Chapter 6

EIGENVALUES AND
EIGENVECTORS
6.1 Motivation
We motivate the chapter on eigenvalues by discussing the equation
ax
2
+ 2hxy +by
2
= c,
where not all of a, h, b are zero. The expression ax
2
+ 2hxy + by
2
is called
a quadratic form in x and y and we have the identity
ax
2
+ 2hxy +by
2
=
_
x y

_
a h
h b
_ _
x
y
_
= X
t
AX,
where X =
_
x
y
_
and A =
_
a h
h b
_
. A is called the matrix of the quadratic
form.
We now rotate the x, y axes anticlockwise through radians to new
x
1
, y
1
axes. The equations describing the rotation of axes are derived as
follows:
Let P have coordinates (x, y) relative to the x, y axes and coordinates
(x
1
, y
1
) relative to the x
1
, y
1
axes. Then referring to Figure 6.1:
115
116 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
E '
T
c

d
d
d
d
d
d
d
ds

d
d
d
d
d
d
d
d

d
d
d
x
y
x
1
y
1
P
Q
R
O

Figure 6.1: Rotating the axes.

x = OQ = OP cos ( +)
= OP(cos cos sin sin )
= (OP cos ) cos (OP sin ) sin
= ORcos PRsin
= x
1
cos y
1
sin .
Similarly y = x
1
sin +y
1
cos .
We can combine these transformation equations into the single matrix
equation:
_
x
y
_
=
_
cos sin
sin cos
_ _
x
1
y
1
_
,
or X = PY , where X =
_
x
y
_
, Y =
_
x
1
y
1
_
and P =
_
cos sin
sin cos
_
.
We note that the columns of P give the directions of the positive x
1
and y
1
axes. Also P is an orthogonal matrix we have PP
t
= I
2
and so P
1
= P
t
.
The matrix P has the special property that det P = 1.
A matrix of the type P =
_
cos sin
sin cos
_
is called a rotation matrix.
We shall show soon that any 2 2 real orthogonal matrix with determinant
6.1. MOTIVATION 117
equal to 1 is a rotation matrix.
We can also solve for the new coordinates in terms of the old ones:
_
x
1
y
1
_
= Y = P
t
X =
_
cos sin
sin cos
_ _
x
y
_
,
so x
1
= xcos +y sin and y
1
= xsin +y cos . Then
X
t
AX = (PY )
t
A(PY ) = Y
t
(P
t
AP)Y.
Now suppose, as we later show, that it is possible to choose an angle so
that P
t
AP is a diagonal matrix, say diag(
1
,
2
). Then
X
t
AX =
_
x
1
y
1

_

1
0
0
2
_ _
x
1
y
1
_
=
1
x
2
1
+
2
y
2
1
(6.1)
and relative to the new axes, the equation ax
2
+ 2hxy + by
2
= c becomes

1
x
2
1
+
2
y
2
1
= c, which is quite easy to sketch. This curve is symmetrical
about the x
1
and y
1
axes, with P
1
and P
2
, the respective columns of P,
giving the directions of the axes of symmetry.
Also it can be veried that P
1
and P
2
satisfy the equations
AP
1
=
1
P
1
and AP
2
=
2
P
2
.
These equations force a restriction on
1
and
2
. For if P
1
=
_
u
1
v
1
_
, the
rst equation becomes
_
a h
h b
_ _
u
1
v
1
_
=
1
_
u
1
v
1
_
or
_
a
1
h
h b
1
_ _
u
1
v
1
_
=
_
0
0
_
.
Hence we are dealing with a homogeneous system of two linear equations in
two unknowns, having a nontrivial solution (u
1
, v
1
). Hence

a
1
h
h b
1

= 0.
Similarly,
2
satises the same equation. In expanded form,
1
and
2
satisfy

2
(a +b) +ab h
2
= 0.
This equation has real roots
=
a +b
_
(a +b)
2
4(ab h
2
)
2
=
a +b
_
(a b)
2
+ 4h
2
2
(6.2)
(The roots are distinct if a = b or h = 0. The case a = b and h = 0 needs
no investigation, as it gives an equation of a circle.)
The equation
2
(a+b)+abh
2
= 0 is called the eigenvalue equation
of the matrix A.
118 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
6.2 Denitions and examples
DEFINITION 6.2.1 (Eigenvalue, eigenvector)
Let A be a complex square matrix. Then if is a complex number and
X a nonzero complex column vector satisfying AX = X, we call X an
eigenvector of A, while is called an eigenvalue of A. We also say that X
is an eigenvector corresponding to the eigenvalue .
So in the above example P
1
and P
2
are eigenvectors corresponding to
1
and
2
, respectively. We shall give an algorithm which starts from the
eigenvalues of A =
_
a h
h b
_
and constructs a rotation matrix P such that
P
t
AP is diagonal.
As noted above, if is an eigenvalue of an n n matrix A, with
corresponding eigenvector X, then (A I
n
)X = 0, with X = 0, so
det (AI
n
) = 0 and there are at most n distinct eigenvalues of A.
Conversely if det (AI
n
) = 0, then (AI
n
)X = 0 has a nontrivial
solution X and so is an eigenvalue of Awith X a corresponding eigenvector.
DEFINITION 6.2.2 (Characteristic equation, polynomial)
The equation det (A I
n
) = 0 is called the characteristic equation of A,
while the polynomial det (AI
n
) is called the characteristic polynomial of
A. The characteristic polynomial of A is often denoted by ch
A
().
Hence the eigenvalues of A are the roots of the characteristic polynomial
of A.
For a 2 2 matrix A =
_
a b
c d
_
, it is easily veried that the character-
istic polynomial is
2
(trace A)+det A, where trace A = a+d is the sum
of the diagonal elements of A.
EXAMPLE 6.2.1 Find the eigenvalues of A =
_
2 1
1 2
_
and nd all eigen-
vectors.
Solution. The characteristic equation of A is
2
4 + 3 = 0, or
( 1)( 3) = 0.
Hence = 1 or 3. The eigenvector equation (AI
n
)X = 0 reduces to
_
2 1
1 2
_ _
x
y
_
=
_
0
0
_
,
6.2. DEFINITIONS AND EXAMPLES 119
or
(2 )x +y = 0
x + (2 )y = 0.
Taking = 1 gives
x +y = 0
x +y = 0,
which has solution x = y, y arbitrary. Consequently the eigenvectors
corresponding to = 1 are the vectors
_
y
y
_
, with y = 0.
Taking = 3 gives
x +y = 0
x y = 0,
which has solution x = y, y arbitrary. Consequently the eigenvectors corre-
sponding to = 3 are the vectors
_
y
y
_
, with y = 0.
Our next result has wide applicability:
THEOREM 6.2.1 Let A be a 2 2 matrix having distinct eigenvalues
1
and
2
and corresponding eigenvectors X
1
and X
2
. Let P be the matrix
whose columns are X
1
and X
2
, respectively. Then P is nonsingular and
P
1
AP =
_

1
0
0
2
_
.
Proof. Suppose AX
1
=
1
X
1
and AX
2
=
2
X
2
. We show that the system
of homogeneous equations
xX
1
+yX
2
= 0
has only the trivial solution. Then by theorem 2.5.10 the matrix P =
[X
1
|X
2
] is nonsingular. So assume
xX
1
+yX
2
= 0. (6.3)
Then A(xX
1
+yX
2
) = A0 = 0, so x(AX
1
) +y(AX
2
) = 0. Hence
x
1
X
1
+y
2
X
2
= 0. (6.4)
120 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
Multiplying equation 6.3 by
1
and subtracting from equation 6.4 gives
(
2

1
)yX
2
= 0.
Hence y = 0, as (
2

1
) = 0 and X
2
= 0. Then from equation 6.3, xX
1
= 0
and hence x = 0.
Then the equations AX
1
=
1
X
1
and AX
2
=
2
X
2
give
AP = A[X
1
|X
2
] = [AX
1
|AX
2
] = [
1
X
1
|
2
X
2
]
= [X
1
|X
2
]
_

1
0
0
2
_
= P
_

1
0
0
2
_
,
so
P
1
AP =
_

1
0
0
2
_
.
EXAMPLE 6.2.2 Let A =
_
2 1
1 2
_
be the matrix of example 6.2.1. Then
X
1
=
_
1
1
_
and X
2
=
_
1
1
_
are eigenvectors corresponding to eigenvalues
1 and 3, respectively. Hence if P =
_
1 1
1 1
_
, we have
P
1
AP =
_
1 0
0 3
_
.
There are two immediate applications of theorem 6.2.1. The rst is to the
calculation of A
n
: If P
1
AP =diag (
1
,
2
), then A = Pdiag (
1
,
2
)P
1
and
A
n
=
_
P
_

1
0
0
2
_
P
1
_
n
= P
_

1
0
0
2
_
n
P
1
= P
_

n
1
0
0
n
2
_
P
1
.
The second application is to solving a system of linear dierential equations
dx
dt
= ax +by
dy
dt
= cx +dy,
where A =
_
a b
c d
_
is a matrix of real or complex numbers and x and y
are functions of t. The system can be written in matrix form as

X = AX,
where
X =
_
x
y
_
and

X =
_
x
y
_
=
_
dx
dt
dy
dt
_
.
6.2. DEFINITIONS AND EXAMPLES 121
We make the substitution X = PY , where Y =
_
x
1
y
1
_
. Then x
1
and y
1
are also functions of t and

X = P

Y = AX = A(PY ), so

Y = (P
1
AP)Y =
_

1
0
0
2
_
Y.
Hence x
1
=
1
x
1
and y
1
=
2
y
1
.
These dierential equations are wellknown to have the solutions x
1
=
x
1
(0)e

1
t
and y
1
= y
1
(0)e

2
t
, where x
1
(0) is the value of x
1
when t = 0.
[If
dx
dt
= kx, where k is a constant, then
d
dt
_
e
kt
x
_
= ke
kt
x +e
kt
dx
dt
= ke
kt
x +e
kt
kx = 0.
Hence e
kt
x is constant, so e
kt
x = e
k0
x(0) = x(0). Hence x = x(0)e
kt
.]
However
_
x
1
(0)
y
1
(0)
_
= P
1
_
x(0)
y(0)
_
, so this determines x
1
(0) and y
1
(0) in
terms of x(0) and y(0). Hence ultimately x and y are determined as explicit
functions of t, using the equation X = PY .
EXAMPLE 6.2.3 Let A =
_
2 3
4 5
_
. Use the eigenvalue method to
derive an explicit formula for A
n
and also solve the system of dierential
equations
dx
dt
= 2x 3y
dy
dt
= 4x 5y,
given x = 7 and y = 13 when t = 0.
Solution. The characteristic polynomial of Ais
2
+3+2 which has distinct
roots
1
= 1 and
2
= 2. We nd corresponding eigenvectors X
1
=
_
1
1
_
and X
2
=
_
3
4
_
. Hence if P =
_
1 3
1 4
_
, we have P
1
AP = diag (1, 2).
Hence
A
n
=
_
Pdiag (1, 2)P
1
_
n
= Pdiag ((1)
n
, (2)
n
)P
1
=
_
1 3
1 4
_ _
(1)
n
0
0 (2)
n
_ _
4 3
1 1
_
122 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
= (1)
n
_
1 3
1 4
_ _
1 0
0 2
n
_ _
4 3
1 1
_
= (1)
n
_
1 3 2
n
1 4 2
n
_ _
4 3
1 1
_
= (1)
n
_
4 3 2
n
3 + 3 2
n
4 4 2
n
3 + 4 2
n
_
.
To solve the dierential equation system, make the substitution X =
PY . Then x = x
1
+ 3y
1
, y = x
1
+ 4y
1
. The system then becomes
x
1
= x
1
y
1
= 2y
1
,
so x
1
= x
1
(0)e
t
, y
1
= y
1
(0)e
2t
. Now
_
x
1
(0)
y
1
(0)
_
= P
1
_
x(0)
y(0)
_
=
_
4 3
1 1
_ _
7
13
_
=
_
11
6
_
,
so x
1
= 11e
t
and y
1
= 6e
2t
. Hence x = 11e
t
+ 3(6e
2t
) = 11e
t
+
18e
2t
, y = 11e
t
+ 4(6e
2t
) = 11e
t
+ 24e
2t
.
For a more complicated example we solve a system of inhomogeneous
recurrence relations.
EXAMPLE 6.2.4 Solve the system of recurrence relations
x
n+1
= 2x
n
y
n
1
y
n+1
= x
n
+ 2y
n
+ 2,
given that x
0
= 0 and y
0
= 1.
Solution. The system can be written in matrix form as
X
n+1
= AX
n
+B,
where
A =
_
2 1
1 2
_
and B =
_
1
2
_
.
It is then an easy induction to prove that
X
n
= A
n
X
0
+ (A
n1
+ +A+I
2
)B. (6.5)
6.2. DEFINITIONS AND EXAMPLES 123
Also it is easy to verify by the eigenvalue method that
A
n
=
1
2
_
1 + 3
n
1 3
n
1 3
n
1 + 3
n
_
=
1
2
U +
3
n
2
V,
where U =
_
1 1
1 1
_
and V =
_
1 1
1 1
_
. Hence
A
n1
+ +A+I
2
=
n
2
U +
(3
n1
+ + 3 + 1)
2
V
=
n
2
U +
(3
n1
1)
4
V.
Then equation 6.5 gives
X
n
=
_
1
2
U +
3
n
2
V
__
0
1
_
+
_
n
2
U +
(3
n1
1)
4
V
__
1
2
_
,
which simplies to
_
x
n
y
n
_
=
_
(2n + 1 3
n
)/4
(2n 5 + 3
n
)/4
_
.
Hence x
n
= (2n 1 + 3
n
)/4 and y
n
= (2n 5 + 3
n
)/4.
REMARK 6.2.1 If (A I
2
)
1
existed (that is, if det (A I
2
) = 0, or
equivalently, if 1 is not an eigenvalue of A), then we could have used the
formula
A
n1
+ +A+I
2
= (A
n
I
2
)(AI
2
)
1
. (6.6)
However the eigenvalues of Aare 1 and 3 in the above problem, so formula 6.6
cannot be used there.
Our discussion of eigenvalues and eigenvectors has been limited to 2 2
matrices. The discussion is more complicated for matrices of size greater
than two and is best left to a second course in linear algebra. Nevertheless
the following result is a useful generalization of theorem 6.2.1. The reader
is referred to [28, page 350] for a proof.
THEOREM 6.2.2 Let A be an n n matrix having distinct eigenvalues

1
, . . . ,
n
and corresponding eigenvectors X
1
, . . . , X
n
. Let P be the matrix
whose columns are respectively X
1
, . . . , X
n
. Then P is nonsingular and
P
1
AP =
_

1
0 0
0
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
n
_

_
.
124 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
Another useful result which covers the case where there are multiple eigen-
values is the following (The reader is referred to [28, pages 351352] for a
proof):
THEOREM 6.2.3 Suppose the characteristic polynomial of A has the fac-
torization
det (I
n
A) = ( c
1
)
n
1
( c
t
)
nt
,
where c
1
, . . . , c
t
are the distinct eigenvalues of A. Suppose that for i =
1, . . . , t, we have nullity (c
i
I
n
A) = n
i
. For each i, choose a basis X
i1
, . . . , X
in
i
for the eigenspace N(c
i
I
n
A). Then the matrix
P = [X
11
| |X
1n
1
| |X
t1
| |X
tnt
]
is nonsingular and P
1
AP is the following diagonal matrix
P
1
AP =
_

_
c
1
I
n
1
0 0
0 c
2
I
n
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 c
t
I
nt
_

_
.
(The notation means that on the diagonal there are n
1
elements c
1
, followed
by n
2
elements c
2
,. . . , n
t
elements c
t
.)
6.3 PROBLEMS
1. Let A =
_
4 3
1 0
_
. Find a nonsingular matrix P such that P
1
AP =
diag (1, 3) and hence prove that
A
n
=
3
n
1
2
A+
3 3
n
2
I
2
.
2. If A =
_
0.6 0.8
0.4 0.2
_
, prove that A
n
tends to a limiting matrix
_
2/3 2/3
1/3 1/3
_
as n .
6.3. PROBLEMS 125
3. Solve the system of dierential equations
dx
dt
= 3x 2y
dy
dt
= 5x 4y,
given x = 13 and y = 22 when t = 0.
[Answer: x = 7e
t
+ 6e
2t
, y = 7e
t
+ 15e
2t
.]
4. Solve the system of recurrence relations
x
n+1
= 3x
n
y
n
y
n+1
= x
n
+ 3y
n
,
given that x
0
= 1 and y
0
= 2.
[Answer: x
n
= 2
n1
(3 2
n
), y
n
= 2
n1
(3 + 2
n
).]
5. Let A =
_
a b
c d
_
be a real or complex matrix with distinct eigenvalues

1
,
2
and corresponding eigenvectors X
1
, X
2
. Also let P = [X
1
|X
2
].
(a) Prove that the system of recurrence relations
x
n+1
= ax
n
+by
n
y
n+1
= cx
n
+dy
n
has the solution
_
x
n
y
n
_
=
n
1
X
1
+
n
2
X
2
,
where and are determined by the equation
_

_
= P
1
_
x
0
y
0
_
.
(b) Prove that the system of dierential equations
dx
dt
= ax +by
dy
dt
= cx +dy
has the solution
_
x
y
_
= e

1
t
X
1
+e

2
t
X
2
,
126 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
where and are determined by the equation
_

_
= P
1
_
x(0)
y(0)
_
.
6. Let A =
_
a
11
a
12
a
21
a
22
_
be a real matrix with nonreal eigenvalues =
a + ib and = a ib, with corresponding eigenvectors X = U + iV
and X = U iV , where U and V are real vectors. Also let P be the
real matrix dened by P = [U|V ]. Finally let a + ib = re
i
, where
r > 0 and is real.
(a) Prove that
AU = aU bV
AV = bU +aV.
(b) Deduce that
P
1
AP =
_
a b
b a
_
.
(c) Prove that the system of recurrence relations
x
n+1
= a
11
x
n
+a
12
y
n
y
n+1
= a
21
x
n
+a
22
y
n
has the solution
_
x
n
y
n
_
= r
n
{(U +V ) cos n + (U V ) sin n},
where and are determined by the equation
_

_
= P
1
_
x
0
y
0
_
.
(d) Prove that the system of dierential equations
dx
dt
= ax +by
dy
dt
= cx +dy
6.3. PROBLEMS 127
has the solution
_
x
y
_
= e
at
{(U +V ) cos bt + (U V ) sin bt},
where and are determined by the equation
_

_
= P
1
_
x(0)
y(0)
_
.
[Hint: Let
_
x
y
_
= P
_
x
1
y
1
_
. Also let z = x
1
+iy
1
. Prove that
z = (a ib)z
and deduce that
x
1
+iy
1
= e
at
( +i)(cos bt +i sinbt).
Then equate real and imaginary parts to solve for x
1
, y
1
and
hence x, y.]
7. (The case of repeated eigenvalues.) Let A =
_
a b
c d
_
and suppose
that the characteristic polynomial of A,
2
(a +d) +(ad bc), has
a repeated root . Also assume that A = I
2
. Let B = AI
2
.
(i) Prove that (a d)
2
+ 4bc = 0.
(ii) Prove that B
2
= 0.
(iii) Prove that BX
2
= 0 for some vector X
2
; indeed, show that X
2
can be taken to be
_
1
0
_
or
_
0
1
_
.
(iv) Let X
1
= BX
2
. Prove that P = [X
1
|X
2
] is nonsingular,
AX
1
= X
1
and AX
2
= X
2
+X
1
and deduce that
P
1
AP =
_
1
0
_
.
8. Use the previous result to solve system of the dierential equations
dx
dt
= 4x y
dy
dt
= 4x + 8y,
128 CHAPTER 6. EIGENVALUES AND EIGENVECTORS
given that x = 1 = y when t = 0.
[To solve the dierential equation
dx
dt
kx = f(t), k a constant,
multiply throughout by e
kt
, thereby converting the lefthand side to
dx
dt
(e
kt
x).]
[Answer: x = (1 3t)e
6t
, y = (1 + 6t)e
6t
.]
9. Let
A =
_
_
1/2 1/2 0
1/4 1/4 1/2
1/4 1/4 1/2
_
_
.
(a) Verify that det (I
3
A), the characteristic polynomial of A, is
given by
( 1)(
1
4
).
(b) Find a nonsingular matrix P such that P
1
AP = diag (1, 0,
1
4
).
(c) Prove that
A
n
=
1
3
_
_
1 1 1
1 1 1
1 1 1
_
_
+
1
3 4
n
_
_
2 2 4
1 1 2
1 1 2
_
_
if n 1.
10. Let
A =
_
_
5 2 2
2 5 2
2 2 5
_
_
.
(a) Verify that det (I
3
A), the characteristic polynomial of A, is
given by
( 3)
2
( 9).
(b) Find a nonsingular matrix P such that P
1
AP = diag (3, 3, 9).