0% found this document useful (0 votes)
250 views110 pages

Selected Solutions To Hoffman and Kunze's Linear Algebra Second Edition

Uploaded by

PARAMITA DUTTA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
250 views110 pages

Selected Solutions To Hoffman and Kunze's Linear Algebra Second Edition

Uploaded by

PARAMITA DUTTA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 110

Selected Solutions to Hoffman and Kunze’s

Linear Algebra Second Edition

Greg Kikola

August 28, 2020


ii

Copyright © 2020 Greg Kikola.


License CC BY-SA 4.0:
Creative Commons Attribution-ShareAlike 4.0 International.

This document lives at:


https://www.gregkikola.com/projects/guides/
You can find the LATEX source code on GitHub at:

https://github.com/gkikola/sol-hoffman-kunze
Contents

Preface v

1 Linear Equations 1
1.2 Systems of Linear Equations . . . . . . . . . . . . . . . . . . . . . 1
1.3 Matrices and Elementary Row Operations . . . . . . . . . . . . . 5
1.4 Row-Reduced Echelon Matrices . . . . . . . . . . . . . . . . . . . 11
1.5 Matrix Multiplication . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6 Invertible Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2 Vector Spaces 27
2.1 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3 Bases and Dimension . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 Computations Concerning Subspaces . . . . . . . . . . . . . . . . 50

3 Linear Transformations 55
3.1 Linear Transformations . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 The Algebra of Linear Transformations . . . . . . . . . . . . . . 63
3.3 Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.4 Representation of Transformations by Matrices . . . . . . . . . . 74
3.5 Linear Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.6 The Double Dual . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.7 The Transpose of a Linear Transformation . . . . . . . . . . . . . 101

iii
iv CONTENTS
Preface

This is an unofficial solution guide to the book Linear Algebra, Second Edition,
by Kenneth Hoffman and Ray Kunze. It is intended for students who are study-
ing linear algebra using Hoffman and Kunze’s text. I encourage students who
use this guide to first attempt each exercise on their own before looking up the
solution, as doing exercises is an essential part of learning mathematics.
In writing this guide, I have avoided using techniques or results before the
point at which they are introduced in the text. My solutions should therefore
be accessible to someone who is reading through Hoffman and Kunze for the
first time.
Given the large number of exercises, errors are unavoidable in a work such
as this. I have done my best to proofread each solution, but mistakes will get
through nonetheless. If you find one, please feel free to tell me about it via
email: [email protected]. I appreciate any corrections or feedback.
Please know that this guide is currently unfinished. I am slowly working on
adding the remaining chapters, but this will be done at my own pace. If you
need a solution to an exercise that I have not included, try typing the problem
statement into a web search engine such as Google; it is likely that someone else
has already posted a solution.
This guide is licensed under the Creative Commons Attribution-ShareAlike
4.0 International License. To view a copy of this license, visit

http://creativecommons.org/licenses/by-sa/4.0/
I am deeply grateful to the authors, Kenneth Hoffman and Ray Kunze, for
producing a well-organized and comprehensive book that is a pleasure to read.

Greg Kikola
www.gregkikola.com
[email protected]

v
vi PREFACE
Chapter 1

Linear Equations

1.2 Systems of Linear Equations


1.2.1 Exercise 1
Verify that the set of complex numbers described in Example 4 is a subfield of
C.
Solution.
√ The set in Example 4 consisted of all complex numbers of the form
x + y 2, where x and y √are rational. Call this set
√ F. √
Note that
√ 0 = 0 + 0 2 ∈ F and 1 = 1 + 0 2 ∈ F . If α = a + b 2 and
β = c + d 2 are any elements of F , then

α + β = (a + c) + (b + d) 2 ∈ F,
and √
−α = −a − b 2 ∈ F.
We also have
√ √
αβ = ac + ad 2 + bc 2 + 2bd

= (ac + 2bd) + (ad + bc) 2 ∈ F
and, provided α is nonzero,

−1 1 a−b 2 a b √
α = √ = 2 2
= 2 2
− 2 2
2 ∈ F.
a+b 2 a − 2b a − 2b a − 2b
Since F contains 0 and 1 and is closed under addition, multiplication, additive
inverses, and multiplicative inverses, F is a subfield of C.

1.2.2 Exercise 2
Let F be the field of complex numbers. Are the following two systems of linear
equations equivalent? If so, express each equation in each system as a linear
combination of the equations in the other system.
x1 − x2 = 0 3x1 + x2 = 0
2x1 + x2 = 0 x1 + x2 = 0

1
2 CHAPTER 1. LINEAR EQUATIONS

Solution. The systems are equivalent. For the first system, we can write

x1 − x2 = (3x1 + x2 ) − 2(x1 + x2 ) = 0,
2x1 + x2 = 21 (3x1 + x2 ) + 12 (x1 + x2 ) = 0.

And for the second system,

3x1 + x2 = 31 (x1 − x2 ) + 43 (2x1 + x2 ) = 0,


x1 + x2 = − 31 (x1 − x2 ) + 32 (2x1 + x2 ) = 0.

1.2.3 Exercise 3
Test the following systems of equations as in Exercise 1.2.2.

−x1 + x2 + 4x3 = 0 x1 − x3 = 0
x1 + 3x2 + 8x3 = 0 x2 + 3x3 = 0
1 5
2 x1 + x2 + 2 x3 =0

Solution. For the first system, we have

−x1 + x2 + 4x3 = −(x1 − x3 ) + (x2 + 3x3 ) = 0,


x1 + 3x2 + 8x3 = (x1 − x3 ) + 3(x2 + 3x3 ) = 0,
1
2 x1 + x2 + 52 x3 = 21 (x1 − x3 ) + (x2 + 3x3 ) = 0.

For the second system, we have

x1 − x3 = − 43 (−x1 + x2 + 4x3 ) + 14 (x1 + 3x2 + 8x3 ) + 0( 12 x1 + x2 + 25 x3 ),


x2 + 3x3 = 14 (−x1 + x2 + 4x3 ) + 14 (x1 + 3x2 + 8x3 ) + 0( 21 x1 + x2 + 52 x3 ).

So, the two systems are equivalent.

1.2.4 Exercise 4
Test the following systems as in Exercise 1.2.2.
 
i
2x1 + (−1 + i)x2 + x4 = 0 1+ x1 + 8x2 − ix3 − x4 = 0
2
2
3x2 − 2ix3 + 5x4 = 0 3 x1 − 21 x2 + x3 + 7x4 = 0

Solution. Call the equations in the system on the left L1 and L2 , and the equa-
tions on the right R1 and R2 . If R1 = aL1 +bL2 then, by equating the coefficients
of x3 , we get
−i = −2ib,
which implies that b = 1/2. By equating the coefficients of x1 , we get
i
1+ = 2a,
2
so that
1 1
a= + i.
2 4
1.2. SYSTEMS OF LINEAR EQUATIONS 3

Now, comparing the coefficients of x4 , we find that

1 1 5 1
−1 = a + 5b = + i + = 3 + i,
2 4 2 4

which is clearly a contradiction. Therefore the two systems are not equivalent.

1.2.5 Exercise 5
Let F be a set which contains exactly two elements, 0 and 1. Define an addition
and multiplication by the tables:

+ 0 1 · 0 1
0 0 1 0 0 0
1 1 0 1 0 1

Verify that the set F , together with these two operations, is a field.

Solution. From the symmetry in the tables, we see that both operations are
commutative.
By considering all eight possibilities, one can see that (a+b)+c = a+(b+c).
And one may in a similar way verify that (ab)c = a(bc), so that associativity
holds for the two operations.
0 + 0 = 0 and 0 + 1 = 1 so F has an additive identity. Similarly, 1 · 0 = 0
and 1 · 1 = 1 so F has a multiplicative identity.
The additive inverse of 0 is 0 and the additive inverse of 1 is 1. The multi-
plicative inverse of 1 is 1. So F has inverses.
Lastly, by considering the eight cases, one may verify that a(b + c) = ab + ac.
Therefore distributivity of multiplication over addition holds and F is a field.

1.2.7 Exercise 7
Prove that each subfield of the field of complex numbers contains every rational
number.

Proof. Let F be a subfield of C and let r = m/n be any rational number, written
in lowest terms. F must contain 0 and 1, so if r = 0 then we are done. Now
assume r is nonzero.
Since 1 ∈ F , and F is closed under addition, we know that 1 + 1 = 2 ∈ F .
And, if the integer k is in F , then k + 1 is also in F . By induction, we see that
all positive integers belong to F . We also know that all negative integers are
in F because F is closed under additive inverses. So, in particular, m ∈ F and
n ∈ F.
Now F is closed under multiplicative inverses, so n ∈ F implies 1/n ∈ F .
Finally, closure under multiplication shows that m · (1/n) = m/n = r ∈ F .
Since r was arbitrary, we can conclude that all rational numbers are in F .
4 CHAPTER 1. LINEAR EQUATIONS

1.2.8 Exercise 8
Prove that each field of characteristic zero contains a copy of the rational number
field.
Proof. Let F be a field of characteristic zero. Define the map f : Q → F (where
Q denotes the rational numbers) as follows. Let f (0) = 0F and f (1) = 1F ,
where 0F and 1F are the additive and multiplicative identities, respectively, of
F . Given a positive integer n, define f (n) = f (n − 1) + 1F and f (−n) = −f (n).
−1
If a rational number r = m/n is not an integer, define f (r) = f (m) · (f (n)) .
First we show that the function f preserves addition and multiplication. A
simple induction argument will show that, in the case of integers m and n, we
have
f (m + n) = f (m) + f (n) and f (mn) = f (m)f (n).
Now let r1 = m1 /n1 and r2 = m2 /n2 be rational numbers in lowest terms.
Then, by the definition of f ,

f (r1 + r2 ) = f ((m1 n2 + m2 n1 )/(n1 n2 ))


= f (m1 n2 + m2 n1 )f (n1 n2 )−1
= (f (m1 )f (n2 ) + f (m2 )f (n1 ))f (n1 )−1 f (n2 )−1
= f (m1 )f (n1 )−1 + f (m2 )f (n2 )−1
= f (r1 ) + f (r2 ).

Likewise,

f (r1 r2 ) = f (m1 )f (m2 )f (n1 )−1 f (n2 )−1 = f (r1 )f (r2 ).

(Formally, this shows that f is a ring homomorphism.)


We will next show that the function f is one-to-one. If r1 = m1 /n1 and
r2 = m2 /n2 are rational numbers in lowest terms, then f (r1 ) = f (r2 ) implies

f (m1 )f (n1 )−1 = f (m2 )f (n2 )−1

or
f (m1 )f (n2 ) = f (m2 )f (n1 ).
This implies
f (m1 n2 ) = f (m2 n1 ).
Now if m1 n2 6= m2 n1 , then this would imply that F does not have characteristic
zero. So m1 n2 = m2 n1 and so r1 = r2 .
What we have shown is that every rational number corresponds to a distinct
element of F , and that the operations of addition and multiplication of rational
numbers is preserved by this correspondence. So F contains a copy of Q.
1.3. MATRICES AND ELEMENTARY ROW OPERATIONS 5

1.3 Matrices and Elementary Row Operations


1.3.1 Exercise 1
Find all solutions to the system of equations
(1 − i)x1 − ix2 = 0
(1.1)
2x1 + (1 − i)x2 = 0.
Solution. Using elementary row operations, we get
     
1 − i −i (1) 1−i −i (2) 0 0
−−→ 1 1 −
−→ 1 .
2 1−i 1 2 − 2 i 1 2 − 12 i
So the system in (1.1) is equivalent to
 
1 1
x1 + − i x2 = 0.
2 2
Therefore, if c is any complex scalar, then x1 = (−1 + i)c and x2 = 2c is a
solution to (1.1).

1.3.2 Exercise 2
If  
3 −1 2
A = 2 1 1
1 −3 0
find all solutions of AX = 0 by row-reducing A.
Solution. We get
1 − 31 23 1 − 31 2
     
3 −1 2 3
 (1)   (2)  5  (1)
2 1 1 −−→ 2 1 1  −−→ 0 − 13  −−→

3
1 −3 0 1 −3 0 0 − 38 − 23
1 − 13 2 3 3
     
3 1 0 5 1 0 5
1  (2)  1  (1)   (2)
0 1 − 5  −−→ 0 1 − 5  −−→ 0 1 − 15  −−→

0 − 83 − 23 0 0 − 65 0 0 1
 
1 0 0
0 1 0 .
0 0 1
Thus AX = 0 has only the trivial solution.

1.3.3 Exercise 3
If  
6 −4 0
A= 4 −2 0
−1 0 3
find all solutions of AX = 2X and all solutions of AX = 3X. (The symbol cX
denotes the matrix each entry of which is c times the corresponding entry of
X.)
6 CHAPTER 1. LINEAR EQUATIONS

Solution. The matrix equation AX = 2X corresponds to the system of linear


equations

6x1 − 4x2 = 2x1


4x1 − 2x2 = 2x2
−1x1 + 3x3 = 2x3 ,

or, equivalently,

4x1 − 4x2 =0
4x1 − 4x2 =0
−1x1 + x3 = 0.

This system is homogeneous, and can be represented by the equation BX = 0,


where B is given by
 
4 −4 0
B =  4 −4 0 .
−1 0 1

B can be row-reduced:
   
4 −4 0 1 0 −1
 4 −4 0 → 0 1 −1 .
−1 0 1 0 0 0

Therefore any solution of AX = 2X will have the form

(x1 , x2 , x3 ) = (a, a, a) = a(1, 1, 1),

where a is a scalar.
Similarly, the equation AX = 3X can be solved by row-reducing
   
3 −4 0 1 0 0
4 −5 0 → 0 1 0 .
−1 0 0 0 0 0

So, solutions of AX = 3X have the form

(x1 , x2 , x3 ) = (0, 0, b) = b(0, 0, 1),

where b is a scalar.

1.3.4 Exercise 4
Find a row-reduced matrix which is row-equivalent to
 
i −(1 + i) 0
A = 1 −2 1 .
1 2i −1
1.3. MATRICES AND ELEMENTARY ROW OPERATIONS 7

Solution. Using the elementary row operations, we get


     
i −(1 + i) 0 1 −1 + i 0 1 −1 + i 0
(1) (2) (1)
1 −2 1  −−→ 1 −2 1  −−→ 0 −1 − i 1  −−→
1 2i −1 1 2i −1 0 1+i −1
   
1 −1 + i 0 1 0 i
 (2) 
0 1 − 21 + 12 i −−→ 0 1 − 12 + 21 i
 

0 1+i −1 0 0 0

The last matrix is row-equivalent to A.

1.3.5 Exercise 5
Prove that the following two matrices are not row-equivalent:
   
2 0 0 1 1 2
a −1 0 , −2 0 −1 .
b c 3 1 3 5

Proof. By performing row operations on the first matrix, we get


     
2 0 0 1 0 0 1 0 0
(1) (2) (1)
a −1 0 − −→ a −1 0 −−→ 0 −1 0 −−→
b c 3 b c 3 0 c 3
     
1 0 0 1 0 0 1 0 0
(2) (1)
0 1 0 − −→ 0 1 0 −−→ 0 1 0 .
0 c 3 0 0 3 0 0 1

We see that this matrix is row-equivalent to the identity matrix. The corre-
sponding system of equations has only the trivial solution.
For the second matrix, we get
 

1 1 2
 
1 1 2
 1 1 2
(2) (1)   (2)
−2 0 −1 − −→ 0 2 3 −−→ 0 1 23  −−→
1 3 5 0 2 3 0 2 3
1
   
1 1 2 1 0 2
 (2) 
0 1 32  −−→ 0 1 32  .
 

0 0 0 0 0 0

The system of equations corresponding to this matrix has nontrivial solutions.


Therefore the two matrices are not row-equivalent.

1.3.6 Exercise 6
Let  
a b
A=
c d
be a 2 × 2 matrix with complex entries. Suppose that A is row-reduced and also
that a + b + c + d = 0. Prove that there are exactly three such matrices.
8 CHAPTER 1. LINEAR EQUATIONS

Proof. One possibility is the zero matrix,


 
0 0
A= .
0 0

If A is not the zero matrix, then it has at least one nonzero row. If it has
exactly one nonzero row, then in order to satisfy the given constraints, the
nonzero row will have a 1 in the first column and a −1 in the second. This gives
two possibilities,
   
1 −1 0 0
A= or A = .
0 0 1 −1

Finally, if A has two nonzero rows, then it must be the identity matrix or the
matrix [ 01 10 ], but neither of these are valid since the sum of the entries is nonzero
in each case. Thus there are only the three possibilities given above.

1.3.7 Exercise 7
Prove that the interchange of two rows of a matrix can be accomplished by a
finite sequence of elementary row operations of the other two types.

Proof. We can, without loss of generality, assume that the matrix has only two
rows, since any additional rows could just be ignored in the procedure that
follows. Let this matrix be given by
 
a a2 a3 · · · an
A0 = 1 .
b1 b2 b3 · · · bn

First, add −1 times row 2 to row 1 to get the matrix


 
a1 − b1 a2 − b2 a3 − b3 · · · an − bn
A1 = .
b1 b2 b3 ··· bn

Next, add row 1 to row 2 to get


 
a − b1 a2 − b2 a3 − b3 ··· an − bn
A2 = 1 ,
a1 a2 a3 ··· an

and then add −1 times row 2 to row 1, which gives


 
−b1 −b2 −b3 · · · −bn
A3 = .
a1 a2 a3 · · · an

For the final step, multiply row 1 by −1 to get


 
b b2 b3 · · · bn
A4 = 1 .
a1 a2 a3 · · · an

We can see that A4 has the same entries as A0 but with the rows interchanged.
And only a finite number of elementary row operations of the first two kinds
were performed.
1.3. MATRICES AND ELEMENTARY ROW OPERATIONS 9

1.3.8 Exercise 8
Consider the system of equations AX = 0 where
 
a b
A=
c d

is a 2 × 2 matrix over the field F . Prove the following.


(a) If every entry of A is 0, then every pair (x1 , x2 ) is a solution of AX = 0.

Proof. This is clear, since the equation 0x1 + 0x2 = 0 is satisfied for any
(x1 , x2 ) ∈ F 2 (note that in any field, 0x = (1 − 1)x = x − x = 0).

(b) If ad−bc 6= 0, the system AX = 0 has only the trivial solution x1 = x2 = 0.

Proof. First suppose bd 6= 0. Then we can perform the following row-


reduction.
     
a b (1) ad bd (2) ad − bc 0 (1)
−−→ −−→ −−→
c d bc bd bc bd
     
1 0 (2) 1 0 (1) 1 0
−−→ −−→ .
bc bd 0 bd 0 1

In this case, A is row-equivalent to the 2 × 2 identity matrix.


On the other hand, if bd = 0 then one of b or d is zero (but not both). If
b = 0, then ad 6= 0 and we get
       
a 0 (1) 1 0 (2) 1 0 (1) 1 0
−−→ −−→ −−→ .
c d c d 0 d 0 1

If d = 0, then bc 6= 0 and we have


         
a b (3) c 0 (1) 1 0 (2) 1 0 (1) 1 0
−−→ −−→ −−→ −−→ .
c 0 a b a b 0 b 0 1

We see that, in every case, A is row-equivalent to the identity matrix.


Therefore AX = 0 has only the trivial solution.

(c) If ad − bc = 0 and some entry of A is different from 0, then there is a


solution (x01 , x02 ) such that (x1 , x2 ) is a solution if and only if there is a
scalar y such that x1 = yx01 , x2 = yx02 .

Proof. Since one of the entries a, b, c, d is nonzero, we can assume without


loss of generality that a is nonzero (because, if the first row is zero then we
could simply interchange the rows and relabel the entries; and if the only
nonzero entry occurs in the second column, then we could interchange the
columns which would correspond to relabeling x1 and x2 ).
Keeping in mind that a is nonzero, we perform the following row-reduction.
" # " #
b b
a b (1) 1 a (2) 1
 
a
−−→ −−→ .
c d c d 0 ad−bc
a
10 CHAPTER 1. LINEAR EQUATIONS

Since ad − bc = 0, the second row of this final matrix is zero, and we see
that there are nontrivial solutions. If we let

x01 = b and x02 = −a,

then (x1 , x2 ) is a solution if and only if x1 = yx01 and x2 = yx02 for some
y ∈ F.
1.4. ROW-REDUCED ECHELON MATRICES 11

1.4 Row-Reduced Echelon Matrices


1.4.1 Exercise 1
Find all solutions to the following system of equations by row-reducing the
coefficient matrix:

1
3 x1 + 2x2 − 6x3 = 0
−4x1 + 5x3 = 0
−3x1 + 6x2 − 13x3 = 0
− 37 x1 + 2x2 − 8
3 x3 =0

Solution. The coefficient matrix reduces as follows:


 1
2 −6 1 6 −18 1 6 −18
    
3
 −4 0 5  (1)  −4 0 5  (2) 0 24 −67 
 (1)
 −−→   −−→   −−→
    
 −3 6 −13  −3 6 −13 −67 

0 24
− 73 2 − 38 − 73 2− 38 0 16 − 134
3
1 6 −18 1 0 − 45
   
67  67 
0 1 − 24  (2) 0 1
 − 24
 −−→  .
 
− 67

0 1 24
 0 0 0 
0 16 − 134
3 0 0 0

Setting x3 = 24t, we see that all solutions have the form

x1 = 30t, x2 = 67t, and x3 = 24t,

where t is an arbitrary scalar.

1.4.2 Exercise 2
Find a row-reduced echelon matrix which is row-equivalent to
 
1 −i
A = 2 2 .
i 1+i

What are the solutions of AX = 0?

Solution. Performing row-reduction on A gives


       
1 −i 1 −i 1 −i 1 0
(2) (1) (2)
2 2  −−→ 0 2 + 2i −−→ 0 1  −−→ 0 1 ,
i 1+i 0 i 0 i 0 0

and this last matrix is in row-reduced echelon form. Therefore the homogeneous
system AX = 0 has only the trivial solution x1 = x2 = 0.
12 CHAPTER 1. LINEAR EQUATIONS

1.4.3 Exercise 3
Describe explicitly all 2 × 2 row-reduced echelon matrices.

Solution. If a 2 × 2 matrix has no nonzero rows, then it is the zero matrix,


 
0 0
0= ,
0 0

which is in row-reduced echelon form.


Next, if a 2 × 2 matrix has exactly one nonzero row, then in order to be in
row-reduced echelon form, the nonzero row must be in row 1 and it must start
with an entry of 1. There are two possibilities,
   
1 a 0 1
or ,
0 0 0 0

where a is an arbitrary scalar.


Lastly, if a 2 × 2 matrix in row-reduced echelon form has two nonzero rows,
then the diagonal entries must be 1 and the other entries 0, so we get the identity
matrix  
1 0
I= .
0 1
These are the only possibilities.

1.4.4 Exercise 4
Consider the system of equations

x1 − x2 + 2x3 = 1
2x1 + 2x3 = 1
x1 − 3x2 + 4x3 = 2.

Does this system have a solution? If so, describe explicitly all solutions.

Solution. We perform row-reduction on the augmented matrix:


   
1 −1 2 1 1 −1 2 1
(2) (1)
2 0 2 1 − −→ 0 2 −2 −1 −−→
1 −3 4 2 0 −2 2 1
1
 
1 0 1
 
1 −1 2 1 2
(2) 
0 1 −1 − 1  − −→ 0 1 −1 − 21  .

2
0 −2 2 1 0 0 0 0

From this we see that the original system of equations has solutions. All solu-
tions are of the form
1 1
x1 = −t + , x2 = t − , and x3 = t,
2 2
for some scalar t.
1.4. ROW-REDUCED ECHELON MATRICES 13

1.4.5 Exercise 5
Give an example of a system of two linear equations in two unknowns which
has no solution.

Solution. We can find such a system by ensuring that the coefficients in one
equation are a multiple of the other, while the constant term is not the same
multiple. For example, one such system is

x1 + 2x2 = 3
−3x1 − 6x2 = 5.

This system has no solutions since the augmented matrix is row-equivalent to a


matrix in which one row consists of zero entries everywhere but the rightmost
column.

1.4.6 Exercise 6
Show that the system

x1 − 2x2 + x3 + 2x4 = 1
x1 + x2 − x3 + x4 = 2
x1 + 7x2 − 5x3 − x4 = 3

has no solution.

Solution. Row-reduction on the augmented matrix gives


   
1 −2 1 2 1 1 −2 1 2 1
(2) (1)
1 1 −1 1 2 − −→ 0 3 −2 −1 1 −−→
1 7 −5 −1 3 0 9 −6 −3 2
1 0 − 31 4 5
   
1 −2 1 2 1 3 3
 (2) 
0 1 − 32 − 13 13  −−→ 0 1 − 23 − 31 1 
3 .

0 9 −6 −3 2 0 0 0 0 −1

Since the first nonzero entry in the bottom row of the last matrix is in the right-
most column, the corresponding system of equations has no solution. Therefore
the original system of equations also has no solution.

1.4.7 Exercise 7
Find all solutions of

2x1 − 3x2 − 7x3 + 5x4 + 2x5 = −2


x1 − 2x2 − 4x3 + 3x4 + x5 = −2
2x1 − 4x3 + 2x4 + x5 = 3
x1 − 5x2 − 7x3 + 6x4 + 2x5 = −7.
14 CHAPTER 1. LINEAR EQUATIONS

Solution. The augmented matrix can be row-reduced as follows:


   
2 −3 −7 5 2 −2 1 −2 −4 3 1 −2
1 −2 −4 3 1 −2 (3) 2 −3 −7 5 2 −2 (2)
2 0 −4 2 1 3  −−→ 2 0 −4
   − −→
2 1 3
1 −5 −7 6 2 −7 1 −5 −7 6 2 −7
   
1 −2 −4 3 1 −2 1 0 −2 1 1 2
0 1 1 −1 0 2  (2) 0 1 1
 −−→  −1 0 2− (1)
 −→
0 4 4 −4 −1 7  0 0 0 0 −1 −1
0 −3 −3 3 1 −5 0 0 0 0 1 1
   
1 0 −2 1 1 2 1 0 −2 1 0 1
0 1 1 −1 0 2 (2) 0 1 1 −1 0 2
  −−→  .
0 0 0 0 1 1 0 0 0 0 1 1
0 0 0 0 1 1 0 0 0 0 0 0
The columns with a leading 1 correspond to the variables x1 , x2 , and x5 , so
these variables will depend on the remaining two variables, which can take any
value. Therefore all solutions have the form
x1 = 2s − t + 1, x2 = t − s + 2, x3 = s, x4 = t, and x5 = 1,
where s and t are arbitrary scalars.

1.4.8 Exercise 8
Let  
3 −1 2
A = 2 1 1 .
1 −3 0
For which triples (y1 , y2 , y3 ) does the system AX = Y have a solution?
Solution. We will perform row-reduction on the augmented matrix:
1 − 31 2 1
 
3 y1
 
3 −1 2 y1 3
(1)   (2)
2 1 1 y2  −−→ 2 1 1 y2  −−→
1 −3 0 y3 1 −3 0 y3
1 2 1
1 − 31 2 1
   
1 −3 3 3 y1 3 3 y1
5  (1)   (2)
0 − 13 − 23 y1 + y2  −−→ 0 1 − 15 − 52 y1 + 53 y2  −−→

3
0 − 38 − 23 − 13 y1 + y3 0 − 38 − 23 − 31 y1 + y3
3 1 1 3 1 1
   
1 0 5 5 y1 + 5 y2 1 0 5 5 y1 + 5 y2
 (1)   (2)
0 1 − 15 − 52 y1 + 35 y2  −−→ 0 1 − 15 − 25 y1 + 35 y2  −−→

0 0 − 65 − 75 y1 + 85 y2 + y3 0 0 1 7 4 5
6 y1 − 3 y2 − 6 y3
1 0 0 − 12 y1 + y2 + 21 y3
 

0 1 0 − 16 y1 + 13 y2 − 16 y3  .
 
7 4 5
0 0 1 6 y1 − 3 y2 − 6 y3

Since every row contains a nonzero entry in the first three columns, the system
of equations AX = Y is consistent regardless of the values of y1 , y2 , and y3 .
Therefore AX = Y has a unique solution for any triple (y1 , y2 , y3 ).
1.4. ROW-REDUCED ECHELON MATRICES 15

1.4.9 Exercise 9
Let  
3 −6 2 −1
−2 4 1 3
A=
0
.
0 1 1
1 −2 1 0
For which (y1 , y2 , y3 , y4 ) does the system of equations AX = Y have a solution?
Solution. Row-reduction on the augmented matrix gives
2
1 −2 − 13 1
3 y1
   
3 −6 2 −1 y1 3
−2 4 1 3 y2  (1)  −2 4 1 3 y2 
 (2)
  −−→   −−→
0 0 1 1 y3 

0 0 1 1 y3 
1 −2 1 0 y4 1 −2 1 0 y4
2 1 1 2
1 −2 3 − 3 3 y1 1 −2 − 31 1
3 y1
   
3
0 0 7 7 2 2 3
3 y1 + y2  (1) 0 0 1 1 7 y1 + 7 y2  (2)
  
3 3
 −−→   −−→


0 0 1 1 y3  0 0 1 1 y3 
1 1
0 0 3 3 − 13 y1 + y4 0 0 1
3
1
3 − 31 y1 + y4
1
1 −2 0 −1 7 y1 − 27 y2
 
2
0 0 1 1 7 y1 + 37 y2 
.
 
− 27 y1 − 73 y2 + y3 

0 0 0 0
0 0 0 0 − 37 y1 − 71 y2 + y4

So, in order for the system AX = Y to have a solution, we need (y1 , y2 , y3 , y4 )


to satisfy

− 72 y1 − 37 y2 + y3 =0
− 37 y1 − 1
7 y2 + y4 = 0.

To determine the conditions on Y , we row-reduce the coefficient matrix for this


system.
" # " # " #
− 72 − 37 1 0 (1) 1 3
2 − 7
2 0 (2) 1 3
2 − 7
2 0 (1)
−−→ −−→ −−→
− 37 − 17 0 1 − 37 − 17 0 1 0 12 − 23 1
" #
1 32 − 72 0 (2) 1 0 1 −3
 
−−→ .
0 1 −3 2 0 1 −3 2

From this we see that in order for AX = Y to have a solution, (y1 , y2 , y3 , y4 )


must take the form

(y1 , y2 , y3 , y4 ) = (3t − s, 3s − 2t, s, t),

where s and t are arbitrary.


16 CHAPTER 1. LINEAR EQUATIONS

1.5 Matrix Multiplication


1.5.1 Exercise 1
Let

  3
2 −1 1  
A= , B =  1 , C= 1 −1 .
1 2 1
−1

Compute ABC and CAB.

Solution. We get
  
  3 
2 −1 1   
ABC = 1 1 −1 
1 2 1
−1
 
  3 −3  
2 −1 1  4 −4
= 1 −1 = ,
1 2 1 4 −4
−1 1

and
  
  3
  2 −1 1  
CAB = 1 −1  1
1 2 1
−1
 
  4  
= 1 −1 = 0 .
4

1.5.2 Exercise 2
Let
   
1 −1 1 2 −2
A = 2 0 1 , B = 1 3 .
3 0 1 4 4

Verify directly that A(AB) = A2 B.

Solution. We have
    
1 −1 1 1 −1 1 2 −2
A(AB) = 2 0 1 2 0 1 1 3 
3 0 1 3 0 1 4 4
    
1 −1 1 5 −1 7 −3
= 2 0 1  8 0  = 20 −4 ,
3 0 1 10 −2 25 −5
1.5. MATRIX MULTIPLICATION 17

and
 2  
1 −1 1 2 −2
A2 B = 2 0 1 1 3 
3 0 1 4 4
  
2 −1 1 2 −2
= 5 −2 3 1 3 
6 −3 4 4 4
 
7 −3
= 20 −4 .
25 −5
So A(AB) = A2 B as expected.

1.5.3 Exercise 3
Find two different 2 × 2 matrices A such that A2 = 0 but A 6= 0.
Solution. Two possibilities are
   
0 1 0 0
and .
0 0 1 0
Both of these are nonzero matrices that satisfy A2 = 0.

1.5.4 Exercise 4
For the matrix A of Exercise 1.5.2, find elementary matrices E1 , E2 , . . . , Ek such
that
Ek · · · E2 E1 A = I.
Solution. We want to reduce
 
1 −1 1
A = 2 0 1
3 0 1
to the identity matrix. To start, we can use two elementary row operations of
the second kind to get 0 in the bottom two entries of column 1. Performing the
same operations on the identity matrix gives
   
1 0 0 1 0 0
E1 = −2 1 0 and E2 =  0 1 0 .
0 0 1 −3 0 1
Then  
1 −1 1
E2 E1 A = 0 2 −1 .
0 3 −2
Next, we can use a row operation of the first kind to make the central entry into
a 1:    
1 0 0 1 −1 1
E3 = 0 21 0 , so that E3 E2 E1 A = 0 1 − 21  .
   

0 0 1 0 3 −2
18 CHAPTER 1. LINEAR EQUATIONS

Continuing in this way, we get


   
1 1 0 1 0 0
E4 = 0 1 0 and E5 = 0 1 0 ,
0 0 1 0 −3 1

so that
1
 
1 0 2
E5 E4 E3 E2 E1 A = 0 1 − 12  .
 

0 0 − 12
Then
1
 

1 0 0
 1 0 2
E6 = 0 1 0 so that E6 E5 E4 E3 E2 E1 A = 0 1 − 12  .
 
0 0 −2 0 0 1

Finally,
− 21
   
1 0 0 1 0
1
E7 = 0 1 and E8 = 0 1 0 ,
  
2
0 0 1 0 0 1
which gives  
1 0 0
E8 E7 E6 E5 E4 E3 E2 E1 = 0 1 0 = I.
0 0 1
Thus each of E1 , E2 , . . . , E8 are elementary matrices, and they are such that
E8 · · · E2 E1 A = I.

1.5.5 Exercise 5
Let  
1 −1  
3 1
A = 2 2 and B = .
−4 4
1 0
Is there a matrix C such that CA = B?

Solution. Suppose there is, and let


 
c c2 c3
C= 1 .
c4 c5 c6

Then  
  1 −1  
c1 c2 c3  3 1
2 2 = .
c4 c5 c6 −4 4
1 0
This leads to the following system of equations:

c1 + 2c2 + c3 = 3, c4 + 2c5 + c6 = −4,


−c1 + 2c2 = 1, −c4 + 2c5 = 4.
1.5. MATRIX MULTIPLICATION 19

This system has solutions

(c1 , c2 , c3 , c4 , c5 , c6 ) = (1 − 2s, 1 − s, 4s, 2t − 4, t, −4t).

For example, taking s = t = −1, we get the matrix


 
3 2 −4
C= ,
−6 −1 4

and one can easily verify that CA = B.

1.5.6 Exercise 6
Let A be an m × n matrix and B an n × k matrix. Show that the columns
of C = AB are linear combinations of the columns of A. If α1 , . . . , αn are the
columns of A and γ1 , . . . , γk are the columns of C, then
n
X
γj = Brj αr .
r=1

Proof. Let A, B, C be as stated. By the definition of matrix multiplication, we


have
 
A11 B1j + A12 B2j + · · · + A1n Bnj
 A21 B1j + A22 B2j + · · · + A2n Bnj 
γj = 
 
.. 
 . 
Am1 B1j + Am2 B2j + · · · + Amn Bnj
     
A11 A12 A1n
 A21   A22   A2n 
= B1j  .  + B2j  .  + · · · + Bnj  . 
     
.
 .  .
 .  .
 . 
Am1 Am2 Amn
n
X
= B1j α1 + B2j α2 + · · · + Bnj αn = Brj αr .
r=1

Therefore the columns of C = AB are linear combinations of the columns of


A.
20 CHAPTER 1. LINEAR EQUATIONS

1.6 Invertible Matrices


1.6.1 Exercise 1
Let
 
1 2 1 0
A = −1 0 3 5 .
1 −2 1 1

Find a row-reduced echelon matrix R which is row-equivalent to A and an


invertible 3 × 3 matrix P such that R = P A.

Solution. We can perform elementary row operations on A, while performing


the same operations on I, in order to find R and P :
   
1 2 1 0 1 0 0
−1 0 3 5 , 0 1 0
1 −2 1 1 0 0 1
   
1 2 1 0 1 0 0
0 2 4 5 ,  1 1 0
0 −4 0 1 −1 0 1
   
1 0 −3 −5 0 −1 0
0 2 4 5 , 1 1 0
0 0 8 11 1 2 1
   
1 0 −3 −5 0 −1 0
5  1 1
0 1 2 2 , 0
 
2 2
11 1 1 1
0 0 1 8
 83 4 8
1 0 0 − 87 − 41 3
  
8 8
0 1 0 − 14  ,
1
0 − 41  .
  
4
0 0 1 11
8
1
8
1
4
1
8

Therefore,

− 78
3
− 14 3
  
1 0 0 8 8

3 −2 3

1
− 14  ,
1
R = 0 1 0 P = 0 − 14  = 2 0 −2 ,
  
4
8
0 0 1 11 1 1 1 1 2 1
8 8 4 8

and R = P A.

1.6.2 Exercise 2
Do Exercise 1.6.1, but with
 
2 0 i
A = 1 −3 −i .
i 1 1
1.6. INVERTIBLE MATRICES 21

Solution. We proceed as before:


   
2 0 i 1 0 0
1 −3 −i , 0 1 0
i 1 1 0 0 1
   
1 −3 −i 0 1 0
2 0 i , 1 0 0
i 1 1 0 0 1
   
1 −3 −i 0 1 0
0 6 3i  , 1 −2 0
0 1 + 3i 0 0 −i 1
   
1 −3 −i 0 1 0
1  1
0 1 2 i , − 31 0
 
6
0 1 + 3i 0 0 −i 1
1 1
   
1 0 2i 2 0 0
1 1
0 1 2i  , − 13 0
   
 6
0 0 32 − 12 i −1 − 1i 1
1
 16 2 3
1 0 21 i
  
2 0 0
 1
0 1 12 i , − 13 0
  
 6 
0 0 1 −1i 1
+1
15 i
3
+ 15 i
   13 5
1 1
5
1 3

1 0 0 3 30 − 10 i 10 − 10 i
3 1 1 3 
0 1 0 ,  0 − 10 − 10 i − 10 i .
  
10
0 0 1 − 13 i 1
5 + 1
15 i 3
5 + 1
5i

So    
1 0 0 10 1 − 3i 3 − 9i
1 
R = 0 1 0 = I, P = 0 −9 − 3i 3 − 9i  ,
30
0 0 1 −10i 6 + 2i 18 + 6i
and R = P A.

1.6.3 Exercise 3
For each of the two matrices
   
2 5 −1 1 −1 2
4 −1 2 , 3 2 4
6 4 1 0 1 −2

use elementary row operations to discover whether it is invertible, and to find


the inverse in case it is.

Solution. For the first matrix, row-reduction gives

1 25 − 21 5
− 21 5
− 12
     
1 1
 
2 5 −1 2 2
(1)  (2)  (2) 
4 −1 2  −−→ 4 −1 2  −−→ 0 −11 4  −−→ 0
 
−11 4 ,

6 4 1 6 4 1 0 −11 4 0 0 0
22 CHAPTER 1. LINEAR EQUATIONS

and we see that the original matrix is not invertible since it is row-equivalent to
a matrix having a row of zeros.
For the second matrix, we get
   
1 −1 2 1 0 0
3 2 4  , 0 1 0
0 1 −2 0 0 1
   
1 −1 2 1 0 0
0 5 −2 , −3 1 0
0 1 −2 0 0 1
   
1 −1 2 1 0 0
0 1 −2 ,  0 0 1
0 5 −2 −3 1 0
   
1 0 0 1 0 1
0 1 −2 ,  0 0 1 
0 0 8 −3 1 −5
   
1 0 0 1 0 1
0 1 −2 ,  0 0 1 
   

0 0 1 − 3 1 − 58
   8 8 
1 0 0 1 0 1
0 1 0 , − 34 41 − 14  .
   

0 0 1 − 38 81 − 58

From this we see that the original matrix is invertible and its inverse is the
matrix  
8 0 8
1
−6 2 −2 .
8
−3 1 −5

1.6.4 Exercise 4
Let  
5 0 0
A = 1 5 0 .
0 1 5
For which X does there exist a scalar c such that AX = cX?

Solution. Let  
x1
X = x2  .
x3
Then AX = cX implies

5x1 = cx1
x1 + 5x2 = cx2
x2 + 5x3 = cx3 ,
1.6. INVERTIBLE MATRICES 23

and this is a homogeneous system of equations with coefficient matrix


 
5−c 0 0
B= 1 5−c 0 .
0 1 5−c
If c = 5 then (x1 , x2 , x3 ) = (0, 0, t) for some scalar t, so this gives one possibility
for X. If we assume c 6= 5, then the matrix B can be row-reduced to the identity
matrix, so that X = 0 is then the only possibility. Therefore, there is a scalar
c with AX = cX if and only if
 
0
X = 0 ,
t
for some arbitrary scalar t.

1.6.5 Exercise 5
Discover whether  
1 2 3 4
0 2 3 4
A=
 
0 0 3 4
0 0 0 4
is invertible, and find A−1 if it exists.
Solution. We proceed in the usual way:
   
1 2 3 4 1 0 0 0
0 2 3 4 0 1 0 0
0 0 3 4 , 0 0 1
   
0
0 0 0 4 0 0 0 1
1 2 3 0 1 0 0 −1
   
0 2 3 0 0 1 0 −1
, 
   
−1
 
0 0 3 0 0 0 1
1
0 0 0 1 0 0 0 4
1 2 0 0 1 0 −1 0
   
0 2 0 0 0 1 −1 0 
,
   
0 0 13 − 13 
  
0 0 1 0
1
0 0 0 1 0 0 0 4
1 0 0 0 1 −1 0 0
   
0 1 0 0 0 1 − 1 0 
2 2
, .
   
1
− 13 
 
0 0 1 0 0 0 3
1
0 0 0 1 0 0 0 4
Thus A is invertible and
1 −1 0 0
 
1
0 − 21 0 
A−1 = 2
.
 
1
0 0 3 − 13 
1
0 0 0 4
24 CHAPTER 1. LINEAR EQUATIONS

1.6.6 Exercise 6
Suppose A is a 2 × 1 matrix and that B is a 1 × 2 matrix. Prove that C = AB
is not invertible.

Proof. Let
 
a  
A= and B = c d
b

so that  
ac ad
C = AB = .
bc bd

Suppose C has an inverse. Then C is row-equivalent to the identity matrix, and


so cannot be row-equivalent to a matrix having a row of zeros. Consequently,
each of a, b, c, and d must be nonzero, since otherwise C would be row-equivalent
to such a matrix.
But, since a and b are nonzero, we can multiply the second row of C by a/b
to get the row-equivalent matrix
     
ac ad (1) ac ad (2) ac bd
−−→ −−→ ,
bc bd ac ad 0 0

which is clearly not invertible. Therefore C cannot have an inverse.

1.6.7 Exercise 7
Let A be an n × n (square) matrix. Prove the following two statements:

(a) If A is invertible and AB = 0 for some n × n matrix B, then B = 0.

Proof. Since AB = 0 and A is invertible, we can multiply on the left by


A−1 to get
B = A−1 0.

But the product on the right is clearly the n×n zero matrix, so B = 0.

(b) If A is not invertible, then there exists an n×n matrix B such that AB = 0
but B 6= 0.

Proof. If A is not invertible, then the homogeneous system of equations


AX = 0 has a nontrivial solution X0 . Let B be the matrix whose first
column is X0 and whose other entries are all zero, and consider the product
AB.
The entries in the first column of AB must be zero since the first column is
just AX0 , and the remaining entries must be zero since all other columns
are the product of A with a zero column. Thus the proof is complete.
1.6. INVERTIBLE MATRICES 25

1.6.8 Exercise 8
Let  
a b
A= .
c d
Prove, using elementary row operations, that A is invertible if and only if

ad − bc 6= 0.

Proof. First, if A is invertible, then one of a, b, c, or d must be nonzero. If


a 6= 0, then we can reduce
" # " #
b b
a b (1) 1 a (2) 1
 
a
−−→ −−→ ,
c d c d 0 ad−bc
a

and we must have ad − bc 6= 0 since otherwise A could not be row-equivalent to


the identity matrix, contradicting Theorem 12.
If, instead, a = 0 then we must have b 6= 0 since otherwise A would have a
row of zeros and could not be row-equivalent to the identity matrix. So we can
proceed:
       
0 b (1) 0 1 (3) c d (2) c 0
−−→ −−→ −−→ ,
c d c d 0 1 0 1

and we see that we must have c 6= 0. Thus ad − bc = −bc 6= 0. This completes


the first half of the proof.
Conversely, assume that ad − bc 6= 0. If d 6= 0 then A can be reduced to get
     
a b (1) ad bd (2) ad − bc 0 (1)
−−→ −−→ −−→
c d c d c d
     
1 0 (2) 1 0 (1) 1 0
−−→ −−→
c d 0 d 0 1

and A is row-equivalent to the identity matrix. On the other hand, if d = 0


then b and c must be nonzero and we get
         
a b (1) a b (3) 1 0 (2) 1 0 (1) 1 0
−−→ −−→ −−→ −−→
c 0 1 0 a b 0 b 0 1

so that A is again row-equivalent to the identity matrix. In either case, A must


be invertible by Theorem 12.

1.6.9 Exercise 9
An n × n matrix A is called upper-triangular if Aij = 0 for i > j, that is,
if every entry below the main diagonal is 0. Prove that an upper-triangular
(square) matrix is invertible if and only if every entry on its main diagonal is
different from 0.

Proof. Let A be an n × n upper-triangular matrix.


26 CHAPTER 1. LINEAR EQUATIONS

First, suppose every entry on the main diagonal of A is nonzero, and consider
the homogeneous linear system AX = 0:

A11 x1 + A12 x2 + · · · + A1n xn = 0


A22 x2 + · · · + A2n xn = 0
..
.
Ann xn = 0.

Since Ann is nonzero, the last equation implies that xn = 0. Then, since
An−1,n−1 is nonzero, the second-to-last equation implies that xn−1 = 0. Con-
tinuing in this way, we see that xi = 0 for each i = 1, 2, . . . , n. Therefore the
system AX = 0 has only the trivial solution, hence A is invertible.
Conversely, suppose A is invertible. Then A cannot contain any zero rows,
nor can A be row-equivalent to a matrix with a row of zeros. This implies that
Ann 6= 0. Consider An−1,n−1 . If An−1,n−1 is zero, then by dividing row n by
Ann , and then by adding −An−1,n times row n to row n−1, we see that A is row-
equivalent to a matrix whose (n − 1)st row is all zeros. This is a contradiction,
so An−1,n−1 6= 0. In the same manner, we can show that Aii 6= 0 for each
i = 1, 2, . . . , n. Thus all entries on the main diagonal of A are nonzero.

1.6.11 Exercise 11
Let A be an m×n matrix. Show that by means of a finite number of elementary
row and/or column operations one can pass from A to a matrix R which is both
‘row-reduced echelon’ and ‘column-reduced echelon,’ i.e., Rij = 0 if i 6= j,
Rii = 1, 1 ≤ i ≤ r, Rii = 0 if i > r. Show that R = P AQ, where P is an
invertible m × m matrix and Q is an invertible n × n matrix.
Proof. By Theorem 5, A is row-equivalent to a row-reduced echelon matrix R0 .
And, by the second corollary to Theorem 12, there is an invertible m×m matrix
P such that R0 = P A.
Results that are analogous to Theorems 5 and 12 (with similar proofs) hold
for column-reduced echelon matrices, so there is a matrix R which is column-
equivalent to R0 and an invertible n × n matrix Q such that R = R0 Q. Then
R = P AQ and we see that, through a finite number of elementary row and/or
column operations, A passes to a matrix R that is both row- and column-reduced
echelon.
Chapter 2

Vector Spaces

2.1 Vector Spaces


2.1.1 Exercise 1
If F is a field, verify that F n (as defined in Example 1) is a vector space over
the field F .
Proof. We need to check that addition and scalar multiplication, as defined in
Example 1, satisfy conditions (3) and (4) of the definition. Let

α = (α1 , α2 , . . . , αn ),
β = (β1 , β2 , . . . , βn ),

and

γ = (γ1 , γ2 , . . . , γn )

be arbitrary vectors in F n . From the commutativity of addition in F , we have

α + β = (α1 + β1 , . . . , αn + βn ) = (β1 + α1 , . . . , βn + αn ) = β + α,

so addition is commutative in F n . Similarly, by associativity of addition in F ,


we have

α + (β + γ) = (α1 + (β1 + γ1 ), . . . , αn + (βn + γn ))


= ((α1 + β1 ) + γ1 , . . . , (αn + βn ) + γn )
= (α + β) + γ,

and associativity holds in F n . The unique 0 vector is

0 = (0, 0, . . . , 0),

and it is clear that α + 0 = α. The unique additive inverse of α is given by

−α = (−α1 , −α2 , . . . , −αn ),

and certainly α + (−α) = 0. The conditions in (3) are satisfied.

27
28 CHAPTER 2. VECTOR SPACES

Now let c and d be scalars in F . Then

1α = (1α1 , . . . , 1αn ) = (α1 , . . . , αn ) = α,


(cd)α = ((cd)α1 , . . . , (cd)αn ) = (c(dα1 ), . . . , c(dαn )) = c(dα),
c(α + β) = (c(α1 + β1 ), . . . , c(αn + βn ))
= (cα1 + cβ1 , . . . , cαn + cβn )
= cα + cβ,

and

(c + d)α = ((c + d)α1 , . . . , (c + d)αn )


= (cα1 + dα1 , . . . , cαn + dαn )
= cα + dα,

so the conditions in (4) are satisfied. Therefore F n is a vector space over F .

2.1.2 Exercise 2
If V is a vector space over the field F , verify that

(α1 + α2 ) + (α3 + α4 ) = [α2 + (α3 + α1 )] + α4

for all vectors α1 , α2 , α3 , and α4 in V .

Proof. We only need to make use of commutativity and associativity of vector


addition:

(α1 + α2 ) + (α3 + α4 ) = (α2 + α1 ) + (α3 + α4 )


= α2 + [α1 + (α3 + α4 )]
= α2 + [(α1 + α3 ) + α4 ]
= [α2 + (α1 + α3 )] + α4
= [α2 + (α3 + α1 )] + α4 .

2.1.3 Exercise 3
If C is the field of complex numbers, which vectors in C 3 are linear combinations
of (1, 0, −1), (0, 1, 1), and (1, 1, 1)?

Solution. A vector α = (y1 , y2 , y3 ) is a linear combination of (1, 0, −1), (0, 1, 1),


and (1, 1, 1) if there are scalars x1 , x2 , x3 such that

x1 (1, 0, −1) + x2 (0, 1, 1) + x3 (1, 1, 1) = α,

which leads to the following system of equations:

x1 + x3 = y1
x2 + x3 = y2
−x1 + x2 + x3 = y3 .
2.1. VECTOR SPACES 29

Since the coefficient matrix


 
1 0 1
A= 0 1 1
−1 1 1

is row-equivalent to the identity matrix, this system of equations has a solution


for each α. Therefore, all vectors in C 3 are linear combinations of the vectors
(1, 0, −1), (0, 1, 1), and (1, 1, 1).

2.1.4 Exercise 4
Let V be the set of all pairs (x, y) of real numbers, and let F be the field of real
numbers. Define

(x, y) + (x1 , y1 ) = (x + x1 , y + y1 )
c(x, y) = (cx, y).

Is V , with these operations, a vector space over the field of real numbers?
Solution. No, V is not a vector space. Most of the conditions are satisfied, but
distributivity over scalar addition fails when y is nonzero:

(c + d)(x, y) = ((c + d)x, y) = (cx + dx, y)

but
c(x, y) + d(x, y) = (cx, y) + (dx, y) = (cx + dx, 2y).

2.1.5 Exercise 5
On Rn , define two operations

α⊕β =α−β
c · α = −cα.

The operations on the right are the usual ones. Which of the axioms for a vector
space are satisfied by (Rn , ⊕, ·)?
Solution. Commutativity of ⊕ fails, since α−β is not, in general, equal to β −α.
Associativity of ⊕ also fails since, for nonzero γ,

(α − β) − γ 6= α − β + γ = α − (β − γ).

The usual zero vector still works, since α ⊕ 0 = α − 0 = α. Additive inverses


also exist, but they are not the usual ones. Instead, each vector is its own
inverse, since α ⊕ α = α − α = 0.
For multiplication ·, it is not the case that 1 · α = α since, for nonzero α,
1 · α = −1α 6= α. Associativity with scalar multiplication does not hold either,
since
(c1 c2 ) · α = −c1 c2 α
while
c1 · (c2 · α) = c1 · (−c2 α) = c1 c2 α.
30 CHAPTER 2. VECTOR SPACES

For the distributive properties, we have

c · (α ⊕ β) = c · (α − β)
= −c(α − β)
= −cα + cβ

and

c · α ⊕ c · β = −cα − (−cβ)
= −cα + cβ,

so the first distributive property holds. And

(c1 + c2 ) · α = −(c1 + c2 )α
= −c1 α − c2 α,

while

c1 · α ⊕ c2 · α = −c1 α − (−c2 α)
= −c1 α + c2 α,

so the second distributive property fails.


To summarize, in the vector space definition, only properties (c) and (d) of
(3) and (c) of (4) hold.

2.1.6 Exercise 6
Let V be the set of all complex-valued functions f on the real line such that
(for all t in R)
f (−t) = f (t).
The bar denotes complex conjugation. Show that V , with the operations

(f + g)(t) = f (t) + g(t)


(cf )(t) = cf (t)

is a vector space over the field of real numbers. Give an example of a function
in V which is not real-valued.

Solution. Commutativity and associativity of addition follow from the proper-


ties of addition in C. Note that the zero function is in V . If f ∈ V , then the
function −f given by
(−f )(t) = −f (t)
is in V since
−f (−t) = −f (t) = −f (t) = (−f )(t).
And f + (−f ) is the zero function.
For scalar multiplication, we have

(1f )(t) = f (t),


2.1. VECTOR SPACES 31

so the first property is satisfied. And

((cd)f )(t) = (cd)f (t) = c((df )(t)) = (c(df ))(t),

so the second property is satisfied. And distributivity holds, since

(c(f + g))(t) = c(f (t) + g(t))


= cf (t) + cg(t)
= (cf )(t) + (cg)(t)
= (cf + cg)(t)

and

((c + d)f )(t) = (c + d)f (t)


= cf (t) + df (t)
= (cf )(t) + (df )(t)
= (cf + df )(t).

Therefore V is a vector space over R.


For an example of a function in V , consider the function f from R to C
given by
f (t) = ti.
Then f (−t) = −ti = ti = f (t) as required.

2.1.7 Exercise 7
Let V be the set of pairs (x, y) of real numbers and let F be the field of real
numbers. Define

(x, y) + (x1 , y1 ) = (x + x1 , 0)
c(x, y) = (cx, 0).

Is V , with these operations, a vector space?


Solution. No, V is not a vector space since 1α = α does not hold for all α
in V . For example, 1(1, 1) = (1, 0) 6= (1, 1). V also fails to have an additive
identity.
32 CHAPTER 2. VECTOR SPACES

2.2 Subspaces
2.2.1 Exercise 1
Which of the following sets of vectors α = (a1 , . . . , an ) in Rn are subspaces of
Rn (n ≥ 3)?
(a) all α such that a1 ≥ 0

Solution. This is not a subspace since it is not closed under scalar multi-
plication (take any negative scalar).

(b) all α such that a1 + 3a2 = a3

Solution. This is a subspace: Let β = (b1 , b2 , . . . , bn ). Then consider the


vector cα + β. We have

(ca1 + b1 ) + 3(ca2 + b2 ) = c(a1 + 3a2 ) + (b1 + 3b2 )


= ca3 + b3 .

Therefore cα + β is in the subset, so it is a subspace by Theorem 1.

(c) all α such that a2 = a21

Solution. This is not a subspace since it is not closed under vector addi-
tion. For example, (1, 1, 1, . . . ) is in the set, but the sum of this vector
with itself is not.

(d) all α such that a1 a2 = 0

Solution. This is not a subspace since it is not closed under vector addi-
tion. For example (1, 0, . . . ) and (0, 1, . . . ) are each in the set, but their
sum is not.

(e) all α such that a2 is rational

Solution. This is not a subspace because it is not closed under scalar


multiplication: Multiplication of any vector in the set having a2 6= 0 by
an irrational scalar produces a vector that is not in the set.

2.2.2 Exercise 2
Let V be the (real) vector space of all functions f from R into R. Which of the
following sets of functions are subspaces of V ?
(a) all f such that f (x2 ) = f (x)2

Solution. The functions f and g given by

f (x) = x and g(x) = 1

each belong to this set, but their sum f + g does not. Therefore this is
not a subspace.
2.2. SUBSPACES 33

(b) all f such that f (0) = f (1)

Solution. Suppose f and g both belong to this set. Then


(cf + g)(0) = cf (0) + g(0)
= cf (1) + g(1)
= (cf + g)(1),
so the set satisfies the subspace criterion of Theorem 1 and is thus a
subspace of V .

(c) all f such that f (3) = 1 + f (−5)

Solution. Take any f and g in this set. Then


(f + g)(3) = f (3) + g(3) = 2 + (f + g)(−5),
which does not belong to the set. Therefore this set is not a subspace.

(d) all f such that f (−1) = 0

Solution. Let f and g be such functions. Then


(cf + g)(−1) = cf (−1) + g(−1) = 0 + 0 = 0,
so this set is a subspace of V by Theorem 1.

(e) all f which are continuous

Solution. If f and g are continuous, then cf + g is also continuous, so this


is a subspace.

2.2.3 Exercise 3
Is the vector (3, −1, 0, −1) in the subspace of R5 spanned by the vectors
(2, −1, 3, 2), (−1, 1, 1, −3), and (1, 1, 9, −5)?
Solution. The subspace spanned by these three vectors consists of all linear
combinations
x1 (2, −1, 3, 2) + x2 (−1, 1, 1, −3) + x3 (1, 1, 9, −5).
Therefore (3, −1, 0, −1) is in this subspace if and only if the system of equations
2x1 − x2 + x3 = 3
−x1 + x2 + x3 = −1
3x1 + x2 + 9x3 = 0
2x1 − 3x2 − 5x3 = −1
has a solution. However, the augmented matrix can be row-reduced to
   
2 −1 1 3 1 0 2 0
−1 1 1 −1 0 1 3 0
 → .
3 1 9 0 0 0 0 1
2 −3 −5 −1 0 0 0 0
Therefore, this system of equations has no solution and the vector (3, −1, 0, −1)
is not in the subspace spanned by the other three given vectors.
34 CHAPTER 2. VECTOR SPACES

2.2.4 Exercise 4
Let W be the set of all (x1 , x2 , x3 , x4 , x5 ) in R5 which satisfy

2x1 − x2 + 34 x3 − x4 =0
2
x1 + 3 x3 − x5 = 0
9x1 − 3x2 + 6x3 − 3x4 − 3x5 = 0.

Find a finite set of vectors which spans W .

Solution. After performing the necessary elementary row operations, the coef-
ficient matrix becomes

2 −1 43 −1 0 1 0 32
   
0 −1
2
1 0 0 −1 → 0 1 0 1 −2 .
   
3
9 −3 6 −3 −3 0 0 0 0 0

So, letting x3 = 3t, x4 = u, and x5 = v, we see that the elements of W have the
form
(v − 2t, 2v − u, 3t, u, v).

Therefore, a spanning set for W is given by the vectors

(−2, 0, 3, 0, 0), (0, −1, 0, 1, 0), and (1, 2, 0, 0, 1).

2.2.5 Exercise 5
Let F be a field and let n be a positive integer (n ≥ 2). Let V be the vector
space of all n × n matrices over F . Which of the following sets of matrices A in
V are subspaces of V ?

(a) all invertible A

Solution. This cannot be a subspace since the zero matrix is not invertible.

(b) all non-invertible A

Solution. This is also not a subspace since it is possible for the sum of two
non-invertible matrices to be invertible. For example, in the 2 × 2 case,
the matrices    
1 0 0 0
and
0 0 0 1
are not invertible, but their sum is the identity matrix, which is invertible.

(c) all A such that AB = BA, where B is some fixed matrix in V


2.2. SUBSPACES 35

Solution. Let A1 and A2 be matrices in V such that A1 B = BA1 and


A2 B = BA2 . Then, since matrix multiplication is distributive,
(cA1 + A2 )B = cA1 B + A2 B
= cBA1 + BA2
= B(cA1 ) + BA2
= B(cA1 + A2 ).
Therefore, by Theorem 1, this set is a subspace of V .

(d) all A such that A2 = A

Solution. We will assume that the field F has more than two elements.
In that case, this set cannot be a subspace since the identity matrix has
the property that I 2 = I, but the sum of the identity with itself does not
have this property.

2.2.6 Exercise 6
(a) Prove that the only subspaces of R1 are R1 and the zero subspace.

Proof. Suppose W is a subspace of R1 . If W = {0} we are done, so


suppose W contains a nonzero element x. Then W must contain cx for
any real number c. In particular, if r is any real number, then W must
contain r since r = (rx−1 )x. This shows that W = R1 .

(b) Prove that a subspace of R2 is R2 , or the zero subspace, or consists of all


scalar multiples of some fixed vector in R2 . (The last type of subspace is,
intuitively, a straight line through the origin.)

Proof. Let W be a subspace of R2 . If W = {(0, 0)} we are done, so assume


W contains a nonzero vector α. Then W must contain all scalar multiples
of α. If these are the only elements in W , then we are again finished. If,
however, W contains two nonzero elements α and β such that β is not a
scalar multiple of α, then we must show that W = R2 .
Let α = (a1 , a2 ) and β = (b1 , b2 ). Also let γ = (c1 , c2 ) be any element in
R2 . Then γ is a linear combination of α and β if and only if the system
of equations
a1 x1 + b1 x2 = c1
a2 x1 + b2 x2 = c2
has a solution. Suppose the coefficient matrix
 
a1 b1
a2 b2
is not invertible. By Exercise 1.6.8, we then know that a1 b2 − a2 b1 = 0.
Now, one of a1 and a2 is nonzero. If a1 6= 0, then
b1
b2 = · a2 .
a1
36 CHAPTER 2. VECTOR SPACES

Also
b1
b1 = · a1 ,
a1
and we have a contradiction since β was assumed to not be a scalar mul-
tiple of α. Similarly a2 6= 0 also leads to a contradiction. This shows that
the system of equations above has a solution, so that W = R2 .

(c) Can you describe the subspaces of R3 ?

Solution. The subspaces of R3 are the zero subspace, the set of all scalar
multiples of a fixed nonzero vector (i.e., a line through the origin), the set
of all linear combinations of two linearly independent vectors (i.e., a plane
through the origin), and R3 itself.

2.2.7 Exercise 7
Let W1 and W2 be subspaces of a vector space V such that the set-theoretic
union of W1 and W2 is also a subspace. Prove that one of the spaces Wi is
contained in the other.
Proof. Let W1 and W2 be as stated, but assume that neither is contained in the
other. Then there is a vector u ∈ W1 such that u 6∈ W2 , and there is a vector
v ∈ W2 such that v 6∈ W1 . Since W1 ∪ W2 is a subspace, u + v ∈ W1 ∪ W2 . Now
either u + v ∈ W1 or u + v ∈ W2 . In the first case, since −u ∈ W1 we must have
(u + v) + (−u) = v ∈ W1 ,
which is a contradiction. But then u + v ∈ W2 leads to a similar contradiction.
Therefore one of the subspaces Wi must be contained in the other.

2.2.8 Exercise 8
Let V be the vector space of all functions from R into R; let Ve be the subset
of even functions,
f (−x) = f (x);
let Vo be the subset of odd functions,
f (−x) = −f (x).
(a) Prove that Ve and Vo are subspaces of V .

Proof. Suppose f and g are even functions. Then for any scalar c,
(cf + g)(−x) = cf (−x) + g(−x)
= cf (x) + g(x)
= (cf + g)(x),
so cf + g is also even and therefore Ve is a subspace of V . Similarly, if f
and g are both odd functions, then
(cf + g)(−x) = cf (−x) + g(−x)
= −cf (x) − g(x)
= −(cf + g)(x),
so Vo is also a subspace.
2.2. SUBSPACES 37

(b) Prove that Ve + Vo = V .

Proof. Let f ∈ V be arbitrary. Let g be the function in V defined by

f (x) + f (−x)
g(x) =
2
and let h be the function given by

f (x) − f (−x)
h(x) = .
2
It is clear that g ∈ Ve and h ∈ Vo . Since f (x) = g(x) + h(x) for all x, we
see that V = Ve + Vo .

(c) Prove that Ve ∩ Vo = {0}.

Proof. Suppose f ∈ Ve ∩ Vo and fix a particular x ∈ R. Since f is even,


f (−x) = f (x). And since f is odd, f (−x) = −f (x). Therefore we have
f (x) = −f (x), which is only possible if f (x) = 0. Since x was arbitrary, f
must be the zero function. This shows that Ve ∩Vo is the zero subspace.

2.2.9 Exercise 9
Let W1 and W2 be subspaces of a vector space V such that W1 + W2 = V and
W1 ∩ W2 = {0}. Prove that for each vector α in V there are unique vectors α1
in W1 and α2 is W2 such that α = α1 + α2 .
Proof. Since W1 + W2 = V , we may find α1 in W1 and α2 in W2 such that
α = α1 +α2 . Now suppose there is also α3 in W1 and α4 in W2 with α = α3 +α4 .
Then
α1 + α2 = α3 + α4 .
Rearranging, we get
α1 − α3 = α4 − α2 .
But the vector on the left-hand side must belong to W1 , and the vector on the
right-hand side must belong to W2 . Therefore α1 −α3 belongs to the intersection
of W1 and W2 , which implies that α1 − α3 = 0 or α1 = α3 . And α4 = α2 also.
This shows that the vectors α1 and α2 are unique.
38 CHAPTER 2. VECTOR SPACES

2.3 Bases and Dimension


2.3.1 Exercise 1
Prove that if two vectors are linearly dependent, one of them is a scalar multiple
of the other.

Proof. Let α1 and α2 be linearly dependent vectors in the space V . Then, by


definition, there are scalars c1 , c2 not both zero such that

c1 α1 + c2 α2 = 0.

If c1 is nonzero, then we may write


c2
α1 = − α2
c1

so that α1 is a scalar multiple of α2 . If c1 = 0, then c2 is nonzero and a similar


argument will do.

2.3.2 Exercise 2
Are the vectors

α1 = (1, 1, 2, 4), α2 = (2, −1, −5, 2)


α3 = (1, −1, −4, 0), α4 = (2, 1, 1, 6)

linearly independent in R4 ?

Solution. Suppose c1 α1 + c2 α2 + c3 α3 + c4 α4 = 0. This leads to the system of


equations

c1 + 2c2 + c3 + 2c4 = 0
c1 − c2 − c3 + c4 = 0
2c1 − 5c2 − 4c3 + c4 = 0
4c1 + 2c2 + 6c4 = 0.

Using the method of elimination developed in Chapter 1, we find that this


system has the general solution
 
s − 4t −2s − t
(c1 , c2 , c3 , c4 ) = , , s, t ,
3 3

where s, t ∈ R4 are arbitrary. For example, we may take s = 3 and t = 0 to get


c1 = 1, c2 = −2, c3 = 3, and c4 = 0. This shows that the vectors α1 , α2 , α3 , α4
are linearly dependent.

2.3.3 Exercise 3
Find a basis for the subspace of R4 spanned by the four vectors of Exercise 2.3.2.
2.3. BASES AND DIMENSION 39

Solution. Since α2 is not a scalar multiple of α1 , the set {α1 , α2 } is linearly


independent (by Exercise 2.3.1). We also see that it spans the subspace since
we can write
2 1
α3 = α2 − α1
3 3
and
4 1
α4 = α1 + α2 .
3 3
So {α1 , α2 } is a basis for the subspace.

2.3.4 Exercise 4
Show that the vectors

α1 = (1, 0, −1), α2 = (1, 2, 1), α3 = (0, −3, 2)

form a basis for R3 . Express each of the standard basis vectors as linear com-
binations of α1 , α2 , and α3 .

Solution. Since dim R3 = 3, we need only show that the three vectors are inde-
pendent. Let c1 , c2 , c3 be scalars such that

c1 α1 + c2 α2 + c3 α3 = 0.

Then we arrive at the homogeneous system of equations Ax = 0, where A is the


3 × 3 matrix whose jth column is αj . By row-reducing this matrix, we see that
it is row-equivalent to the identity matrix. Hence the system Ax = 0 has only
the trivial solution, i.e. c1 = c2 = c3 = 0. Therefore {α1 , α2 , α3 } is a basis for
R3 .
To write the standard basis vectors as linear combinations of α1 , α2 , α3 , we
may solve the systems Ax = i . This gives

7 3 1
(1, 0, 0) = α1 + α2 + α3
10 10 5
1 1 1
(0, 1, 0) = − α1 + α2 − α3
5 5 5
3 3 1
(0, 0, 1) = − α1 + α2 + α3 .
10 10 5

2.3.5 Exercise 5
Find three vectors in R3 which are linearly dependent, and are such that any
two of them are linearly independent.

Solution. Consider the vectors

α1 = (1, 0, 0), α2 = (0, 1, 0), and α3 = (1, 1, 0).

These vectors are pairwise-independent since neither is a scalar multiple of an-


other. But they are clearly linearly dependent since α1 + α2 − α3 = 0.
40 CHAPTER 2. VECTOR SPACES

2.3.6 Exercise 6
Let V be the vector space of all 2 × 2 matrices over the field F . Prove that V
has dimension 4 by exhibiting a basis for V which has four elements.
Proof. We may simply take the standard basis:
       
1 0 0 1 0 0 0 0
, , , .
0 0 0 0 1 0 0 1

Later, in Exercise 2.3.12, we will prove that this set is a basis for V in the more
general case where V is the space of m × n matrices.
Since V has a basis with four elements, it has dimension 4.

2.3.7 Exercise 7
Let V be the vector space of Exercise 6. Let W1 be the set of matrices of the
form  
x −x
y z
and let W2 be the set of matrices of the form
 
a b
.
−a c

(a) Prove that W1 and W2 are subspaces of V .

Proof. Both sets are nonempty. Consider the arbitrary matrices


   
x1 −x1 x2 −x2
A1 = and A2 =
y1 z1 y2 z2

in W1 and let c be an arbitrary scalar. Then


   
cx1 + x2 −cx1 − x2 cx1 + x2 −(cx1 + x2 )
cA1 + A2 = =
y1 + y2 z1 + z2 y1 + y2 z1 + z2

which is again in W1 . This shows that W1 is a subspace of V .


A similar argument will show that W2 is a subspace of V .

(b) Find the dimensions of W1 , W2 , W1 + W2 , and W1 ∩ W2 .

Solution. First we find bases for W1 and W2 . We may take


     
1 −1 0 0 0 0
, ,
0 0 1 0 0 1

as a basis for W1 and


     
1 0 0 1 0 0
, ,
−1 0 0 0 0 1

as a basis for W2 . Consequently, we see that dim W1 = dim W2 = 3.


2.3. BASES AND DIMENSION 41

Next, observe that matrices in W1 ∩ W2 must have the form


 
x −x
.
−x y
A basis for this space is then
   
1 −1 0 0
,
−1 0 0 1
so that dim(W1 ∩W2 ) = 2. Finally, we may apply Theorem 6 to determine
that
dim(W1 + W2 ) = dim W1 + dim W2 − dim(W1 ∩ W2 ) = 3 + 3 − 2 = 4.
It follows that W1 + W2 = V .

2.3.8 Exercise 8
Again let V be the space of 2 × 2 matrices over F . Find a basis {A1 , A2 , A3 , A4 }
for V such that A2j = Aj for each j.
Solution. Let
       
1 0 0 0 1 1 0 0
A1 = , A2 = , A3 = , and .
0 0 0 1 0 0 1 1
A simple check will show that A2j = Aj for each j. To show that {A1 , A2 , A3 , A4 }
is a basis for V , we need only show that it spans V (since any spanning set with
four vectors must be linearly independent).
Let  
x y
B=
z w
be an arbitrary 2×2 matrix over F . Then we can write B as a linear combination
of A1 , A2 , A3 , A4 as follows:
B = (x − y)A1 + (w − z)A2 + yA3 + zA4 .
Therefore the set {A1 , A2 , A3 , A4 } is indeed a basis for V .

2.3.9 Exercise 9
Let V be a vector space over a subfield F of the complex numbers. Suppose α,
β, and γ are linearly independent vectors in V . Prove that (α + β), (β + γ),
and (γ + α) are linearly independent.
Proof. Let c1 , c2 , and c3 be scalars in F such that
c1 (α + β) + c2 (β + γ) + c3 (γ + α) = 0.
By rearranging, this becomes
(c1 + c3 )α + (c1 + c2 )β + (c2 + c3 )γ = 0.
Since α, β, and γ are linearly independent, we must have
c1 + c3 = 0, c1 + c2 = 0, and c2 + c3 = 0.
But this system of equations has the unique solution (c1 , c2 , c3 ) = (0, 0, 0).
Therefore (α + β), (β + γ), and (γ + α) are linearly independent.
42 CHAPTER 2. VECTOR SPACES

2.3.10 Exercise 10
Let V be a vector space over the field F . Suppose there are a finite number of
vectors α1 , . . . , αr in V which span V . Prove that V is finite-dimensional.

Proof. We know by Theorem 4 that any independent set of vectors in V can


have at most a finite number r of elements. Thus, if a basis exists, it must be
finite.
We can explicitly construct such a basis: if α1 , . . . , αr are linearly indepen-
dent, then we are done. If not, one of the vectors αi can be written in terms of
the other αj . So remove αi from the set. This will not affect the span. If the
set is now linearly independent, then we have a basis. If not, continue removing
elements that are linear combinations of the remaining vectors. This process
must eventually terminate since we started with a finite number of vectors in
the set. Consequently, a finite basis exists.

2.3.11 Exercise 11
Let V be the set of all 2 × 2 matrices A with complex entries which satisfy
A11 + A22 = 0.

(a) Show that V is a vector space over the field of real numbers, with the
usual operations of matrix addition and multiplication of a matrix by a
scalar.

Proof. Let A and B be members of V . Then

(A + B)11 + (A + B)22 = (A11 + A22 ) + (B11 + B22 ) = 0 + 0 = 0.

And for any scalar c in R,

(cA)11 + (cA)22 = c(A11 + A22 ) = c0 = 0.

This shows that V is closed under matrix addition and scalar multiplica-
tion.
Next, we already know that matrix addition is commutative and associa-
tive. The zero matrix belongs to V , and for any A in V , the matrix −A
is also in V .
The remaining vector space axioms follow from the properties of matrix
addition and scalar multiplication. Therefore V is a vector space.

(b) Find a basis for this vector space.

Solution. One basis is given by


           
1 0 i 0 0 1 0 i 0 0 0 0
B= , , , , , .
0 −1 0 −i 0 0 0 0 1 0 i 0

It is fairly straightforward to check that B both spans V and is linearly


independent.
2.3. BASES AND DIMENSION 43

(c) Let W be the set of all matrices A in V such that A21 = −A12 (the bar
denotes complex conjugation). Prove that W is a subspace of V and find
a basis for W .

Solution. First, the zero matrix belongs to W so W is nonempty. Now for


any A and B in V and c in R, consider the matrix cA + B. We must have
(cA + B)21 = cA21 + B21 = −cA12 − B 12 = −(cA + B)12 .
This shows that W is a subspace of V . A basis for W is given by
       
1 0 i 0 0 1 0 i
, , , .
0 −1 0 −i −1 0 i 0

2.3.12 Exercise 12
Prove that the space of all m × n matrices over the field F has dimension mn,
by exhibiting a basis for this space.
Proof. Let F m×n denote the space of m × n matrices over F .
For each i and j with 1 ≤ i ≤ m and 1 ≤ j ≤ n, let ij denote the m × n
matrix over F whose ijth entry is 1, with all other entries 0. Let B denote the
set of all ij . We will show that B is a basis for F m×n , so that the dimension of
this space is mn.
First, let
Xm X n
A= cij ij ,
i=1 j=1
where each cij is an arbitrary scalar in F . Then A is the matrix whose ijth entry
is cij . By choosing these scalars appropriately, we see that any m × n matrix
over F can be written as a linear combination of the matrices in B. Therefore
B spans F m×n .
Moreover, A = 0 if and only if each cij = 0, so B is linearly independent.
This shows that B is a basis for F m×n .

2.3.13 Exercise 13
Discuss Exercise 2.3.9, when V is a vector space over the field with two elements
described in Exercise 1.2.5.
Solution. In Exercise 2.3.9, it was stated that the field F should be a subfield
of the complex numbers (in particular, a field with characteristic 0). When this
restriction is taken away, the result does not necessarily hold, as we will now
demonstrate.
Let V be the vector space F 3 , where F is the field with 2 elements. Let
α = (1, 0, 0), β = (0, 1, 0), and γ = (0, 0, 1). We see that α, β, and γ are linearly
independent (in fact they form the standard basis of F 3 ).
Now consider the vectors
α + β = (1, 1, 0), β + γ = (0, 1, 1), and γ + α = (1, 0, 1).
These are not linearly independent, since
(1, 1, 0) + (0, 1, 1) + (1, 0, 1) = (0, 0, 0).
So the result from Exercise 2.3.9 does not hold in this more general setting.
44 CHAPTER 2. VECTOR SPACES

2.3.14 Exercise 14
Let V be the set of real numbers. Regard V as a vector space over the field of
rational numbers, with the usual operations. Prove that this vector space is not
finite-dimensional.
Proof. Assume the contrary, and let {x1 , x2 , . . . , xn } be a finite basis for V .
Then every real number can be expressed as a linear combination

c1 x1 + c2 x2 + · · · + cn xn ,

where c1 , . . . , cn are rational numbers. Thus we can establish a one-to-one cor-


respondence between the n-tuples of rational numbers and the set of real num-
bers. Since the rational numbers are countable, this implies that the reals are
also countable. But this is clearly a contradiction. Therefore V is not finite-
dimensional.
2.4. COORDINATES 45

2.4 Coordinates
2.4.1 Exercise 1
Show that the vectors

α1 = (1, 1, 0, 0), α2 = (0, 0, 1, 1)


α3 = (1, 0, 0, 4), α4 = (0, 0, 0, 2)

form a basis for R4 . Find the coordinates of each of the standard basis vectors
in the ordered basis {α1 , α2 , α3 , α4 }.

Solution. Let  
1 0 1 0
1 0 0 0
P =
0
.
1 0 0
0 1 4 2
P is invertible and has inverse
 
0 1 0 0
0 0 1 0
P −1 = .
1 −1 0 0
−2 2 − 12 1
2

By Theorem 8, {α1 , α2 , α3 , α4 } is a basis for R4 . Moreover, the jth column of


P −1 gives the coordinates of the standard basis vector j in the ordered basis
{α1 , α2 , α3 , α4 }.

2.4.2 Exercise 2
Find the coordinate matrix of the vector (1, 0, 1) in the basis of C 3 consisting
of the vectors (2i, 1, 0), (2, −1, 1), (0, 1 + i, 1 − i), in that order.

Solution. Let  
2i 2 0
P =  1 −1 1 + i .
0 1 1−i
Then
1
− 21 i
 
2 −i −1
P −1 =  − 12 i −1 i .
 

− 14 + 14 i 1
2 + 12 i 1
Since
   1 − 1i −i

−1 1
 1 1 
−2 − 2i
1 2 2
P −1 0 =  − 12 i
   1
−1 i  0 =  2i  ,
 
1 −1 + 1i 1 1 1 3 1
4 4 2 + 2i 1 4 + 4i

the vector (1, 0, 1) has coordinates − 12 − 12 i, 21 i, 34 + 41 i in the given basis.



46 CHAPTER 2. VECTOR SPACES

2.4.3 Exercise 3
Let B = {α1 , α2 , α3 } be the ordered basis for R3 consisting of

α1 = (1, 0, −1), α2 = (1, 1, 1) α3 = (1, 0, 0).

What are the coordinates of the vector (a, b, c) in the ordered basis B?

Solution. Let  
1 1 1
P = 0 1 0 .
−1 1 0
Then  
0 1 −1
P −1 = 0 1 0
1 −2 1
and       
a 0 1 −1 a b−c
P −1  b  = 0 1 0  b =  b .
c 1 −2 1 c a − 2b + c

So (a, b, c) has coordinates (b − c, b, a − 2b + c) in the ordered basis B.

2.4.4 Exercise 4
Let W be the subspace of C 3 spanned by α1 = (1, 0, i) and α2 = (1 + i, 1, −1).

(a) Show that α1 and α2 form a basis for W .

Solution. Since neither α1 nor α2 is a scalar multiple of the other, the set
{α1 , α2 } is linearly independent. Hence this set is a basis for W .

(b) Show that the vectors β1 = (1, 1, 0) and β2 = (1, i, 1 + i) are in W and
form another basis for W .

Solution. If c1 (1, 0, i) + c2 (1 + i, 1, −1) = (1, 1, 0), then equating coordi-


nates and solving the resulting system gives c1 = −i and c2 = 1. Therefore
β1 is in W and its coordinates in {α1 , α2 } are (−i, 1).
In a similar way, we can determine that β2 is in W and has coordinates
(2 − i, i) in the given basis.
Neither β1 nor β2 is a scalar multiple of the other, so the set {β1 , β2 } is
linearly independent. Since W has dimension 2, the set {β1 , β2 } is also a
basis for W .

(c) What are the coordinates of α1 and α2 in the ordered basis {β1 , β2 } for
W?
2.4. COORDINATES 47

Solution. From the coordinates for β1 and β2 that we found previously,


we get the transition matrix
 
−i 2 − i
P = .
1 i
This matrix has inverse
" #
1
−1 2 − 21 i 3 1
2 + 2i
P = 1
,
2 + 21 i − 12 + 12 i
so
   
1 1 1 1
α1 = − i β1 + + i β2
2 2 2 2

and
   
3 1 1 1
α2 = + i β1 + − + i β2 .
2 2 2 2

2.4.6 Exercise 6
Let V be the vector space over the complex numbers of all functions from R into
C, i.e., the space of all complex-valued functions on the real line. Let f1 (x) = 1,
f2 (x) = eix , f3 (x) = e−ix .
(a) Prove that f1 , f2 , and f3 are linearly independent.

Proof. Let c1 , c2 , and c3 be complex numbers such that


c1 f1 (x) + c2 f2 (x) + c3 f3 (x) = 0
for all x in R. Then
c1 + c2 eix + c3 e−ix = 0.
Using Euler’s formula, we can write
c1 + c2 (cos x + i sin x) + c3 (cos x − i sin x) = 0
or, rearranging,
c1 + (c2 + c3 ) cos x + (c2 − c3 )i sin x = 0. (2.1)
If x = 0, then
c1 + c2 + c3 = 0, (2.2)
while if x = π we get
c1 − c2 − c3 = 0. (2.3)
Equations (2.2) and (2.3) together imply that c1 = 0.
Next, letting x = π/2 in (2.1), we get
(c2 − c3 )i = 0, (2.4)
which implies that c2 = c3 . Equation (2.2) then implies that c2 = c3 = 0.
Since it is necessary that c1 = c2 = c3 = 0, it follows that {f1 , f2 , f3 } is a
linearly independent set.
48 CHAPTER 2. VECTOR SPACES

(b) Let g1 (x) = 1, g2 (x) = cos x, g3 (x) = sin x. Find an invertible 3×3 matrix
P such that
X3
gj = Pij fi .
i=1

Solution. First, we have g1 = f1 . Next, since

f2 (x) + f3 (x) = (cos x + i sin x) + (cos x − i sin x) = 2 cos x,

we have g2 = 21 f2 + 12 f3 . And since

f2 (x) − f3 (x) = 2i sin x,

we see that g3 = − 21 if2 + 12 if3 . Therefore the desired matrix is


 
1 0 0
1
P = 0 − 12 i ,
 
2
1 1
0 2 2i

and this matrix is invertible.

2.4.7 Exercise 7
Let V be the (real) vector space of all polynomial functions from R into R of
degree 2 or less, i.e., the space of all functions f of the form

f (x) = c0 + c1 x + c2 x2 .

Let t be a fixed real number and define

g1 (x) = 1, g2 (x) = x + t, g3 (x) = (x + t)2 .

Prove that B = {g1 , g2 , g3 } is a basis for V . If

f (x) = c0 + c1 x + c2 x2

what are the coordinates of f in this ordered basis B?

Solution. Let

A = a + b(x + t) + c(x + t)2 = (a + bt + ct2 ) + (b + 2ct)x + cx2 ,

where a, b, c are real numbers. First, if A = 0 then, by equating coefficients, we


have

a + bt + ct2 = 0,
b + 2ct = 0,
c = 0.

Working backward through the equations, we see that a, b, and c must all be 0.
This shows that B is linearly independent.
2.4. COORDINATES 49

If we now set A = c0 + c1 x + c2 x2 , where c1 , c2 , c3 are arbitrary, and equate


coefficients, we get

a + bt + ct2 = c0 ,
b + 2ct = c1 ,
c = c2 .

Through back-substitution, we find that

b = c1 − 2tc2 and a = c0 − tc1 + t2 c2 .

This shows that any polynomial of degree 2 or less can be written as a linear
combination of g1 , g2 , and g3 . B is therefore a basis for V .
Moreover, we have also shown that the polynomial f (x) = c0 + c1 x + c2 x2
has coordinates
(c0 − tc1 + t2 c2 , c1 − 2tc2 , c2 )
in the ordered basis {g1 , g2 , g3 }.
50 CHAPTER 2. VECTOR SPACES

2.6 Computations Concerning Subspaces


2.6.1 Exercise 1
Let s < n and A an s × n matrix with entries in the field F . Use Theorem 4
(not its proof) to show that there is a non-zero X in F n×1 such that AX = 0.
Proof. Let α1 , . . . , αn denote the columns of A. Then each αi is a member of
the vector space F s , which has dimension strictly less than n. Therefore, by
Theorem 4, the αi are necessarily linearly dependent. Thus we can write

c1 α1 + c2 α2 + · · · + cn αn = 0

for c1 , . . . , cn in F not all 0. If we let


 
c1
 c2 
X =  . ,
 
 .. 
cn

then AX = 0 as required.

2.6.2 Exercise 2
Let
α1 = (1, 1, −2, 1), α2 = (3, 0, 4, −1), α3 = (−1, 2, 5, 2).
Let
α = (4, −5, 9, −7), β = (3, 1, −4, 4), γ = (−1, 1, 0, 1).
(a) Which of the vectors α, β, γ are in the subspace of R4 spanned by the αi ?

Solution. Let A be the 3 × 4 matrix whose ith row is αi . By performing


row reduction on A, we get
   
1 1 −2 1 1 0 0 −3/13
3 0 4 −1 → 0 1 0 14/13  .
−1 2 5 2 0 0 1 −1/13

A vector ρ is in the row space of A if and only if


 
3 14 1
ρ = c1 α1 + c2 α2 + c3 α3 = c1 , c2 , c3 , − c1 + c2 − c3 .
13 13 13
Checking each of α, β, and γ, we see that only α is in the row space.
So α is in the subspace spanned by the αi , while β and γ are not in the
subspace.

(b) Which of the vectors α, β, γ are in the subspace of C 4 spanned by the αi ?

Solution. Our work above is still valid in C 4 . α1 , α2 , and α3 will span


a larger subspace due to the scalars being taken from C instead of R,
but the members of this subspace will still have the same form as before.
Thus, of the three vectors α, β, and γ, only α is in the subspace.
2.6. COMPUTATIONS CONCERNING SUBSPACES 51

(c) Does this suggest a theorem?

Solution. This suggests the following theorem: let F be a subfield of the


field E. Let α be a vector in F n , and let β1 , . . . , βn in F n span some
subspace. Then α is in this subspace of F n if and only if it is in the
subspace of E n spanned by the same vectors βi .

2.6.3 Exercise 3
Consider the vectors in R4 defined by

α1 = (−1, 0, 1, 2), α2 = (3, 4, −2, 5), α3 = (1, 4, 0, 9).

Find a system of homogeneous linear equations for which the space of solutions
is exactly the subspace of R4 spanned by the three given vectors.

Solution. Let A be the 3 × 4 matrix whose ith row is αi . We can perform


row-reduction on A to get
 
1 0 −1 −2
1 11 
R = 0 1 4 .

4
0 0 0 0

Then a vector ρ in R4 is in the row space of A if and only if it has the form
 
1 11
ρ= r1 , r2 , r2 − r1 , r2 − 2r1 ,
4 4

where r1 and r2 are real numbers. If we label the components of ρ as

ρ = (x1 , x2 , x3 , x4 ),

then we get the following system of equations:

1
x3 = x2 − x1
4
11
x4 = x2 − 2x1 .
4

Or, we can rearrange these equations to write

1
x1 − x2 + x3 = 0
4
11
2x1 − x2 + x4 = 0.
4

This system of equations is homogeneous and its solution set is precisely the
subspace spanned by α1 , α2 , and α3 .
52 CHAPTER 2. VECTOR SPACES

2.6.4 Exercise 4
In C 3 , let
α1 = (1, 0, −i), α2 = (1 + i, 1 − i, 1), α3 = (i, i, i).
Prove that these vectors form a basis for C 3 . What are the coordinates of the
vector (a, b, c) in this basis?
Solution. Let  
1 0 −i
A = 1 + i 1 − i 1 .
i i i
By performing row-reduction, one can verify that A is row-equivalent to the
identity matrix. So A has rank 3 and α1 , α2 , and α3 are linearly independent
and span C 3 , as required to be a basis.
Let the coordinates of (a, b, c) in this basis be (x, y, z). This leads to the
following system of equations.
x + (1 + i)y + iz = a
(1 − i)y + iz = b
−ix + y + iz = c.
With a bit of effort, one may determine this system to have the solution
   a+b−2c + 4c−2a−2b i
x 5 5
 a+b−2c
y  = 
5 + 3b−2a−c 
5 i.
z 3a−2b−c
− a+b+3c
i
5 5

2.6.5 Exercise 5
Give an explicit description for the vectors
β = (b1 , b2 , b3 , b4 , b5 )
5
in R which are linear combinations of the vectors
α1 = (1, 0, 2, 1, −1), α2 = (−1, 2, −4, 2, 0)
α3 = (2, −1, 5, 2, 1), α4 = (2, 1, 3, 5, 2).
Solution. Performing row-reduction on the augmented matrix
 
1 −1 2 2 b1
0
 2 −1 1 b2  
A= 2 −4 5 3 b3  
1 2 2 5 b4 
−1 0 1 2 b5
produces
2 1
− 61 b4 − 12 b5
 
1 0 0 0 3 b1 + 2 b2
7 5
0 1 0 0 6 b4 − − 23 b2 − 12 b5 
3 b1
 
 3

0 0 1 0
R= 2 b4 − 2b1 − 25 b2 − 12 b5 
.
4 3 5 1 
0 0 0 1 3 b1 + b − b + b

2 2 6 4 2 5
0 0 0 0 b3 + b2 − 2b1
2.6. COMPUTATIONS CONCERNING SUBSPACES 53

Thus we see that β = (b1 , b2 , b3 , b4 , b5 ) is in the subspace spanned by α1 , α2 ,


α3 , α4 if and only if b3 + b2 − 2b1 = 0. For any such β, we have
   
2 1 1 1 7 5 3 1
β= b1 + b2 − b4 − b5 α1 + b4 − b1 − b2 − b5 α2
3 2 6 2 6 3 2 2
   
3 5 1 4 3 5 1
+ b4 − 2b1 − b2 − b5 α3 + b1 + b2 − b4 + b5 α4 .
2 2 2 3 2 6 2

2.6.6 Exercise 6
Let V be the real vector space spanned by the rows of the matrix
 
3 21 0 9 0
1 7 −1 −2 −1
A= 2 14 0
.
6 1
6 42 −1 13 0
(a) Find a basis for V .

Solution. If we perform row-reduction on the matrix A, we get the row-


reduced echelon matrix
 
1 7 0 3 0
0 0 1 5 0
R= 0 0 0 0 1 .

0 0 0 0 0
The three nonzero rows ρ1 , ρ2 , and ρ3 of R form a basis for V .

(b) Tell which vectors (x1 , x2 , x3 , x4 , x5 ) are elements of V .

Solution. If we take linear combinations of ρ1 , ρ2 , and ρ3 , we can see that


the vector (x1 , x2 , x3 , x4 , x5 ) is in V if and only if
x2 = 7x1 and x4 = 3x1 + 5x3 .

(c) If (x1 , x2 , x3 , x4 , x5 ) is in V what are its coordinates in the basis chosen


in part (a)?

Solution. Let x = (x1 , x2 , x3 , x4 , x5 ) be in V . If


x = c1 ρ1 + c2 ρ2 + c3 ρ3 ,
then we get the following system of equations:
c1 = x1
7c1 = x2
c2 = x3
3c1 + 5c2 = x4
c3 = x5 .
We see that x has coordinates (x1 , x3 , x5 ) in the basis {ρ1 , ρ2 , ρ3 }.
54 CHAPTER 2. VECTOR SPACES

2.6.7 Exercise 7
Let A be an m × n matrix over the field F , and consider the system of equations
AX = Y . Prove that this system of equations has a solution if and only if the
row rank of A is equal to the row rank of the augmented matrix of the system.
Proof. Let R be the row-reduced echelon matrix that is row-equivalent to A.
Form the augmented matrix A0 and let R0 be the row-reduced echelon matrix
row-equivalent to A0 . Then the nonzero rows of R form a basis for the row space
of A, and the nonzero rows of R0 form a basis for the row space of A0 . We want
to show that these bases have the same number of elements.
By the nature of the process of row reduction, it must be that the first n
columns of R0 will be identical to the n columns of R. Consequently, R0 cannot
have fewer nonzero rows than R, as any nonzero row of R must correspond to a
nonzero row in R0 . However, it might be possible for R0 to have more nonzero
rows than R. Such nonzero rows would need to have zeros in every column
except the last. But then such a row would indicate that the system AX = Y
has no solutions, which we know to be false. Therefore A and A0 have the same
row rank.
Now let us consider the converse. Let R and R0 be as before, and suppose
that the row ranks of A and A0 are equal. If AX = Y has no solutions, then R0
would necessarily have a row consisting of zeros in every column but the last.
But then the corresponding row in R would have only zero entries, resulting in
R0 having a larger row rank than R. This is impossible, so AX = Y must have
a solution.
Chapter 3

Linear Transformations

3.1 Linear Transformations


3.1.1 Exercise 1
Which of the following functions T from R2 into R2 are linear transformations?

(a) T (x1 , x2 ) = (1 + x1 , x2 );

Solution. T cannot be a linear transformation since

T (0, 0) = (1, 0) 6= (0, 0).

(b) T (x1 , x2 ) = (x2 , x1 );

Solution. T is a linear transformation: let c be a scalar and let α = (x1 , x2 )


and β = (y1 , y2 ). Then

T (cα + β) = T (cx1 + y1 , cx2 + y2 )


= (cx2 + y2 , cx1 + y1 )
= c(x2 , x1 ) + (y2 , y1 )
= cT (α) + T (β).

(c) T (x1 , x2 ) = (x21 , x2 );

Solution. T is not a linear transformation. For example,

T (1, 0) + T (2, 0) = (1, 0) + (4, 0) = (5, 0)

but
T ((1, 0) + (2, 0)) = T (3, 0) = (9, 0).

(d) T (x1 , x2 ) = (sin x1 , x2 );

55
56 CHAPTER 3. LINEAR TRANSFORMATIONS

Solution. T is not a linear transformation since


π  π 
T ,0 + T , 0 = (1, 0) + (1, 0) = (2, 0)
2 2
while  π   π 
T ,0 + , 0 = T (π, 0) = (0, 0).
2 2

(e) T (x1 , x2 ) = (x1 − x2 , 0).

Solution. Let α = (x1 , x2 ) and β = (y1 , y2 ). Since

cT (α) + T (β) = c(x1 − x2 , 0) + (y1 − y2 , 0)


= (cx1 − cx2 + y1 − y2 , 0)
= ((cx1 + y1 ) − (cx2 + y2 ), 0)
= T (cx1 + y1 , cx2 + y2 )
= T (cα + β),

we see that T is a linear transformation.

3.1.2 Exercise 2
Find the range, rank, null space, and nullity for the zero transformation and
the identity transformation on a finite-dimensional space V .

Solution. Let V be a vector space of finite dimension n, let T : V → V be the


zero transformation, and let U : V → V be the identity transformation.
Then the range of T is clearly the set consisting of the zero vector alone,
and the null space is V itself. The rank of T is then 0 (the zero subspace has
the empty set as a basis) and the nullity of T is n.
For the identity transformation U , we see that the range is all of V , while
the null space is the zero subspace {0}. Then the rank of U is n and the nullity
of U is 0.
Note that in each case, the rank plus the nullity is n, in agreement with
Theorem 2.

3.1.3 Exercise 3
Describe the range and the null space for the differentiation transformation of
Example 2. Do the same for the integration transformation of Example 5.

Solution. Let V be the space of polynomial functions from F into F and let D
be the differentiation transformation. Given a polynomial

f (x) = c0 + c1 x + · · · + ck xk ,

we can always find another polynomial g(x) such that (Dg)(x) = f (x), namely
the polynomial
1 1
g(x) = c0 x + c1 x2 + · · · + ck xk+1 .
2 k+1
3.1. LINEAR TRANSFORMATIONS 57

Therefore the range of D is V .


A function f has zero derivative if and only if f is constant,

f (x) = c0 , for some c0 in F .

So the null space of D is the space of constant functions.


Now, let T be the integration transformation defined in Example 5. We can
integrate any polynomial to get another polynomial, but because of the manner
in which T was defined, the resulting polynomial will always have a constant
term of 0. Let f be a polynomial with constant term zero,

f (x) = c1 x + c2 x2 + · · · + ck xk .

Then the function g(x) = (Df )(x) is such that (T g)(x) = f (x). So we see that
the range of T is the space of polynomials with zero constant term.
Lastly, if f is a polynomial such that (T f )(x) = 0, then f must be the zero
polynomial. The null space of T is therefore the trivial subspace {0}.

3.1.4 Exercise 4
Is there a linear transformation T from R3 into R2 such that T (1, −1, 1) = (1, 0)
and T (1, 1, 1) = (0, 1)?

Solution. Yes. In fact, there are infinitely many such transformations, as we


will now show.
Let α = (1, −1, 1) and let β = (1, 1, 1). Since neither α nor β is a multiple of
the other, the set {α, β} is linearly independent. Therefore it can be extended to
a basis for R3 . The existence of a linear transformation T such that T (α) = (1, 0)
and T (β) = (0, 1) now follows from Theorem 1.
To find such a transformation explicitly, we will form a basis for R3 . If we
let γ = (1, 0, 0), for example, it is not difficult to show that {α, β, γ} is a linearly
independent set of vectors which spans R3 .
We want to be able to write a vector ρ = (b1 , b2 , b3 ) as a linear combination
of α, β, and γ. To do this, we will set up an augmented matrix and perform
row-reduction, as we have done before in Chapter 2:

1
− 12 y2
 

1 1 1 y1
 1 0 0 2 y3
1
−1 1 0 y2 → 0 1 0 2 y2 + 12 y3  .
  
1 1 0 y3 0 0 1 y1 − y3

So we may write

1 1
ρ = (b1 , b2 , b3 ) = (b3 − b2 )α + (b2 + b3 )β + (b1 − b3 )γ.
2 2

Now suppose the transformation T is such that T (γ) = (x1 , x2 ), for some x1
58 CHAPTER 3. LINEAR TRANSFORMATIONS

and x2 in R. Then we have


 
1 1
T (b1 , b2 , b3 ) = T (b3 − b2 )α + (b2 + b3 )β + (b1 − b3 )γ
2 2
1 1
= (b3 − b2 )T (α) + (b2 + b3 )T (β) + (b1 − b3 )T (γ)
2 2
1 1
= (b3 − b2 , 0) + (0, b2 + b3 ) + (x1 b1 − x1 b3 , x2 b1 − x2 b3 )
2 2
1
= (2x1 b1 − b2 + (1 − 2x1 )b3 , 2x2 b1 + b2 + (1 − 2x2 )b3 ).
2
So, for example, taking (x1 , x2 ) = 21 , 12 , we get


 
1 1 1 1
T (b1 , b2 , b3 ) = b1 − b2 , b1 + b2 .
2 2 2 2
By picking different values for x1 and x2 , we see that there are infinitely many
possibilities for T .

3.1.5 Exercise 5
If
α1 = (1, −1), β1 = (1, 0)
α2 = (2, −1), β2 = (0, 1)
α3 = (−3, 2), β3 = (1, 1)
is there a linear transformation T from R2 into R2 such that T αi = βi for i = 1,
2, and 3?
Solution. No. To see why, observe that
(−3, 2) = −(1, −1) − (2, −1).
We have
T (−α1 − α2 ) = (1, 1)
but
−T (α1 ) − T (α2 ) = −(1, 0) − (0, 1) = (−1, −1).
So T cannot be a linear transformation.

3.1.6 Exercise 6
Describe explicitly the linear transformation T from F 2 into F 2 such that
T 1 = (a, b), T 2 = (c, d).
Solution. For any (x1 , x2 ) in F 2 , we have
T (x1 , x2 ) = T (x1 1 + x2 2 )
= x1 T (1 ) + x2 T (2 )
= x1 (a, b) + x2 (c, d)
= (x1 a + x2 c, x1 b + x2 d).
3.1. LINEAR TRANSFORMATIONS 59

3.1.7 Exercise 7
Let F be a subfield of the complex numbers and let T be the function from F 3
into F 3 defined by

T (x1 , x2 , x3 ) = (x1 − x2 + 2x3 , 2x1 + x2 , −x1 − 2x2 + 2x3 ).

(a) Verify that T is a linear transformation.

Solution. Let α = (x1 , x2 , x3 ) and β = (y1 , y2 , y3 ). Also for x in F 3 , let


π1 (x) denote the first coordinate of x, π2 (x) the second coordinate, and
π3 (x) the third coordinate. Then

π1 (cT (α) + T (β)) = c(x1 − x2 + 2x3 ) + (y1 − y2 + 2y3 )


= (cx1 + y1 ) − (cx2 + y2 ) + 2(cx3 + y3 )
= π1 (T (cα + β)),
π2 (cT (α) + T (β)) = c(2x1 + x2 ) + (2y1 + y2 )
= 2(cx1 + y1 ) + (cx2 + y2 )
= π2 (T (cα + β)),

and

π3 (cT (α) + T (β)) = c(−x1 − 2x2 + 2x3 ) + (−y1 − 2y2 + 2y3 )


= −(cx1 + y1 ) − 2(cx2 + y2 ) + 2(cx3 + y3 )
= π3 (T (cα + β)).

This shows that T is a linear transformation.

(b) If (a, b, c) is a vector in F 3 , what are the conditions on a, b, and c that


the vector be in the range of T ? What is the rank of T ?

Solution. If (a, b, c) is in the range of T , then

x1 − x2 + 2x3 = a
2x1 + x2 =b
−x1 − 2x2 + 2x3 = c.

In performing row-reduction on the augmented matrix for the above sys-


tem, we get
2 1 1
 
1 0 3a + 3b
 
1 −1 2 a 3
2 1 0 b  → 0 1 − 43 − 32 a + 13 b  .
 
−1 −2 2 c 0 0 0 −a + b + c

From this latter matrix, we see that this system of equations has a solution
if and only if
−a + b + c = 0.
60 CHAPTER 3. LINEAR TRANSFORMATIONS

We also see that the coefficient matrix


 
1 −1 2
A= 2 1 0
−1 −2 2

has a row rank (and thus column rank) of 2. But the column space of A
is precisely the range of T , so we may conclude that T has rank 2.

(c) What are the conditions on a, b, and c that (a, b, c) be in the null space
of T ? What is the nullity of T ?

Solution. (a, b, c) is in the null space of T if and only if

a − b + 2c = 0
2a + b =0
−a − 2b + 2c = 0.

We have already seen above that the coefficient matrix reduces to


2
 
1 0
 
1 −1 2 3
2 1 0 → 0 1 − 34  .
 
−1 −2 2 0 0 0

So (a, b, c) is in the null space if and only if a + 2c/3 = 0 and b − 4c/3 = 0.


Letting c = −3, for example, we find one possible basis for the null space
of T to be {(2, −4, −3)}. We see that the nullity of T is therefore 1, which
is as we should expect since F 3 has dimension 3 and the rank of T is 2.

3.1.8 Exercise 8
Describe explicitly a linear transformation from R3 into R3 which has as its
range the subspace spanned by (1, 0, −1) and (1, 2, 2).

Solution. Let {1 , 2 , 3 } denote the standard ordered basis for R3 . Theorem 1
allows us to find infinitely many linear transformations satisfying the given
criterion. For example, we may take some linear combination of the two given
vectors, say (2, 2, 1), and then look for a linear transformation T such that

T 1 = (1, 0, −1), T 2 = (1, 2, 2), and T 3 = (2, 2, 1).

Evidently, the transformation

T (x1 , x2 , x3 ) = (x1 + x2 + 2x3 , 2x2 + 2x3 , −x1 + 2x2 + x3 )

does the job. The range of T is precisely the subspace of R3 spanned by (1, 0, −1)
and (1, 2, 2). Of course, as noted, there are infinitely many other transformations
that would work.
3.1. LINEAR TRANSFORMATIONS 61

3.1.9 Exercise 9
Let V be the vector space of all n × n matrices over the field F , and let B be a
fixed n × n matrix. If
T (A) = AB − BA

verify that T is a linear transformation from V into V .

Proof. Let A1 and A2 be members of V and let c be a scalar in F . Using the


ordinary properties of matrix addition and multiplication, we have

cT (A1 ) + T (A2 ) = c(A1 B − BA1 ) + (A2 B − BA2 )


= (cA1 + A2 )B − B(cA1 + A2 )
= T (cA1 + A2 ).

Therefore T is a linear transformation from V into V .

3.1.10 Exercise 10
Let V be the set of all complex numbers regarded as a vector space over the
field of real numbers (usual operations). Find a function from V into V which
is a linear transformation on the above vector space, but which is not a linear
transformation on C 1 , i.e., which is not complex linear.

Solution. Define T (a + bi) = a. Then for any real number c and any complex
numbers z = a1 + b1 i and w = a2 + b2 i, we have

cT (z) + T (w) = ca1 + a2 = T (cz + w),

so T is a linear transformation from V into V . However,

iT (1) = i 6= 0 = T (i),

so T is not linear on C 1 .

3.1.11 Exercise 11
Let V be the space of n × 1 matrices over F and let W be the space of m × 1
matrices over F . Let A be a fixed m × n matrix over F and let T be the linear
transformation from V into W defined by T (X) = AX. Prove that T is the
zero transformation if and only if A is the zero matrix.

Proof. First suppose that T is the zero transformation. Then AX = 0 for all X
in V . In particular, let X = j , where j is the column vector whose jth entry
is 1 and all other entries zero. Then AX = 0 implies that the jth column of A
has only zero entries. Since this is true for all j with 1 ≤ j ≤ n, we see that A
is the m × n zero matrix.
Conversely, let A be the zero matrix. Then AX = 0 for all X so T is clearly
the zero transformation.
62 CHAPTER 3. LINEAR TRANSFORMATIONS

3.1.12 Exercise 12
Let V be an n-dimensional vector space over the field F and let T be a linear
transformation from V into V such that the range and null space of T are
identical. Prove that n is even. (Can you give an example of such a linear
transformation T ?)
Solution. This result follows directly from Theorem 2: if the rank of T is k,
then the nullity is also k and we have

k + k = n,

or n = 2k. Hence n is even.


As an example, let V = R2 and consider the linear transformation T given
by
T (x, y) = (y, 0).
Since

cT (x1 , y1 ) + T (x2 , y2 ) = c(y1 , 0) + (y2 , 0)


= (cy1 + y2 , 0)
= T (c(x1 , y1 ) + (x2 , y2 )),

T is a linear transformation. And both the range and the null space of T is the
x-axis.

3.1.13 Exercise 13
Let V be a vector space and T a linear transformation from V into V . Prove
that the following two statements about T are equivalent.
(a) The intersection of the range of T and the null space of T is the zero
subspace of V .

(b) If T (T α) = 0, then T α = 0.
Proof. Assume that (a) is true. If T (T α) = 0, then T α belongs to the null space
of T . But T α is also in the range of T , so T α = 0 by assumption.
Conversely, assume that (b) holds. Let β belong to the intersection of the
range of T with the null space of T . Then T (β) = 0 and there is some α in V
such that T (α) = β. Then T (T α) = T (β) = 0, so that T α = 0 by assumption.
But T α = β, so β = 0. Therefore the specified intersection is the zero subspace
and (a) holds.
3.2. THE ALGEBRA OF LINEAR TRANSFORMATIONS 63

3.2 The Algebra of Linear Transformations


3.2.1 Exercise 1
Let T and U be the linear operators on R2 defined by

T (x1 , x2 ) = (x2 , x1 ) and U (x1 , x2 ) = (x1 , 0).

(a) How would you describe T and U geometrically?

Solution. T mirrors points across the line y = x. U projects points onto


the x-axis.

(b) Give rules like the ones defining T and U for each of the transformations
(U + T ), U T , T U , T 2 , U 2 .

Solution. We have

(U + T )(x1 , x2 ) = (x1 + x2 , x1 ),
U T (x1 , x2 ) = (x2 , 0),
T U (x1 , x2 ) = (0, x1 ),
T 2 (x1 , x2 ) = (x1 , x2 ),

and

U 2 (x1 , x2 ) = (x1 , 0).

3.2.2 Exercise 2
Let T be the (unique) linear operator on C 3 for which

T 1 = (1, 0, i), T 2 = (0, 1, 1), T 3 = (i, 1, 0).

Is T invertible?

Solution. Let α = (z1 , z2 , z3 ) be a vector in C 3 such that T α = 0. Then

z1 (1, 0, i) + z2 (0, 1, 1) + z3 (i, 1, 0) = (0, 0, 0),

or

z1 + z3 i = 0
z2 + z3 = 0
z1 i + z2 = 0.

This system of equations has infinitely many solutions, each of the form

(z1 , z2 , z3 ) = (−ti, −t, t).

Therefore the null space of T is not {0}, so T is singular and not invertible.
64 CHAPTER 3. LINEAR TRANSFORMATIONS

3.2.3 Exercise 3
Let T be the linear operator on R3 defined by

T (x1 , x2 , x3 ) = (3x1 , x1 − x2 , 2x1 + x2 + x3 ).

Is T invertible? If so, find a rule for T −1 like the one which defines T .

Solution. It is not difficult to see that T is non-singular since

T (x1 , x2 , x3 ) = (0, 0, 0) if and only if x1 = x2 = x3 = 0.

By Theorem 9, T is invertible.
Let α = (y1 , y2 , y3 ) in R3 be such that T α = (x1 , x2 , x3 ). Then

3y1 = x1
y1 − y2 = x2
2y1 + y2 + y3 = x3 .

We see that this system of equations has the unique solution


 
1 1
(y1 , y2 , y3 ) = x1 , x1 − x2 , x3 − x1 + x2 .
3 3

So we have
 
−1 1 1
T (x1 , x2 , x3 ) = x1 , x1 − x2 , x3 − x1 + x2 .
3 3

3.2.4 Exercise 4
For the linear operator T of Exercise 3.2.3, prove that

(T 2 − I)(T − 3I) = 0.

Proof. We have

T 2 (x1 , x2 , x3 ) = T (3x1 , x1 − x2 , 2x1 + x2 + x3 )


= (9x1 , 2x1 + x2 , 9x1 + x3 ),

so
(T 2 − I)(x1 , x2 , x3 ) = (8x1 , x1 , 8x1 ).
Also,
(T − 3I)(x1 , x2 , x3 ) = (0, x1 − 4x2 , 2x1 + x2 − 2x3 ).
Consequently,

(T 2 − I)(T − 3I)(x1 , x2 , x3 ) = (T 2 − I)(0, x1 − 4x2 , 2x1 + x2 − 2x3 )


= (0, 0, 0).

Therefore (T 2 − I)(T − 3I) = 0.


3.2. THE ALGEBRA OF LINEAR TRANSFORMATIONS 65

3.2.5 Exercise 5
Let C 2×2 be the complex vector space of 2 × 2 matrices with complex entries.
Let  
1 −1
B=
−4 4

and let T be the linear operator on C 2×2 defined by T (A) = BA. What is the
rank of T ? Can you describe T 2 ?

Solution. Let  
a b
A=
c d

be a matrix in C 2×2 . Then


 
a−c b−d
T (A) = BA = .
−4a + 4c −4b + 4d

We see that T (A) = 0 if and only if both a = c and b = d. Consequently, a


basis for the null space of T is
   
1 0 0 1
, .
1 0 0 1

Therefore the nullity of T is 2. By Theorem 2, the rank of T is 4 − 2 = 2.


Since    
2 5 −5 1 −1
B = =5 = 5B,
−20 20 −4 4

we see that T 2 (A) = 5BA = 5T (A).

3.2.6 Exercise 6
Let T be a linear transformation from R3 into R2 , and let U be a linear trans-
formation from R2 into R3 . Prove that the transformation U T is not invertible.
Generalize the theorem.

Solution. We will state and prove the more general result directly. Let V and
W be finite-dimensional vector spaces over the same field F , and suppose that
dim V > dim W . Let T be a linear transformation from V into W and let U
be a linear transformation from W into V . Then the transformation U T is not
invertible, as we will now show.
First, since the rank of T is at most dim W < dim V , it follows that the
nullity of T is greater than 0. Thus T is not one to one, and there are distinct
vectors α and β in V such that T α = T β. Then we have

U T (α − β) = U (T (α) − T (β)) = U (0) = 0,

so U T is not one to one. This shows that U T is not invertible.


66 CHAPTER 3. LINEAR TRANSFORMATIONS

3.2.7 Exercise 7
Find two linear operators T and U on R2 such that T U = 0 but U T 6= 0.
Solution. Let T and U be given by
T (x1 , x2 ) = (x1 , 0) and U (x1 , x2 ) = (0, x1 ).
Then T U (x1 , x2 ) = T (0, x1 ) = (0, 0) as required. We also have U T 6= 0 since
U T (1, 1) = U (1, 0) = (0, 1) 6= (0, 0).

3.2.8 Exercise 8
Let V be a vector space over the field F and T a linear operator on V . If T 2 = 0,
what can you say about the relation of the range of T to the null space of T ?
Give an example of a linear operator T on R2 such that T 2 = 0 but T 6= 0.
Solution. Let β be in the range of T . Then there is an α in V such that T α = β.
But then
T β = T (T α) = T 2 (α) = 0,
so β is in the null space of T . This shows that the range of T is contained in
the null space of T .
On R2 , define T by
T (x1 , x2 ) = (x2 , 0).
Then
T 2 (x1 , x2 ) = T (x2 , 0) = (0, 0),
so T 2 = 0 but T 6= 0.

3.2.9 Exercise 9
Let T be a linear operator on the finite-dimensional space V . Suppose there
is a linear operator U on V such that T U = I. Prove that T is invertible
and U = T −1 . Give an example which shows that this is false when V is not
finite-dimensional.
Solution. Let α be in the null space of U , i.e., let U α = 0. Then
T U (α) = T (U α) = T (0) = 0.
Thus α is in the null space of T U . But T U = I, so this implies that α = 0.
This shows that U is non-singular. By Theorem 9, U is invertible.
Since T U = I, we have by the associativity of function composition that
U −1 = (T U )U −1 = T (U U −1 ) = T.
But if T = U −1 , then by definition T is invertible and U = T −1 .
To show that the original statement is not true when we remove the require-
ment that V be finite-dimensional, let V be the space of polynomial functions
over F , where F has characteristic zero. Let T = D, the differentiation opera-
tor, and let U = E, the integration operator, as defined in Example 11. Then T
and U are linear operators on V such that T U = I, but T is not invertible since
the differentiation operator is singular (its null space consists of all constant
functions).
3.2. THE ALGEBRA OF LINEAR TRANSFORMATIONS 67

3.2.10 Exercise 10
Let A be an m×n matrix with entries in F and let T be the linear transformation
from F n×1 into F m×1 defined by T (X) = AX. Show that if m < n it may
happen that T is onto without being non-singular. Similarly, show that if m > n
we may have T non-singular but not onto.

Solution. Let B = {1 , . . . , n } be the standard ordered basis for F n×1 and let
B 0 = {01 , . . . , 0m } be the standard ordered basis for F m×1 .
Now suppose m < n. Let A be the m × n matrix whose jth column, for
1 ≤ j ≤ m, is 0j , and whose remaining columns are zero. Then the linear
transformation T (X) = AX is such that

T (j ) = 0j , for each j with 1 ≤ j ≤ m,

and
T (j ) = 0 for m < j ≤ n.

Since every vector in B 0 is in the range of T , we see that T is onto. However


it is not possible for T to be non-singular, since by Theorem 2 the nullity of T
must be n − m > 0.
On the other hand, assume m > n. Take A to be the m × n matrix whose
ith row is i for 1 ≤ i ≤ n, with remaining rows zero. Then AX = 0 if and only
if X = 0, so that the nullspace of T is {0}. But the rank of T is n < m, so T is
non-singular but not onto.

3.2.11 Exercise 11
Let V be a finite-dimensional vector space and let T be a linear operator on V .
Suppose that rank(T 2 ) = rank(T ). Prove that the range and null space of T
are disjoint, i.e., have only the zero vector in common.

Proof. Note that the null space of T is contained in the null space of T 2 , since
if T α = 0 then T 2 (α) = T (0) = 0. But T and T 2 have the same rank, so by
Theorem 2 they must have the same nullity. So any basis for the null space of T
must also be a basis for the null space of T 2 . It follows that the two null spaces
are exactly equal.
Now let β be in the intersection of the range and null space of T . Then there
is an α in V with T α = β. This implies that

T 2 (α) = T β = 0,

so α is in the null space of T 2 . But T and T 2 have the same null space, so α is
in the null space of T . Hence

β = T α = 0.

This shows that the intersection of the range of T and the null space of T is
precisely the set {0}.
68 CHAPTER 3. LINEAR TRANSFORMATIONS

3.2.12 Exercise 12
Let p, m, and n be positive integers and F a field. Let V be the space of m × n
matrices over F and W the space of p × n matrices over F . Let B be a fixed
p × m matrix and let T be the linear transformation from V into W defined
by T (A) = BA. Prove that T is invertible if and only if p = m and B is an
invertible m × m matrix.
Proof. First assume that p = m (so that V = W ) and that B is invertible.
Define the linear transformation U from V into V by U (A) = B −1 A. Then for
any A in V , we have

T U (A) = T (B −1 A) = B(B −1 A) = (BB −1 )A = A

and
U T (A) = U (BA) = B −1 (BA) = (B −1 B)A = A.
This shows that T U = U T = I, so by definition T is invertible and T −1 = U .
The first half of the proof is complete.
Next, for the converse, assume that T is invertible. Then T is non-singular,
so the nullity of T is 0. By Theorem 2, we have

rank(T ) = dim V = mn.

On the other hand, T is onto, so

rank(T ) = dim W = pn.

Therefore mn = pn and we see that p = m and V = W . Now B is an m × m


matrix. If we define C = T −1 (I), then

T (CA) = B(CA) = (BC)A = T (C) · A = IA = A,

so T −1 (A) = CA. Since T T −1 = T −1 T = I, it follows that BC = CB = I so


that B is invertible. This completes the proof.
3.3. ISOMORPHISM 69

3.3 Isomorphism
3.3.1 Exercise 1
Let V be the set of complex numbers and let F be the field of real numbers.
With the usual operations, V is a vector space over F . Describe explicitly an
isomorphism of this space onto R2 .
Solution. Define the map T from V into R2 by

T (a + bi) = (a, b), where a and b belong to R.

Then for any c in R,

T (c(a1 + b1 i) + (a2 + b2 i)) = T ((ca1 + a2 ) + (cb1 + b2 )i)


= (ca1 + a2 , cb1 + b2 )
= c(a1 , b1 ) + (a2 , b2 )
= cT (a1 + b1 i) + T (a2 + b2 i),

so T is a linear transformation. It is one to one, since

(a, b) = (c, d) implies a + bi = c + di,

and it is onto since (a, b) is evidently in the range of T for all a, b in R. Therefore
T is an isomorphism and V and R2 are isomorphic.

3.3.2 Exercise 2
Let V be a vector space over the field of complex numbers, and suppose there is
an isomorphism T of V onto C 3 . Let α1 , α2 , α3 , α4 be vectors in V such that

T α1 = (1, 0, i) T α2 = (−2, 1 + i, 0),



T α3 = (−1, 1, 1), T α4 = ( 2, i, 3).

(a) Is α1 in the subspace spanned by α2 and α3 ?

Solution. If T α1 is in the subspace of C 3 spanned by T α2 and T α3 , then


there is x1 , x2 in C with

−2x1 − x2 = 1
(1 + i)x1 + x2 = 0
x2 = i.

We reduce the augmented matrix for this system of equations to get

1 0 − 21 − 12 i
   
−2 −1 1
1 + i 1 0 →  0 1 i ,

0 1 i 0 0 0
so  
1 1
T α1 = − − i T α2 + iT α3 .
2 2
70 CHAPTER 3. LINEAR TRANSFORMATIONS

Since T is an isomorphism, it follows that


  
1 1
T α1 = T − − i α2 + iα3
2 2
or  
1 1
α1 = − − i α2 + iα3 .
2 2
Therefore α1 is in the subspace spanned by α2 and α3 .

(b) Let W1 be the subspace spanned by α1 and α2 , and let W2 be the subspace
spanned by α3 and α4 . What is the intersection of W1 and W2 ?

Solution. Let α be in W1 ∩ W2 . Then α is a linear combination of α1 and


α2 , and also a linear combination of α3 and α4 . We can therefore find c1 ,
c2 , c3 , and c4 in C with
α = c1 α1 + c2 α2 = c3 α3 + c4 α4 .
So
T (c1 α1 + c2 α2 − c3 α3 − c4 α4 ) = 0
which implies that
c1 T α1 + c2 T α2 − c3 T α3 − c4 T α4 = 0.
This then leads to the system of equations

c1 − 2c2 + c3 − 2c4 = 0
(1 + i)c2 − c3 − ic4 = 0
ic1 − c3 − 3c4 = 0.
The coefficient matrix for this system reduces to
 √  
1 0 i 0

1 −2 1 − 2
0 1 + i −1 −i  →  1
0 1 − 2 + 21 i 0 .

i 0 −1 −3 0 0 0 1
So, letting t = −2c3 , we get
2itα1 + (i − 1)tα2 + 2tα3 = 0.
In particular, c4 = 0 and we see that α = −2tα3 where t is arbitrary. The
space W1 ∩ W2 therefore has {α3 } as a basis. Consequently, this space is
the one-dimensional subspace consisting of scalar multiples of α3 .

(c) Find a basis for the subspace of V spanned by the four vectors αj .

Solution. We have already seen previously that α3 can be written as a


linear combination of α1 and α2 :
1−i
α3 = −iα1 + α2 .
2
A check will show that the remaining vectors are linearly independent.
Therefore {α1 , α2 , α4 } forms a basis for the subspace of V spanned by
αj .
3.3. ISOMORPHISM 71

3.3.3 Exercise 3
Let W be the set of all 2 × 2 complex Hermitian matrices, that is, the set
of 2 × 2 complex matrices A such that Aij = Aji (the bar denoting complex
conjugation). As we pointed out in Example 6 of Chapter 2, W is a vector space
over the field of real numbers, under the usual operations. Verify that
 
t + x y + iz
(x, y, z, t) →
y − iz t − x
is an isomorphism of R4 onto W .
Proof. Denote this mapping by T . Then for any
α = (x1 , y1 , z1 , t1 ) and β = (x2 , y2 , z2 , t2 )
in R4 and any c in R, we have
T (cα + β) = T (cx1 + x2 , cy1 + y2 , cz1 + z2 , ct1 + t2 )
 
(ct1 + t2 ) + (cx1 + x2 ) (cy1 + y2 ) + i(cz1 + z2 )
=
(cy1 + y2 ) − i(cz1 + z2 ) (ct1 + t2 ) − (cx1 + x2 )
   
t + x1 y1 + iz1 t + x2 y2 + iz2
=c 1 + 2
y1 − iz1 t1 − x1 y2 − iz2 t2 − x2
= cT α + T β.
This shows that T is a linear transformation.
Next, if T α = T β then t1 + x1 = t2 + x2 and t1 − x1 = t2 − x2 , which
together imply that t1 = t2 and x1 = x2 . Similarly, y1 + iz1 = y2 + iz2 implies
that y1 = y2 and z1 = z2 . Therefore T is one to one.
Finally, let  
a b + ci
A=
b − ci d
be any 2 × 2 Hermitian matrix. Then we see that
   
1 1 1 1 a b + ci
T a + d, b, c, a − d = = A,
2 2 2 2 b − ci d
so T is onto. This shows that T is an isomorphism and R4 is isomorphic to
W.

3.3.4 Exercise 4
Show that F m×n is isomorphic to F mn .
Proof. An obvious isomorphism is the map T from F m×n onto F mn given by
 
A11 A12 · · · A1n
 A21 A22 · · · A2n 
T .
 
.. .. .. 
 .. . . . 
Am1 Am2 · · · Amn
= (A11 , A12 , . . . , A1n , A21 , A22 , . . . , A2n , . . . , Am1 , Am2 , . . . , Amn ).
That is, the jth coordinate of T (A) is the jth entry of A when the entries are
ordered from left-to-right and then top-to-bottom. It should be evident that T
is a linear transformation that is both one to one and onto.
72 CHAPTER 3. LINEAR TRANSFORMATIONS

3.3.5 Exercise 5
Let V be the set of complex numbers regarded as a vector space over the field
of real numbers. We define a function T from V into the space of 2 × 2 real
matrices, as follows. If z = x + iy with x and y real numbers, then
 
x + 7y 5y
T (z) = .
−10y x − 7y
(a) Verify that T is a one-one (real) linear transformation of V into the space
of 2 × 2 real matrices.

Proof. Let z1 = x1 + iy1 and z2 = x2 + iy2 where x1 , x2 , y1 , y2 are real


numbers. Let c be any real number. Then
T (cz1 + z2 ) = T (cx1 + x2 + i(cy1 + y2 ))
 
(cx1 + x2 ) + 7(cy1 + y2 ) 5(cy1 + y2 )
=
−10(cy1 + y2 ) (cx1 + x2 ) − 7(cy1 + y2 )
   
x1 + 7y1 5y1 x2 + 7y2 5y2
=c +
−10y1 x1 − 7y1 −10y2 x2 − 7y2
= cT (z1 ) + T (z2 )
and we see that T is a linear transformation.
Moreover, if T (z1 ) = T (z2 ) then 5y1 = 5y2 so that y1 = y2 and x1 = x2 ,
or z1 = z2 . Thus T is one to one.

(b) Verify that T (z1 z2 ) = T (z1 )T (z2 ).

Proof. As above, we let z1 = x1 + iy1 and z2 = x2 + iy2 . Then


  
x + 7y1 5y1 x2 + 7y2 5y2
T (z1 )T (z2 ) = 1 .
−10y1 x1 − 7y1 −10y2 x2 − 7y2
If we calculate the upper-left entry of this matrix product, we find
[T (z1 )T (z2 )]11 = (x1 + 7y1 )(x2 + 7y2 ) − 50y1 y2
= x1 x2 + 7x1 y2 + 7x2 y1 + 49y1 y2 − 50y1 y2
= (x1 x2 − y1 y2 ) + 7(x1 y2 + x2 y1 )
= [T (z1 z2 )]11 .
The remaining entries are calculated in the same manner, and are all
straightforward. Therefore T (z1 z2 ) = T (z1 )T (z2 ).

(c) How would you describe the range of T ?

Solution. We have shown above that the range of T is isomorphic to V .


A basis for V is {1, i}, so we may compute
   
1 0 7 5
T (1) = and T (i) = .
0 1 −10 −7
Since T preserves linear independence (Theorem 8), we see that the set
   
1 0 7 5
,
0 1 −10 −7
is a basis for the range of T .
3.3. ISOMORPHISM 73

3.3.6 Exercise 6
Let V and W be finite-dimensional vector spaces over the field F . Prove that
V and W are isomorphic if and only if dim V = dim W .
Proof. First suppose that V and W are isomorphic via the isomorphism T .
Suppose V has dimension n and let {α1 , . . . , αn } be a basis for V . Then by
Theorem 8, the set {T α1 , . . . , T αn } is linearly independent and thus forms a
basis for the range of T . But T is onto, so the range of T is W . Therefore
dim W = n as required.
Conversely, suppose dim V = dim W = n. By Theorem 10, V and W are
both isomorphic to F n . Thus V is isomorphic to W and the proof is complete.

3.3.7 Exercise 7
Let V and W be vector spaces over the field F and let U be an isomorphism of V
onto W . Prove that T → U T U −1 is an isomorphism of L(V, V ) onto L(W, W ).
Proof. Let S denote the stated map from L(V, V ) to L(W, W ). If c is in F , then

S(cT1 + T2 ) = U (cT1 + T2 )U −1
= (cU T1 + U T2 )U −1
= cU T1 U −1 + U T2 U −1
= cS(T1 ) + S(T2 ),

so S is a linear transformation.
Next, suppose S(T1 ) = S(T2 ). That is, U T1 U −1 = U T2 U −1 . Then

T1 = (U −1 U )T1 (U −1 U )
= U −1 (U T1 U −1 )U
= U −1 (U T2 U −1 )U
= (U −1 U )T2 (U −1 U )
= T2 ,

showing that S is one to one.


Finally, let T be any linear operator in L(W, W ). Then

S(U −1 T U ) = U (U −1 T U )U −1 = (U U −1 )T (U U −1 ) = T,

so S is onto. This shows that S is an isomorphism, so that L(V, V ) is isomorphic


to L(W, W ).
74 CHAPTER 3. LINEAR TRANSFORMATIONS

3.4 Representation of Transformations by Ma-


trices
3.4.1 Exercise 1
Let T be the linear operator on C 2 defined by T (x1 , x2 ) = (x1 , 0). Let B be
the standard ordered basis for C 2 and let B 0 = {α1 , α2 } be the ordered basis
defined by α1 = (1, i), α2 = (−i, 2).
(a) What is the matrix of T relative to the pair B, B 0 ?

Solution. We have

T (1, 0) = (1, 0) and T (0, 1) = (0, 0).

Now let  
1 −i
P = .
i 2
Then  
−1 2 i
P =
−i 1
and we get
    
−1 2 i 1 2
[(1, 0)] B0 =P [(1, 0)]B = = .
−i 1 0 −i

Of course, the zero vector has the same coordinates in every basis, so we
see that the matrix of T relative to B, B 0 is
 
0 2 0
[T ]B
B = .
−i 0

(b) What is the matrix of T relative to the pair B 0 , B?

Solution. We have

T (1, i) = (1, 0) and T (−i, 2) = (−i, 0).

So the matrix of T relative to B 0 , B is


 
1 −i
[T ]B
B 0 = .
0 0

(c) What is the matrix of T in the ordered basis B 0 ?

Solution. We have
     
−1 2 i 1 0 1 −i 2 −2i
[T ]B0 = P [T ]B P = = .
−i 1 0 0 i 2 −i −1

(d) What is the matrix of T in the ordered basis {α2 , α1 }?


3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 75

Solution. The change-of-basis matrix P such that

P [α]{α2 ,α1 } = [α]B0

is given by  
0 1
P = ,
1 0
so
     
−1 0 1 2 −2i 0 1 −1 −i
[T ]{α2 ,α1 } = P [T ]B0 P = = .
1 0 −i −1 1 0 −2i 2

3.4.2 Exercise 2
Let T be the linear transformation from R3 into R2 defined by

T (x1 , x2 , x3 ) = (x1 + x2 , 2x3 − x1 ).

(a) If B is the standard ordered basis for R3 and B 0 is the standard ordered
basis for R2 , what is the matrix of T relative to the pair B, B 0 ?

Solution. Since

T (1, 0, 0) = (1, −1), T (0, 1, 0) = (1, 0), and T (0, 0, 1) = (0, 2),

we see that the matrix of T relative to B, B 0 is


 
B0 1 1 0
[T ]B = .
−1 0 2

(b) If B = {α1 , α2 , α3 } and B 0 = {β1 , β2 }, where

α1 = (1, 0, −1), α2 = (1, 1, 1), α3 = (1, 0, 0),


β1 = (0, 1), β2 = (1, 0)

what is the matrix of T relative to the pair B, B 0 ?

Solution. We have

T α1 = (1, −3) = −3β1 + β2 ,


T α2 = (2, 1) = β1 + 2β2 ,

and

T α3 = (1, −1) = −β1 + β2 ,

so the corresponding matrix is


 
0 −3 1 −1
[T ]B
B = .
1 2 1
76 CHAPTER 3. LINEAR TRANSFORMATIONS

3.4.3 Exercise 3
Let T be a linear operator on F n , let A be the matrix of T in the standard
ordered basis for F n , and let W be the subspace of F n spanned by the column
vectors of A. What does W have to do with T ?

Solution. W is simply the range of T , as we will now show.


Let {1 , . . . , n } denote the standard ordered basis for F n . Note that the jth
column of A is simply T j . Take any vector α in F n . Then α belongs to W if
and only if

α = x1 T 1 + x2 T 2 + · · · + xn T n
= T (x1 1 + x2 2 + · · · + xn n )
= T (x1 , x2 , · · · , xn ),

for some vector (x1 , x2 , . . . , xn ) in F n . That is, α is in W if and only if α is in


the range of T .

3.4.4 Exercise 4
Let V be a two-dimensional vector space over the field F , and let B be an
ordered basis for V . If T is a linear operator on V and
 
a b
[T ]B =
c d

prove that T 2 − (a + d)T + (ad − bc)I = 0.

Proof. By Theorem 12, the function which assigns a linear operator on V to its
matrix relative to B is an isomorphism between L(V, V ) and F 2×2 . Theorem 13
shows that this function preserves products also. Thus we can operate on T by
simply performing the corresponding operations on its matrix and vice versa.
So consider the following computation.

[T ]2B − (a + d)[T ]B + (ad − bc)I


 2   2   
a + bc ab + bd a + ad ab + bd ad − bc 0
= − +
ac + cd bc + d2 ac + cd ad + d2 0 ad − bc
 
0 0
= .
0 0

From this we see that T 2 − (a + d)T + (ad − bc)I = 0.

3.4.5 Exercise 5
Let T be the linear operator on R3 , the matrix of which in the standard ordered
basis is  
1 2 1
A =  0 1 1 .
−1 3 4
Find a basis for the range of T and a basis for the null space of T .
3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 77

Solution. By Exercise 3.4.3, we know that the column space of A is the range of
T . We can find a basis for the column space of A by row-reducing its transpose
AT and taking the nonzero rows. We get
   
1 0 −1 1 0 −1
AT = 2 1 3  → 0 1 5  ,
1 1 4 0 0 0

so a basis for the range of T is given by

{(1, 0, −1), (0, 1, 5)}.

For the nullspace, we seek column vectors X for which AX = 0. Row-


reducing A gives    
1 2 1 1 0 −1
A =  0 1 1 → 0 1 1 ,
−1 3 4 0 0 0
so (x1 , x2 , x3 ) is in the null space of T if and only if x1 = x3 and x2 = −x3 .
That is, if and only if (x1 , x2 , x3 ) has the form (t, −t, t) for some scalar t. A
basis for the null space of T is therefore given by

{(1, −1, 1)}.

3.4.6 Exercise 6
Let T be the linear operator on R2 defined by

T (x1 , x2 ) = (−x2 , x1 ).

(a) What is the matrix of T in the standard ordered basis for R2 ?

Solution. Since

T (1, 0) = (0, 1) and T (0, 1) = (−1, 0),

the matrix of T relative to the standard ordered basis is


 
0 −1
[T ]{1 ,2 } = .
1 0

(b) What is the matrix of T in the ordered basis B = {α1 , α2 }, where

α1 = (1, 2) and α2 = (1, −1)?

Solution. The transition matrix P from B to the standard basis is


" #
  1 1
1 1 −1 3 3
P = , with inverse P = 2 .
2 −1 3 − 31

So,
" #  " 1 #
1 1 2
−3

−1 3 3 0 −1 1 1 3
[T ]B = P [T ]{1 ,2 } P = = .
2
3 − 31 1 0 2 −1 − 53 1
3
78 CHAPTER 3. LINEAR TRANSFORMATIONS

(c) Prove that for every real number c the operator (T − cI) is invertible.

Proof. Fix c in R and let U = T − cI. If B is the standard ordered basis


for R2 , then
     
0 −1 1 0 −c −1
[U ]B = [T ]B − cI = −c = .
1 0 0 1 1 −c
Since c2 + 1 must be nonzero, it is easily verified that
" #
− c2c+1 1  
c2 +1 1 c −1
= −
− c21+1 − c2c+1 c2 + 1 1 c

is an inverse for [U ]B . It follows that U is invertible, as required.

(d) Prove that if B is any ordered basis for R2 and [T ]B = A, then A12 A21 6= 0.

Proof. Let B = {α1 , α2 } where


α1 = (a, c) and α2 = (b, d)
so that
[T ]B = P −1 [T ]{1 ,2 } P,
where  
a b
P = .
c d
Since P is invertible, we know from Exercise 1.6.8 that ad − bc 6= 0, and
through direct computation we can find that
"
d b
#  " −cd−ab −d2 −b2 #


0 −1 a b
A = [T ]B = ad−bc c
ad−bc
a = cad−bc
2
+a2
ad−bc
.
− ad−bc ad−bc
1 0 c d cd+ab
ad−bc ad−bc
2 2
Now, if A12 = 0, then b +d = 0 so that we must have b = d = 0. But this
is impossible since ad − bc 6= 0. Similarly if A21 = 0 then a = c = 0 which
is again a contradiction. So we find that A12 and A21 are each nonzero,
and their product must be nonzero also.

3.4.7 Exercise 7
Let T be the linear operator on R3 defined by
T (x1 , x2 , x3 ) = (3x1 + x3 , −2x1 + x2 , −x1 + 2x2 + 4x3 ).
(a) What is the matrix of T in the standard ordered basis for R3 ?

Solution. Since
T (1, 0, 0) = (3, −2, −1),
T (0, 1, 0) = (0, 1, 2),
T (0, 0, 1) = (1, 0, 4),
the matrix of T in the standard ordered basis is
 
3 0 1
[T ]{1 ,2 ,3 } = −2 1 0 .
−1 2 4
3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 79

(b) What is the matrix of T in the ordered basis

{α1 , α2 , α3 }

where α1 = (1, 0, 1), α2 = (−1, 2, 1), and α3 = (2, 1, 1)?

Solution. The matrix of T in this basis is

[T ]{α1 ,α2 ,α3 } = P −1 [T ]{1 ,2 ,3 } P,

where  
1 −1 2
P = 0 2 1 .
1 1 1

So
 1
− 34 5

−4 4

3

0 1 1 −1 2

 1 1 1 
[T ]{α1 ,α2 ,α3 } = − 4 4 4  −2 1 0 0 2 1
1 1
− 21 −1 2 4 1 1 1
2 2
17 35 11
 
4 4 2
 3 15
= − 4 − 23  .

4
− 12 − 72 0

(c) Prove that T is invertible and give a rule for T −1 like the one which defines
T.

Proof. We already saw that the matrix A of T relative to the standard


ordered basis of R3 is  
3 0 1
A = −2 1 0 .
−1 2 4

Using the same methods as we used in Chapter 1, we can see that A is


invertible and  4 2
− 19

9 9
A−1 =  89 13
− 29  .
 
9
− 31 − 23 1
3

This implies that T is invertible and

T −1 (x1 , x2 , x3 )
 
4 2 1 8 13 2 1 2 1
= x1 + x2 − x3 , x1 + x2 − x3 , − x1 − x2 + x3 .
9 9 9 9 9 9 3 3 3
80 CHAPTER 3. LINEAR TRANSFORMATIONS

3.4.8 Exercise 8
Let θ be a real number. Prove that the following two matrices are similar over
the field of complex numbers:
   iθ 
cos θ − sin θ e 0
,
sin θ cos θ 0 e−iθ

Proof. Let T be the linear operator on C 2 which is represented by the first


matrix in the standard ordered basis. Let

α1 = (1, −i) and α2 = (−i, 1),

so that B = {α1 , α2 } is an ordered basis for C 2 . Then

T α1 = (cos θ + i sin θ, sin θ − i cos θ) = (eiθ , −ieiθ ) = eiθ α1

and

T α2 = (−i cos θ − sin θ, −i sin θ + cos θ) = (−ie−iθ , e−iθ ) = e−iθ α2 .

Thus the second matrix represents T in the ordered basis B. By Theorem 14,
the two matrices are similar.

3.4.9 Exercise 9
Let V be a finite-dimensional vector space over the field F and let S and T be
linear operators on V . We ask: When do there exist ordered bases B and B 0 for
V such that [S]B = [T ]B0 ? Prove that such bases exist if and only if there is an
invertible linear operator U on V such that T = U SU −1 .
Proof. Assume such bases exist, so that [S]B = [T ]B0 . Let U be the operator
which carries B onto B 0 . Then by Theorem 14, we have

[S]B = [T ]B0 = [U ]−1


B [T ]B [U ]B = [U
−1
T U ]B .

Since S and U −1 T U have the same matrix relative to B, it follows by Theorem 12


that S = U −1 T U . Thus we have shown that there is an invertible operator U
with T = U SU −1 .
Conversely, assume that T = U SU −1 for some invertible U and let B be
any ordered basis for V . Let B 0 be the image of B under U . Then, again by
Theorem 14, we have

[T ]B0 = [U ]−1
B [T ]B [U ]B = [U
−1
T U ]B = [S]B .

Thus the proof is complete.

3.4.10 Exercise 10
We have seen that the linear operator T on R2 defined by T (x1 , x2 ) = (x1 , 0) is
represented in the standard ordered basis by the matrix
 
1 0
A= .
0 0
3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 81

This operator satisfies T 2 = T . Prove that if S is a linear operator on R2 such


that S 2 = S, then S = 0, or S = I, or there is an ordered basis B for R2 such
that [S]B = A (above).

Proof. Assume that S is such that S 2 = S. Certainly 02 = 0 and I 2 = I, so if


S = 0 or S = I there is nothing left to prove. We will therefore assume that
S 6= 0 and S 6= I. Since S 6= 0, there is nonzero α1 in the range of S. So
Sβ1 = α1 for some β1 in R2 and we have

Sα1 = S(Sβ1 ) = Sβ1 = α1 . (3.1)

On the other hand, since S 6= I, there exist distinct β2 and β3 such that

Sβ2 = β3 .

Applying S to both sides, we get S 2 β2 = Sβ3 or Sβ2 = Sβ3 . Now letting


α2 = β3 − β2 , we get
Sα2 = Sβ3 − Sβ2 = 0. (3.2)
Next, if c1 α1 + c2 α2 = 0 for scalars c1 and c2 , then

c1 Sα1 + c2 Sα2 = 0.

Substituting equations (3.1) and (3.2), we see c1 α1 = 0, which implies that


c1 = 0 since α1 is nonzero. So c2 α2 = 0 and we must have c2 = 0 since α2 is
also nonzero. Thus the set B = {α1 , α2 } is a set of two linearly independent
vectors in R2 . Therefore B spans R2 and is a basis.
Finally, we see that (3.1) and (3.2) together imply that
 
1 0
[S]B = = A,
0 0

so the proof is complete.

3.4.11 Exercise 11
Let W be the space of all n × 1 column matrices over a field F . If A is an
n × n matrix over F , then A defines a linear operator LA on W through left
multiplication: LA (X) = AX. Prove that every linear operator on W is left
multiplication by some n × n matrix, i.e., is LA for some A.
Now suppose V is an n-dimensional vector space over the field F , and let B
be an ordered basis for V . For each α in V , define U α = [α]B . Prove that U is
an isomorphism of V onto W . If T is a linear operator on V , then U T U −1 is a
linear operator on W . Accordingly, U T U −1 is left multiplication by some n × n
matrix A. What is A?

Solution. Let C be the standard ordered basis for W = F n×1 , i.e. the jth vector
in the basis has a 1 in its jth row and all other entries zero. Then if S is any
linear operator on W , we may take the n × n matrix A to be

A = [S]C .
82 CHAPTER 3. LINEAR TRANSFORMATIONS

Then LA and S have the same matrix relative to the ordered basis C, and
therefore LA = S. Therefore every linear operator on W is left multiplication
by some n × n matrix.
Now let V be an n-dimensional vector space over F and B = {α1 , . . . , αn }
an ordered basis for V . Define U α = [α]B , so that U is a map of V into W . We
will show that U is an isomorphism.
First, for any β1 , β2 in V and scalar c in F , we have

U (cβ1 + β2 ) = [cβ1 + β2 ]B
= c[β1 ]B + [β2 ]B
= cU β1 + U β2 ,

and U is a linear transformation.


Now choose Y in W , where
 
y1
 y2 
Y =  . .
 
 .. 
yn

Define
β = y1 α1 + y2 α2 + · · · + yn αn .

Then
U β = [β]B = Y.

So Y is in the range of U , and U is therefore onto. By Theorem 9, U is invertible


and therefore an isomorphism of V onto W .
Finally, if T is a linear operator on V , then U T U −1 is a linear operator on
W and hence U T U −1 = LA for some n × n matrix A. Choose any X in W ,
with
 
x1
 x2 
X =  . .
 
 .. 
xn

Then

U T U −1 (X) = U T (x1 α1 + x2 α2 + · · · + xn αn )
= U (x1 T α1 + x2 T α2 + · · · + xn T αn )
= x1 [T α1 ]B + x2 [T α2 ]B + · · · + xn [T αn ]B
= x1 [T ]B [α1 ]B + x2 [T ]B [α2 ]B + · · · + xn [T ]B [αn ]B
= [T ]B (x1 U α1 + x2 U α2 + · · · + xn U αn )
= [T ]B X.

We see that A = [T ]B , so U T U −1 = L[T ]B .


3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 83

3.4.12 Exercise 12
Let V be an n-dimensional vector space over the field F , and let

B = {α1 , . . . , αn }

be an ordered basis for V .


(a) According to Theorem 1, there is a unique linear operator T on V such
that
T αj = αj+1 , j = 1, . . . , n − 1, T αn = 0.
What is the matrix A of T in the ordered basis B?

Solution. For each j = 1, . . . , n − 1, we have Aj+1,j = 1 and all other


entries 0. That is,
 
0 0 0 ··· 0 0
1 0
 0 ··· 0 0 
0 1 0 ··· 0 0
A = [T ]B = 0 0 .
 
 1 ··· 0 0 
 .. .. .. .. .... 
. . . . . .
0 0 0 ··· 1 0

(b) Prove that T n = 0 but T n−1 6= 0.

Proof. Since T n−1 α1 = αn 6= 0, we see that T n−1 6= 0. On the other


hand,
T n α1 = T (T n−1 α1 ) = T αn = 0,
and we can similarly see that T n αj = 0 for all j with 1 ≤ j ≤ n. Since
any vector α can be written as a linear combination of the basis vectors
α1 , . . . , αn , it follows by the linearity of T that T n α = 0. Therefore
T n = 0.

(c) Let S be any linear operator on V such that S n = 0 but S n−1 6= 0. Prove
that there is an ordered basis B 0 for V such that the matrix of S in the
ordered basis B 0 is the matrix A of part (a).

Proof. Since S n−1 6= 0, there is a nonzero β1 in V such that S n−1 β1 is


nonzero. Now, for each j with 2 ≤ j ≤ n, define

βj = S j−1 β1 .

Let B 0 = {β1 , . . . , βn }.
Now suppose c1 , c2 , . . . , cn are scalars in F such that

c1 β1 + c2 β2 + · · · + cn βn = 0.

Taking S n−1 of both sides gives c1 βn = 0, which implies that c1 = 0. So

c2 β2 + c3 β3 + · · · + cn βn = 0,
84 CHAPTER 3. LINEAR TRANSFORMATIONS

and we can take S n−2 of both sides to get c2 βn = 0, implying that c2 is


zero. More generally, assuming that c1 , . . . , ck are all zero for some k with
1 ≤ k ≤ n − 1, we have

ck+1 βk+1 + · · · + cn βn = 0.

Taking S n−k−1 of both sides (in the case where k = n−1, we take S 0 = I)
then gives ck+1 βn = 0, so that ck+1 = 0. Therefore c1 = c2 = · · · = cn = 0
and B 0 is linearly independent. Since dim V = n and B 0 is a linearly
independent set of n vectors in V , it follows that B 0 is a basis for V .
Finally, we have defined β1 , . . . , βn so that

Sβj = βj+1 , j = 1, . . . , n − 1, Sβn = 0.

Therefore we have [S]B0 = A.

(d) Prove that if M and N are n × n matrices over F such that M n = N n = 0


but M n−1 6= 0 6= N n−1 , then M and N are similar.

Proof. Let P and Q be the linear operators on V whose matrices relative to


B are, respectively, M and N . Then P n = Qn = 0 but P n−1 6= 0 6= Qn−1 .
We have shown above that there are bases C and C 0 for V such that

[P ]C = A = [Q]C 0 .

By Exercise 3.4.9, there is an invertible linear transformation U such that


Q = U P U −1 . Therefore

N = [Q]B = [U ]B [P ]B [U ]−1 −1
B = [U ]B M [U ]B

and we see that M and N are similar matrices.

3.4.13 Exercise 13
Let V and W be finite-dimensional vector spaces over the field F and let T be
a linear transformation from V into W . If

B = {α1 , . . . , αn } and B 0 = {β1 , . . . , βm }

are ordered bases for V and W , respectively, define the linear transformations
E p,q as in the proof of Theorem 5: E p,q (αi ) = δiq βp . Then the E p,q , 1 ≤ p ≤ m,
1 ≤ q ≤ n, form a basis for L(V, W ), and so
m X
X n
T = Apq E p,q
p=1 q=1

for certain scalars Apq (the coordinates of T in this basis for L(V, W )). Show
that the matrix A with entries A(p, q) = Apq is precisely the matrix of T relative
to the pair B, B 0 .
3.4. REPRESENTATION OF TRANSFORMATIONS BY MATRICES 85

Solution. Note that for each j = 1, . . . , n,


m X
X n
T αj = Apq E p,q (αj )
p=1 q=1
Xm X n
= Apq δjq βp
p=1 q=1
Xm
= Apj βp .
p=1

Thus the ith coordinate of T αj is Aij and we see that the matrix of T relative
to B, B 0 is precisely the matrix A.
86 CHAPTER 3. LINEAR TRANSFORMATIONS

3.5 Linear Functionals


3.5.1 Exercise 1
In R3 , let α1 = (1, 0, 1), α2 = (0, 1, −2), α3 = (−1, −1, 0).
(a) If f is a linear functional on R3 such that
f (α1 ) = 1, f (α2 ) = −1, f (α3 ) = 3,
and if α = (a, b, c), find f (α).

Solution. Suppose f (x1 , x2 , x3 ) = c1 x1 + c2 x2 + c3 x3 . Then


f (α1 ) = c1 + c3 = 1,
f (α2 ) = c2 − 2c3 = −1,
f (α3 ) = −c1 − c2 = 3.
Row-reducing the augmented matrix for this system gives
   
1 0 1 1 1 0 0 4
0 1 −2 −1 → 0 1 0 −7 ,
−1 −1 0 3 0 0 1 −3
so c1 = 4, c2 = −7, and c3 = −3. Therefore
f (α) = 4a − 7b − 3c.

(b) Describe explicitly a linear functional f on R3 such that


f (α1 ) = f (α2 ) = 0 but f (α3 ) 6= 0.

Solution. For example, suppose f (α1 ) = f (α2 ) = 0 but f (α3 ) = 1. As


above, this leads to a system of linear equations having augmented matrix
   
1 0 1 0 1 0 0 1
0 1 −2 0 → 0 1 0 −2 .
−1 −1 0 1 0 0 1 −1
So we may write
f (x1 , x2 , x3 ) = x1 − 2x2 − x3 .

(c) Let f be any linear functional such that


f (α1 ) = f (α2 ) = 0 and f (α3 ) 6= 0.
If α = (2, 3, −1), show that f (α) 6= 0.

Solution. By inspection, we see that


α = −α1 − 3α3 .
Therefore
f (α) = f (−α1 − 3α3 )
= −f (α1 ) − 3f (α3 )
= −3f (α3 ) 6= 0.
3.5. LINEAR FUNCTIONALS 87

3.5.2 Exercise 2
Let B = {α1 , α2 , α3 } be the basis for C 3 defined by

α1 = (1, 0, −1), α2 = (1, 1, 1), α3 = (2, 2, 0).

Find the dual basis of B.

Solution. Let {f1 , f2 , f3 } be the dual basis of B, and let


 
1 1 2
P = 0 1 2 ,
−1 1 0

so that P is the transition matrix from B to the standard ordered basis of C 3 .


We find  
1 −1 0
P −1 =  1 −1 1  .
 

− 12 1 − 12

So, given a vector α = (x1 , x2 , x3 ) in C3 , we can write


 
x1
[α]B = P −1 x2  .
x3

Therefore
  
1 −1 0 x1
 
f1 (x1 , x2 , x3 ) = 1 0 0  1 −1 1  x2  = x1 − x2 .

− 21 1 −1 x3
2

Similarly, we get
  
1 −1 0 x1
 
f2 (x1 , x2 , x3 ) = 0 1 0  1 −1 1  x2  = x1 − x2 + x3 ,

− 21 1 −1 x3
2

and
  
1 −1 0 x1
  1 1
f3 (x1 , x2 , x3 ) = 0 0 1  1 −1 1  x2  = − x1 + x2 − x3 .

2 2
− 21 1 −1 x3
2

3.5.3 Exercise 3
If A and B are n×n matrices over the field F , show that trace(AB) = trace(BA).
Now show that similar matrices have the same trace.
88 CHAPTER 3. LINEAR TRANSFORMATIONS

Proof. We may directly compute


n
X
tr(AB) = (AB)ii
i=1
n X
X n n X
X n
= Aij Bji = Bji Aij
i=1 j=1 j=1 i=1
Xn
= (BA)jj
j=1

= tr(BA).

So tr(AB) = tr(BA).
Next, suppose A and B are similar, and let P be an invertible n × n matrix
such that B = P −1 AP . Using the fact that was proven above, we get

tr(B) = tr(P −1 AP )
= tr((P −1 A)P )
= tr(P (P −1 A))
= tr((P P −1 )A)
= tr(A).

This shows that similar matrices have the same trace.

3.5.4 Exercise 4
Let V be the vector space of all polynomial functions p from R into R which
have degree 2 or less:
p(x) = c0 + c1 x + c2 x2 .
Define three linear functionals on V by
Z 1 Z 2 Z −1
f1 (p) = p(x) dx, f2 (p) = p(x) dx, f3 (p) = p(x) dx.
0 0 0

Show that {f1 , f2 , f3 } is a basis for V ∗ by exhibiting the basis for V of which it
is the dual.
Solution. First we evaluate,
  1
1 2 1 3 1 1
f1 (p) = c0 x + c1 x + c2 x = c0 + c1 + c2 ,
2 3 2 3
0
  2
1 1 8
f2 (p) = c0 x + c1 x2 + c2 x3 = 2c0 + 2c1 + c2 ,

2 3 3
0
  −1

1 1 1 1
f3 (p) = c0 x + c1 x2 + c2 x3 = −c0 + c1 − c2 .

2 3 2 3
0

Now, let {p1 , p2 , p3 } be the basis for V of which {f1 , f2 , f3 } is the dual. To
determine pi , we want to find values for the coefficients c1 , c2 , and c3 so that
3.5. LINEAR FUNCTIONALS 89

fi (pi ) = 1 and fj (pi ) = 0 for j 6= i. This gives three systems of linear equations,
having augmented matrices
1 1 1 1 1 1
     
1 2 3 1 1 2 3 0 1 2 3 0
8 8 8
2 2 0 ,  2 2 1 , and  2 2 0 .
     
3 3 3
1
−1 2 − 13 0 −1 1
2 − 31 0 −1 1
2 − 13 1
We can combine these into one augmented matrix and perform row-reduction,
which gives
1 1
− 61 − 31
   
1 2 3 1 0 0 1 0 0 1
8
2 2 0 1 0 → 0 1 0 1 0 1 .
   
3
1 1
−1 2 − 3 0 0 1 0 0 1 − 32 1
2 − 21
So we see that
3
p1 (x) = 1 + x − x2 ,
2
1 1 2
p2 (x) = − + x ,
6 2
1 1
p3 (x) = − + x − x2 ,
3 2
and {f1 , f2 , f3 } is the dual basis of {p1 , p2 , p3 }.

3.5.5 Exercise 5
If A and B are n × n complex matrices, show that AB − BA = I is impossible.
Proof. In Example 19, the trace function was shown to be a linear functional
on the space of n × n matrices. And in Exercise 3.5.3, we proved that, given
two matrices A and B, tr(AB) = tr(BA). It now follows that

tr(AB − BA) = tr(AB) − tr(BA) = 0.

But tr(I) = n 6= 0, so it cannot be the case that AB − BA = I.

3.5.6 Exercise 6
Let m and n be positive integers and F a field. Let f1 , . . . , fm be linear func-
tionals on F n . For α in F n define

T α = (f1 (α), . . . , fm (α)).

Show that T is a linear transformation from F n into F m . Then show that every
linear transformation from F n into F m is of the above form, for some f1 , . . . , fm .
Proof. First, for any α1 , α2 in F n and any c in F , we have

T (cα1 + α2 ) = (f1 (cα1 + α2 ), . . . , fm (cα1 + α2 ))


= (cf1 (α1 ) + f1 (α2 ), . . . , cfm (α1 ) + fm (α2 ))
= c(f1 (α1 ), . . . , fm (α1 )) + (f1 (α2 ), . . . , fm (α2 ))
= cT α1 + T α2 .
90 CHAPTER 3. LINEAR TRANSFORMATIONS

This shows that T is a linear transformation from F n into F m .


Next, let T be any linear transformation from F n into F m . For each i with
1 ≤ i ≤ m, define fi (α) to be the ith coordinate of T α. Then

T α = (f1 (α), . . . , fm (α))

and, moreover, each fi is a linear transformation because T itself is linear.


Therefore every linear transformation from F n into F m can be written this
way.

3.5.7 Exercise 7
Let α1 = (1, 0, −1, 2) and α2 = (2, 3, 1, 1), and let W be the subspace of R4
spanned by α1 and α2 . Which linear functionals f :

f (x1 , x2 , x3 , x4 ) = c1 x1 + c2 x2 + c3 x3 + c4 x4

are in the annihilator of W ?

Solution. Let f be in W 0 . We want

f (1, 0, −1, 2) = f (2, 3, 1, 1) = 0.

This leads to a system of equations in c1 , c2 , c3 , c4 having coefficient matrix


   
1 0 −1 2 1 0 −1 2
→ .
2 3 1 1 0 1 1 −1

From the reduced form we see that c3 and c4 can be arbitrary, with

c1 = c3 − 2c4 and c2 = c4 − c3 .

Therefore W 0 consists of the linear functionals f having the form

f (x1 , x2 , x3 , x4 ) = (s − 2t)x1 + (t − s)x2 + sx3 + tx4 ,

where s and t are scalars in F . Note that we can also find a basis for W 0 by
first taking s = 1, t = 0 and then by taking s = 0, t = 1.

3.5.8 Exercise 8
Let W be the subspace of R5 which is spanned by the vectors

α1 = 1 + 22 + 3 , α2 = 2 + 33 + 34 + 5


α3 = 1 + 42 + 63 + 44 + 5 .

Find a basis for W 0 .

Solution. Take f in W 0 , and write

f (x1 , x2 , x3 , x4 , x5 ) = c1 x1 + c2 x2 + c3 x3 + c4 x4 + c5 x5 .
3.5. LINEAR FUNCTIONALS 91

Since f annihilates W , we have

f (α1 ) = c1 + 2c2 + c3 =0
f (α2 ) = c2 + 3c3 + 3c4 + c5 = 0
f (α3 ) = c1 + 4c2 + 6c3 + 4c4 + c5 = 0.

The coefficient matrix for this system reduces as follows:


   
1 2 1 0 0 1 0 0 4 3
0 1 3 3 1 → 0 1 0 −3 −2 .
1 4 6 4 1 0 0 1 2 1
Since the latter matrix has three nonzero rows, we see that W has dimension 3
and {α1 , α2 , α3 } is a basis for W . We also see that

c1 = −4c4 − 3c5 ,
c2 = 3c4 + 2c5 ,

and

c3 = −2c4 − c5 .

Therefore the set {f1 , f2 }, where

f1 (x1 , x2 , x3 , x4 , x5 ) = −4x1 + 3x2 − 2x3 + x4

and
f2 (x1 , x2 , x3 , x4 , x5 ) = −3x1 + 2x2 − x3 + x5 ,
is a basis for W . Note that dim W 0 = 2, which agrees with Theorem 16.
0

3.5.9 Exercise 9
Let V be the vector space of all 2 × 2 matrices over the field of real numbers,
and let  
2 −2
B= .
−1 1
Let W be the subspace of V consisting of all A such that AB = 0. Let f be a
linear functional on V which is in the annihilator of W . Suppose that f (I) = 0
and f (C) = 3, where I is the 2 × 2 identity matrix and
 
0 0
C= .
0 1

Find f (B).
Solution. Note that
    
a b 2 −2 2a − b −2a + b
= =0
c d −1 1 2c − d −2c + d
if and only if 2a = b and 2c = d. Therefore, a basis for W is
   
1 2 0 0
, .
0 0 1 2
92 CHAPTER 3. LINEAR TRANSFORMATIONS

Now, let
f (A) = c1 A11 + c2 A12 + c3 A21 + c4 A22 ,
where A is any 2 × 2 matrix. We know that f annihilates the two basis vectors
for W found above, and we also know that f (I) = 0 and f (C) = 3. This leads
to the following system of linear equations:

c1 + 2c2 = 0
c3 + 2c4 = 0
c1 + c4 = 0
c4 = 3.

This system has the unique solution

3
c1 = −3, c2 = , c3 = −6, c4 = 3,
2
so
3
f (A) = −3A11 + A12 − 6A21 + 3A22 .
2
Therefore
3
f (B) = −3(2) + (−2) − 6(−1) + 3(1) = 0.
2

3.5.10 Exercise 10
Let F be a subfield of the complex numbers. We define n linear functionals on
F n (n ≥ 2) by
n
X
fk (x1 , . . . , xn ) = (k − j)xj , 1 ≤ k ≤ n.
j=1

What is the dimension of the subspace annihilated by f1 , . . . , fn ?

Solution. Call the subspace W . We first want to find dim W 0 . Fix some n ≥ 2
and consider the set B = {f1 , f2 }. Since f1 (α) = f2 (α) = 0 for any α in W , we
have

−x2 − 2x3 − 3x4 − · · · − (n − 1)xn = 0,


x1 − x3 − 2x4 − · · · − (n − 2)xn = 0.

The coefficient matrix for the above system of equations is given by


 
0 −1 −2 −3 · · · 1 − n
A= .
1 0 −1 −2 · · · 2 − n

Observe that by multiplying the first row by −1 and interchanging the two
rows, we can put A in row-reduced echelon form. Since A is row-equivalent to
a row-reduced matrix having two nonzero rows, we see that the set B is linearly
independent.
3.5. LINEAR FUNCTIONALS 93

Next, consider fk for some k with 3 ≤ k ≤ n. Then we can write


n
X
fk (x1 , . . . , xn ) = (k − j)xj
j=1
Xn n
X
= (−k − 2j + 2 + jk)xj + (2k + j − 2 − jk)xj
j=1 j=1
n
X n
X
= (2 − k) (1 − j)xj + (k − 1) (2 − j)xj
j=1 j=1

= (2 − k)f1 (x1 , . . . , xn ) + (k − 1)f2 (x1 , . . . , xn ).

This shows that B spans the annihilator of W . Therefore B is a basis for W 0


and dim W 0 = 2. It now follows from Theorem 16 that dim W = n − 2.

3.5.11 Exercise 11
Let W1 and W2 be subspaces of a finite-dimensional vector space V .

(a) Prove that (W1 + W2 )0 = W10 ∩ W20 .

Proof. If a linear functional f belongs to (W1 + W2 )0 , then it annihilates


every vector in W1 + W2 . But W1 and W2 are subspaces of W1 + W2 , so
f must belong to W10 ∩ W20 .
Conversely, if f belongs to W10 ∩ W20 , then f annihilates all vectors in W1 ,
and also annihilates all vectors in W2 . Since f is linear, it must therefore
annihilate sums of these vectors, so f is in (W1 + W2 )0 .
We have shown that members of (W1 + W2 )0 are members of W10 ∩ W20
and vice versa, so these spaces are equal.

(b) Prove that (W1 ∩ W2 )0 = W10 + W20 .

Proof. Let B = {α1 , . . . , αm } be a basis for W1 ∩ W2 . Extend this basis


to a basis B 0 for W1 , where

B 0 = {α1 , . . . , αm , β1 , . . . , βn }.

Also extend B to a basis B 00 for W2 , where

B 00 = {α1 , . . . , αm , γ1 , . . . , γp }.

Now take any linear functional f belonging to (W1 ∩ W2 )0 . Let g be the


linear functional on V such that

g(α1 ) = · · · = g(αm ) = 0,
g(β1 ) = · · · = g(βn ) = 0,
g(γ1 ) = f (γ1 ), ..., g(γp ) = f (γp ).
94 CHAPTER 3. LINEAR TRANSFORMATIONS

Notice that g belongs to W10 . Similarly, define h to be the linear functional


given by

h(α1 ) = · · · = h(αm ) = 0,
h(γ1 ) = · · · = h(γp ) = 0,
h(β1 ) = f (β1 ), ..., h(βn ) = f (βn ).

Notice that h belongs to W20 . Moreover, f = g + h. This shows that f


belongs to the sum W10 + W20 , and we see that (W1 ∩ W2 )0 is a subset of
W10 + W20 .
Next, suppose f = f1 + f2 , where f1 is in W10 and f2 is in W20 . Since
W1 ∩ W2 is a subspace of W1 and also a subspace of W2 , it follows that
both f1 and f2 annihilate W1 ∩ W2 , i.e. f belongs to (W1 ∩ W2 )0 . This
completes the proof that (W1 ∩ W2 )0 = W10 + W20 .

3.5.12 Exercise 12
Let V be a finite-dimensional vector space over the field F and let W be a
subspace of V . If f is a linear functional on W , prove that there is a linear
functional g on V such that g(α) = f (α) for each α in the subspace W .

Proof. Let B = {α1 , . . . , αm } be a basis for W and extend this basis to a basis

B 0 = {α1 , . . . , αm , β1 , . . . , βn }

for V .
Let f be any linear functional on W . Define the linear functional g on V by

g(αi ) = f (αi ), for 1 ≤ i ≤ m, and g(βj ) = 0, for 1 ≤ j ≤ n.

We know g exists by Theorem 1. So g is a linear functional on V that agrees


with f on W , as we wanted to show.

3.5.13 Exercise 13
Let F be a subfield of the field of complex numbers and let V be any vector
space over F . Suppose that f and g are linear functionals on V such that the
function h defined by h(α) = f (α)g(α) is also a linear functional on V . Prove
that either f = 0 or g = 0.

Proof. Let f , g, and h be the linear functionals on V with the properties de-
scribed above.
Choose any α, β in V . Then

h(α + β) = f (α + β)g(α + β)
= (f (α) + f (β))(g(α) + g(β))
= f (α)g(α) + f (α)g(β) + f (β)g(α) + f (β)g(β)
= h(α) + h(β) + f (α)g(β) + f (β)g(α)
= h(α + β) + f (α)g(β) + f (β)g(α).
3.5. LINEAR FUNCTIONALS 95

This shows that for any pair of vectors α, β in V ,

f (α)g(β) + f (β)g(α) = 0. (3.3)

Taking β = α, we also see that h = 0.


Now, either f = 0 or f 6= 0. If f = 0 then there is nothing left to prove, so
we will assume that f 6= 0. Again, let α in V be arbitrary, and let β in V be
such that f (β) is nonzero. We know that

f (β)g(β) = 0,

and it follows that g(β) = 0 (since f (β) 6= 0). Substituting zero for g(β) in
equation (3.3) then gives
f (β)g(α) = 0.
But, again, f (β) is nonzero, so we must have g(α) = 0. Since α was chosen
arbitrarily, it follows that g = 0 and the proof is complete.

3.5.14 Exercise 14
Let F be a field of characteristic zero and let V be a finite-dimensional vector
space over F . If α1 , . . . , αm are finitely many vectors in V , each different from
the zero vector, prove that there is a linear functional f on V such that

f (αi ) 6= 0, i = 1, . . . , m.

Proof. Let B = {β1 , . . . , βn } be a basis for V . We will use induction on m to


show that we can always find a linear functional f such that f (αi ) 6= 0 for all i.
First, when m = 1, we have a single nonzero vector α1 . Writing α1 as a
linear combination of the vectors in B gives

α 1 = c1 β1 + c2 β2 + · · · + cn βn ,

for some scalars c1 , . . . , cn in F . Since α1 6= 0, there is an index k such that


ck 6= 0. Now define f to be the linear functional on V such that

f (βi ) = δik c−1


k ,

where δik is the Kronecker delta. We now have

f (α1 ) = ck c−1
k = 1 6= 0.

This shows that the statement holds for the base case of m = 1.
Now assume that the statement is true when m = k for some k ≥ 1, and
let k + 1 nonzero vectors α1 , . . . , αk+1 be given. We may apply the inductive
hypothesis to find a linear functional f0 on V such that f0 (αi ) 6= 0 for all i with
1 ≤ i ≤ k.
If f0 (αk+1 ) happens to be nonzero, then we are done. So we will assume
that f0 (αk+1 ) = 0. Again the inductive hypothesis, when applied to the single
vector αk+1 , allows us to find a linear functional f1 such that f1 (αk+1 ) 6= 0.
For each i with 1 ≤ i ≤ k + 1, define the number Ni as follows. First, if
there is no positive integer n such that

f0 (αi ) + nf1 (αi ) = 0, (3.4)


96 CHAPTER 3. LINEAR TRANSFORMATIONS

then set Ni = 1. Otherwise, define Ni to be the unique positive integer such


that
f0 (αi ) + Ni f1 (αi ) = 0.
In this second case, we know that Ni is unique for the following reason. Suppose
n = M and n = Ni both satisfy (3.4). By construction, it is not possible for
f0 (αi ) and f1 (αi ) to both be zero. But if one is zero, then (3.4) implies that
the other is as well (because F has characteristic zero). So we see that neither
f0 (αi ) nor f1 (αi ) is zero, hence
f0 (αi )
M =− = Ni .
f1 (αi )
So Ni is well-defined for all i = 1, . . . , k + 1.
Now, the set
A = {N1 , N2 , . . . , Nk+1 }
is a finite set of natural numbers, so we can find a largest element in A. Take
any natural number P greater than this largest element, and define the linear
functional f on V by
f = f0 + P f1 .
Then we see that f (αi ) 6= 0 for each i = 1, . . . , k + 1, completing the inductive
step of the proof.
By induction, the original statement must be true for all positive integers
m.

3.5.15 Exercise 15
According to Exercise 3.5.3, similar matrices have the same trace. Thus we
can define the trace of a linear operator on a finite-dimensional space to be the
trace of any matrix which represents the operator in an ordered basis. This is
well-defined since all such representing matrices for one operator are similar.
Now let V be the space of all 2 × 2 matrices over the field F and let P be a
fixed 2 × 2 matrix. Let T be the linear operator on V defined by T (A) = P A.
Prove that trace(T ) = 2 trace(P ).
Proof. Let  
a b
P =
c d
and let        
1 0 0 1 0 0 0 0
B= , , ,
0 0 0 0 1 0 0 1
be an ordered basis for V . We calculate
       
1 0 a 0 1 0 0 0
T = =a +c ,
0 0 c 0 0 0 1 0
       
0 1 0 a 0 1 0 0
T = =a +c ,
0 0 0 c 0 0 0 1
       
0 0 b 0 1 0 0 0
T = =b +d ,
1 0 d 0 0 0 1 0
       
0 0 0 b 0 1 0 0
T = =b +d .
0 1 0 d 0 0 0 1
3.5. LINEAR FUNCTIONALS 97

From this, we see that the matrix for T relative to B is the 4 × 4 matrix
 
a 0 b 0
0 a 0 b
[T ]B = 
c
.
0 d 0
0 c 0 d

We can now readily see that

trace(T ) = tr([T ]B ) = 2a + 2d = 2 tr(P ).

3.5.16 Exercise 16
Show that the trace functional on n × n matrices is unique in the following
sense. If W is the space of n × n matrices over the field F and if f is a linear
functional on W such that f (AB) = f (BA) for each A and B in W , then f is
a scalar multiple of the trace function. If, in addition, f (I) = n, then f is the
trace function.
Proof. Let
B = {11 , 12 , . . . , 1n , . . . , n1 , n2 , . . . , nn }
be the basis for W where ij is the n × n matrix having a 1 in the i, jth entry
and all other entries 0. Since f is linear, we may write it as
n X
X n
f (A) = Cij Aij , (3.5)
i=1 j=1

where each Cij is a fixed constant and Aij is the i, jth entry of A.
Note that the p, qth entry of ij is δip δqj , where δij is the Kronecker delta.
So, if we fix some indices i, j, a, b, each in the range from 1 to n, then we have
X n
n X n
X
f (ij ab ) = Cpq (δip δkj )(δak δqb ) = Cib δaj . (3.6)
p=1 q=1 k=1

From this, we see that

f (ij ji ) = Cii and f (ji ij ) = Cjj .

Since these must be equal, we see that C11 = C22 = · · · = Cnn . On the other
hand, (3.6) also gives

f (i1 1j ) = Cij and f (1j i1 ) = C11 δij .

These must be equal, so by looking at values of i and j where i 6= j, we see that


Cij = 0 whenever i 6= j. Therefore, equation (3.5) can be simplified to
n
X
f (A) = CAkk = C(A11 + A22 + · · · + Ann ) = C tr(A),
k=1

where C = C11 is a constant. We conclude that f is a scalar multiple of the


trace function. If we require f (I) = n, then we must have C = 1 and f = tr.
98 CHAPTER 3. LINEAR TRANSFORMATIONS

3.5.17 Exercise 17
Let W be the space of n × n matrices over the field F , and let W0 be the
subspace spanned by the matrices C of the form C = AB − BA. Prove that W0
is exactly the subspace of matrices which have trace zero.
Proof. Let W1 be the subspace of matrices having trace zero. From Exer-
cise 3.5.3, we know that

tr(AB − BA) = tr(AB) − tr(BA) = 0,

so every matrix of the form AB − BA must be in W1 . If we can show that


dim W0 = dim W1 , then the proof will be complete.
Since W1 is the null space of a nonzero linear functional, it must have di-
mension equal to dim(W ) − 1 = n2 − 1. Since W0 is a proper subset of W ,
it must have dimension at most n2 − 1. So we need only find a set of n2 − 1
linearly independent matrices in W0 .
Let ij denote the matrix whose i, jth entry is 1 and all other entries 0. For
each i with 2 ≤ i ≤ n, let

Si = ii − 11 = i1 1i − 1i i1 .

Then the set S = {S2 , . . . , Sn } is a subset of W0 . Let T be the set of matrices


ij with i 6= j. Then T is also a subset of W0 , since we can write

ij = ij jj − jj ij , i 6= j

(note that jj ij = 0). Now the set S ∪ T consists of

(n − 1) + (n2 − n) = n2 − 1

linearly independent vectors belonging to W0 , as needed to complete the proof.


3.6. THE DOUBLE DUAL 99

3.6 The Double Dual


3.6.1 Exercise 1
Let n be a positive integer and F a field. Let W be the set of all vectors
(x1 , . . . , xn ) in F n such that x1 + · · · + xn = 0.

(a) Prove that W 0 consists of all linear functionals f of the form


n
X
f (x1 , . . . , xn ) = c xj .
j=1

Proof. If n = 1, then W is the zero subspace and the result is trivial, so


we will suppose n > 1.
Let f be in W 0 . We can find scalars c1 , . . . , cn in F such that

f (x1 , . . . , xn ) = c1 x1 + · · · + cn xn .

We want to show that c1 = c2 = · · · = cn .


Since f annihilates W , in particular we know that

f (1, −1, 0, 0, . . . , 0) = c1 − c2 = 0.

Therefore c1 = c2 . Likewise, we know that

f (0, 1, −1, 0, 0, . . . , 0) = c2 − c3 = 0,

so c2 = c3 . Continuing in this way, we see that the ci must all be identical.


Thus f has the form that was specified.

(b) Show that the dual space W ∗ of W can be ‘naturally’ identified with the
linear functionals

f (x1 , . . . , xn ) = c1 x1 + · · · + cn xn

on F n which satisfy c1 + · · · + cn = 0.

Proof. Let U be the space of linear functionals

f (x1 , . . . , xn ) = c1 x1 + · · · + cn xn

on F n which satisfy c1 + · · · + cn = 0. We will show that U is isomorphic


to W ∗ .
Let T be the function from U into W ∗ such that T (f ) is the restriction
of f to W . Then T is a linear transformation. We will show that it is
non-singular. Suppose T (f ) = 0. Then f (α) = 0 for all α in W , hence f
belongs to W 0 . By the result from part (a), we know that f has the form
n
X
f (x1 , . . . , xn ) = c xj .
j=1
100 CHAPTER 3. LINEAR TRANSFORMATIONS

But f belongs to U , so c must be 0 and f is the zero functional. That


is, we have shown that T (f ) = 0 implies f = 0, so T is non-singular and
hence one-to-one.
Now, it can be shown that U and W both have dimension n − 1. For
example, if {1 , . . . , n } is the standard ordered basis for F n , then

{1 − i | 2 ≤ i ≤ n}

is a set of n − 1 linearly independent vectors which span W . We can


find a similar basis for U . Since T is a one-to-one linear transformation
between vector spaces of equal dimension, T must be invertible and thus
is an isomorphism.
3.7. THE TRANSPOSE OF A LINEAR TRANSFORMATION 101

3.7 The Transpose of a Linear Transformation


3.7.1 Exercise 1
Let F be a field and let f be the linear functional on F 2 defined by

f (x1 , x2 ) = ax1 + bx2 .

For each of the following linear operators T , let g = T t f , and find g(x1 , x2 ).
(a) T (x1 , x2 ) = (x1 , 0)

Solution. We have

g(x1 , x2 ) = (T t f )(x1 , x2 )
= f (T (x1 , x2 ))
= f (x1 , 0)
= ax1 .

(b) T (x1 , x2 ) = (−x2 , x1 )

Solution. In this case, we get

g(x1 , x2 ) = f (T (x1 , x2 ))
= f (−x2 , x1 )
= −ax2 + bx1 .

(c) T (x1 , x2 ) = (x1 − x2 , x1 + x2 )

Solution.

g(x1 , x2 ) = f (x1 − x2 , x1 + x2 )
= (b + a)x1 + (b − a)x2 .

3.7.2 Exercise 2
Let V be the vector space of all polynomial functions over the field of real
numbers. Let a and b be fixed real numbers and let f be the linear functional
on V defined by
Z b
f (p) = p(x) dx.
a

If D is the differentiation operator on V , what is Dt f ?


Solution. From the definition, we have
Z b
(Dt f )(p) = f (Dp) = (Dp)(x) dx.
a

So, the fundamental theorem of calculus gives

(Dt f )(p) = p(b) − p(a).


102 CHAPTER 3. LINEAR TRANSFORMATIONS

3.7.3 Exercise 3
Let V be the space of all n × n matrices over a field F and let B be a fixed n × n
matrix. If T is the linear operator on V defined by T (A) = AB − BA, and if f
is the trace function, what is T t f ?

Solution. From Exercise 3.5.3, we know that f (AB) = f (BA). So,

(T t f )(A) = f (T A)
= f (AB − BA)
= f (AB) − f (BA)
= 0,

and we see that T t f = 0.

3.7.4 Exercise 4
Let V be a finite-dimensional vector space over the field F and let T be a linear
operator on V . Let c be a scalar and suppose there is a non-zero vector α in
V such that T α = cα. Prove that there is a non-zero linear functional f on V
such that T t f = cf .

Proof. Let n = dim V and let α be a nonzero vector in V with T α = cα. Define
the linear operator U on V by

U = T − cI.

Then U α = T α − cα = 0. Therefore α belongs to the null space of U , which


implies (by Theorem 2) that rank(U ) < n. Now consider the linear operator U t
on V ∗ . By Theorem 22, we have

rank(U t ) = rank(U ) < n,

so the nullspace of U t has dimension greater than zero. Therefore we can find
a nonzero linear functional f in V ∗ such that U t f = 0. Then for any β in V ,

0 = (U t f )(β)
= f (U β)
= f (T β − cβ)
= f (T β) − cf (β)
= (T t f )(β) − cf (β).

So, we have T t f = cf as required.

3.7.5 Exercise 5
Let A be an m × n matrix with real entries. Prove that A = 0 if and only if
trace(At A) = 0.
3.7. THE TRANSPOSE OF A LINEAR TRANSFORMATION 103

Proof. Certainly if A = 0, then At A = 0 and trace(At A) = 0. We now need


only prove the converse. Let B = At and suppose trace(BA) = 0. Then
n
X n X
X m n X
X m
0= (BA)jj = Bji Aij = A2ij .
j=1 j=1 i=1 j=1 i=1

Since we have a sum of squares (of real numbers) equal to zero, it must be the
case that each squared number is zero. In particular Aij = 0 for each i, j, since
every such entry appears in the sum.

3.7.6 Exercise 6
Let n be a positive integer and let V be the space of all polynomial functions
over the field of real numbers which have degree at most n, i.e., functions of the
form
f (x) = c0 + c1 x + · · · + cn xn .
Let D be the differentiation operator on V . Find a basis for the null space of
the transpose operator Dt .

Solution. Let W denote the range of D. By Theorem 22, the null space of Dt
is the annihilator of W . We know that W is the space of polynomials having
degree at most n − 1, so dim W = n − 1. By Theorem 16, the annihilator
W 0 must have dimension n − (n − 1) = 1, so we may take any nonzero linear
functional in W 0 as a basis vector. Let g be the unique linear functional such
that
g(xk ) = δnk , 1 ≤ k ≤ n.
That is, g sends every polynomial to the coefficient for its xn term. In particular,
g annihilates W , so {g} is a basis for W 0 and therefore also a basis for the null
space of Dt .

3.7.7 Exercise 7
Let V be a finite-dimensional vector space over the field F . Show that T → T t
is an isomorphism of L(V, V ) onto L(V ∗ , V ∗ ).

Proof. Suppose V has dimension n and let U be the function from L(V, V )
into L(V ∗ , V ∗ ) given by U (T ) = T t . We could show that U is an isomorphism
by appealing to the definition of the transpose. Instead, we will write U as a
composition of three linear transformations U1 , U2 , and U3 , mapping
U U U
1
L(V, V ) −−→ F n×n −−→
2
F n×n −−→
3
L(V ∗ , V ∗ ),

as follows. U1 sends an operator T to its matrix [T ]B in some fixed basis B, U2


sends an n × n matrix A to its transpose At , and U3 sends an n × n matrix
[T 0 ]B∗ to its corresponding linear operator T 0 on V ∗ . Then U = U3 U2 U1 and
it follows that U is a linear transformation. Moreover, Theorem 12 shows that
U1 and U3 are isomorphisms, and U2 is an isomorphism since it is obviously
invertible (it is its own inverse). Therefore U is an isomorphism from L(V, V )
onto L(V ∗ , V ∗ ).
104 CHAPTER 3. LINEAR TRANSFORMATIONS

3.7.8 Exercise 8
Let V be the vector space of n × n matrices over the field F .
(a) If B is a fixed n × n matrix, define a function fB on V by fB (A) =
trace(B t A). Show that fB is a linear functional on V .

Proof. Since the trace function is linear, we have

fB (A1 + cA2 ) = trace(B t (A1 + cA2 ))


= trace(B t A1 + cB t A2 )
= trace(B t A1 ) + c trace(B t A2 )
= fB (A1 ) + cfB (A2 ).

Therefore fB is linear.

(b) Show that every linear functional on V is of the above form, i.e., is fB for
some B.

Proof. Let g be a linear functional on V . Then g can be written in the


form
Xn X n
g(A) = cij Aij ,
i=1 j=1

where each cij is a fixed scalar in F . Now let B be the matrix whose i, j
entry is cij . Then for each matrix A in V ,
n
X
t
fB (A) = trace(B A) = (B t A)jj
j=1
n X
X n n X
X n
= (B t )ji Aij = cij Aij = g(A).
j=1 i=1 j=1 i=1

Therefore g = fB .

(c) Show that B → fB is an isomorphism of V onto V ∗ .

Proof. Let T denote the function B → fB . Then

T (A1 + cA2 )(A) = trace((A1 + cA2 )t A)


= trace(At1 A + cAt2 A)
= trace(At1 A) + c trace(At2 A)
= fA1 (A) + cfA2 (A)
= T (A1 )(A) + cT (A2 )(A),

so T is a linear transformation. Since we have already proven that every


linear functional on V can be written as fB for some matrix B, it follows
that T is onto. And since dim V = dim V ∗ = n2 , this is enough to show
that T is an isomorphism (by Theorem 9).

You might also like