14
= 17
14 (3, 1, 2) 13 (2, 0, 3)
1
(271, 221, 1030).
= 182
Then x2 = x x1 =
is in U ).

1
182 (93, 402, 62).

As a check: x2 is orthogonal to both e1 and e2 (and so

(d) If e1 = (1, 1, 1, 1), e2 = (1, 1, 1, 1), e3 = (1, 1, 1, 1) and x = (2, 0, 1, 6), then {e1 , e2 , e3 }
is orthogonal so take
x e1
x e2
x e3
x1 = projU (x) =
e3
2 e1 +
2 e2 +
e1 
e2 
e3 2
= 94 (1, 1, 1, 1) 45 (1, 1, 1, 1) 34 (1, 1, 1, 1)
= 41 (1, 7, 11, 17).

Then, x2 = x x1 = 41 (7, 7, 7, 7) = 74 (1, 1, 1, 1). Check: x2 is orthogonal to each ei ,

hence x2 is in U .
(f) If e1 = (1, 1, 2, 0) and e2 = (1, 1, 1, 1) then (as x = (a, b, c, d))

ab+2c
(1, 1, 2, 0) + a+b+c+d
(1, 1, 1, 1)
6
4
5a5b+c3d 5a+5bc+3d ab+11c+3d 3a+3b+3c+3d
(
,
,
,
)
12
12
12
12
7a+5bc+3d 5a+7b+c3d a+b+c3d 3a3b3c+9d
x x1 = (
,
,
,
).
12
12
12
12

x1 = projU (x) =
=
x2 =

132

Section 8.1: Orthogonal Complements and Projections

3(a) Write e1 = (2, 1, 3, 4) and e2 = (1, 2, 0, 1), so {e1 , e2 } is orthogonal.

As x= (1, 2, 1, 6)
x e1
x e2
e2
2 e1 +
e1 
e2 2
3
= 21
30 (2, 1, 3, 4) + 6 (1, 2, 0, 1) =

projU (x) =

15
(c) projU (x) = 14
(1, 0, 2, 3) +

3
70 (4, 7, 1, 2)

3
10 (3, 1, 7, 11).

3
10 (3, 1, 7, 11).

4(b) U = span{(1, 1, 0), (1, 0, 1)} but this basis is not orthogonal. By Gram-Schmidt:
e1 = (1, 1, 0)
e2 = (1, 0, 1)

(1, 0, 1) (1, 1, 0)
(1, 1, 0) = 12 (1, 1, 2).
(1, 1, 0)2

So we use U = span{(1, 1, 0), (1, 1, 2)} . Then the vector x1 in U closest to x= (2, 1, 0) is
x1 = projU (x) =

21+0
2+1+0
(1, 1, 0) +
(1, 1, 2) = (1, 0, 1).
2
6

(d) The given basis of U is not orthogonal. The Gram-Schmidt algorithm gives
e1 = (1, 1, 0, 1)
e2 = (1, 1, 0, 0) = (1, 1, 0, 0) 30 e1 = (1, 1, 0, 0)
e3 = (1, 1, 0, 1) 13 (1, 1, 0, 1) 22 (1, 1, 0, 0) = 13 (1, 1, 0, 2).
Given x= (2, 0, 3, 1), we get (using e3 = (1, 1, 0, 2) for convenience)
projU (x) = 33 (1, 1, 0, 1) + 22 (1, 1, 0, 0) + 06 (1, 1, 0, 2) = (2, 0, 0, 1).









1 1
2
1
1 1
2
1
1 0 1 1
5(b) Here A =

. Hence, AxT = 0 has

1

solution x = (s t, 3s, s, t) = s(1, 3, 1, 0) + t(1, 0, 0, 1).

Thus U = span{(1, 3, 1, 0), (1, 0, 0, 1)} .

8. If x= projU (x) then x is in U by Theorem 3. Conversely, if x is in U, let {f1 , , fm } be an

orthogonal basis of U. Then the expansion theorem (applied to the space U) gives
xfi
x= i f
f = projU (x) by the definition of the projection.
2 i
i

10. Let {f1 , . . . , fm } be an orthonormal basis of U. If X is in U then, since fi  = 1 for each i, so
x = (x f1 )f1 + + (x fm )fm = projU (x) by the expansion theorem (applied to the space
U).

14. If {y1 , , ym } is a basis of U , take A =

y1T
.
..

T
ym

. Then Ax = 0 if and only if yTi x = 0 for

each i; if and only if yi x = 0 for each i; if and only if xis in (U ) = U = U. This shows
that U = {x in Rn | Ax = 0}.

133

Section 8.2: Orthogonal Diagonalization

17(d) If AAT is invertible and E= AT (AAT )1 A, then
E 2 = AT (AAT )1 A AT (AAT )1 A = AT I(AAT )1 A = E
T

T
E T = AT (AAT )1 A = AT (AAT )1 (AT )T

1
= AT (AAT )T
A = AT (AT )T AT
A


1
A = E.
= AT AAT

Thus, E 2 = E = E T .

Exercises 8.2
1(b) Since

32

+ 42

Orthogonal Diagonalization
=

52 ,

each row has length 5. So

(d) Each row has length

a2 + b2 = 0, so

(f) The rows have length

3
5
4
5

4
5
3
5

6, 3, 2 respectively, so
2

1
1

is orthogonal.
(h) Each row has length

1
a2 +b2

6
1

6
1
3
1
2

1
3
1
2

1
6

4 + 36 + 9 = 49 = 7. Hence
2 6

is orthogonal.

7
3
7
67

7
2
7
3
7

6
7
2
7

1
7

1
5

3
4

4
3

is orthogonal.

is orthogonal.

3
6

2. Let P be orthogonal, so P 1 = P T . If P is upper triangular, so also is P 1 , so P 1 = P T is

both upper triangular (P 1 ) and lower triangular P T ). Hence, P 1 = P T is diagonal, whence
P = (P 1 )1 is diagonal. In particular, P is symmetric so P 1 = P T = P. Thus P 2 = I.
Since P is diagonal, this implies that all diagonal entries are 1.


x1

1

= x(x 2).
5(b) cA (x) =

x1

Hence the eigenvalues are 1 = 0, 2 = 2.

1 = 0 :
2 = 2 :

1
0

'
(
1
; E0 (A) = span
.
0
1

'

(
1
; E2 (A) = span
.
1

Note that these eigenvectors are orthogonal (as Theorem 4 asserts). Normalizing them gives
an orthogonal matrix

1





1
1 1
1
2
2
P =
= 2
.
1
1
Then P 1 = P T and P T AP =

134

Section 8.2: Orthogonal Diagonalization

x3
0
7


= (x 5)(x2 6x 40) = (x 5)(x + 4)(x 10). Hence the
x5
0
(d) cA (x) = 0

7
0
x3
eigenvalues

are 1 = 5, 2 =
10, 3 =4.
2
0 7
1 0 0
0

0
0
0
1 = 5 :
0 0 1 ; E5 (A) = span 1 .

7 0
2
0 0 0
0

7
0 7
1 0 1
1

2 = 10 :
0
5
0
0 1 0
; E10 (A) = span 0 .

7 0
7
0 0
0
1

7
0
7
1 0 1
1
3 = 4 : 0 9 0 0 1 0 ; E4 (A) = span 0 .

Note that the three eigenvectors are pairwise orthogonal (as Theorem 4 asserts). Normalizing
them gives an orthogonal matrix

P =

1
2

1
2

1
2

1
2

Then P 1 = P T and P T AP =

(f) cA (x) =

x5
2

= (x 9)
2 = 9.

1 = 0 :

5
2
4

x8

x5

x8

x1

2 = 9 :

10

1
2

x9


0
9x
0
0

x9

= 2
= 2

x

8
2
x

8
4



4


2
x5
4
2
x1


= (x 9)(x2 9x) = x(x 9)2 . The eigenvalues are 1 = 0,

2
E0 (A) = span 1 .

18

18

1
2

0
0

4
1

; E9 (A) = span

12
0

1
0
1

1
2
0

12
0

However, these are not orthogonal and the Gram-Schmidt algorithm replaces

Z2 =

4
1

P T AP =

. Hence P =
0

2
3
1
3
2
3

0
1

3 2
4

3 2
1

3 2

3 2

2 2

2 2

3
0

4
1

1
2
0

with

is orthogonal and satisfies

135

Section 8.2: Orthogonal Diagonalization

We note in passing that

Q=

1
3

2
1

and

1
2
2

are another orthogonal basis of E9 (A), so

also satisfies QT AQ =

(h) To evaluate
cA (x), we begin adding
rows
2, 3 and 4 to row 1.
x 3 5

1
1

x8 x8 x8 x8
5 x 3 1


1
1
= 5 x 3 1
cA (x) =
1
1
1
x

3
5
1
x

3
5


1
1
1
5
x3
1
5
x3




=

x8

x2

5
1

= (x 8)

1 = 0 :

3 = 8 :

6
6

4
x

x2

1
1

5
5
1

5
5

1
1

1
1

1
0

1
1
5

x4

x2

x2 4
6




(x 8) = 0
x
0




6
4 x 2

x+2

= x(x 8)(x2 4x 32) = x(x + 4)(x 8)2 .
x2

1 1 3 5
3 5
1
1
1

0 8 8 16

1
8 8 0 0
1 1 3 5
5
0 16 24 40

0
0
8 8
3
0
0
1
1

1
1 0 0 1

0 1 0 1
1
; E0 (A) = span
.

0 0 1 1

x2

2 = 4 :

x2

= (x 8)

x4

x+2

= x(x 8)

0
0

12

48

12

12

5
5

1
0

36

36

E4 (A) = span

24
24
0

1
1

1
1

24

12

5
24

24
0

0
0

136

Section 8.2: Orthogonal Diagonalization

E8 (A) = span

1
0
0

2
12
12
1
2

Hence, P =

6. cA (x) =

12

0
1
1

1
2
1
2

1
2
12
1
2

2
1

= x

1
2

1
1

1
1

+a

a
c

0
0

0
x

gives P T AP =

0
0

= x(x2 c2 ) a2 x = x(x2 k2 ).

Hence cA (x) = x(x k)(x + k), where k 2 = a2 + c2 , so the eigenvalues are 1 = 0, 2 = k,

3 = k. They are all distinct (k = 0, and a = 0 or c = 0) so the eigenspaces are all one
dimensional.

0
c
0
a
0
c
1 = 0 : a 0 c 0 = 0 ; E0 (A) = span 0 .

a
0
a
0
c
0

0
a
k a 0
a

2 = k : a k c k = 0 ; Ek (A) = span k .

c
0
c
0 c k

0
a
k a
0
a
3 = k : a k c k = 0 ; Ek (A) = span k ,

c
0
c
0
c k

These
are
eigenvalues
orthogonal and have length,
k, 2k, 2k respectively. Hence, P =

c 2

1
2k

a 2

a
k
c

k
c

is orthogonal and P T AP =

10. Similar to Example 6, q has matrix A =

with eigenvalues 1 = 3 and 2 =
2 2





1
2
2 and corresponding eigenvectors x1 =
and x2 =
respectively. Hence P =
2
1






1 2
3 0
1
. Let
is orthogonal and P T AP =
5
2

y1
y2

= y = PTx =

1
5

x1 + 2x2
2x1 + x2

; so y1 =

1 (x1
5

+ 2x2 ) and y2 =

1 (2x1
5

+ x2 ).

Then q = 3y12 + 2y22 is diagonalized by these variables.

11. (c)(a). By Theorem 1 let P 1 AP = D = diag(1 , , n ) where the i are the eigenvalues
of A. By (c) we have i = 1 for each i. It follows that
D2 = diag(21 , , 2n ) = diag(1, , 1) = I.
Since A = P DP 1 , we obtain A2 = (P DP 1 )2 = P D2 P 1 = P IP 1 = I. Since A is
symmetric, this proves (a).

137

Section 8.2: Orthogonal Diagonalization

13(b) Let A and B be orthogonally similar, say B = P T AP where P T = P 1 . Then

B 2 = P T AP P T AP = P T AIAP = P T A2 P. Hence A2 and B 2 are orthogonally similar.
15. Assume that (Ax) y = x Ay for all columns x and y; we must show that AT = A. We
have (Ax) y = xT AT y and x Ay = xT Ay, so the given condition asserts that
xT AT y = xT Ay

for all columns x and y.

((*))

But if Ej denotes column j of the identity matrix, then writing A = [aij ] we have
eTi Aej = aij

for all i and j.

Since (*) shows that AT and A have the same (i, j)-entry for each i and j. In other words,
AT = A.
Note that the same argument shows that if A and B are matrices with the property that
= xT Ay for all columns x and y, then B = A.

xT By

18(b) If P =

cos

sin

sin

cos

and Q =

cos

sin

sin

cos

then P and Q are orthogonal matrices,

det P = 1 and det Q = 1. (We note that every 2 2 orthogonal matrix has the form of P
or Q for some .)
(d) Since P is orthogonal, P T = P 1 . Hence
P T (I P ) = P T P T P = P T I = (I P T ) = (I P )T .
Since P is n n, taking determinants gives
det P T det(I P ) = (1)n det[(I P )T ] = (1)n det(I P ).
Hence, if I P is invertible, then det(I P ) = 0 so this gives det P T = (1)n ; that is
det P = (1)n , contrary to assumption.
21. By the definition of matrix multiplication, the [i, j]-entry of AAT is ri rj . This is zero if
i = j, and equals ri 2 if i = j. Hence, AAT = D = diag(r1 2 , r2 2 , . . . , rn 2 ). Since D
is invertible (ri 2 = 0 for each i), it follows that A is invertible and, since row i of AT is
[a1i a2i . . . aji . . . ani ]

A1 = AT D1 =

Thus, the (i, j)-entry of A1 is

..
.
a1i
..
.

...
...
...

aji
.
rj 2

..
.
aji
..
.

...
...
...

..
.
ani
..
.

1
r1 2

0
..
.

...

1
r2 2

...

..
.

...

..
.

...

1
rn 2

23(b) Observe first that I A and I +A commute, whence I A and (I +A)1 commute. Moreover,

138

Section 8.3: Positive Definite Matrices

1
(I + A)1 = (I + A)T
= (I T + AT )1 = (I A)1 . Hence,

P P T = (I A)(I + A)1 [(I A)(I + A)1 ]T

= (I A)(I + A)1 [(I + A)1 ]T (I A)T

= (I A)(I + A)1 (I A)1 (I + A)

= (I + A)1 (I A)(I A)1 (I + A)

= (I + A)1 I(I + A)
= I.

Exercises 8.3
1(b)

(d)

2
1

1
1

20

and A =

Positive Definite Matrices

2
0

U T U.

1
1
2

. Then A = U T U where U =

20

6
5

15
4

20

6
5

5
12

2
0

1
2

. Hence, U =


2
5

0
0

2
2
1
5
6
30

2
0

2 5
10

30
5
2 15

1
30

60 5
0

12 5

6 30

2. (b) If k is positive and k is odd, then is positive.

4. Assume x = 0 is a column. If A and B are positive definite then xT Ax > 0 and xT Bx > 0
so
xT (A + B)x = xT Ax + xT Bx > 0 + 0 = 0
Thus A + B is positive definite. Now suppose r > 0. Then xT (rA)x = r(xT Ax) > 0, proving
that rA is positive definite.
6. Given x in Rn , xT (U T AU)x = (Ux)T A(Ux) > 0 provided Ux = 0 (because A is positive
n
T
definite). Write
 U = [c1 cm ] where cj in R is column j of U. If 0 = x = [x1 . . . xm ] ,
then Ux = xj cj = 0 because the cj are independent [rank of U is m].

10. Since A is symmetric, the principal axis theorem asserts that an orthogonal matrix P exists
such that P TAP = D = diag(1 , 2 , . . . , n ) where the
of A. Since
i are
the eigenvalues

each i > 0, i is real and positive, so define B = diag 1 , 2 , . . . , n . Then B 2 = D.
As A = P DP T , take C = P BP T . Then
C 2 = P BP T P BP T = P B 2 P T = P DP T = A.


Finally, C issymmetric because B is symmetric C T = P T T B T P T = P BP T = C and C has
eigenvalues i > 0 (C is similar to B). Hence C is positive definite.

12(b) Suppose that A is positive definite so A = U0T U0 where U0 is upper triangular with positive
diagonal entries d1 , d2 , . . . , dn . Put D0 = diag(d1 , d2 , . . . , dn ) . Then L = U0T D01 is lower
triangular with 1s on the diagonal, U = D01 U0 is upper triangular with 1s on the diagonal,
and A = LD02 U. Take D = D02 .
Conversely, if A = LDU as in (a), then AT = U T DLT . Hence, AT = A implies that U T DLT =
T = L and LT = U by (a). Hence, A = U T DU. If D = diag(d , d , . . . , d ), let
LDU, so U
1 2
n


D1 = diag d1 , d2 , . . . , dn . Then D = D12 so A = U T D12 U = (D1 U)T (D1 U). Hence, A
is positive definite.

15 5

10 30

5 15

139

Section 8.4: QR-Factorization

Exercises 8.4

QR-Factorization

1(b) The columns of A are c1 =

2
1

f1 = c1 =

and c2 =

2
1

c2 f1
f2 = c2
f1 =
f1 2

1
1

1
1

. First apply the Gram-Schmidt algorithm

3
5

2
1

15
2
5

Now normalize to obtain

1
f1 =
f1 
1
q2 =
f2 =
f2 

q1 =

Hence Q = [q1

1
5

L=

q2 ] =

preceding Theorem 1:

2
1

f1 
0

1
2

1
5
1
5

2
1

1
2

is an orthogonal matrix. We obtain R from equation (*)

c2 q1
f2 

5
0

3
5
1
5

1
5

Then A = QR.
(d) The columns of A are c1 = [1 1 0 1]T , c2 = [1 0 1
Apply the Gram-Schmidt algorithm
f1 = c1 = [1 1 0 1]T
c2 f1
f2 = c2
f1 = [1 0 1
f1 2
c3 f1
c3 f2
f3 = c3
f2
2 f1
f1 
f2 2
= [0 1 1 0]T
=

2
3

1
3

[1

[0 1 1 1]T .

1]T and c3 = [0 1 1 0]T .

1]T 30 F1 = [1 0 1

1 0 1]T 31 [1 0 1

Normalize
1
f1 =
f1 
1
Q2 =
f2 =
f2 
1
Q3 =
f3 =
f3 
Q1 =

1 0 1]T

1
3

[1

1
3

[1 0 1

1
3

[0 1 1 1]T .

1]T

1]T

1]T

140

Section 8.5: Computing Eigenvalues

Hence Q = [q1

q3 ] =

q2

1
3

equation (*) preceding Theorem 1:

R=

f1 

c2 q1

c3 q1

f3 

f2 

has orthonormal columns. We obtain R from

c3 q2

Then A = QR.

3
0
0

1
3

3
0

3
1
3
2
3

1
3

1
1
2

2(b) If A = QR is a QR-factorization of A, then R has independent columns (it is invertible)

as does Q (its columns are orthonormal). Hence A has independent columns by (a). The
converse is by Theorem 1.

Exercises 8.5

Computing Eigenvalues

x 5 2

= (x + 1)(x 4), so 1 = 1, 2 = 4.
1(b) A =
. Then cA (x) =
3 2
3
x+2









6 2
3 1
1
If 1 = 1 :

: eigenvector =
3
1
3




0 0



1 2
1 2
2
If 2 = 4 :

; dominant eigenvector =
.
3
6
0 0
1


1
Starting with x0 =
, the power method gives x1 = Ax0 , x2 = Ax1 , . . . :

x1 =

7
5

x2 =

25
11

x3 =

103
53

x4 =

409
203

These are approaching (scalar multiples of) the dominant eigenvector

. The Rayleigh
1
xk xk+1
quotients are rk =
, k = 0, 1, 2, . . . , so r0 = 1, r1 = 3.29, r2 = 4.23, r3 = 3.94. These
xk 2
are approaching the dominant eigenvalue 4.





x 3 1

3 1
= x2 3x 1, so the eigenvalues are 1 = 1 (3 + 13),
(d) A =
; cA (x) =

2
1
x
1 0

1
2 = 2 (3 13). Thus the dominant eigenvalue is 1 = 12 (3 + 13). Since 1 2 = 1 and
1 + 2 = 3, we get






1 3

so a dominant eigenvector is

1
1

1
0

. We start with x0 =

Then xk+1 = Axk , k = 0, 1, . . . gives




4
x1 =
, x2 =
1

13
4

x3 =

43
13

1
1

x4 =

142
43

141

Section 8.6: Complex Matrices

These are approaching scalar multiples of the dominant eigenvector
xk xk+1
The Rayleigh quotients are rk =
:
xk 2
r0 = 2.5,

r1 = 3.29,

r2 = 3.30270,

1
1

3.302776
1

r3 = 3.30278.

These are rapidly approaching the dominant eigenvalue 1 = 3.302776.


x 3 1


3 1

= x2 3x 3; 1 = 1 3 + 13 = 3.302776 and
2(b) A =
; cA (x) =
2

1

2 = 12 3 13 = 0.302776. The QR-algorithm proceeds as follows:







3 1
3
1
10
1
1

A1 =
= Q1 R1 where Q1 = 10
, R1 = 10
1

A2 = R1 Q1 =

1
10

A3 = R2 Q2 =

1
109

33

360

33

= Q2 R2 where Q2 =


1
1090

3.302752

0.009174

0.009174

0.302752

33

33

3
1

, R2 =

1
1090

109

10

The diagonal entries already approximate 1 and 2 to 4 decimal places.

4. We prove that ATk = Ak for each k by induction in k. If k = 1, then A1 = A is symmetric by
T
hypothesis, so assume ATk = Ak for some k 1. We have Ak = Qk Rk so Rk = Q1
k Ak = Qk Ak
because Qk is orthogonal. Hence
Ak+1 = Rk Qk = QTk Ak Qk
so
ATk+1 = (QTk Ak Qk )T = QTk ATk QTk T = QTk Ak Qk = Ak+1 .
The eigenvalues of A are all real as A is symmetric, so the QR-algorithm asserts that the Ak
converge to an upper triangular matrix T. But T is symmetric (it is the limit of symmetric
matrices), so it is diagonal.

Exercises 8.6
1(b)

Complex Matrices


|1 i|2 + |1 + i|2 + 12 + (1)2 =

(1 + 1) + (1 + 1) + 1 + 1 =


(d) 4 + |i|2 + |1 + i|2 + |1 i|2 + |2i|2 =

4 + 1 + (1 + 1) + (1 + 1) + 4 =

13

2(b) Not orthogonal: (i, i, 2 + i), (i, i, 2 i) = i(i) + (i)(i) + (2 + i)(2 + i) = 3 + 4i
(d) Orthogonal: 4 + 4i, 2 + i, 2i), (1 + i, 2, 3 2i) = (4 + 4i)(1 i) + (2 + i)2 + (2i)(3 + 2i)
= (8i) + (4 + 2i) + (4 + 6i) = 0.
3(b) Not a subspace. For example, i(0, 0, 1) = (0, 0, i) is not in U.

142

Section 8.6: Complex Matrices

(d) If v = (v + w, v 2w, v) and w = (v + w , v 2w , v ) are in U then

v + w = ((v + v  ) + (w + w ), (v + v ) 2(w + w ), (v + v )) is in U
zv = (zv + zw, zv 2zw, zv) is in U
0 = (0 + 0, 0 20, 0) is in U.
Hence U is a subspace.
4(b) Here U = {(iv + w, 0, 2v w) | v, w C} = {v(i, 0, 2) + w(1, 0, 1) | v, w C}
= span{(i, 0, 2), (1, 0, 1)} .

If z(i, 0, 2) + t(1, 0, 1) = (0, 0, 0) with z, t C, then iz + t = 0, 2z t = 0. Adding gives

(2 + i)z = 0, so z = 0; and so t = iz = 0. Thus {(i, 0, 2), (1, 0, 1)} is independent over C,
and so is a basis of U. Hence dimC U = 2.
(d) U = {(u, v, w) | 2u + (1 + i)v iw = 0; u, v, w C)} . The condition is w = 2iu + (1 i)v,
so
U = {(u, v, 2iu + (1 i)v) | u, v C} = span {(1, 0, 2i), (0, 1, 1 i)} .
If z(1, 0, 2i) + t(0, 1, i 1) = (0, 0, 0) then components 1 and 2 give z = 0 and t = 0. Thus
{(1, 0, 2i), (0, 1, 1 i)} is independent over C, and so is a basis of U. Hence dimC U = 2.









2
3
2 3
2 3
1
5(b) A =
, AH = AT =
, A1 = 13
. Hence, A is not hermitian
3

(A = AH ) and not unitary (A1 = AH ). However, AAH = 13I = AH A, so A is normal.







T
1 i
1 i
1
i
H
T

=
= A. Thus A is hermitian and so is
(d) A =
, A = (A) =
i

AAH

A2

normal. But,
=
= 2I so A is not unitary.






1
1+i
1
1i
(f) A =
. Here A = AT so AH = A =
= A (thus A is not hermitian).
1+i
i
1i
i






3
2 2i
3
2 + 2i
H
H
Next, AA =
= I so A is not unitary. Finally, A A =
=
2+i

AAH , so A is not normal.


z
z
1
(h) A = 2|z|
. Here A =
z

2 2i

1
2|z|

so

AH

1
2|z|

. Thus A = AH

if and
if z 
= z; that is Ais hermitian if and only if z is real. We have AAH =

only
2
2 |z|
0
1
= I, and similarly, AH A = I. Thus it is unitary (and hence normal).
2
2|z|2

2 |z|

3i

x4

3 + i

= x2 5x 6 = (x + 1)(x 6).







2
5
3 + i
3+i 2

; an eigenvector is x1 =
.
3 i
1
0
0





3+i 
2
3 + i
2 3 + i
3i

; an eigenvector is x2 =
.
0
0
2
3 i
5




2
3i
As x1 and x2 are orthogonal and x1 = x2  = 14, U = 114
is unitary and
3+i
2



1 0
U H AU =
.

8(b) A =

, cA (x) =
3+i
1


Eigenvectors for 1 = 1 :


Eigenvectors for 2 = 6 :

3 i

x1

143

Section 8.6: Complex Matrices

x 2 1 i

= x2 5x + 4 = (x 1)(x 4).
(d) A =
; cA (x) =
1i
3
1 + i
x3









1
1 i
1 1+i
1+i
Eigenvectors for 1 = 1 :

; an eigenvector is x1 =
.
1 + i
2
0
0
1









2
1 i
1 + i 1
1
Eigenvectors for 2 = 4 :

; an eigenvector is x2 =
.
1 + i
1
0
0
1i




1+i
1
is unitary and
Since x1 and x2 are orthogonal and x1  = x2  = 3, U = 13
1
1i



1 0
U H AU =
.

1+i

(f) A =

1+i

1i

cA (x) =

x1
0

x1

1 + i

0
1 i
x2

Eigenvectors for 1 = 1 :

If 2 = 0 :

1
0

x3 =

0
1

1i

0
0

1 i

1 + i

Eigenvectors for 3 = 3 :

= (x 1)(x2 3x) = (x 1)x(x 3).

1 i

1 + i

1 + i

1 i

1+i

; an eigenvector is x2 =

1 + i

is orthogonal and U AU =

; an eigenvector is x1 =

1+i

1
3

0
0

; an eigenvector is

. Since {x1 , x2 , x3 } is orthogonal and x2  = x3  = 3, U =

1+i

1i

10(b) (1) If z = (z1 , z2 , . . . , zn ) then z2 = |z1 |2 + |z2 |2 + + |zn |2 . Thus z = 0 if and only if
|z1 | = = |zn | = 0, if and only if z = (0, 0, . . . , 0).

(2) By Theorem 1, we have Z, W  = Z, W  and Z, W  = Z,

W . Hence

Z2 = Z, Z = Z, Z = Z,

Z = ||2 Z2 .
Taking positive square roots gives Z = || Z .
11(b) If A is hermitian then A = AT . If A = [aij ] , the (k, k)-entry of A is a
kk , and the (k, k)-entry
of AT is akk . Thus, A = AT implies that a
kk = akk for each k; that is akk is real.
14(b) Let B be skew-hermitian, that is B H = B. Then Theorem 3 gives
(B 2 )H = (B H )2 = (B)2 = B 2 , so B 2 is hermitian
(iB)H = (i)B H = (i)(B) = iB, so iB is hermitian.

144

Section 8.7: An Application to Linear codes over Finite Fields

(d) If Z = A + B where AH = A and B H = B, then Z H = AH + B H = V B. Solving gives

Z + Z H = 2V and Z Z H = 2B, so V = 12 (Z + Z H ) and S = 12 (Z + Z H ). Hence the matrices
A and B are uniquely determined by the conditions Z = A+B, AH = A, B H = B, provided
such A and B exist. But always,
Z = 12 (Z + Z H ) + 21 (Z Z H )
and the matrices A = 12 (Z + Z H ) and B = 12 (Z Z H ) are hermitian and skew-hermitian
respectively:
AH = 12 (Z H + Z HH ) = 21 (Z H + Z) = A
B H = 12 (Z H Z HH ) = 21 (Z H Z) = B.
16(b) If U is unitary, then U 1 = U H . We must show that U 1 is unitary, that is (U 1 )1 = (U 1 )H .
But
(U 1 )1 = U = (U H )H = (U 1 )H .

18(b) If V =

then V is hermitian because V =

1
i

i
0

V T,

but iV =

i
1

1
0


is

not hermitian (it has a nonreal entry on the main diagonal).





0
1
a b
21(b) Given A =
, let U =
be invertible and real, and assume that U 1 AU =
1 0
c d








. Thus, AU = U
so
0

a + b

c + d

Equating first column entries gives c = a and a = c. Thus, a = (a) = a2 so

(1 + 2 )a = 0. Now is real (a and c are not both zero so either = ac or = ac ), so
1 + 2 = 0. Thus a = 0 (because (1 + 2 )a = 0) whence c = a = 0. This contradicts the
assumption that A is invertible.

Exercises 8.7

An Application to Linear Codes over Finite Fields

1(b) The elements with inverses are 1, 3, 7, 9 : 1 and 9 are self-inverse; 3 and 7 are inverses of each
other.
As for the rest, 2 5 = 4 5 = 6 5 = 8 5 = 0 in Z10 so 2, 5, 4, 6 and 8 do not have inverses
in Z10 .
(d) The powers of 2 computed in Z10 are: 2, 4, 8, 16 = 6, 32 = 2, ,so the sequence repeats: 2,
4, 8, 16, 2, 4, 8, 16, .
2(b) If 2a = 0 in Z10 then 2a = 10k for some integer k. Thus a = 5k so a = 0 or a = 5 in Z10 .
Conversely, it is clear that 2a = 0 in Z10 if a = 0 or a = 5.

145

Section 8.7: An Application to Linear codes over Finite Fields

3(b) We want a number a in Z19 such that 11a = 1. We could try all 19 elements in Z19 , the one
that works is a = 7. However the euclidean algorithm is a systematic method for finding a.
As in Example 2, first divide 19 by 11 to get
19 = 1 11 + 8.
Then divide 11 by 8 to get
11 = 1 8 + 3.
Now divide 8 by 3 to get
8 = 2 3 + 2.
Finally divide 3 by 2 to get
3 = 1 2 + 1.
The process stops here since a remainder of 1 has been reached. Now eliminate remainders
from the bottom up:
1 = 3 1 2 = 3 (8 2 3) = 3 3 8
= 3(11 1 8) 8 = 3 11 4 8
= 3 11 4(19 1 11) = 7 11 4 19.
Hence 1 = 7 11 4 19 = 7 11 in Z19 because 19 = 0 in Z19 .
1
1
6(b) Working
in Z7 , we have det

A =15

24 = 1 + 4 = 5 = 0, so A exists. Since 5 = 3 in Z7 ,
3
6
3 1
2 3
A1 = 3
=3
=
.
4

7(b) Gaussian elimination works over any field F in the same way that we have been using it over
R. In this case we have F = Z7 , and we reduce the augmented matrix of the system as follows.
We have 5 3 = 1 in Z7 so the first step in the reduction is to multiply row 1 by 5 in Z7 :















3 1 4 3
1 5 6 1
1 5 6 1
1 5 6 1
1 0 5 3

.
4

Hence x and y are the leading variables, and the non-leading variable z is assigned as a
parameter, say z = t. Then, exactly as in the real case, we obtain x = 3 + 2t, y = 1 + 4t,
z = t where t is arbitrary in Z7 .
9(b) If the inverse is a + bt then 1 = (1 + t)(a + bt) = (a b) + (a + b)t.This certainly holds if
a b = 1and a + b = 0.Adding gives 2a = 1,that is a = 1inZ3 ,that is a = 1 = 2.Hence
a + b = 0gives b = a = 1,so a + bt = 2 + t;that is (1 + t)1 = 2 + t.Of course it is easily
checked directly that (1 + t)(2 + t) = 1.
10(b) The minimum weight of C is 5, so it detects 4 errors and corrects 2 errors by Theorem 5.
11(b) The linear (5, 2)-code {00000, 01110, 10011, 11101} has minimum weight 3 so it corrects 1
error by Theorem 5.
12(b) The code is {0000000000, 1001111000, 0101100110, 0011010111,

1100011110, 1010101111, 0110110001, 1111001001}.

This has minimum distance 5 and so corrects 2 errors.

146

Section 8.8: An Application to Quadratic Forms

13(b) C = {00000, 10110, 01101, 11011} is a linear (5, 2)-code of minimal weight 3, so it corrects
single errors.





u
14(b) G = 1 u where u is any nonzero vector in the code. H =
.
In1

Exercises 8.8
1(b) A =

(d) A =

An Application to Quadratic Forms

1
(1
2

1
1
(1
2

+ 1)

2
1
(2
2

1
1
(4
2
1
(5
2

1)

+ 2)

1
1
(2
2

+ 0)

1
(1 + 5)
2
1
(0 2)
2

+ 4)

1)

1
3

x 1 2
= x2 2x 3 = (x + 1)(x 3)
2(b) q =
where A =
. cA (x) =
2 1
2
x1








2
2
1 1
1
1 = 3 :

; so an eigenvector is X1 =
.
2
2
0
0
1









2 2
1 1
1
2 = 1 :

; so an eigenvector is X2 =
.
2 2
1


0 0



1
1
3
0
is orthogonal and P T AP =
. As in Theorem 1, take
Hence, P = 12
1 1
0 1








1
1
x1
x1 + x2
= 12
. Then
Y = P T X = 12

X T AX

x2

y1 =

x1 x2

1 (x1
2

+ x2 ) and y2 =

1 (x1
2

x2 ).

Finally, q = 3y12 y22 , the index of q is 1 (the number of positive eigenvalues) and the rank of
q is 2 (the number of nonzero eigenvalues).

(d) q = X T AX where A =

1 = 9 :

2
4

2 = 9 :

16
4

8
8

8
1

x7

cA (x) = 4
4

x7

= 4
0

4
8
8

10
8

4
8

10

0
0

x1

x9

x1

x+7

. To find cA (x), subtract row 2 from row 3 :

0
0

2
0
0

x7
4
4

= 4
x1
8

0
x + 9 x 9



= (x 9)2 (x + 9)

; orthogonal eigenvectors
1

9
9

1
0

2
2
1

; eigenvector

2
1
2

1
2
2

. These

147

Section 8.8: An Application to Quadratic Forms

eigenvectors are orthogonal and each has length 3. Hence, P =

and P T AP =

1
3

. Thus

Y = PTX =

1
3

so

x1
x2
x3

1
3

1
2

2x1 + 2x2 x3

2x1 x2 + 2x3

x1 + 2x2 + 2x3

is orthogonal

y1 = 31 [2x1 + 2x2 x3 ]
y2 = 31 [2x1 x2 + 2x3 ]
y3 = 31 [x1 + 2x2 + 2x3 ]
will give q = 9y12 + 9y22 9y32 . The index of q is 2 and the rank of q is 3.

(f) q = X T AX where A =

1 = 9 :

2 = 0 :

5
2
4

2
8
2

2
5

cA (x) =




=

x5

4
x9

0
0

x5

0
4
x1

x8

x9
0

= 2
x

8

4
2



= x(x 9)2 .

x8

. To find cA (x), subtract row 3 from row 1:

x + 9
2

x5

; orthogonal eigenvectors are

18

18

1
1
0

2
2
1

and

orthogonal and P T AP =

. If

Y = PTX =
then

1
3

2
2

y1 = 13 (2x1 + 2x2 + x3 )
y2 = 13 (x1 + 2x2 2x3 )
y3 = 13 (2x1 + x2 + 2x3 )

1
2
2

x1
x2
x3

1
3

; an eigenvector is

These eigenvectors are orthogonal and each has length 3. Hence P =

1
2

is

148

Section 8.8: An Application to Quadratic Forms

gives q = 9y12 + 9y22 . The rank and index of q are both 2.

(h) q = X T AX where A =

1 = 2 :

2 = 1;

3 = 1 :

2
1
0

1
1
1

1
0

cA (x) =




=

x1

1
2

Hence,

x1
1

0
1

x1

0
1
1

is orthogonal and P T AP =

1
2

1
3
1
3

x1

P =

. To find cA (x), add row 3 to row 1:

0
1
2

Y = PTX =
then

1
2
2

0
0

;an eigenvector is

1
6

y2 =

1 (x1 + x2
3
1 (x1 + x3 )
2

y3 =

1 (x1
6

y1 =

; an eigenvector is

. If

3
1

1
0
1

1
1
1

; an eigenvector is

3
0

x1
x2
x3

1
2
1

1
2
1

+ x3 )

+ 2x2 x3 )

gives q = 2y12 + y22 y32 . Here q has index 2 and rank 3.






x
3
2
2
T
3(b) q = 3x 4xy = X AX where X =
,A=
. cA (t) =
y
2
0









1 2
1 2
2
1 = 4 :

; an eigenvector is
.
2

1
6
2
6
1
6

1
6

x1 0 x1


= 1
x
1

0
1 x 1



= (x 1)(x 2)(x + 1)

t3
2

2
t

= (t4)(t+1)

149

Section 8.8: An Application to Quadratic Forms

2 = 1 :

4
2

1
5

Hence, P =
x1 =

1 (2xy)
5

1
2
1


1

; an eigenvector is
.
2





4
0
gives P T AP =
. If Y = P T X =

1 (x+2y).
5

and y1 =

1 = 6 :

2 = 1 :

1
2

Hence, P =
1 (x + 2y),
5

2
1

2
4
1
5

t5

; an eigenvector is

; an eigenvector is

gives P T AP =

2 1
1 (2x y)
5

y1 =

x1
y1

, then

The equation q = 2 becomes 4x21 y12 = 2, a hyperbola.




x
(d) q = 2x2 + 4xy + 5y 2 = X T AX where X =
,A=
y


t 2 2
= (t 1)(t 6).
In this case cA (t) =

1
2

2
1

.
.

. If Y = P T X =

x1
y1

, then x1 =

and q = 1 becomes 6x21 + y12 = 1. This is an ellipse.






x
x1
are related to X =
by X = AX1
4. After the rotation, the new variables X1 =
y
y1



cos sin
where A =
(this is equation (**) preceding Theorem 2, or see Theorem 4
sin

cos

2.6). Thus x = x1 cos y1 sin and y = x1 sin + y1 cos . If these are substituted in the
equation ax2 + bxy + cy 2 = d, the coefficient of x1 y1 is
2a sin cos + b(cos2 sin2 ) + 2c sin cos = b cos 2 (a c) sin 2.
This is zero if is chosen so that
cos 2 =

b2

ac
+ (a c)2

Such an angle 2 exists because

and

2 ac 2
b +(ac)

2

7(b) The equation is X T AX+BX = 7 where X =

cA (x) =

t1
2
2

t3
0

2
0
t3

t1
2
0

sin 2 =

+
x1
x2
x3

b2

2b 2
b (ac)

, A =
2

t3

t3

2
0
t3

2

b
.
+ (a c)2

= 1.

2
0

t1
2
0

= (t 3)(t2 4t 5) = (t 3)(t 5)(t + 1).

1 = 3 :

2
2
2

1
0

; an eigenvector is

0
1
1

, B = [5 0
4

t3
0

2
0
t3

6] .

150

Section 8.10: An Application to Statistical Principal Component Analysis

2 = 5 :

4
2
2

3 = 1 :

Hence, P =

4
0

1
2
1
2

then

2
0
4

3
1
3
+ 1
3

2
6
1
6
1
6

Y =

=
y1
y2
y3

1
6

1
2

2
0

y2 =
y3 =

= PTX =
y1 =

1
6

; an eigenvector is

2
1
1

2
1
0

1
1
1

; an eigenvector is

satisfies P T AP =

3(x2 + x3 )

2(x1 + x2 x3 )
2x1 x2 + x3

2
1
1

. If

1 (x2 + x3 )
2
1
(x1 + x2 x3 )
3
1 (2x1 x2 + x3 ).
6

As P 1 = P T , we have X = P Y so substitution in X T AX + BX = 7 gives

Y T (P T AP )Y + (BP )Y = 7.


 
As BP = 16 6 3 11 2 4 = 3 2 113 3 2 3 6 , this is

3y12 + 5y22 y32 (3 2)y1 + ( 11

3)y2 + ( 23 6)y3 = 7.
3

9(b) We have A = U T U where U is upper triangular with positive diagonal entries. Hence
q(x) = xT U T Ux = (Ux)T (Ux) = Ux2
So take y = Ux as the new column of variables.

Exercises 8.9
Exercises 8.10

An Application to Constrained Optimization

An Application to Statistical Principal
Component Analysis

151

Section 9.1: The Matrix of a Linear Transformation

Chapter 9: Change of Basis

Exercises 9.1

1(b) CB (v) =
(d) CB (v) =

1
2

The Matrix of a Linear Transformation

because v = ax2 + bx + c = ax2 + (2b c)(x + 1) + (c b)(x + 2).

2b c
cb

ab

a+b

a + 3b + 2c
1
2

because

[(a b)(1, 1, 2) + (a + b)(1, 1, 1) + (a + 3b + 2c)(0, 0, 1)]




2
1
3
2
2(b) MDB (T ) = CD [T (1)] CD [T (x)] CD [T (x )] =
. Comparing columns gives
v = (a, b, c) =

CD [T (1)] =

2
1

CD [T (x)] =

1
0

CD [T (x )] =

3
2

Hence
T (1) = 2(1, 1) (0, 1) = (2, 1)
T (x) = 1(1, 1) + 0(0, 1) = (1, 1)
T (x2 ) = 3(1, 1) 2(0, 1) = (3, 1).
Thus
T (a + bx + cx2 ) = aT (1) + bT (x) + cT (x2 )
= a(2, 1) + b(1, 1) + c(3, 1)
= (2a + b + 3c, a + b + c).


'

CD T


'

1
= CD
0

3(b) MDB (T ) =

(d) MDB (T ) =

0
0
0

(

0
(

'

(
'

(
'

(
0 1
0 0
0 0
CD T
CD
CD T
0 0
1 0
0 1
(
(
(
'

'

'

0 0
0 1
0 0
CD
CD
CD
1


CD [T (1)] CD [T (x)] CD [T (x2 )] = CD (1) CD (x + 1) CD (x2 + 2x + 1)

152

Section 9.1: The Matrix of a Linear Transformation

4(b) MDB (T ) = [CD [T (1, 1)] CD [T (1, 0)]] = [CD (1, 5, 4, 1) CD (2, 3, 0, 1)] =

We have v = (a, b) = b(1, 1) + (a b)(1, 0) so CB (v) =

CD [T (v)] = MDB (T )CB (v) =

Finally, we recover the action of T :

ab

b
ab

. Hence,

2a b

3a + 2b
4b
a

T (v) = (2a b)(1, 0, 0, 0) + (3a + 2b)(0, 1, 0, 0) + 4b(0, 0, 1, 0) + a(0, 0, 0, 1)

= (2a b, 3a + 2b, 4b, a).


CD [T (1)] CD [T (x)] CD [T (x2 )]

(d) MDB (T ) =

= [CD (1, 0) CD (1, 0) CD (0, 1)]



1 1
21
2
2
=
1
1
1

2 2 2 
1 1 1
= 12
.
1 1
1

a

We have v = a + bx + cx2 so CB (v) =

. Hence

1
2

CD [T (v)] = MDB (T )CB (v) =

1
1

Finally, we recover the action of T :

a
b
c

1
2

a+bc

a+b+c

T (v) = 12 (a + b c)(1, 1) + 21 (a + b + c)(1, 1) = (a + b, c).

'

(
'

(
'

(
'

1 0
0 1
0 0
(f) MDB (T ) = CD T
CD T
CD T
CD T
0

= CD

We have v =

so CB (v) =

a
b
c
d

1

=a

CD

CD

CD

+d



+b

+c

. Hence CD [T (v)] = MDB (T )CB (v) =

0
0

a
b
c
d

a
b+c
b+c
d

(

153

Section 9.1: The Matrix of a Linear Transformation

Finally, we recover the action of T :





1 0
0
T (v) = a
+ (b + c)
0

+ (b + c)

+d

b+c

b+c

5(b) Have R3 R4 R2 . Let B, D, E be the standard bases. Then

MED (S) = [CE [S(1, 0, 0, 0)] CE [S(0, 1, 0, 0)] CE [S(0, 0, 1, 0)] CE S(0, 0, 0, 1)]]
= [CE (1, 0) CE (1, 0) CE (0, 1) CE (0, 1)]



1 1 0
0
=
0

MDB (T ) = [CD [T (1, 0, 0)] CD [T (0, 1, 0)] CD [T (0, 0, 1)]]

= [CD (1, 0, 1, 1) CD (1, 1, 0, 1) CD (0, 1, 1, 0)]

We have ST (a, b, c) = S(a + b, c + b, a + c, b a) = (a + 2b + c, 2a b + c). Hence

MEB (ST ) = [CE [ST (1, 0, 0)] CE [ST (0, 1, 0)]
= [CE (1, 2)] CE (2, 1) CE (1, 1)



1
2
1
=
.
2

With this we confirm Theorem 3 as follows:



1 1 0
0

MED (S)MDB (T ) =

CE [ST (0, 0, 1)]]

= MEB (ST ).

)
*
T
S
(d) Have R3 P2 R2 with bases B = {(1, 0, 0), (0, 1, 0), (0, 0, 1)} , D = 1, x, x2 ,
E = {(1, 0), (0, 1)} .


CE S(x)] CE [S(x2 )]

MED (S) = [CE [S(1)]

= [CE (1, 0) CE (1, 0) CE (0, 1)]


1 1 0
=
0

MDB (T ) = [CD [T (1, 0, 0)] CD [T (0, 1, 0)] CD [T (0, 0, 1)]]


= CD (1 x) CD (1 + x2 ) CD (x)

1
0

154

Section 9.1: The Matrix of a Linear Transformation


The action of ST is ST (a, b, c) = S (a b) + (c a)x + bx2 = (2a b c, b). Hence,
MEB (ST ) = [CE [ST (1, 0, 0)] CE [ST (0, 1, 0)] CE [ST (0, 0, 1)]]
= [CE (2, 0) CE (1, 1) CE (1, 0)]



2 1 1
=
0

Hence, we verify Theorem 3 as follows:


1 1 0

MED (S)MDB (T ) =
0

1
1
0

7(b) MDB (T ) = [CD [T (1, 0, 0)] CD [T (0, 1, 0)]

2
0

1
1

1
0

= MEB (ST ).

CD [T (0, 0, 1)]]

= [CD (0, 1, 1) CD (1, 0, 1) C(1, 1, 0)]

If T 1 (a, b, c) = (x, y, z) then (a, b, c) = T (x, y, z) = (y + z, x + z, x + y). Hence, y + z = a,

x + z = b, x + y = c. The solution is
T 1 (a, b, c) = (x, y, z) = 21 (b + c a, a + c b, a + b c).
Hence,


MBD (T 1 ) = CB T 1 (1, 0, 0)
CB T 1 (0, 1, 0)
CB T 1 (0, 0, 1)







= CB 12 , 12 , 12
CB 21 , 12 , 12
CB 21 , 12 , 12

1
2

1
1

This matrix is MDB (T )1 as Theorem 4 asserts.


(d) MDB (T ) = CD [T (1)] CD [T (x)] C T (x2 )

= [CD (1, 0, 0) CD (1, 1, 0) CD (1, 1, 1)]

If T 1 (a, b, c) = r + sx + tx2 , then (a, b, c) = T (r + sx + tx2 ) = (r + s + t, s + t, t). Hence,

r + s + t = a, s + t = b, t = c; the solution is t = c, s = b c, r = a b. Thus,
T 1 (a, b, c) = r + sx + tx2 = (a b) + (b c)x + cx2 .

Hence,


MBD (T 1 ) = CB T 1 (1, 0, 0)
CB T 1 (0, 1, 0)
CB T 1 (0, 0, 1)


= CB (1) CB (1 + x) CB (x + x2 )

1
1

1
1

This matrix is MDB (T )1 as Theorem 4 asserts.

155

Section 9.1: The Matrix of a Linear Transformation

'

8(b) MDB (T ) = CD T

(

'

CD T

(

'

CD T

(

'

CD T

= [CD (1, 0, 0, 0) CD (1, 1, 0, 0) CD (1, 1, 1, 0) CD (0, 0, 0, 1)]

= =

(

This is invertible and the matrix inversion algorithm (and Theorem 4) gives

MDB (T 1 ) = [MDB (T )]1 =

If v = (a, b, c, d) then


CB T 1 (v) = MDB (T 1 )CD (v) =

1
0

1
1

a
b
c
d

ab
bc
c

Hence, we get a formula for the action of T 1 :

T

(a, b, c, d) = T


=

(v) = (a b)

ab bc
.
c

+ (b c)

+c

+d

12. Since D = {T (e1 ), . . . , T (en )} , we have CD [T (ej )] = Cj = column j of In . Hence,

MDB (T ) = [CD [T (e1 )]
= [C1

C2

...

CD [T (e2 )]

...

CD [T (en )]]

Cn ] = In .

16(b) Define T : Pn Rn+1 by T [p(x)] = (p(a0 ), p(a1 ), . . . , p(an )), where a0 , . . . , an are fixed
distinct real numbers. If B = {1, x, . . . , xn } and D Rn+1 is the standard basis,



MDB (T )( = CD [T (1)] CD [T (x)] CD T (x2 )
. . . CD [T (xn )]

= CD (1, 1, . . . , 1) CD (a0 , a1 , . . . , an ) CD (a20 , a21 , . . . , a2n ) . . .

a0

a20

...

an
0

1
.
..

a1
.
..

a21
.
..

...
..
.

an
1
.
..

an

a2n

...

an
n

CD (an0 , an1 , . . . , ann )

Since the ai are distinct, this matrix has nonzero determinant by Theorem 7, 3.2. Hence, T
is an isomorphism by Theorem 4.

156

Section 9.1: The Matrix of a Linear Transformation

R

S,T

20(d) Assume that V W U. Recall that the sum S + T : W U of two operators is defined
by (S + T ) (w) = S(w) + T (w) for all w in W. Hence, for v in V :
[(S + T ) R] (v) = (S + T )[R(v)]
= S[R(v)] + T [R(v)]
= (SR)(v) + (T R)(v)
= (SR + T R)(v).
Since this holds for all v in V, it shows that (S + T )R = SR + T R.
21(b) If P and Q are subspaces of a vector space W, recall that P + Q = {p + q | p in P, q in Q} is
a subspace of W (Exercise 25, 6.4). Now let w be any vector in im(S + T ) . Then
w = (S + T )(v) = S(v) + T (v) for some v in V, whence w is in im S+ im T. Thus,
im(S + T ) im S + im T.
22(b) If T is in X10 , then T (v) = 0 for all v in X1 . As X X1 , this implies that T (v) = 0 for all v
in X; that is T is in X 0 . Hence, X10 X 0 .
24(b) We have R : V L(R, V ) defined by R(v) = Sv . Here Sv : R V is defined by Sv (r) = rv.
R is a linear transformation: The requirements that R(v + w) = R(v) + R(w) and
R(av) = aR(v) translate to Sv+w = Sv + Sw and Sav = aSv . If r is arbitrary in R:
Sv+w (r) = r(v + w) = rv + rw = Sv (r) + Sw (r) = (Sv + Sw )(r)
Sav (r) = r(av) = a(rv) = a [Sv (r)] = (aSv )(r).
Hence, Sv+w = Sv + Sw and Sav = aSv so R is linear.
R is one-to-one: If R(v) = 0 then Sv = 0 is the zero transformation R V. Hence we have
0 = Sv (r) = rv for all r; taking r = 1 gives v = 0. Thus ker R = 0.
R is onto: Given T in L(R, V ), we must find v in V such that T = R(v); that is T = Sv .
Now T : R V is a linear transformation and we take v = T (1). Then, for r in R:
Sv (r) = rv = rT (1) = T (r 1) = T (r).
Hence, Sv = T as required.
25(b) Given the linear transformation T : R V and an ordered basis B = {b1 , b2 , . . . , bn }
of V , write T (1) = a1 b1 + a2 b2 + + an bn where the ai are in R. We must show that
T = a1 S1 + a2 S2 + +an Sn where Si (r) = rbi for all r in R. We have
(a1 S1 + a2 S2 + + an Sn )(r) = a1 S1 (r) + a2 S2 (r) + + an Sn (r)
= a1 (rb1 ) + a2 (rb2 ) + + an (rbn )
= rT (1)
= T (r)
for all r in R. Hence a1 S1 + a2 S2 + + an Sn = T.

157

Section 9.2: Operators and Similarity

27(b) Given v in V, write v = r1 b1 + r2 b2 + + rn bn , ri in R. We must show that rj = Ej (v) for

each j. To see this, apply the linear transformation Ej :
Ej (v) = Ej (r1 b1 + r2 b2 + + rj bj + + rn bn )
= r1 Ej (b1 ) + r2 Ej (b2 ) + + rj Ej (bj ) + + rn Ej (bn )
= r1 0 + r2 0 + + rj 1 + + rn 0
= rj
using the definition of Ej .

Exercises 9.2

1(b) PDB

Operators and Similarity


= CD (x) CD (1 + x) CD (x2 ) =

32

1
1

1
2

1
2

x = 23 2 + 1(x + 3) + 0(x2 1)

because

1 + x = (1) 2 + 1(x + 3) + 0(x2 1)

x2 =

1
2

2 + 0(x + 3) + 1(x2 1).

Given v = 1 + x + x2 , we have

CB (v) =

0
1
1

and CD (v) =

21
1
1

because v = 0 x + 1(1 + x) + 1 x2 and v = 21 2 + 1 (x + 3) + 1(x2 1). Hence

PDB CB (v) =
as expected.
4(b) PBD

1
2

0
1
1

1
2


= CB (1 + x + x2 ) CB (1 x) CB (1 + x2 ) =


PDB = CD (1) CD (x) CD (x2 ) =
1=
x=
x2 =

1
3
1
3
1
3

1
3

2
2

because

= CD (v)
1
0


(1 + x + x2 ) + (1 x) (1 + x2 )


(1 + x + x2 ) 2(1 x) (1 + x2 )


(1 + x + x2 ) + (1 x) + 2(1 + x2 ) .

158

Section 9.2: Operators and Similarity

The fact that PDB = (PBD )1 is verified by multiplying these matrices. Next:

1
0
1


PED = CE (1 + x + x2 ) CE (1 x) CE (1 + x2 ) = 1 1 0
PEB


= CE (1) CE (x) CE (x2 ) =

)
*
where we note the order of the vectors in E = x2 , x, 1 . Finally, matrix multiplication
verifies that PED PDB = PEB .
5(b) Let B = {(1, 2, 1), (2, 3, 0), (1, 0, 2)} be the basis formed by the transposes of the columns
of A. Since D is the standard basis:

PDB = [CD (1, 2, 1) CD (2, 3, 0) CD (1, 0, 2)] =

Hence Theorem 2 gives

= A.

A1 = (PDB )1 = PBD = [CB (1, 0, 0) CB (0, 1, 0) CB (0, 0, 1)] =

because

(1, 0, 0) = 6(1, 2, 1) 4(2, 3, 0) + 3(1, 0, 2)

(0, 1, 0) = 4(1, 2, 1) + 3(2, 3, 0) 2(1, 0, 2)
(0, 0, 1) = 3(1, 2, 1) + 2(2, 3, 0) 1(1, 0, 2).
)
*
7(b) Since B0 = 1, x, x2 , we have
P = PB0 B


= CB0 (1 x2 ) CB0 (1 + x) CB0 (2x + x2 ) =


MB0 (T ) = CB0 [T (1)] CB0 [T (x)] CB0 T (x2 )


= CB0 (1 + x2 ) CB0 (1 + x) CB0 (x + x2 )

Finally


MB (T ) = CB T (1 x2 )
CB [T (1 + x)] CB T (2x + x2 )


= CB (1 x) CB (2 + x + x2 ) CB (2 + 3x + x2 )

159

Section 9.2: Operators and Similarity

because
1 x = 2(1 x2 ) + 3(1 + x) 2(2x + x2 )

2 + x + x2 = 3(1 x2 ) + 5(1 + x) 2(2x + x2 )

2 + 3x + x2 = 1(1 x2 ) + 3(1 + x) + 0(2x + x2 ).

The verification that P 1 MB0 (T )P = MB (T ) is equivalent to checking that

MB0 (T )P = P MB (T ), and so can be seen by matrix multiplication.













5
2
29 12
3 2
5 2
3 2
1
0
1
8(b) P AP =
=
=
.
7
3
70 29
7 5
7 3
7 5
0 1
'

(
3
2
Let B =
,
consist of the columns of P. These are eigenvectors of A correspond7

ing to the eigenvalues 1, 1 respectively. Hence,




 



3
2
MB (TA ) = CB TA
CB TA
= CB
7

CB

2
5

9(b) Choose a basis of R2 , say B = {(1, 0), (0, 1)} , and compute
MB (T ) = [CB [T (1, 0)]

CB [T (0, 1)]] = [CB (3, 2) CB (5, 3)] =





x 3 5

= x2 6x 1. Note that the calculation is easy
Hence, cT (x) = cMB (T ) (x) =
2
x3
because B is the standard basis, but any basis could be used.
)
*
(d) Use the basis B = 1, x, x2 of P2 and compute


MB (T ) = CB [T (1)] CB [T (x)] CB T (x2 )


= CB (1 + x 2x2 ) CB (1 2x + x2 ) CB (2 + x)

Hence,

(f) Use B =

'

2
1

cT (x) = cMB (T ) (x) =

1

x1

x+2

1
2

= x3 + x2 8x 3.






0 1
0 0
,
,
,
0

1
x

(

'

(
'

(
1 0
0 1
MB (T ) = CB T
CB T
0 0
0 0










1 0
0 1
1
= CB
CB
CB
1 0
0 1
1

1
0

1
0

1
0

x1

x+2

x + 3

x2

and compute
'

0
CB T
1



0
0
CB
0

0
0

(
1



'

CB T

(

160

Section 9.3: Invariant Subspaces and Direct Sums

Hence,




cT (x) = cMB (T ) (x) =



x1

= (x 1) 0
1

x1
0

1
0

x1

x+1

x+1

x+1

x+1

0
1
0

x1

x+1

= x4 .

12. Assume that A and B are both n n and that null A = null B. Define TA : Rn Rn by
TA (x) = Ax for all x in Rn ; similarly for TB . Then let T = TA and S = TB . Then ker S =
null B and ker T = null A so, by Exercise 28, 7.3 there is an isomorphism R : Rn Rn
such that T = RS. If B0 is the standard basis of Rn , we have
A = MB0 (T ) = MB0 (RS) = MB0 (R)MB0 (S) = U B
where U = MB0 (R). This is what we wanted because U is invertible by Theorem 4, 9.1.
Conversely, assume that A = U B with U invertible. If x is in null A then Ax = 0, so
U Bx = 0, whence Bx = 0 (because U is invertible), that is x is in null B. In other words null
A null B. But B = U 1 A so null B null A by the same argument. Hence null A = null
B.
16(b) We verify first that S is linear. Showing S(w + v) = S(w) + S(v) means showing that
MB (Tw+v ) = MB (Tw ) + MB (Tv ). If B = {b1 , b2 } then column j of MB (Tw+v ) is
CB [Tw+v (bj )] = CB [(w + v)bj ] = CB (wbj + vbj ) = CB (wbj ) + CB (vbj )
because CB is linear. This is column j of MB (Tw ) + MB (Tv ), which shows that S(w + v) =
S(w) + S(v). A similar argument shows that MB (Taw ) = a MB (Tw ), so S(aw) = a S(w), and
hence that S is linear.
To see that S is one-to-one, let S(w) = 0; by Theorem 2 7.2 we must show that w = 0. We
have MB (Tw ) = S(w) = 0 so, comparing jth columns, we see that CB [Tw (bj )] = CB [wbj ] = 0
for j = 1, 2. As CB is an isomorphism, this means that wbj = 0 for each j. But B is
a basis of C and 1 is in C, so there exist r and s in R such that 1 = rb1 + sb2 . Hence
w = w 1 = rwb1 + swb2 = 0, as required.
Finally, to show that S(wv) = S(w)S(v) we first show that Tw Tv = Twv . Indeed, given z
in C, we have
(Tw Tv )(z) = Tw (Tv (z)) = w(vz) = (wv)z = Twv (z).
Since this holds for all z in C, it shows that Tw Tv = Twv . But then Theorem 1 shows that
S(wv) = MB (Tw Tv ) = MB (Tw )MB (Tv ) = S(w)S(v).
This is what we wanted.

Exercises 9.3

Invariant Subspaces and Direct Sums

161

Section 9.3: Invariant Subspaces and Direct Sums

2(b) Let v T (U), say v = T (u) where u U. Then T (v) = T [T (u)] T (U) because T (u) U.
This shows that T (U) is T -invariant.
3(b) Given v in S(U), we must show that T (v) is also in S(U). We have v = S(u) for some u in
U. As ST = T S, we compute:
T (v) = T [S(u)] = (T S)(u) = (ST )(u) = S[T (u)].
As T (u) is in U (because U is T -invariant), this shows that T (v) = S[T (u)] is in S(U).
6. Suppose that a subspace U of V is T -invariant for every linear operator T : V V ; we must
show that either U = 0 or U = V. Assume that U = 0; we must show that U = V. Choose
u = 0 in U, and (by Theorem 1, 6.4) extend {u} to a basis {u, e2 , . . . , en } of V . Now let v be
any vector in V. Then (by Theorem 3, 7.1) there is a linear transformation T : V V such
that T (u) = v and T (ei ) = 0 for each i. Then v = T (u) lies in U because U is T -invariant.
As v was an arbitrary vector in V , it follows that V = U.
[Remark: The only place we used the hypothesis that V is finite dimensional is in extending
{u} to a basis of V. In fact, this is true for any vector space, even of infinite dimension.]
)
*
8(b) We have U = span 1 2x2 , x + x2 . To show that U is T -invariant, it suffices (by Example
3) to show that T (1 2x2 ) and T (x + x2 ) both lie in U. We have

T (1 2x2 ) = 3 + 3x 3x2 = 3(1 2x2 ) + 3(x + x2 )

(*)

T (x + x2 ) = 1 + 2x2 = (1 2x2 )

So both T (1 2x2 ))and T (x + x2 ), *so U is T -invariant. To get a block triangular matrix for
T extend the basis 1 2x2 , x + x2 of U to a basis B of V in any way at all, say
)
*
B = 1 2x2 , x + x2 , x2 .
Then, using (*), we have


MB (T ) = CB T (1 2x2 )


CB T (x + x2 )


CB T (x2 ) =

3
3
0

where the last column is because T (x2 ) = 1 + x + 2x2 = (1 2x2 ) + (x + x2 ) + 3(x2 ). Finally,




x 3 1 1
x3 1


= (x 3)(x2 3x + 3).



cT (x) = 3 x 1 = (x 3)
3
x
0

0 x3

9(b) Algebraic Solution. If U is TA -invariant and U = {0}, U = R2 , then dim U = 1. Thus

U = Ru where u = 0. Thus TA (u) is in Ru (because U is T -invariant), say TA (u) = ru, that
is Au = ru, whence (rI A) u = 0. But


r cos sin

= (r cos )2 + sin2 = 0 as sin = 0 (0 < < ).
det(rI A) =
sin
r cos
Hence, (rI A)u = 0 implies u = 0, a contradiction. So U = 0 or U = R2 .

162

Section 9.3: Invariant Subspaces and Direct Sums

Geometric Solution. If we view R2 as the euclidean plane, and U = 0, R2 , is a TA -invariant
subspace, then U must have dimension 1 and so be a line through the origin (Example 13
5.2). But TA is rotation through counterclockwise about the origin (Theorem 4 2.6), so
it will move the line U unless = 0 or = , contrary to our assumption that 0 < < . So
no such line U can exist.

10(b) If v is in U W, then v = (a, a, b, b) = (c, d, c, d) for some a, b, c, d. Hence a = c, a = d,

b = c and b = d. It follows that d = d so a = b = c = d = 0; that is U W = {0}. To see
that R4 = U + W, we have (after solving systems of equations)
(1, 0, 0, 0) = 12 (1, 1, 1, 1) + 12 (1, 1, 1, 1) is in U + W
(0, 1, 0, 0) = 12 (1, 1, 1, 1) + 12 (1, 1, 1, 1) is in U + W
(0, 0, 1, 0) = 12 (1, 1, 1, 1) + 12 (1, 1, 1, 1) is in U + W
(0, 0, 0, 1) = 12 (1, 1, 1, 1) + 12 (1, 1, 1, 1) is in U + W.

Hence, R4 = U + W. A simpler argument is as follows. As dim U = 2 = dim W, the subspace

U W has dimension 2 + 2 = 4 by Theorem 6. Hence U W = R4 because dim R4 = 4.





a a
c
d
(d) If A is in U W, then A =
=
for some a, b, c, d, whence a = b = c = d = 0.
b

Thus, U W = {0}. Thus, by Theorem 7

dim(U W ) = dim U + dim W = 2 + 2 = 4.

Since dim

M22 =
4, we have

U W
= M22. Again, as in (b), we could show directly that
1 0
0 1
0 0
0 0
each of
,
,
,
is in U + W.
0

14. First U is a subspace because 0E = 0, and AE = A and A1 E = A1 implies that

(A + A1 )E = AE + A1 E = A + A1 and (rA)E = r(AE) = rA for all r R.
Similarly, W is a subspace because 0E = 0, and BE = 0 = B1 E implies that we have
(B + B1 )E = BE + B1 E = 0 + 0 = 0 and (rB)E = r(BE) = r0 = 0 for all r R.

These calculations hold for any matrix E; but if E 2 = E we get M22 = U W. First
U W = {0} because X in U W implies X = XE because X is in U and XE = 0 because
X is in W, so X = XE = 0. To prove that U + W = M22 let X be any matrix in M22 . Then:
XE is in U
X XE is in W

because (XE)E = XE 2 = XE
because (X XE)E = XE XE 2 = XE XE = 0.

Hence X = XE + (X XE) where XE is in U and (X XE) is in W ; that is X is in U + W.

Thus M22 = U + W.
17 By Theorem 5 6.4, we have dim(U W )+ dim(U + W ) = dim U+ dim W = n by hypothesis.
So if U + W = V then dim(U + W ) = n, whence dim (U W ) = 0. This means that
U W = {0} so, since U + W = V, we have proved that V = U W.
18(b) First, ker TA is TA -invariant by Exercise 2. Now suppose that U is any TA -invariant subspace,
U = 0, U = R2 . Then dim U = 1, say U = Rp, p = 0. Thus p is in U so Ap = TA (p) is in U,
say Ap = p where is a real number. Applying A again, we get A2 p = Ap = 2 p. But
A2 = 0, so this gives 0 = 2 p. Thus 2 = 0, whence = 0 and Ap = p = 0. Hence p is in
ker TA , whence U ker TA . But dim U = 1 = dim(ker TA ), so U = ker TA .

163

Section 9.3: Invariant Subspaces and Direct Sums

20. Let B1 be
a basis of U and extend it (using Theorem 1 6.4) to a basis B of V . Then
MB1 (T ) Y
MB (T ) =
by Theorem 1. Since we are writing T1 for the restriction of T to
0

U, MB1 (T ) = MB1 (T1 ). Hence,

cT (x) = det[xI MB (T )] = det

xI MB1 (T )
0

xI Z

= det [xI MB1 (T1 )] det[xI Z] = cT1 (x) q(x)

where q(x) = det[xI Z].

22(b) We have T : P3 P3 given by T [p(x)] = p(x) for all p(x) in P3 . We leave it to the reader
to verify that T is a linear operator. We have
T 2 [p(x)] = T {T [p(x)]} = T [p(x)] = p((x)) = p(x) = 1P3 (p(x)).
Hence, T 2 = 1P3 . As in Example 10, let
U1 = {p(x) | T [p(x)] = p(x)} = {p(x) | p(x) = p(x)}
U2 = {p(x) | T [p(x)] = p(x)} = {p(x) | p(x) = p(x)} .
These are the subspaces of even and odd polynomials in P3 , respectively, and have bases
B1 = {1, x2 } and B2 = {x, x3 }. Hence, use the ordered basis B = {1, x2 ; x, x3 } of P3 . Then





MB1 (T )
0
I2
0
MB (T ) =
=
MB2 (T )

I2

as in Example 10. More explicitly,


MB (T ) = CB [t(1)] CB [T (x2 )] CB [T (x)] CB [T (x3 )]
= [CB (1) CB (x2 ) CB (x) CB (x3 )]

I2

I2

22(d) Here T 2 (a, b, c) = [(a + 2b + c) + 2((b + c) + (c), (b + c) c), (c)] = (a, b, c), so
T 2 = 2R3 .
Note that T (1, 1, 0) = (1, 1, 0), while T (1, 0, 0) = (1, 0, 0) and T (0, 1, 2) = (0, 1, 2).
Hence, in the notation of Theorem 10, let B1 = {(1, 1, 0)} and B2 = {(1, 0, 0), (0, 1, 2)}.
These are bases of U1 = R(1, 1, 0) and U2 = R(1, 0, 0) + R(0, 1, 2), respectively.


So if we

take B = {(1, 1, 0); (1, 0, 0), (0, 1, 2)} then MB1 (T ) = [1] and MB2 (T ) =


1
0
0
MB1 (T )
0
MB (T ) =
= 0 1 0 .
0

MB2 (T )

1
0

. Hence

164

Section 9.3: Invariant Subspaces and Direct Sums

23(b) Given v, T [v T (v)] = T (v) T 2 (v) = T (v) T (v) = 0, so v T (v) lies in ker T . Hence
v = (v T (v)) + T (v) is in ker T + im T for all v, that is V = ker T + im T . If v lies
in ker T im T, write v = T (w), w in V . Then 0 = T (v) = T 2 (w) = T (w) = v, so
ker T im T = 0.
25(b) We first verify that T 2 = T. Given (a, b, c) in R3 , we have
T 2 (a, b, c) = T (a + 2b, 0, 4b + c) = (a + 2b, 0, 4b + c) = T (a, b, c)
Hence T 2 = T . As in the preceding exercise, write
U1 = {v | T (v) = v}

and

U2 = {v | T (v) = 0} = ker(T ).

Then we claim that R3 = U1 U2 . To show R3 = U1 + U2 , observe that v = T (v) + [v T (v)]

for each v in R3 , and T (v) is in U1 [because T [T (v)] = T 2 (v) = T (v)] while v T (v) is in
U2 [because T [v T (v)] = T (v) T 2 (v) = 0]. Finally we show that U1 U2 = {0}. For if v
is in U1 U2 then T (v) = v and T (v) = 0 so certainly v = 0.
Next, we show that U1 and U2 are T -invariant. If v is in U1 then T (v) is also in U1
because T [T (v)] = T 2 (v) = T (v). Similarly U2 is T -invariant because, if v is in U2 , that is
T (v) = 0), then T [T (v)] = T 2 (v) = R(v) = 0; that is T (v) is also in U2 .
It is clear that T (a, b, c) = (a, b, c) if and only if b = 0; that is U1 = {(a, 0, c) | b, c in R},
so B1 = {(1, 0, 0), (0, 0, 1)} is a basis of U1 . Since T (v) = v for all v in U1 the restriction of T
to U1 is the identity transformation on U1 , and so has matrix I2 .
Similarly, T (a, b, c) = (0, 0, 0) holds if and only if a = 2b and c = 4b for some b, so
U1 = R(2, 1, 4) and B2 = {(2, 1, 4)} is a basis of U2 . Clearly the restriction of T to U2 is
the zero transformation, and so has matrix 02 a 1 1 matrix.

Finally then, B = B1 B2 = {(1, 0, 0), (0,

of R3 (since we have

0, 1); (2, 1, 4)} is
 a basis


MB1 (T )
0
I2
0
shown that R3 = U1 U2 ), so T has matrix
=
.
0

MB2 (T )

01

2 [v] = T [T (v)] = T [f(v)z] = f [f(v)z]z = f(v)f (z)z. This expression

29(b) We have Tf,z
f,z f,z
f,z
equals Tf,z (v) = f(v)z for all v if and only if

f(v)(z f(z)z) = 0
for all v. Since f = 0, f(v) = 0 for some v, so this holds if and only if
z = f(z)z.
As z = 0, this holds if and only if f(z) = 1.
30(b) Let be an eigenvalue of T . If A is in E (T ) then T (A) = A; that is U A = A. If we write
A = [p1 p2 . . . pn ] in terms of its columns p1 , p2 , . . . , pn , then U A = A becomes
U [p1
[Up1

p2

Up2

...
...

pn ] = [p1
Upn ] = [p1

p2 . . .
p2

pn ]
...

pn ] .

Comparing columns gives Upi = pi for each i; that is pi is in E (U) for each i. Conversely,
if p1 , p2 , . . . , pn are all in E (U) then Upi = pi for each i, so T (A) = U A = A as above.
Thus A is in E (T ).

Section 10.1: Inner Products and Norms

165

Chapter 10: Inner Product Spaces

Exercises 10.1

Inner Products and Norms

1(b) P 5 fails: (0, 1, 0), (0, 1, 0) = 1.

The other axioms hold. Write x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ) and z = (z1 , z2 , z3 ).
P 1 holds: x, y = x1 y1 x2 y2 + x3 y3 is real for all x, y in Rn .
P 2 holds: x, y = x1 y1 x2 y2 + x3 y3 = y1 x1 y2 x2 + y3 x3 = y, x.
P 3 holds: x + y, z = (x1 + y1 )z1 (x2 + y2 )z2 + (x3 + y3 )z3

= (x1 z1 x2 z2 + x3 z3 ) + (y1 z1 y2 z2 + y3 z3 ) = x, z + y, z.

P 4 holds: rx, y = (rx1 )y1 (rx2 )y2 + (rx3 )y3 = r(x1 y1 x2 y2 + x3 y3 ) = rx, y.
(d) P 5 fails: x 1, x 1 = 0 0 = 0
P 1 holds: p(x), q(x) = p(1)q(1) is real
P 2 holds: p(x), q(x) = p(1)q(1) = q(1)p(1) = q(x), p(x)

P 3 holds: p(x) + r(x), q(x) = [p(1) + r(1)]q(1) = p(1)q(1) + r(1)q(1)

= p(x), q(x) + r(x), q(x)

P 4 holds: rp(x), q(x) = [rp(1)]q(1) = r[p(1)q(1)] = rp(x), q(x)
(f) P 5 fails: Here f, f = 2f (0)f(1) for any f, so if f(x) = x 12 then f, f = 21
P 1 holds: f, g = f (1)g(0) + f(0)g(1) is real
P 2 holds: f, g = f (1)g(0) + f(0)g(1) = g(1)f(0) + g(0)f (1) = g, f 

P 3 holds: f + h, g = (f + h)(1)g(0) + (f + h)(0)g(1) = [f (1) + h(1)]g(0) + [f(0) + h(0)]g(1)

= [f(1)g(0) + f (0)g(1)] + [h(1)g(0) + h(0)g(1)] = f, g + h, g
P 4 holds: rf, h = (rf)(1)g(0) + (rf)(0)g(1) = [r f (1)]g(0) + [rf(0)]g(1)
= r[f(1)g(0) + f(0)g(1)] = rf, g

2. If  ,  denotes the inner product on V, then u1 , u2  is a real number for all u1 and u2 in
U. Moreover, the axioms P 1 P 5 hold for the space U because they hold for V and U is a
subset of V. So  ,  is an inner product for the vector space U.
.
.


3(b) f2 = cos2 x dx = 12 [1 + cos(2x)] dx = 12 x + 12 sin(2x) = . Hence f = 1 f is a
unit vector.





 1 1
3
1
1
2
T
(d) v = v, v = v
v = 3 1 v
= 17.
1
2
1
2
1



3
1
1

Hence v v = 17
is a unit vector in this space.
1

4(b) d(u, v) = u v = (1, 2, 1, 2) (2, 1, 1, 3) = (1, 1, 0, 1) = 3.


.
.

(d) f g2 = (1 cos x)2 dx = 32 2 cos x + 21 cos(2x) dx because we have cos2 (x) =


1
Hence f g2 = 23 x 2 sin(x) + 14 sin(2x) = 32 [ ()] = 3. Hence
2 [1 + cos(2x)].

d(f, g) = 3.

166

Section 10.1: Inner Products and Norms

8. The space Dn uses pointwise addition and scalar multiplication:

(f + g)(k) = f (k) + g(k) and (rf )(k) = rf(k)
for all k = 1, 2, . . . , n.
P 1. f, g = f(1)g(1) + f(2)g(2) + + f (n)g(n) is real
P 2. f, g = f(1)g(1) + f(2)g(2) + + f (n)g(n)) = g(1)f (1) + g(2)f(2) + + g(n)f (n)
P 3.

= g, f .

f + h, g = (f + h)(1)g(1) + (f + h)(2)g(2) + + (f + h)(n)g(n)

= [f (1) + h(1)]g(1) + [f(2) + h(2)]g(2) + + [f (n) + h(n)]g(n)
= [f (1)g(1) + f(2)g(2) + + f(n)g(n)] + [h(1)g(1) + h(2)g(2) + + h(n)g(n)]
= f, g + h, g
P 4. rf, g = (rf)(1)g(1) + (rf)(2)g(2) + + (rf )(n)g(n)
= [rf(1)]g(1) + [rf (2)]g(2) + + [rf(n)]g(n)
= r[f(1)g(1) + f (2)g(2) + + f(n)g(n)] = rf, g
P 5. f, f  = f (1)2 + f (2)2 + + f (n)2 0 for all f. If f, f  = 0 then
f (1) = f(2) = = f(n) = 0 (as the f(k) are real numbers) so f = 0.
12(b) We need only verify P 5. [P 1
P 4 hold for any symmetric matrix A by (the discussion
v1
:
preceding) Theorem 2.] If v =
v2

v, v = v Av = [v1 , v2 ]

5
3

3
2



v1
v2

= 5v12 6v1 v2 + 2v22


9 2
= 5 v12 65 v1 v2 + 25
v2 95 v22 + 2v22

2
= 5 v1 35 v2 + 15 v22


= 15 (5v1 3v2 )2 + v22 .

Thus, v, v 0 for all v; and v, v = 0 if and only if 5v1 3v2 = 0 = v2 ; that is if and only
if v1 = v2 = 0 (v = 0). So P 5 holds.



v1
(d) As in (b), consider v =
.
v2

v, v = [v1
=
=
=
=

v2 ]



v1
v2

3v12 + 8v1 v2 + 6v22

2
16 2
3(v12 + 83 v1 v2 + 16
9 v2 ) 3 v2


2
3 v1 + 34 v2 + 23 v22
1
3



(3v1 + 4v2 )2 + 2v22 .

+ 6v22

Thus, v, v 0 for all v; and (v, v) = 0 if and only if 3v1 + 4v2 = 0 = v2 ; that is if and only
if v = 0. Hence P 5 holds. The other axioms hold because A is symmetric.

167

Section 10.1: Inner Products and Norms

13(b) If A =

and a22

a11

a12

a21

a22

, then aij is the coefficient of vi wj in v, w. Here a11 = 1, a12 = 1 = a21 ,



1
1
= 2. Thus, A =
. Note that a12 = a21 , so A is symmetric.

(d) As in (b): A =

2
0

14. As in the hint, write x, y = xT Ay. Since A is symmetric, this satisfies axioms P 1, P 2, P 3
and P 4 for an inner product on Rn (and only P 2 requires that A be symmetric). Then it
follows that
0 = x + y, x + y = x, x + x, y + y, x + y, y = 2x, y for all x, y in Rn .
Hence x, y = 0 for all x and y in Rn . But if ej denotes column j of In , then
ei , ej  = eTi Aej is the (i, j)-entry of A. It follows that A = 0.

16(b) u 2v w, 3w v = 3u, w 6v, w 3w, w u, v + 2v, v + w, v

= 3u, w 5v, w 3 w2 u, v + 2 v2
= 3 0 5 3 3 3 (1) + 2 4
= 15
P2

P3

P2

20. (1) u, v + w = v + w, u = v, u + w, u = u, v + u, w

P2
P4
P2
(2) v, rw = rw, v = rw, v = rv, w
(1)

(3) By (1): v, 0 = v, 0 + 0 = v, 0 + v, 0. Hence v, 0 = 0. Now 0, v = 0 by P2.
(4) If v = 0 then v, v = 0, 0 = 0 by (3). If v, v = 0 then it is impossible that v = 0 by
P 5, so v = 0.

22(b) 3u 4v, 5u + v = 15u, u + 3u, v 20v, u 4v, v

= 15 u2 17u, v 4 v2 .

22(d) u + v2 = u + v, u + v = u, u + u, v + v, u + 4v, v

= u2 + 2u, v + v2 .

26(b) Here
)
*
W = w | w in R3 and v w = 0
= {(x, y, z) | x y + 2z = 0}
= {(s, s + 2t, t) | s, t in R}
= span B

where B = {(1, 1, 0), (0, 2, 1)} . Then B is the desired basis because B is independent
[In fact, if s(1, 1, 0) + t(0, 2, 1) = (s, s + 2t, t) = (0, 0, 0) then s = t = 0].
28. Write u = v w; we show that u = 0. We are given that
u, vi  = v w, vi  = v, vi  w, vi  = 0

168

Section 10.2: Orthogonal Sets of Vectors

for each i. As V = span{v1 , . . . , vn } , write u = r1 v1 + + rn vn , ri in R. Then
u2 = u, u = u, r1 v1 + + rn vn 
= r1 u, v1 ) + + rn u, vn 
= r1 0 + + rn 0
= 0.

Thus, u = 0, so u = 0.
29(b) If u = (cos , sin ) in R2 (with the dot product), then u = 1. If v = (x, y) the Schwarz
inequality (Theorem 4 10.1) gives
u, v2 u2 v2 1 v2 = v2 .
This is what we wanted.

Exercises 10.2

Orthogonal Sets of Vectors

1(b) B is an orthogonal set because (writing f1 =

e1 , e2  = [1 1 1]

f1 , f3  = [1 1 1]

f2 , f3  = [1 0 1]

1
1

0
1

and f3 =

= [3 1 3]

= [3 1 3]

6
1

, f2 =

6
1

1
0
1
1

6
1

= [1 0 1]

Thus, B is an orthogonal basis of V and the expansion theorem gives

1
6
1

=0

=0
1

6
1

= 0.

v, f1 
v, f2 
v, f3 
f3
2 f1 +
2 f2 +
f1 
f2 
f3 2
3a + b + 3c
ca
3a 6b + 3c
=
e1 +
e2 +
e3
7
2
42
1
= 14
{(6a + 2b + 6c)e1 + (7c 7a)e2 + (a 2b + c)e3 ].

v=

/

(d) Observe first that

c d



1 0
where f1 =
, f2 =
0

because



,

a
c

b

0

= aa + bb + cc + dd . Now write B = {f1 , f2 , f3 , f4 }








0 1
0
1
, f3 =
, f4 =
. Then B is orthogonal
d

f1 , f2  = 1 + 0 + 0 1 = 0

f1 , f3  = 0 + 0 + 0 + 0 = 0
f1 , f4  = 0 + 0 + 0 + 0 = 0

f2 , f3  = 0 + 0 + 0 + 0 = 0

f2 , f4  = 0 + 0 + 0 + 0 = 0
f3 , f4  = 0 + 1 + 0 + 1 = 0.

169

Section 10.2: Orthogonal Sets of Vectors

The expansion theorem gives
v, f1 
v, f2 
v, f3 
v, f4 
f4
2 f1 +
2 f2 +
2 f3 +
f1 
f2 
f3 
f4 2








a+d
ad
b+c
bc
=
f1 +
f2 +
f3 +
f4 .
2
2
2
2

v=

2(b) Write b1 = (1, 1, 1), b2 = (1, 1, 1), b3 = (1, 1, 0). Note that in the Gram-Schmidt algorithm
we may multiply each ei by a nonzero constant and not change the subsequent ei . This avoids
fractions.
f1 = b1 = (1, 1, 1)
b2 , e1 
f2 = b2
f1
e1 2
= (1, 1, 1) 46 (1, 1, 1)
= 31 (1, 5, 1); use e3 = (1, 5, 1) with no loss of generality
b3 , f1 
b3 , f2 
f3 = b3
f2
2 f1
e1 
e2 2
(3)
= (1, 1, 0) 36 (1, 1, 1)
(1, 5, 1)
(30)
1
[(10, 10, 9) (5, 5, 5) + (1, 5, 1)]
= 10
1
= 5 (3, 0, 2); use f3 = (3, 0, 2) with no loss of generality
So the orthogonal basis is {(1, 1, 1), (1, 5, 1), (3, 0, 2)}.
/



0
a b
a b
3(b) Note that
,
= aa + bb + cc + dd . For convenience write
c d
c d












1 1
1 0
1 0
1 0
b1 =
, b2 =
, b3 =
, b4 =
. Then:
0

f1 = b1 =

b2 , f1 
f1
f2 = b2
f1 2





1 0
1
23
=
1

For the rest of the algorithm, use f2 =

1
3

2
1

1
1

1
3

2
1


, the result is the same.

b3 , f1 
b3 , f2 
f3 = b3
f2
2 f1
f1 
f2 2








1 0
1 1
2
=
23
15
0 1
0 1



1
2
= 51
.
2

1
3

2
1

170

Section 10.2: Orthogonal Sets of Vectors

Now use f4 =

1
2

2
1

, the results are unchanged.

b4 , f1 
b4 , f2 
b4 , f3 
f4 = b4
f3
2 f1
2 f2
f1 
f2 
f3 2









1 0
1 1
1 2
1
=
13
15

0 0
0 1
3
1



1
0
1
=2
.
0

Use f4 =

orthogonal basis

'

1
10

1
2

2
1

for convenience. Hence, finally, the Gram-Schmidt algorithm gives the







(
1 1
1 2
1
2
1
0
,
,
,
.
0

4(b) f1 = 1

x, f1 
f1 = x 22 1 = x 1
f1 2
x2 , f2 
x2 , f1 
f

f2 = x2
f3 = x2
1
f1 2
f2 2
f2 = x

6(b) [x y

8/3
2

4/3
2/3

(x 1) = x2 2x + 23 .

w] is in U if and only if
x + y = [x y

w] [1 1 0 0] = 0.

Thus y = x and
U = {[x x z w] | x, z, w in R}
= span {[1 1 0 0] , [0 0 1 0] , [0 0 0 1]} .
Hence dim U = 3 and U = 1.
(d) If p(x) = a + bx + cx2 , p is in U if and only if
1 1
0 = p, x =
(a + bx + cx2 )x dx =
0

a
2

b
3

+ 4c .

Thus a = 2s + t, b)= 3s, c = 2t*where s and t are in R, so p(x) = (2s + t) 3sx 2tx2 .
Hence, U = span 2 3x, 1 2x2 and dim U = 2, dim U = 1.

a b
is in U if and ony if
(f)
c d
0=
0=
0=

/

/

/



,


,


,

0

0

0

=a+b
=a+c
= a + c + d.

171

Section 10.2: Orthogonal Sets of Vectors

U

The solution d = 0, b = c = a, so
1 and dim U = 3.





1 0
7(b) Write b1 =
, b2 =
0

'

a
a

, and b3 =

a
0


(

a in R = span{[]} . Thus dim U =

1
0

but not orthogonal. The Gram-Schmidt algorithm gives

If E3 =

f1 = b1 = []





b2 , f1 
1
1
1
0
2
f2 = b2
2 f1 =
1
1
0
f1 
b3 , f1 
b3 , f2 
f3 = b3
f2
2 f1
f1 
f2 2









1 1
1 0
1
1
1
2
=
2
4
0 0
0 1
1 1



0
1
= 21
.

. Then {b1 , b2 , b3 } is independent

0
1

then {E1 , E2 , E3 } is an orthogonal basis of U. If A =

then

A, E1 
A, E2 
A, E3  
E3
E
+
E
+
1
2
E1 2
E2 2
E3 2









1 0
1
1
0
1
4
4
2
=2
+4
+ 2
0 1
1 1
1 0



3 0
=

projU (A) =

is the vector in U closest to A.

8(b) We are given U = span{1, 1 + x2 }, and applying the Gram-Schmidt algorithm gives an
orthogonal basis consisting of 1 and
2
3
1 + x2 , 1
(1 + 02 )1 + (1 + 12 )1 + (1 + 22 )1
2
(1 + x )
1 = (1 + x2 )
= 35 + x2
2
1
+
1
+
1
1
We use U = span{1, 5 3x2 }. Then Theorem 8 asserts that the closest vector in U to x is
2
3
x, 5 3x2
x, 1
2
2
3
1+
(5 3x2 ) = 33 + 12
projU (x) =
78 (5 3x ) = 13 (1 + 2x ).
12
5 3x2 2
2
3
Here, for example x, 5 3x2 = 0(5) + 1(2) + 2(7) = 12, and the other calculations are
similar.
.1
9(b) {1, 2x 1} is an orthogonal basis of U because 1, 2x 1 = 0 (2x 1) dx = 0. Thus
projU (x2 + 1) =
=

x2 + 1, 1
x2 + 1, 2x 1
1
+
(2x 1)
12
2x 12
3/4
1 1

= x + 56 .

1/6
1/3 (2x 1)

172

Section 10.2: Orthogonal Sets of Vectors

Hence, x2 + 1 = (x + 56 ) + (x2 x + 16 ) is the required decomposition. Check: x2 x +
in U because
1 1
2 2
3
 2

x x + 61 , 1 =
x x + 16 dx = 0
0
1 1
3
 2

2 2
x x + 16 (2x 1)dx = 0.
x x + 61 , 2x 1 =

1
6

is

11(b) We have v + w, v w = v, v v, w + w, v u, u = v2 w2 . But this means
that v + w, v u = 0 if and only if v = w. This is what we wanted.
14(b) If v is in U then v, u = 0 for all u in U. In particular, v, ui  = 0 for 1 i n, so v is in
{u1 , . . . , um } . This shows that U {u1 , . . . , um } . Conversely, if v is in {u1 , . . . , um }
then v, ui  = 0 for each i. If u is in U, write u = r1 u1 + + rm um , ri in R. Then
v, u = v, r1 u1 + + rm um 
= r1 v, u1  + + rm v, um 
= r1 0 + + rm 0
= 0.
As u was arbitrary in U, this shows that v is in U ; that is {u1 , . . . , um } U .
18(b) Write e1 = (3, 2, 5) and e2 = (1, 1, 1), write B = {e1 , e2 } , and write U = span B. Then
B is orthogonal and so is an orthogonal basis of U. Thus if v = (5, 4, 3) then
v e1
v e2
e2
2 e1 +
e1 
e2 2
6
= 38
38 (3, 2, 5) + 3 (1, 1, 1)
= (5, 4, 3)
= v.

projU (v) =

Thus, v is in U. However, if v1 = (1, 0, 2) then

v1 e1
v e2
e2
2 e1 +
e2 
e2 2
7
= 38
(3, 2, 5) + 33 (1, 1, 1)
1
= 38
(17, 24, 73).

projU (v1 ) =

As v1 = projU (v1 ), v1 is not in U by (a).

'

(
nw
19(b) The plane is U = {x | x n = 0}, so span n w, w
n U. Since
n2
'

(
nw
dim U = 2, it suffices to show that B = n w, w
n is independent. These
n2
two vectors are orthogonal (because (n w) n = 0 = (n w) w). Hence B is orthogonal
(and so independent) provided eash of the vectors is nonzero. But: n w = 0 because n
nw
and w are not parallel, and w
n is nonzero because w and n are not parallel, and
n2
nw
n (w
n) = 0.
n2

173

Section 10.3: Orthogonal Diagonalization

20(b) CE (bi ) is column i of P . Since CE (bi ) CE (bj ) = bi , bj  by (a), the result follows.

23(b) Let V be an inner product space, and let U be a subspace of V. If U = span{f1 , ..., fm }, then
m
m


v, fi 2
v, fi 
2 =
f
by
Theorem
7
so
proj
(v)
by Pythagoras theorem.
projU (v) =
i
U
fi 2
fi 2
i=1
i=1
So it suffices to show that projU (v)2 v2 .
Given v in V, write v = u + w where u = projU (v) is in U and w is in U . Since u and
w are orthogonal, Pythagoras theorem (again) gives
v2 = u2 + w2 u2 = projU (v)2 .
This is what we wanted.

Exercises 10.3

Orthogonal Diagonalization

1(b) If B = {E1 , E2 , E3 , E4 } where E1 =

, E2 =
, E3 =
0 0
0 0



0 0
and E4 =
, then B is an orthonormal basis for M22 and
0

T (E1 ) =
T (E2 ) =
T (E3 ) =
T (E4 ) =

0
0

0
1
1

= E1 + E3
= E2 + E4

= E1 + 2E3
= E2 + 2E4 .

Hence,
MB (T ) = [CB [T (E1 )] CB [T (E2 )] CB [T (E3 )] CB [T (E4 )]]

1
0

1
0

As MB (T ) is symmetric, T is a symmetric operator.

4(b) If T is symmetric then v, T (w) = T (v), w holds for all v and w in V. Given r in R:
v, (rT )(w) = v, rT (w) = rv, T (w) = rT (v), w = rT (v), w = (rT )(v), w
for all v and w in V. This shows that rT is symmetric.
(d) Given v and w, write T 1 (v) = v1 and T 1 (w) = w1 . Then
T 1 (v), w = v1 , T (w1 ) = T (v1 ), w1  = v, T 1 (w).
This shows that T 1 is a symmetric operator.

174

Section 10.3: Orthogonal Diagonalization

5(b) If E = {e1 = (1, 0, 0), e2 = (0, 1, 0), e3 = (0, 0, 1)} is the standard basis of R3 :
ME (T ) = [CE [T (e1 )] CE [T (e2 )] CE [T (e3 )]]
= [CE (7, 1, 0) CE (1, 7, 0) CE (0, 0, 2)]

x7

1
0



= (x 6)(x 8)(x 2) so the eigenvalues are 1 = 6,
x7
0
Thus, cT (x) = 1

0
0
x2
2 = 8, and
Corresponding (orthogonal) eigenvectors
3= 2, (real
as M

B0 (T ) is symmetric).
1

are x1 =

1
0

, x2 =

1
0

, and x3 =

1
2

1
1
0

0
1

, so

, 1
2

1
1
0

0
0
1

is an orthonormal basis of eigenvectors of ME (T ) These vectors are equal to CE



CE 12 (1, 1, 0) , and CE [(0, 0, 1)] respectively, so
+

1 (1, 1, 0)
2

1 (1, 1, 0), 1 (1, 1, 0), (0, 0, 1)

2
2

is an orthonormal basis of eigenvectors of T .

)
*
(d) If B0 = 1, x, x2 then


MB0 (T ) = CB0 [T (1)] CB0 [T (x)] CB0 [T (x2 )]


= CB0 (1 + x2 ) CB0 (3x) CB0 (1 x2 )

Hence, cT (x) =

x+1

x3

1
0

x+1

= x(x 3)(x + 2) so the (real) eigenvalues are 1 = 3,




0

2 = 0, 3 = 2. Corresponding (orthogonal) eigenvectors are X1 =

, so

1
0

, X2 =

0
1

1
1
1 , 0 , 0 is an orthonormal basis of eigenvectors of
X3 = 0
2
2

1
0
1
1




MB0 (T ). These have the form CB0 (x), CB0 12 (1 + x2 ) , and CB0 12 (1 x2 ) , respectively,
so
+
,
x, 12 (1 + x2 ), 12 (1 x2 )

is an orthonormal basis of eigenvectors of T .

175

Section 10.4: Isometries

7(b) Write A =

and compute:









1 0
0 0
MB (T ) = CB T
CB T
CB T
0 0
1 0













a 0
b 0
0 a
= CB
CB
CB
CB
c 0
d 0
0 c



CB T



 

Hence,
'



(
xI
0
A 0
cT (x) = det[xI MB (T )] = det

0
xI
0 A



xI A
0
= det
= det(xI A) det(xI A) = [cA (x)]2 .
xI A

12(2) We prove that (1)(2). If B = {f1 , , fn } is an orthonormal basis of V, then MB (T ) = [aij ]

where aij = fi , T (fj ) by Theorem 2. If (1) holds then aji = fj , T (fi ) = T (fj ), fi  =
fi , T (fj ) = aij . Hence [MV (T )]T = MV (T ), proving (2).
14(c) We have

MB (T  ) = CB [T  (f1 )] CB [T  (f2 )]

Hence, column j of MB (T  ) is

CB (T  (fj )) =

fj , T (f1 )

fj , T (f2 )
..
.
fj , T (fn )


CB [T  (fn )] .

by the definition of T  . Hence the (i, j)-entry of MB (T  ) is fj , T (fi ). But this is the (j, i)-entry
of MB (T ) by Theorem 2. Thus, MB (T  ) is the transpose of MB (T ).

Exercises 10.4
2(b) We have T

a
b

Isometries


a
b

1
0

0
1



a
b

so T has matrix

1
0

0
1

, which is orthog-

onal. Hence T is an isometry, and det T = 1 so T is a rotation by Theorem

4. In fact,


 T
cos sin
is counterclockwise rotation through . (Rotation through has matrix
by
sin

cos

Theorem 4 2.6; see also the discussion following Theorem 3). This can also be seen directly
from the diagram.

176

Section 10.4: Isometries

Y
a
b

X
a a
T =
b b

(d) We have T

a
b

0
1

1
0



a
b

so T has matrix

0
1

1
0

. This is orthog-

onal, so T is an isometry. Moreover, det T = 1 so T is a reflection by Theorem 4. In fact,

T is reflection in the line y = x by Theorem 5 2.6. This can also be seen directly from the
diagram.
Y
a
b

y = x
a b
T =
b a

(f) If B0 is the standard basis of R2 , then




 
1
0
MB0 (T ) = CB0 T
CB0 T
0
1






 
1
1
1
1

= CB0
CB0
2
2
1
1



1 1
= 12
.
1

Hence, det T = 1 so T is a rotation. Indeed, (the discussion following) Theorem 3 shows that
T is a rotation through an angle where cos = 12 , sin = 12 ; that is = 4 .

177

Section 10.4: Isometries

3(b) T

a
b
c

cT (x) =

1
2

3c a

3a + c

2b

1
2
23

x+

0
x

3
2
21

1
2

1
2
23

x+

a
b
c

3
2
x2 12

, so T has matrix

3
2
2
x 12

1
2
3
2

x+

1
2

. Thus,



= (x1) x2 + 3 x + 1 .

2

Hence,
we are in (1) of Table 8.1 so T is a rotation about the line Re with direction vector

e=

3(d) T

a
b
c

, where e is an eigenvector corresponding to the eigenvalue 1.

a
b
c

, so T has matrix

. This is orthog-

onal,so T is an isometry. Since cT (x) = (x 1)(x + 1)2 , we are in case (4) of Table 1. Then
1

e=

0
0

is an eigenvector corresponding to 1, so T is a rotation of about the line Re with

direction vector e, that is the x-axis.

3(f) T

1
2

Hence,

cT (x) =

a+c

2b

ca

1
2

=
0

x+1

1
2

1
2

0
x

1
2

2
0

1
0
1

= (x + 1)

, so T has matrix

b
c

1
2

1
2

Thus we are in case (2) of Table 1. Now e =

0
1
0

1
x

2
1
2

1
2

1
0
1

2
0

1
0
1

= (x + 1)(x2 2x + 1).

is an eigenvector corresponding to the



eigenvalue 1, so T is rotation of 3
4 about the line Re (the y-axis) followed by a reflection
in the plane (Re) the x-z plane.
6. Let T be an arbitrary isometry, and let a be a real number. If aT is an isometry then Theorem
2 gives
v = (aT )(v) = a(T (v)) = |a| T (v) = |a|v holds for all v.
Thus |a| = 1 so, since a is real, a = 1. Conversely, if a = 1 then |a| = 1 so we have
(aT )(v) = |a| T (v) = 1 T (v) = v for all v. Hence aT is an isometry by Theorem 2.
12 (b) Assume that S = Su T where u is in V and T is an isometry of V. Since T is onto (by
Theorem 2), let u = T (w) where w V. Then for any v V, we have
(T Sw )(v) =
T (w + v) = T (w) + T (v) = ST (w) (T (v)) = (ST (w) T )(v).
Since this holds for all v V, it follows that T Sw = ST (w) T.

178

Section 10.5: An Application to Fourier Approximation

Exercises 10.5

An Application to Fourier Approximation

The integrations involved in the computation of the Fourier coefficients are omitted in 1(b), 1(d),
and 2(b).
1(b) f5 =


cos x +

cos 3x
32

cos 5x
52


+ sin x sin22x + sin33x


2x
4x
6x
8 cos
+ cos
+ cos
22 1
42 1
62 1

(d) f5 =
2(b)

sin 4x
4

sin 5x
5


cos x +

cos 3x
32

cos 5x
52

4. We use the formula that cos ( ) = cos cos sin sin, so that

2 cos cos = cos ( )cos ( + ). Hence:

1
1
1
{cos[(k E)x] + cos[(k + E)x]} dx
cos(kx) cos(Ex)dx = 2
0

0

sin[(k E)x]
1 sin[(k + E)x]
= 2
+
k+E
(k E)
0
= 0 if k = E.

179

Section 11.1: Block Triangular Form

Chapter 11: Canonical Forms

Exercises 11.1

Block Triangular Form

x + 5 3 1 x + 1 x 1 0 x + 1

0
0









x 2 1 =
4
x2
1 =
4
x + 2 1 = (x + 1)3 .
1(b) cA (x) = 4
4
4
4
3
x
3
x
1
x
Hence, 1 = 1 and we are in case k = 1 of the triangulation algorithm.

I A =

p11

1
1

p12


2 .
(I

A)
Hence, {p11 , p12 } is a basis of null(I
A).
We
now
expand
this
to
a
basis
of
null


However, (I A)2 = 0 so null (I A)2 = R3 . Hence,
inthis case, we expand {p11 , p12 }
0

R3 ,

to any basis {p11 , p12 , p13 } of

P = [p11

p12

say by taking P13

p13 ] =

0
1

. Hence

satisfies P 1 AP =

1
0
0

as may be verified.



x+3
x+3

1
0
1
0






x+1
3
(d) cA (x) = 4 x + 1 3 = 4
4
0
2
x + 1 x 1

x 4
x+3

1
0



= 4 x 2 3 = (x 1)2 (x + 2).
0
0
x1
Hence 1 = 1, 3 = 2, and we are in case k = 2 of the triangulation algorithm.

I A=

3
0

p11 =

1
4
4

(I A)12 =

12

12

3
3

1
0
0

p11 =

1
4
4


Thus, null(I A) = span{p11 } . We enlarge {p11 } to a basis of null (I A)2

12

p12 =

0
1
2

Thus, null (I A)2 = span{p11 , p12 } . As dim[G1 (A)] = 2 in this case (by Lemma 1), we
have G1 (A) = span{p11 , p12 } . However, it is instructive to continue the process:

(I A)2 = 3

1
1
1

180

Section 11.1: Block Triangular Form

whence

(I A)3 = 9

1
1

= 3(I A)2 .

This continues to give (I  A)4 = 32 (I A)2, . . . , and in general (I A)k = 3k2 (I A)2 for
k 2. Thus null (I A)k = null (I A)2 for all k 2, so


G1 (A) = null (I A)2 = span {p11 , p12 } .

as we expected. Turning to 2 = 2 :

2I A =

0
3

p21 =

1
1
1

Hence, null[2I A] = span{p21 } . We need go no further with this as {p11 , p12 , p21 } is a
basis of R3 . Hence

P = [p11

p21 ] =

p12

1
1
1

satisfies P 1 AP =

as may be verified.
(f) To evaluate cA (x), we begin by adding column 4 to column 1:




cA (x) =

x + 1 6 3

0
2
x 3 2 2
x 3 2

=
0
1
3
x
1
3
x



1
1
2
x
x+1
1
2



x 3 2 2
x3



= (x + 1) 3
x
1 = (x + 1) 3
5
5
1
x+2


x 3 2
= (x + 1)2 (x 1)2 .
= (x + 1)1

x+3

1
x

2
x

x+1
0
0

x 1
x+1

x+1

x3

2
1

= (x + 1)

x+2
x3

x+1

x+1

Hence, 1 = 1, 2 = 1 and we are in case k = 2 of the triangulation algorithm. We omit

the details of the row reductions:

I A =

(I A)2 =

13

23

13

13

12

10

0
0

1
0

p11

p11

0
0
1

1
0
0
1

p12 =

1
0
1
0

181

Section 11.2: Jordan Canonical Form

We have dim [G1 (A)] = 2 as 1 = 1 has multiplicity 2 in cA (x), so G1 (A) = span{P11 , P12 } .
Turning to 2 = 1 :

I A=

(I A)2 =

1
1

2
0

p21 =

p21

Hence, G2 (A) = span{p21 , p22 } using Lemma 1. Finally, then

P = [p11 , p12 , p21 , p22 ] =

0
0

as may be verified.

2
1

5
2
1

gives P 1 AP =

p22 =

0
0
0

1
1

0
0

4. Let B be any basis of V and write A = MB (T ). Then cT (x) = cA (x) and this is a polynomial:
cT (x) = a0 + a1 x + + an xn for some ai in R. Now recall that MB : L(V, V ) Mnn
is an isomorphism of vector spaces (Exercise 26, 9.1) with the additional property that
MB (T k ) = MB (T )k for k 1 (Theorem 1 9.2). With this we get
MB [cT (T )] = MB [a0 1V + a1 T + + an T n ]
= a0 MB (1V ) + a1 MB (T ) + + an MB (T )n
= a0 I + a1 A + + an An
= cA (A)
=0
by the Cayley-Hamilton theorem. Hence cT (T ) = 0 because MB is one-to-one.

Exercises 11.2

2.

Jordan Canonical Form

, and

is invertible.

182
.

Section 11.2: Jordan Canonical Form

183

Appendix A: Complex Numbers

APPENDICES
Exercises A

Complex Numbers

1(b) 12 + 5i = (2 + xi)(3 2i) = (6 + 2x) + (4 + 3x)i. Equating real and imaginary parts gives
6 + 2x = 12, 4 + 3x = 5, so x = 3.
(d) 5 = (2 + xi)(2 xi) = (4 + x2 ) + 0i. Hence 4 + x2 = 5, so x = 1.

2(b) (3 2i)(1 + i) + |3 + 4i| = (5 + i) + 9 + 16 = 10 + i.

(d)

3 2i 3 7i

1i
2 3i

=
=
=

(3 2i)(1 + i) (3 7i)(2 + 3i)

(1 i)(1 + i)
(2 3i)(2 + 3i)
5 + i 27 5i

1+1
4+9
11
23
26 + 26 i

(f) (2 i)3 = (2 i)2 (2 i) = (3 4i)(2 i) = 2 11i

(h) (1 i)2 (2 + i)2 = (2i)(3 + 4i) = 8 6i
3(b) iz + 1 = i + z 6i + 3iz = 5i + (1 + 3i)z. Hence 1 + 5i = (1 + 2i)z, so
z=

(1 + 5i)(1 2i)
11 + 3i
1 + 5i
=
=
=
1 + 2i
(1 + 2i)(1 2i)
1+4

11
5

+ 35 i.

(d) z 2 = 3 4i. If z = a + bi the condition is (a2 b2 ) + (2ab)i = 3 4i, whence a2 b2 = 3

4
2
4
2
and ab = 2. Thus b = 2
a , so a a2 = 3. Hence a 3a 4 = 0. This factors as
(a2 4)(a2 + 1) = 0, so a = 2, whence b = 1. Finally, z = a + bi = (2 i).
(f) Write z = a + bi. Then the condition reads
(a + bi)(2 i) = (a bi + 1)(1 + i)
(2a + b) + (2b a)i = (a + 1 + b) + (a + 1 b)i.
Thus 2a + b = a + 1 + b and 2b a = a + 1 b; whence a = 1, b = 1, so z = 1 + i.




4(b) x = 21 (1) (1)2 4 = 12 1 i 3

(d) x =

1
4


(5)


(5)2 4 2 2 =

1
4

5 9 = 2, 21 .

5(b) If x = rei then x3 = 8 becomes r3 e3i = 8ei . Thus r3 = 8 (whence r = 2) and 3 = +2k.
Hence = 3 + k 2
3 , k = 0, 1, 2. The roots are
2ei/3 = 1 +
2ei = 2

5i/3

2e

3i (k = 0)

(k = 1)

= 1 3i (k = 2).

184

Appendix A: Complex Numbers

(d) If x = rei then x4 = 64 becomes r4 e4i = 64ei0 . Hence r4 = 64 (whence r = 2 2) and

4 = 0 + 2k; = k 2 , k = 0, 1, 2, 3. The roots are

2 2e0i = 2 2
(k = 0)
i/2

2 2e
= 2 2i (k = 1)

i
2 2e = 2 2 (k = 2)

2 2e3i/2 = 2 2i (k = 3).
6(b) The quadratic is (x u)(x u
) = x2 (u + u
)x + u u
= x2 4x + 13. The other root is
u
= 2 + 3i.
(d) The quadratic is (x u)(x u
) = x2 (u + u
)x + u u
= x2 6x + 25. The other root is
u
= 3 + 4i.
8. If u = 2 i, then u is a root of (x u)(x u
) = x2 (u + u
)x + u u
= x2 4x + 5.
2
If v = 3 2i, then v is a root of (x v)(x v) = x (v + v)x + v
v = x2 6x + 13.
Hence u and v are roots of
(x2 4x + 5)(x2 6x + 13) = x4 10x3 + 42x2 82x + 65.
10(b) Taking x = u = 2: x2 + ix (4 2i) = 4 2i 4 + 2i = 0. If v is the other root then
u + v = i (i is the coefficient of x) so v = u i = 2 i.
(d) Taking x = u = 2 + i: (2 + i)2 +3(1 i)(2 + i) 5i
= (3 ri) + 3(1 + 3i) 5i
= 0.
If v is the other root then u + v = 3(1 i), so v = 3(1 i) u = 1 + 2i.



11(b) x2 x+(1i) = 0 gives x = 21 1 1 4(1 i) = 21 1 3 + 4i . Write w = 3 + 4i

so w2 = 3 + 4i. If w = a + bi then w2 = (a2 b2 ) + (2ab)i, so a2 b2 = 3, 2ab = 4. Thus
b = a2 , a2 a42 = 3, a4 + 3a2 4 = 0, (a2 + 4)(a2 1) = 0, a = 1, b = 2, w = (1 + 2i).
Finally the roots are 12 [1 w] = 1 + i, i.




(d) x2 3(1 i)x 5i = 0 gives x = 12 3(1 i) 9(1 i)2 + 20i = 21 3(1 i) 2i . If

w = 2i then w2 = 2i. Write w = a + bi so (a2 b2 ) + 2abi = 2i. Hence a2 = b2 and ab = 1;

the solution is a = b = 1 so w = (1 + i). Thus the roots are x = 12 (3(1 i) w) = 2 i,
1 2i.

12(b) |z 1| = 2 means that the distance from z to 1 is 2. Thus the graph is the circle, radius 2,
center at 1.
(d) If z = x + yi, then z =
z becomes x + yi = x + yi. This holds if and only if x = 0; that is
if and only if z = yi. Hence the graph is the imaginary axis.
(f) If z = x + yi, then im z = m re z becomes y = mx. This is the line through the origin with
slope m.
18(b) 4i = 4e3i/2 .

185

Appendix A: Complex Numbers

(d) 4 + 4 3i = 4 1 + 3 = 8 and cos =

4 + 4 3i = 8e2i/3 .

4
8

(f) |6 + 6i| = 6 1 + 1 = 6 2 and cos =

6 + 6i = 6 2e3i/4 .

19(b) e7i/3 = e(/3+2)i = ei/3 = cos 3 + i sin 3 =

(d)

1
2.

6 2

1
2

3
2 i.

i/4   
   1
= 2 2
2e
= 2 cos 4 + i sin
4

Thus =

1 .
2

1

  
 
+
i
sin
=
2
(f) 2 3e2i/6 = 2 3 cos
3 2
3
3
20(b) (1 +

4
3i)
= (2ei/3 )4 = 24 e4i/3
=
=

1
16
1
16

1 i
2

3,

so =

Thus =

2
3 .

and we have

so =

3
4 ;

whence

= 1 i.


3
2 i

3 3i.

[cos(4/3) + i sin(4/3)]


12 + 23 i

1
= 32
+

3
32 i.

i/4 10
2e
= ( 2)10 e5i/2 = ( 2)10 e(/22)i

 
 
+ i sin
= ( 2)10 ei/2 = 25 cos
2
2

(d) (1 i)10 =

= 32(0 i) = 32i.

5 
(f) ( 3 i)9 (2 2i)5 = 2ei/6 ]9 2 2ei/4

= 29 e3i/2 (2 2)5 e5i/4



= 29 (i)25 ( 2)4 2 12 + 12 i
= 216 i(1 + i)

= 216 (1 + i).

23(b) Write z = rei . Then z 4 = 2( 3i 1) becomes r4 e4i = 4e2i/3 . Hence r4 = 4, so r = 2, and

4 = 2
3 + 2k; that is
=

+ 2 k

k = 0, 1, 2, 3.

The roots are

i/6
2e
4i/g
2e
7i/6
2e
10i/6
2e


 3 1  2 
2 2 + 2i = 2
3+i




= 2 21 + 23 i = 22 1 + 3i




= 2 23 12 i = 22 1 + 3i




= 2 12 23 i = 22 1 + 3i
=

186

Appendix A: Complex Numbers

k=2

k =1

/6

k =3
k =4

(d) Write z = r ei . Then z 6 = 64 becomes r6 e6i = 64ei . Hence r6 = 64, so r = 2, and

6 = + 2k; that is = 6 + 3 k where k = 0, 1, 2, 3, 4, 5. The roots are thus z = 2e/6+/3k
for these values of k. In cartesian form they are

k
z

3+i

1
2i

3+i

3i

4
2i

3i

26(b) Each point on the unit circle has polar form ei for some angle . As the n points are equally
spaced, the angle between consecutive points is 2
n . Suppose the first point into the first
i
2i/n
quadrant is z0 = e . Write w = e
. If the points are labeled z1 , z2 , z3 , . . . , zn around the
unit circle, they have polar form

z1 = ei
z2 = e(+2/n)i = ei e2i/n = z1 w
z3 = e[+2(2/n)]i = ei e4i/n = z1 w2
z4 = e[+3(2/n)]i = ei e6i/n = z1 w3
..
.
zn = e[+(n1)(2/n)]i = eai e2(n1)i/n = z1 wn1 .

187

Appendix B: Proofs

z4

z3
z2
2 / n
2 / n
2 / n

z1

Hence the sum of the roots is

z1 + z2 + + zn = z1 (1 + w + + wn1 ).

(*)


n
Now wn = e2i/n = e2i = 1 so

0 = 1 wn = (1 w)(1 + w + w2 + + wn1 ).

As w = 1, this gives 1 + w + + wn1 = 0. Hence (*) gives

z1 + z2 + + zn = z1 0 = 0.

Exercises B

Proofs

1(b) (1). We are to prove that if the statement m is even and n is odd is true then the statement
m + n is odd is also true.
If m is even and n is odd, they have the form m = 2p and n = 2q + 1, where p and q are
integers. But then m + n = 2(p + q) + 1 is odd, as required.
(2). The converse is false. It states that if m + n is odd then m is even and n is odd; and a
counterexample is m = 1, n = 2.
(d) (1). We are to prove that if the statement x2 5x + 6 = 0 is true then the statement x = 2
or x = 3 is also true.
Observe first that x2 5x + 6 = (x 2)(x 3). So if x is a number satisfying x2 5x + 6 = 0
then (x 2)(x 3) 0 so either x = 2 or x = 3. [Note that we are using an important fact
about real numbers: If the product of two real numbers is zero then one of them is zero.]
(2). The converse is true. It states that if x = 2 or x = 3 then x satisfies the equation
x2 5x + 6 = 0. This is indeed the case as both x = 2 or x = 3 satisfy this equation.

188

Appendix C: Mathematical Induction

2(b) The implication here is p q where p is the statement n is any odd integer, and q is the
statement n2 = 8k + 1 for some integer k. We are asked to either prove this implication or
give a counterexample.
This implication is true. If p is true then n is odd, say n = 2t + 1 for some integer t. Then
= (2t)2 + 2(2t) + 1 = 4t(t + 1) + 1. But t(t + 1) is even (because t is either even or odd),
say t(t + 1) = 2k where k is an integer. Hence n2 = 4t(t + 1) + 1 = 4(2k) + 1, as required.

n2

3(b) The implication here is p q where p is the statement n + m = 25, where n and m are
integers, and q is the statement one of m and n is greater than 12 is also true. We are asked
to either prove this implication by the method of contradiction, or give a counterexample.
The implication is true. To prove it by contradiction, we assume that the conclusion q
is false, and look for a contradiction. In this case assuming that q is false means both n 12
and m 12. But then n + m 24, contradicting the hypothesis that n + m = 25. So the
statement is true by the method of proof by contradiction.
The converse is false. It states that q p, that is if one of m and n is greater than 12
then n + m = 25. But n = 13 and m = 13 is a counterexample.
(d) The implication here is p q where p is the statement mn is even, where n and m are
integers, and q is the statement m is even or n is even. We are asked to either prove this
implication by the method of contradiction, or give a counterexample.
This implication is true. To prove it by contradiction, we assume that the conclusion q
is false, and look for a contradiction. In this case assuming that q is false means that m and
n are both odd. But then mn is odd (if either were even the product would be even). This
contradicts the hypothesis, so the statement is true by the method of proof by contradiction.
The converse is true. It states that if m or n is even then mn is even, and this is true (if
m or n is a multiple of 2, then mn is a multiple of 2).
4(b) The implication here is: x is irrational and y is rational x + y is irrational.

To argue by contradiction, assume that x + y is rational. Then x = (x + y) y is the

difference of two rational numbers, and so is rational, contrary to the hypothesis that x is
irrational.

5(b) At first glance the statement does not appear to be an implication. But another way to say
it is that if the statement n 2 is true then the statement n3 2n is also true.
This is not true. In fact, n = 10 is a counterexample because 103 = 1000 while 210 = 1024.
It is worth noting that the statement n3 2n does hold for 2 n < 9.

Exercises C

Mathematical Induction

6. Write Sn for the statement

1
1
1
n
+
+ +
=
.
12 23
n(n + 1)
n+1

(Sn )

1
1
Then S1 is true: It reads 12
= 1+1
, which is true. Now assume Sn is true for some n 1.
We must use Sn to show that Sn+1 is also true. The statement Sn+1 reads as follows:

1
1
1
n+1
+
+ +
=
.
12 23
(n + 1)(n + 2)
n+2

189

Appendix C: Mathematical Induction

The second last term on the left side is

1
n(n+1)

so we can use Sn :


1
1
1
1
1
1
1
+
+ +
=
+
+ +
+
12 23
(n + 1)(n + 2)
12 23
n(n + 1)
(n + 1)(n + 2)
1
n
+
=
n + 1 (n + 1)(n + 2)
n(n + 2) + 1
=
(n + 1)(n + 2)
(n + 1)2
=
(n + 1)(b + 2)
n+1
=
.
n+2
Thus Sn+1 is true and the induction is complete.
14. Write Sn for the statement

1
1
1
+ + + 2 n 1.
(Sn )
n
1
2

Then S1 is true as it asserts that 11 2 1 1, which is true. Now assume that Sn is true
for some n 1. We must use Sn to show that Sn+1 is also true. The statement Sn+1 reads
as follows:



1
1
1
1
1
1
1
+ + +
= + + + +
n
n+1
n+1
1
2
1
2


1
2 n1 +
n+1

2
2 n +n+1

=
1
n+1
2(n + 1)
1
<
n+1

= 2 n+11

where, at the second last step, we used the fact that n2 + n < (n + 1)this follows by
showing that n2 + n < (n + 1)2 , and taking positive square roots. Thus Sn+1 is true and the
induction is complete.
18. Let Sn stand for the statement
n3 n is a multiple of 3.
Clearly S1 is true. If Sn is true, then n3 n = 3k for some integer k. Compute:
(n + 1)3 (n + 1) = (n3 + 3n2 + 3n + 1) (n + 1)
= 3k + 3n2 + 3n

which is clearly a multiple of 3. Hence Sn+1 is true, and so Sn is true for every n by induction.

190

Appendix C: Mathematical Induction

20. Look at the first few values: B1 = 1, B2 = 5, B3 = 23, B4 = 119, . If these are compared
to the factorials: 1! = 1, 2! = 4, 3! = 6, 4! = 24, 5! = 120, , it is clear that Bn = (n+1)!1
holds for n = 1, 2, 3, 4 and 5. So it seems a reasonable conjecture that
Bn = (n + 1)! 1

for

n 1.

Sn

This certainly holds for n = 1 : B1 = 1 = 2! 1. If this is true for some n 1, then

Bn+1 = [1 1! + 2 2! + + n n!] + (n + 1)(n + 1)!
= [(n + 1)! 1] + (n + 1)(n + 1)!
= (n + 1)!{1 + (n + 1)} 1
= (n + 1)!{n + 2} 1
= (n + 2)! 1.
Hence Sn+1 is true and so the induction goes through.
Note that many times mathematical theorems are discovered by experiment", somewhat
as in this example. Several examples are worked out, a pattern is observed and formulated,
and the result is proved (often by induction).
22(b) If we know that Sn Sn+8 then it is enough to verify that S1 , S2 , S3 , S4 , S5 , S6 , S7 , and S8
are all true. Then
S1

S9

S17 S25

S2 S10 S18 S26

S3 S11 S19 S27
S4 S12 S20 S28
S5 S13 S21 S29
S6 S14 S22 S30
S7 S15 S23 S31
S8 S16 S24 S32
Clearly each Sn will appear in this array, and so will be true.

Chapter 5:
Q1. Let U = {u1 = (1, 1, 2, 5), u2 = (4, 1, 1, 1), u3 = (7, 28, 5, 5)}. Using
Maple, verify if the following subset of R4 is independent. Verify if U contains
orthogonal vectors. If so, normalize them.
> with(LinearAlgebra)
> u1 := V ector([1, 1, 2, 5])
> u2 := V ector([4, 1, 1, 1])
> u3 := V ector([7, 28, 5, 5])
> A :=< u1|u2|u3 >
> LU Decomposition(A, output =0 R0 )
> DotP roduct(u1, u2)
> DotP roduct(u1, u3)
> DotP roduct(u2, u3)
> normu1 := (1/N orm(u1, 2)) u1
> normu2 := (1/N orm(u2, 2)) u2
> normu3 := (1/N orm(u3, 2)) u3

1
4
2
4 3
5
1
5
2 1

Q2. Let A =
4 5 19 10 3. Find a bases for the row space and
0 11 10 9 1
column space of A. Determine the rank of A.
> with(LinearAlgebra)
> A := M atrix([[1, 4, 2, 4, 3], [5, 1, 5, 2, 1], [4, 5, 19, 10, 3], [0, 11, 10, 9, 1]])
> RowSpace(A)
(Returns the row space of A)
> N ullSpace(A)
(Returns the null space of A)
> Rank(A)
(Returns the rank of A)

Chapter 6:
Q1. Let V = M2,2 . Determine if the following subset U of V is independent:

 
 

1 0
1 0
1 1
{
,
,
}
0 1
1 0
1 1
Does v =


1
5


4
lie in U ?
3

> with(LinearAlgebra)
> e1 := V ector([1, 0, 0, 1])
> e2 := V ector([1, 0, 1, 0])
> e3 := V ector([1, 1, 1, 1])
> v := V ector([1, 4, 5, 3])
> A :=< e1|e2|e3 >
> LU Decomposition(A, output =0 R0 )
> newA :=< e1|e2|e3|v >
> LU Decomposition(newA, output =0 R0 )
Q2. Let V = P3 . Using Maple, determine if the following set of polynomials,
{1, 1 + x3 , x + x2 , 1 + x + x3 }, spans V .
> with(LinearAlgebra)
> e1 := V ector([1, 0, 0, 0])
> e2 := V ector([1, 0, 0, 1])
> e3 := V ector([0, 1, 0, 1])
> e4 := V ector([1, 1, 0, 1])
> A :=< e1|e2|e3|e4 >
> Rank(A)

Chapter 5:
Q1. Let U = {u1 = (16, 5, 9, 4), u2 = (2, 11, 7, 14), u3 = (3, 10, 6, 15)}. Using
Matlab, verify if the following subset of R4 is independent. Is b = (13, 8, 12, 1)
is the span of U ? Verify if U contains orthogonal vectors.
>> u1 = [16; 5; 9; 4]
>> u2 = [2; 11; 7; 14]
>> u3 = [3; 10; 6; 15]
>> b = [13, 8, 12, 1]
>> A =[u1 u2 u3]
>> rref (A)
>> aug =[A b]
>> rref (aug)
>> u10 u2
>> u10 u3
>> u20 u3

4 0 0
1 0 1
Q2. Let A = 3 4 1 and B = 0 1 0. Using Matlab, compute the
0 0 2
0 0 2
rank and trace of A and B to verify if these two matrices are similar. Using
theorem 3 in section 5.5, find the eigenvalues and eigenvectors of A to determine
if it is diagonalizable. If yes, then find the matrix P such that P 1 AP = D.
>> A =[-1 0 1; -3 4 1; 0 0 2]
>> B =[4 0 0; 0 1 0; 0 0 2]
>> rankA = rank(A)
(Returns the rank of A)
>> rankB = rank(B)
(Returns the rank of B)
>> traceA = trace(A)
(Returns the trace of A)
>> traceA = trace(B)
(Returns the trace of B)
>> [P, D] = eig(A)
>> test = inv(P ) A P

Chapter 11:
Q3. Let T : R3 R3 be defined by T (x, y, z) = (14y + z, 5y, 4x13y + 4z). Use
MATLAB to find the Jordan canonical form of the transformation matrix, A,
with respect to the standard basis. Verify that P transforms A in Jordan form
by showing that P 1 AP = D.
Input the matrix A as follows
>> A =[ 0 14 1; 0 5 0; -4 13 4]
>> [P, J] = jordan(A)
(Returns the invertible matrix P and the Jordan blocks)
>> inv(P ) A P
Q2. Let T : R5 R5 be defined by T (a, b, c, d, e) = (3a + b + 2c + 3d +
4e, 3b + c + 2d + 3e, 3c + d + 3e, 3d + e, e). Use MATLAB to find the Jordan
canonical form of the transformation matrix, A, with respect to the standard
basis. Verify that P transforms A in Jordan form by showing that P 1 AP = D..
>> A =[3 1 2 3 4; 0 3 1 2 3; 0 0 3 1 2; 0 0 0 3 1; 0 0 0 0 3]
>> [P, J] = jordan(A)
>> inv(P ) A P

11