Chapter 7

Eigenvalues and Eigenvectors

7.1 Eigenvalues and Eigenvectors

Homework: [Textbook, §7.1 Ex. 5, 11, 15, 19, 25, 27, 61, 63, 65].

Optional Homework:[Textbook, §7.1 Ex. 53, 59].

In this section, we introduce eigenvalues and eigenvectors. This is

one of most fundamental and most useful concepts in linear algebra.

223
224 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

Definition 7.1.1 Let A be an n × n matrix. A scalar λ is said to be

a eigenvalue of A, if

Ax = λx f or some vector x 6= 0.

The vector x is called an eigenvector corresponding to λ. The zero

vector 0 is never an eigenvectors, by definition.

Reading assignment: Read [Textbook, Examples 1, 2, page 423].

7.1.1 Eigenspaces
Given a square matrix A, there will be many eigenvectors corresponding
to a given eigenvalue λ. In fact, together with the zero vector 0, the
set of all eigenvectors corresponding to a given eigenvalue λ will form
a subspace. We state the same as a theorem:

Theorem 7.1.2 Let A be an n × n matrix and λ is an eigenvalue of

A. Then the set

E(λ) = {0} ∪ {x : x is an eigenvector corresponding to λ}

(of all eigenvalues corresponding to λ, together with 0) is a subspace of

Rn . This subspace E(λ) is called the eigenspace of λ.

Proof. Since 0 ∈ E(λ), we have E(λ) is nonempty. Because of theorem

4.3.3, we need only to check that E(λ) is closed under addition and
scalar multiplication. Suppose x, y ∈ E(λ) and c be a scalar. Then,

Ax = λx and Ay = λy.

So,
A(x + y) = Ax + Ay = λx + λy = λ(x + y).
7.1. EIGENVALUES AND EIGENVECTORS 225

So, x + y is an eigenvalue corresponding to λ or zero. So, x + y ∈ E(λ)

and E(λ) is closed under addition. Also,
A(cx) = c(Ax) = c(λx) = λ(cx).
So, cx ∈ E((λ) and E(λ) is closed under scalar multiplication. There-
fore, E(λ) is a subspace of Rn . The proof is complete.

Reading assignment: Read [Textbook, Examples 3, page 423].

Theorem 7.1.3 Let A be a square matrix of size n × n. Then

1. Then a scalar λ is an eigenvalue of A if and only if

det(λI − A) = 0,

here I denotes the identity matrix.

2. A vector x is an eigenvector, of A, corresponding to λ if and only

if x is a nozero solution

(λI − A)x = 0.

Proof. By definition, λ is an eigenvalue of A if and only if, for some

nonzero x, we have
Ax = λx = λIx ⇔ (λI − A)x = 0 ⇔ det(λI − A) = 0.
The last equivalence is given by [Textbook, §3.3], which we did not
cover. This establishes (1) of the theorem. The proof of (2) is obvious
or same as that of (1). This completes the proof.

Definition 7.1.4 Let A be a square matrix of size n × n. Then the

equation
det(λI − A) = 0
is called the characteristic equation of A. (The German word ’eigen’
roughly means ’characteristic’.)
226 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

1. Using induction and expanding det((λI − A), it follows that

det(λI − A) = λn + cn−1 λn−1 + · · · + c1 λ + c0 ,

which is a polynomial in λ, of degree n. This polynomial is called

the characteristic polynomial of A.

2. If  
a11 a12 a13 · · · a1n
 a21 a22 a23 · · · a2n 
 
 
A=
 a31 a32 a33 · · · a3n 

 ··· ··· ··· ··· ··· 
 

am1 am2 am3 · · · amn

then
 
λ − a11 −a12 −a13 ··· −a1n
 −a21 λ − a22 −a23 ··· −a2n
 

 
(λI − A) = 
 −a31 −a32 λ − a33 · · · −a3n .

 ··· ··· ··· ··· ···
 

−an1 −an2 −an3 · · · λ − ann

So, the characteristic polynomial is the determinant of this ma-

trix.

Method of finding eigenvalues and eigenvectors is as follows: Let

A be an n × n matrix.

1. To find the eigenvalues of A solve the characteristic equation

det(λI − A) = 0.

This is a polynomial equation in λ of degree n. We only consider

real roots of this equation, in this class.
7.1. EIGENVALUES AND EIGENVECTORS 227

2. Given an eigenvalue λi (i.e. a root of the characteristic equation),

to find the eigenspace E(λi ), corresponding to λi , we solve the
linear system
(λi I − A)x = 0.
As usual, to solve this we reduce it to the row echelon form or
Gauss-Jordan form. Since λi is an eigenvalue, at least one row of
the echlon form will be zero.

Reading assignment: Read [Textbook, Examples 4-7, page 426-].

Exercise 7.1.5 (Ex. 6, p. 432) Let

 
−2 2 −3
A= 2 1 −6  .
 

−1 −2 0

1. Verify that λ1 = 5 is a eigenvalue of A and x1 = (1, 2, −1)T is a

corresponding eigenvector.
Solution: We need to check Ax1 = 5x1 , We have
      
−2 2 −3 1 5 1
Ax1 =  2 1 −6   2  =  10  = 5  2  = 5x1 .
      

−1 −2 0 −1 −5 −1
So, assertion is verified.

2. Verify that λ2 = −3 is a eigenvalue of A and x2 = (−2, 1, 0)T is

a corresponding eigenvector.
Solution: We need to check Ax2 = −3x2 , We have
      
−2 2 −3 −2 6 −2
Ax2 =  2 1 −6   1  =  −3  = −3  1  = −3x2 .
      

−1 −2 0 0 0 0
So, assertion is verified.
228 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

3. Verify that λ3 = −3 is a eigenvalue of A and x3 = (3, 0, 1)T is a

corresponding eigenvector.
Solution: We need to check Ax3 = −3x3 , We have
      
−2 2 −3 3 −9 3
Ax3 =  2 1 −6   0  =  0  = −3  0  = −3x3 .
      

−1 −2 0 1 −3 1

So, assertion is verified.

Exercise 7.1.6 (Ex. 14, p. 433) Let

 
1 0 5
A =  0 −2 4  .
 

1 −2 9

1. Determine whether x = (1, 1, 0)T is an eigenvector of A.

Solution: We have
      
1 0 5 1 1 1
Ax =  0 −2 4   1  =  −2  =
6 λ 1 
      

1 −2 9 0 −1 0

for all λ. So, x is not an eigenvector of A.

2. Determine whether x = (−5, 2, 1)T is an eigenvector of A.

Solution: We have
      
1 0 5 −5 0 −5
Ax =  0 −2 4   2  =  0  = 0  2  = 0x..
      

1 −2 9 1 0 1

So, x is an eigenvector and corresponding eigenvalue is λ = 0.

7.1. EIGENVALUES AND EIGENVECTORS 229

3. Determine whether x = (0, 0, 0)T is an eigenvector of A.

Solution: No, 0 is, by definition, never an eigenvector.
√ √
4. Determine whether x = (2 6 − 3, −2 6 + 6, 3)T is an eigenvector
of A.
Solution: We have
  √   √   √ 
1 0 5 2 6−3 2 6 + 12 2 6−3
√ √  √
Ax =  0 −2 4   −2 6 + 6  =  4 6 = 6 λ  −2 6 + 6  .
     
√
1 −2 9 3 6 6 + 12 3

So, x is not an eigenvector of A.

Exercise 7.1.7 (Ex. 20, p. 433) Let

 
−5 0 0
A= 3 7 0 .
 

4 −2 3

1. Find the characteristic equation of A.

Solution: The characteristic polynomial is

λ+5 0 0
det(λI − A) = −3 λ − 7 0 = (λ + 5)(λ − 7)(λ − 3).
−4 2 λ−3

So, the characteristic equation is

(λ + 5)(λ − 7)(λ − 3) = 0.

2. Find eigenvalues (and corresponding eigenvectors) of A.

Solution: Solving the characteristic equation, the eigenvalues
are λ = −5, 7, 3.
230 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

(a) To find an eigenvector corresponding to λ = −5, wehave to

solve (−5I − A)x = 0 or
    
0 0 0 x 0
 −3 −12 0  y  =  0 .
    

−4 2 −8 z 0

Solving, we get
16 4
x=− t y= t z = t.
9 9
So, that eigenspace of λ = −5 is
  
16 4
− t, t, t : t ∈ R .
9 9

In particular, with t = 1, an eigenvector, for eigenvalue λ =

T
−5, is − 16 , 4, 1 .
9 9

(b) To find an eigenvector corresponding to λ = 7, wehave to

solve (7I − A)x = 0 or
    
12 0 0 x 0
 −3 0 0   y  =  0  .
    

−4 2 4 z 0

Solving, we get

x=0 y = −2t z = t.

So, that eigenspace of λ = 7 is

{(0, −2t, t) : t ∈ R} .

In particular, with t = 1, an eigenvector, for eigenvalue λ =

7, is (0, −2, 1)T .
7.1. EIGENVALUES AND EIGENVECTORS 231

(c) To find an eigenvector corresponding to λ = 3, wehave to

solve (3I − A)x = 0 or
    
8 0 0 x 0
 −3 −4 0   y  =  0  .
    

−4 2 0 z 0

Solving, we get

x=0 y=0 z = t.

So, that eigenspace of λ = 3 is

{(0, 0, t) : t ∈ R} .

In particular, with t = 1, an eigenvector, for eigenvalue λ =

3, is (0, 0, 1)T .

Exercise 7.1.8 (Ex. 66, p. 435) Let

 
3 1 1
A =  0 3 1 .
 

0 0 3

Find the dimension of the eigenspace corresponding to the eigenvalue

λ = 3.

Solution: The eigenspace E(3) is the solution space of the system

(3I − A)x = x, or
         
3−3 −1 −1 x 0 0 −1 −1 x 0
0 3−3 −1   y  =  0  or  0 0 −1   y  =  0 
         

0 0 3−3 z 0 0 0 0 z 0
232 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

The coefficient matrix

 
0 −1 −1
C= 0 0 −1  .
 

0 0 0

has rank 2. Since

rank(C) + nullity(C) = 3, we nullity(C) = 1.

Therefore, dim E(3) = 1.

7.2. DIAGONALIZATION 233

7.2 Diagonalization
Homework: [Textbook, Ex. 1, 3, 5, 9, 11, 13, 17, 19; p.444].

In this section, we discuss, given a square matrix A, when or whether

we can find an invertible matrix P such that P −1 AP is a diagonal ma-
trix. This problem is closely associated to eigenvalues and eigenvectors.

First, we recall the definition 6.4.1, as follows:

Definition 7.2.1 Suppose A, B are two square matrices of size n × n.

We say A, B are similar, if A = P −1 BP for some invertible matrix P.

We also define the following:

Definition 7.2.2 Suppose A is a square matrix of size n × n. We say

that A is diagonalizable, if there exists a invertible matrix P such
that P −1 AP is a diagonal matrix.

So, our question is which matrices are diagonalizable? Following

theorem has some answer.

Theorem 7.2.3 Suppose A is a square matrix of size n × n. Then A is

diagonalizable if and only if A has n linearly independent eigenvectors.

Proof. Suppose A is diagonalizable. So, there is an invertible matrix

234 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

P such that P −1 AP = D is a diagonal matrix. Write

 
λ1 0 ··· 0
   0 λ2 ··· 0 
P = p1 p2 · · · pn and D =   ··· ···
,
··· ··· 
0 0 · · · λn

where p1 , p2 , . . . , pn are the cllumns of P. We have AP = P D. So,

 
λ1 0 ··· 0
     0 λ2 · · · 0 
A p1 p2 · · · pn = p1 p2 · · · pn   ··· ··· ··· ··· .


0 0 · · · λn

Therefore, i =, 2, . . . , n we have Api = λi pi and so pi are eigenvectors

of A. Also, since P is invertible p1 , p2 , . . . , pn are linearly independent.
So, A have n linearly independent eigenvectors.
To prove the converse, assume A has n linearly independent eigen-
vectors. Let p1 , p2 , . . . , pn be n linearly independent eigenvectors of A.
Then Then, for i =, 2, . . . , n we have, Api = λi pi for some λi . Write,
 
λ1 0 ··· 0
   0 λ2 · · · 0 
P = p1 p2 · · · pn and D =   ··· ··· ··· ··· .


0 0 · · · λn

It follows easily that AP = P D. Since, columns af P are linearly in-

dependent, it follows that P is invertible. Therefore, P −1 AP = D is a
diagonal matrix. So, the proof is complete.

Steps for Diagonalizing an n × n matrix:

Let A be an n × n matrix.

1. Find n linearly independent eigenvectors p1 , p2 , · · · , pn for A with

corresponding eigenvalues λ1 , λ2 , . . . , λn . If n independent eigen-
vectors do not exists, then A is not diagonalizable.
7.2. DIAGONALIZATION 235

2. If A has n linearly independent eigenvectors as above, write

 
λ1 0 ··· 0
   0 λ2 · · · 0 
P = p1 p2 · · · pn and D =   ··· ··· ··· ···


0 0 · · · λn

3. Then D = P −1 AP is a diagonal matrix.

Theorem 7.2.4 Suppose A is an n × n matrix. If A has n distinct

eigenvalues, then the corresponding eignevectors are linearly indepen-
dent and A is diagonizable.

Proof. Let λ1 , λ2 , . . . , λn be distinct eigenvalues of A and let xi eigen-

vectosr corresponding to λi . So, Axi = λi xi .
We claim that x1 , x2 , . . . , xn are linearly independent. If not, as-
sume form some m < n, we have x1 , x2 , . . . , xm are mutually linearly
independent and xm1 is in Span ({x1 , x2 , . . . , xm }) . So, we can write

xm+1 = c1 x1 + c2 x2 + · · · + cm xm Eqn − I.

Here, at least one ci 6= 0. Multiply by A and use the equation Axi =

λi xi , we have

λm+1 xm+1 = λ1 c1 x1 + λ2 c2 x2 + · · · + λm cm xm Eqn − II.

Multiply Eqn-I by λm+1 , we have

λm+1 xm+1 = λm+1 c1 x1 + λm+1 c2 x2 + · · · + λm+1 cm xm Eqn − III.

Subtract Eqn-II from Eqn -III:

(λm+1 − λ1 )c1 x1 + (λm+1 − λ2 )c2 x2 + · · · + (λm+1 − λm )cm xm = 0.

Since, at least one ci 6= 0. and since λi are distinct, at least one coeffi-
cient (λm+1 −λi )ci 6= 0. This contrdicts that x1 , x2 , . . . , xm are mutually
linearly independent. So, it is established that these n eigenvectors
236 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

x1 , x2 , . . . , xn are mutually linearly independent. (This method of

proof is called a proof by contrapositive argument.) Now, by
theorem 7.2.3, A is diagonizable. So, the proof is complete.

Reading assignment: Read [Textbook, Examples 1-7, page 436-].

Exercise 7.2.5 (Ex. 6, p. 444) Let

   
2 3 1 1 1 5
A =  0 −1 2  and P =  0 −1 1  .
   

0 0 3 0 0 2

Verify that A is diagonalizable, by computing P −1 AP.

Solution: We do it in a two steps.

1. Use TI to compute
 
1 1 −3
P −1 =  0 −1 .5  .
 

0 0 .5

2. Use TI to compute
 
2 0 0
P −1 AP =  0 −1 0  .
 

0 0 3

So, it is verified that P −1 AP is a diagonal matrix.

Exercise 7.2.6 (Ex. 10, p. 444) Let

" #
1 .5
A= .
−2 −1
7.2. DIAGONALIZATION 237

Show that A is not diagonalizable.

Solution: To do this, we have find and count the dimensions of all the
eigenspaces E(λ). We do it in a few steps.

1. First, find all the eigenvalues. To do this, we solve

λ−1 −.5
det(λI − A) = = λ2 = 0.
2 λ+1
So, λ = 0 is the only eigenvalue of A.

2. Now we compute the eigenspace E(0) of the eigenvalue λ = 0.

We have E(0) is solution space of
" # " # " #" # " #
x 0 1 .5 x 0
(0I − A) = or =
y 0 −2 −1 y 0
Using TI (or by hand), a parametric solution of this system is
given by

x = −.5t y = t. so E(0) = {(−.5t, t) : t ∈ R} = R(−.5, 1).

So, the (sum of) dimension(s) of the eigenspace(s)

= dim E(0) = 1 < 2.

Therefore A is not diagonizable.

Exercise 7.2.7 (Ex. 14, p. 444) Let

 
2 1 −1
A =  0 −1 2 .
 

0 0 −1
Show that A is not diagonalizable.
Solution: To do this, we have find and count the dimensions of all the
eigenspaces E(λ). We do it in a few steps.
238 CHAPTER 7. EIGENVALUES AND EIGENVECTORS

1. First, find all the eigenvalues. To do this, we solve

λ−2 −1 1
det(λI − A) = 0 λ+1 −2 = (λ − 2)(λ + 1)2 = 0.
0 0 λ+1
So, λ = −1, 2 are the only eigenvalues of A.

2. Now we compute the dimension dim E(−1) of the eigenspace

E(−1) of the eigenvalue λ = −1. We have E(−1) is solution
space of
        
x 0 −3 −1 +1 x 0
(−I−A)  y  =  0  or 0 0 −2   y  =  0 
        

z 0 0 0 0 z 0
(We will avoid solving this system.) The rank of the coefficient
matrix is 2. So,

dim(E(−1)) = nullity = 3 − rank = 3 − 2 = 1.

3. Now we compute the dimension dim E(2) of eigenspace E(2) of

the eigenvalue λ = 2. We have E(2) is solution space of
        
x 0 0 −1 +1 x 0
(−I − A)  y  =  0  or  0 3 −2   y = 0 
        

z 0 0 0 3 z 0
Use TI (or look at the columns) to see that rank of the coefficient
matrix is 2. So,

dim(E(2)) = nullity = 3 − rank = 3 − 2 = 1.

4. So, the sum of dimensions of the eigenspaces

= dim E(−1) + dim E(2) = 2 < 3.

Therefore A is not diagonizable.

7.2. DIAGONALIZATION 239

Exercise 7.2.8 (Ex. 20, p. 444) Let

 
4 3 −2
A= 0 1 1 .
 

0 0 −2

Find the eigenvalues of A and determine whether there is a sufficient

number of them to guarantee that A is diagonalizable.
Solution: First, find all the eigenvalues. To do this, we solve

λ−4 −3 2
det(λI − A) = 0 λ−1 −1 = (λ − 4)(λ − 1)(λ + 2) = 0.
0 0 λ+2

So, λ = 4, 1, −2 are the eigenvalues of A. This means, A has three

distinct eigenvalues. Therefore, by theorem 7.2.4, A is diagonalizable.
240 CHAPTER 7. EIGENVALUES AND EIGENVECTORS
Bibliography

[Textbook] Ron Larson and David C. Falvo, Elementary Linear Alge-

bra, Houghton Miffin

241