Chapter 04.

10
Eigenvalues and Eigenvectors
After reading this chapter, you should be able to:
1. define eigenvalues and eigenvectors of a square matrix,
2. find eigenvalues and eigenvectors of a square matrix,
3. relate eigenvalues to the singularity of a square matrix, and
4. use the power method to numerically find the largest eigenvalue in magnitude of a
square matrix and the corresponding eigenvector.
What does eigenvalue mean?
The word eigenvalue comes from the German word Eigenwert where Eigen means
characteristic and Wert means value. However, what the word means is not on your mind
!ou want to "now why # need to learn a$out eigenvalues and eigenvectors. %nce # give you
an e&am'le of an a''lication of eigenvalues and eigenvectors, you will want to "now how to
find these eigenvalues and eigenvectors.
Can you give me a physical example application of eigenvalues and eigenvectors?
(oo" at the s'ring)mass system as shown in the 'icture $elow.
*ssume each of the two mass)dis'lacements to $e denoted $y
1
x and
2
x , and let us assume
each s'ring has the same s'ring constant k . Then $y a''lying +ewton,s 2
nd
and 3
rd
law of
motion to develo' a force)$alance for each mass we have
- .
1 2 1 2
1
2
1
x x k kx
dt
x d
m +
/4.1/.1
&
1
&
2
m
1
m

k k
/4.1/.2 0ha'ter /4.1/
- .
1 2 2
2
2
2
x x k
dt
x d
m
1ewriting the e2uations, we have
/ - 2 .
2 1 2
1
2
1
+ x x k
dt
x d
m
/ - .
2 1 2
2
2
2
x x k
dt
x d
m
(et 13 , 2/ , 1/
2 1
k m m
/ - . 13 2/
2 1 2
2
2
x x
dt
x d
4rom vi$ration theory, the solutions can $e of the form
( ) / sin / t A x
i i

where
i
A
5 am'litude of the vi$ration of mass i ,

5 fre2uency of vi$ration,
// 5 'hase shift.
then
- / .
2
2
2
/ t !in w A
dt
x d
i
i

6u$stituting
i
x
and
2
2
dt
x d
i
in e2uations,
/ - 2 . 13 1/
2 1
2
1
+ A A A
/ - . 13 2/
2 1
2
2
A A A
gives
/ 13 - 3/ 1/ .
2 1
2
+ A A
/ - 13 2/ . 13
2
2
1
+ + A A
or
/ 3 . 1 - 3 .
2 1
2
+ A A
/ - 73 . / . 73 . /
2
2
1
+ + A A
#n matri& form, these e2uations can $e rewritten as
1
]
1

1
]
1

1
]
1

+
+
/
/
73 . / 73 . /
3 . 1 3
2
1
2
2
A
A

1
]
1

1
]
1

1
]
1

1
]
1

/
/
73 . / 73 . /
3 . 1 3
2
1 2
2
1
A
A
A
A

(et
2
1
]
1

73 . / 73 . /
3 . 1 3
8 9A
/ - 2 . 13 1/
2 1 2
1
2
+ x x
dt
x d
Eigenvalues and Eigenvectors /4.1/.3
1
]
1

2
1
8 9
A
A
"
/ 8 9 8 9 8 9 " " A
8 9 8 9 8 9 " " A
#n the a$ove e2uation, is the eigenvalue and
8 9"
is the eigenvector corres'onding to .
*s you can see, if we "now for the a$ove e&am'le we can calculate the natural fre2uency
of the vi$ration

Why are the natural fre2uencies of vi$ration im'ortant: ;ecause you do not want to have a
forcing force on the s'ring)mass system close to this fre2uency as it would ma"e the
am'litude i
A
very large and ma"e the system unsta$le.
What is the general definition of eigenvalues and eigenvectors of a square matrix?
#f
8 9 A
is a
n n
matri&, then
/ 8 9

"
is an eigenvector of
8 9 A
if
8 9 8 9 8 9 " " A
where is a scalar and
/ 8 9 "
. The scalar is called the eigenvalue of
8 9 A
and
8 9 "
is
called the eigenvector corres'onding to the eigenvalue .
How do I find eigenvalues of a square matrix?
To find the eigenvalues of a n n matri&
8 9 A
, we have
8 9 8 9 8 9 " " A
/ 8 9 8 9 8 9 " " A
/ 8 89 9 8 9 8 9 " # " A
/ 8 8-9 89 9 8 .9 " # A
+ow for the a$ove set of e2uations to have a non<ero solution,
/ 8- 9 8 det.9 # A
This left hand side can $e e&'anded to give a 'olynomial in and solving the a$ove
e2uation would give us values of the eigenvalues. The a$ove e2uation is called the
characteristic e2uation of
8 9 A
.
4or a
8 9 A

n n
matri&, the characteristic 'olynomial of A is of degree
n
as follows
/ 8- 9 8 det.9 # A
giving
/
2
2
1
1
+ + + +

n
n n n
c c c
Hence. this 'olynomial has
n
roots.
Example 1
4ind the eigenvalues of the 'hysical 'ro$lem discussed in the $eginning of this cha'ter, that
is, find the eigenvalues of the matri&
1
]
1

73 . / 73 . /
3 . 1 3
8 9A
Solution
1
]
1

73 . / 73 . /
3 . 1 3
8 9 8 9 # A
/4.1/.4 0ha'ter /4.1/
[ ] [ ] / - 3 . 1 -. 73 . / . - 73 . / -. 3 . - det. # A
/ 123 . 1 3 73 . / 23 . 2
2
+
/ 123 . 1 73 . 3
2
+
- 1 . 2
- 123 . 1 -. 1 . 4 - 73 . 3 . - 73 . 3 .
2
t


2
/=2 . 3 73 . 3 t

32>> . / , 421 . 3
6o the eigenvalues are 3.421 and /.32>>.
Example
4ind the eigenvectors of
1
]
1

73 . / 73 . /
3 . 1 3
A
6olution
The eigenvalues have already $een found in E&am'le 1 as
32>> . / , 421 . 3
2 1

(et
1
]
1

2
1
8 9
x
x
"
$e the eigenvector corres'onding to
421 . 3
1

Hence
/ 8 8-9 9 8 .9
1
" # A
/
1 /
/ 1
421 . 3
73 . / 73 . /
3 . 1 3
2
1

1
]
1

'

1
]
1

1
]
1

x
x
1
]
1

1
]
1

1
]
1

/
/
?71 . 2 73 . /
3 . 1 421 . /
2
1
x
x
#f
s x
1
then
s x
x s
2>/> . /
/ 3 . 1 421 . /
2
2


The eigenvector corres'onding to 421 . 3
1
then is
1
]
1

s
s
"
2>/> . /
8 9

1
]
1

2>/> . /
1
s
The eigenvector corres'onding to
421 . 3
1

is
Eigenvalues and Eigenvectors /4.1/.3
1
]
1

2>/> . /
1
6imilarly, the eigenvector corres'onding to
32>> . /
2

is
1
]
1

7>1 . 1
1
Example !
4ind the eigenvalues and eigenvectors of
1
1
1
]
1

/ / 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
8 9 A
Solution
The characteristic e2uation is given $y
/ 8- 9 8 det.9 # A
/
/ 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
det
1
1
1
]
1

/ -8 3 . / -. 3 . / . - / -. 3 . / -9. 1 . -8 / -. 3 . / . - -. 3 . / -9. 3 . 1 . +
/ 23 . / 23 . 1 2
2 3
+ +
To find the roots of the characteristic 'olynomial e2uation
/ 23 . / 23 . 1 2
2 3
+ +
we find that the first root $y o$servation is
1
as su$stitution of 1 gives
/ 23 . / - 1 . 23 . 1 - 1 . 2 - 1 .
2 3
+ +
/ /
6o
- 1 .
is a factor of
23 . / 23 . 1 2
2 3
+ + .
To find the other factors of the characteristic 'olynomial, we first conduct long division
23 . /
23 . / 23 . 1 2 1
2
2 3
+ +
+ +



2 3
+
@@@@@@@@@@@@@@@@@@@@@@
/4.1/.? 0ha'ter /4.1/

+
2
2
23 . / 23 . 1

23 . / 23 . /
23 . / 23 . /
+
+

Hence
- 23 . / -. 1 . 23 . / 23 . 1 2
2 2 3
+ + + +
To find <eroes of 23 . /
2
+ + , we solve the 2uadratic e2uation,
/ 23 . /
2
+ +
to give

- 1 . 2
- 23 . / -. 1 -. 4 . - 1 . - 1 .
2

t

2
/ 1

3 . / , 3 . /
6o
3 . / and 3 . / are the <eroes of
3 . /
2
+ +
giving
- 3 . / -. 3 . / . 23 . /
2
+ +
Hence
/ 23 . / 23 . 1 2
2 3
+ +
can $e rewritten as
/ - 3 . / -. 3 . / -. 1 .
to give the roots as
3 . / , 3 . / , 1
These are the three roots of the characteristic 'olynomial e2uation and hence the eigenvalues
of matri& 9*8.
+ote that there are eigenvalues that are re'eated. 6ince there are only two distinct
eigenvalues, there are only two eigens'aces. ;ut, corres'onding to 3 . / there should $e
two eigenvectors that form a $asis for the eigens'ace corres'onding to 3 . / .
Given
/ 8 9 -8 9. " # A
then
Eigenvalues and Eigenvectors /4.1/.7
1
1
1
]
1

1
1
1
]
1

1
1
1
]
1

/
/
/
/ 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
3
2
1
x
x
x

4or 3 . / ,
1
1
1
]
1

1
1
1
]
1

1
1
1
]
1

/
/
/
3 . / / 3 . /
3 . / / 3 . /
1 / 1
3
2
1
x
x
x
6olving this system gives
a x b x a x
3 2 1
, ,
6o
1
1
1
]
1

1
1
1
]
1

a
b
a
x
x
x
3
2
1

1
1
1
]
1

+
1
1
1
]
1

/
/
/ b
a
a


1
1
1
]
1

+
1
1
1
]
1

/
1
/
1
/
1
b a
6o the vectors
1
1
1
]
1

1
/
1
and
1
1
1
]
1

/
1
/
form a $asis for the eigens'ace for the eigenvalue 3 . /
and are the two eigenvectors corres'onding to 3 . / .
4or 1 ,
1
1
1
]
1

1
1
1
]
1

1
1
1
]
1

/
/
/
1 / 3 . /
3 . / 3 . / 3 . /
1 / 3 . /
3
2
1
x
x
x
6olving this system gives
a x a x a x 3 . / , 3 . / ,
3 2 1

The eigenvector corres'onding to 1 is
1
1
1
]
1

1
1
1
]
1

3 . /
3 . /
1
3 . /
3 . / a
a
a
a
Hence the vector
1
1
1
]
1

3 . /
3 . /
1

is a $asis for the eigens'ace for the eigenvalue of 1 , and is the eigenvector corres'onding
to 1 .
/4.1/.> 0ha'ter /4.1/
What are some of the theorems of eigenvalues and eigenvectors?
Theorem 1A #f
8 9 A
is a
n n
triangular matri& B u''er triangular, lower triangular or
diagonal, the eigenvalues of
8 9 A
are the diagonal entries of
8 9 A
.
Theorem 2A / is an eigenvalue of
8 9 A
if
8 9 A
is a singular .noninverti$le- matri&.
Theorem 3A
8 9 A
and
T
8 9 A have the same eigenvalues.
Theorem 4A Eigenvalues of a symmetric matri& are real.
Theorem 3A Eigenvectors of a symmetric matri& are orthogonal, $ut only for distinct
eigenvalues.
Theorem ?A
- det. A
is the 'roduct of the a$solute values of the eigenvalues of
8 9 A
.
Example "
What are the eigenvalues of
1
1
1
1
]
1

2 . 7 / ? 2
/ 3 . 7 3 =
/ / 3 7
/ / / ?
8 9 A
Solution
6ince the matri&
8 9 A
is a lower triangular matri&, the eigenvalues of
8 9 A
are the diagonal
elements of
8 9 A
. The eigenvalues are
2 . 7 , 3 . 7 , 3 , ?
4 3 2 1

Example #
%ne of the eigenvalues of
1
1
1
]
1

7 1 2
= 3 3
2 ? 3
8 9 A
is <ero. #s
8 9 A
inverti$le:
Solution
/ is an eigenvalue of
8 9 A
, that im'lies
8 9 A
is singular and is not inverti$le.
Example $
Given the eigenvalues of
1
1
1
]
1

3 . > 1 >
2 3 3 . 3
? 3 . 3 2
8 9 A
are
711 . 4 , 33 . 12 , 347 . 1
3 2 1

What are the eigenvalues of
8 9$
if
1
1
1
]
1

3 . > 2 ?
1 3 3 . 3
> 3 . 3 2
8 9$
Eigenvalues and Eigenvectors /4.1/.=
Solution
6ince
%
A $ 8 9 8 9 , the eigenvalues of
8 9 A
and
8 9$
are the same. Hence eigenvalues of
8 9$
also are
711 . 4 , 33 . 12 , 347 . 1
3 2 1

Example %
Given the eigenvalues of
1
1
1
]
1

3 . > 1 >
2 3 3 . 3
? 3 . 3 2
8 9 A
are
711 . 4 , 33 . 12 , 347 . 1
3 2 1

0alculate the magnitude of the determinant of the matri&.
Solution
6ince
3 2 1
8 det9 A

711 . 4 33 . 12 347 . 1
>> . >=
How does one find eigenvalues and eigenvectors numerically?
%ne of the most common methods used for finding eigenvalues and eigenvectors is the
'ower method. #t is used to find the largest eigenvalue in an a$solute sense. +ote that if this
largest eigenvalues is re'eated, this method will not wor". *lso this eigenvalue needs to $e
distinct. The method is as followsA
1. *ssume a guess 8 9
- / .
" for the eigenvector in
8 9 8 9 8 9 " " A
e2uation. %ne of the entries of 8 9
- / .
" needs to $e unity.
2. 4ind
8 9 8 9 8 9
- / . - 1 .
" A &
3. 6cale 8 9
- 1 .
& so that the chosen unity com'onent remains unity.
8 9 8 9
- 1 . - 1 . - 1 .
" &
4. 1e'eat ste's .2- and .3- with
8 9 8 9
- 1 .
" " to get 8 9
- 2 .
" .
3. 1e'eat the ste's 2 and 3 until the value of the eigenvalue converges.
#f s
E
is the 're)s'ecified 'ercentage relative error tolerance to which you would li"e the
answer to converge to, "ee' iterating until
s
i
i i
E

+
+
1//
- 1 .
- . - 1 .

where the left hand side of the a$ove ine2uality is the definition of a$solute 'ercentage
relative a''ro&imate error, denoted generally $y
s
E
* 're)s'ecified 'ercentage relative
tolerance of
m

2
1/ 3 . / im'lies at least
m
significant digits are current in your answer.
/4.1/.1/ 0ha'ter /4.1/
When the system converges, the value of is the largest .in a$solute value- eigenvalue of
8 9 A
.
Example &
Csing the 'ower method, find the largest eigenvalue and the corres'onding eigenvector of
1
1
1
]
1

/ / 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
8 9 A
Solution
*ssume
1
1
1
]
1

1
1
1
8 9
- / .
"
1
1
1
]
1

1
1
1
]
1

1
1
1
/ / 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
8 9 8 9
- / .
" A

1
1
1
]
1

3 . /
3 . /
3 . 2
1
1
1
]
1

2 . /
2 . /
1
3 . 2
- 1 .
&
3 . 2
- 1 .

We will choose the first element of 8 9
- / .
" to $e unity.
1
1
1
]
1

2 . /
2 . /
1
8 9
- 1 .
"
1
1
1
]
1

1
1
1
]
1

2 . /
2 . /
1
/ / 3 . /
3 . / 3 . / 3 . /
1 / 3 . 1
8 9 8 9
- 1 .
" A

1
1
1
]
1

3 . /
3 . /
3 . 1
1
1
1
]
1

3>4? . /
3>4? . /
1
3 . 1 8 9
- 2 .
"
3 . 1
- 2 .

1
1
1
]
1

3>4? . /
3>4? . /
1
8 9
- 2 .
"
The a$solute relative a''ro&imate error in the eigenvalues is
Eigenvalues and Eigenvectors /4.1/.11
1//
- 2 .
- 1 . - 2 .

a

1//
3 . 1
3 . 1 3 . 1

D 3/7 . =2
0onducting further iterations, the values of
- .i
and the corres'onding eigenvectors is given
in the ta$le $elow
i
- .i

8 9
- .i
"
.D-
a

1 2.3
1
1
1
]
1

2 . /
2 . /
1
@@@@@
2 1.3
1
1
1
]
1

3>4?2 . /
3>4?2 . /
1
=2.3/7
3 1.1134
1
1
1
]
1

44>27 . /
44>27 . /
1
1?.332
4 1./317
1
1
1
]
1

47341 . /
47341 . /
1
?./32=
3 1./243=
1
1
1
]
1

4>>// . /
4>>// . /
1
1.2441
The e&act value of the eigenvalue is 1
and the corres'onding eigenvector is
1
1
1
]
1

3 . /
3 . /
1
8 9 "
'ey (erms)
Eigenvalue
Eigenvectors
'ower method