SMA 2216 APPLIED LINEAR ALGEBRA II

LINEAR MAPPINGS
Definition: Let U and V be vector spaces and let f be a function that associates a unique
vector v ∈ V with each
vector u ∈ U. Then f is said to map U into V, denoted f : U −→ V.
In this case, v = f u is called the image of u under f . The vector space U is called the
e e
e domf , while V is callede the codomain (image space) of f , denoted
domain of f ,edenoted
codf .
Definition: Let f : U −→ V be a function from a vector space U into a vector space V.
Then f is said to be
 linear
 if:
  
(i ) f u1 + u2 = f u1 + f u2 for all u1 , u2 ∈ U (i.e. imf is closed under addition).
f
f
f f f f
(ii ) f λu = λf u for all u ∈ U and all scalars λ (i.e. imf is closed under scalar
multiplication).
e e e

A linear mapping that maps a vector space U into itself is called a linear operator on U.
Examples:
1. Define a mapping f : R2 −→ R3 by f (x, y) = (x − y, x + y, 0). Determine whether f
is linear or not.
Solution: Let u = (x1 , y1 ), v = (x2 , y2 ) ∈ R2 and λ ∈ R. Then:
(i ) e e
 
f u + v = f (x1 + x2 , y1 + y2 )
e e
= (x1 + x2 − y1 − y2 , x1 + x2 + y1 + y2 , 0)
= (x1 − y1 , x1 + y1 , 0) + (x2 − y2 , x2 + y2 , 0)
= f (x1 , y1 ) + f (x2 , y2 )
   
=f u +f v
e e
(ii )
 
f λu = f (λx1 , λy1 )
e
= (λx1 − λy1 , λx1 + λy1 , 0)
= λ (x1 − y1 , x1 + y1 , 0)
= λf (x1 , y1 )
 
= λf u
e
Therefore, f is linear.
2. Show that the function g : R3 −→ R2 given by g (x, y, z) = (x + y + z, x − y − z) is
linear.
Solution: Let u = (a, b, c), v = (x, y, z) ∈ R3 and λ ∈ R. Then:
e e
1
 (i )
 
g u + v = g (a + x, b + y, c + z)
e e
= (a + x + b + y + c + z, a + x − b − y − c − z)
= (a + b + c, a − b − c) + (x + y + z, x − y − z)
= g (a, b, c) + g (x, y, z)
   
=g u +g v
e e
(ii )
 
g λu = g (λa, λb, λc)
e
= (λa + λb + λc, λa − λb − λc)
= λ (a + b + c, a − b − c)
= λg (a, b, c)
 
= λg u
e
Therefore, g is linear.
3. Consider the function h : R3 −→ R3 defined by h (x, y, z) = (2x + y + 1, −x, z). Let
u = (−1, 0, 0) and v = (1, 1, 1). Then
e e



h u + h v = (−1, 1, 0) + (4, −1, 1) = (3, 0, 1)
e e
while

h u + v = h (0, 1, 1) = (2, 0, 1)


e e

Since h u + v 6= h u + h v , h is not linear.
e e e e
Alternatively, let u = (−1, 0, 0) and λ = 2. Then
e


h 2u = h (−2, 0, 0) = (−3, 2, 0)
e
while

2h u = 2 (−1, 1, 0) = (−2, 2, 0)


e
Since 2h u 6= h 2u , h is not linear.
e e
4. The function f : R −→ R where f (x) = sin x is not linear.
Proof :
f (π/2 + π/2) = f (π) = sin π = 0
but
f (π/2) + f (π/2) = sin π/2 + sin π/2 = 1 + 1 = 2.
Now, f is not linear since f (π/2) + f (π/2) 6= f (π/2 + π/2).
Note: A function f (x, y, z, w, · · · ) = (· · · · · · ) is linear if the vector (· · · · · · ) = f (x, y, z, w, · · · )
is a linear combination of the componenets x, y, z, w, · · · .

2
Examples:
1. In Example 1 above,

f (x, y) = (x − y, x + y, 0)
= x (1, 1, 0) + y (−1, 1, 0) .

Since f (x, y) is a linear combination of x and y, then f is linear.

2. In Example 2 above,

g (x, y, z) = (x + y + z, x − y − z)
= x (1, 1) + y (1, −1) + z (1, −1)

Since g (x, y, z) is a linear combination of x, y, and z, then g is linear.

3. The function f : R3 −→ R3 , f (x, y, z) = (sin x)2 , z 3 + 3, xy − 1 is not linear since

(sin x)2 , z 3 + 3, xy − 1 cannot be expressed a linear combination of x, y, and z.

Further Examples:
1. Let U and V be vector spaces and define T : U −→ V by T u = 0 for every u ∈ U.
Then    
e e
 
e
T u1 + u2 = 0 = 0 + 0 = T u1 + T u2
f f e e e f f
and


T αu = 0 = α0 = αT u .
e e e e
Thus T is a linear transformation, called the zero transformation.
2. Let V be a vector space. Define I : V −→ V by Iv = v for all v ∈ V. In this case, I is
a linear transformation, called the identity transformation/operator.
e e e

3. Let f : R2 −→ R2 be defined by f (x, y) = (−x, y). Geometrically, f is a reflection

about the y-axis and can be shown to be linear.

Non-Linear Transformations
Not every transformation that looks linear is actually linear.
Example:
Let T : R −→ R where T (x) = 2x + 3. Then the graph of {(x, T (x) : x ∈ R)} is a straight
line in the xy-plane. However, T is not linear since T (x + y) = 2 (x + y) + 3 = 2x + 2y + 3
and T (x) + T (y) = (2x + 3) + (2y + 3) = 2x + 2y + 6 so that T (x + y) 6= T (x) + T (y).
Note: The only linear transformation from R to R are functions of the form f (x) = mx,
m ∈ R.
Exercise:
Determine whether the following functions are linear or not:
1. f : R2 −→ R2 , f (x, y) = (2x + y, x)
2. f : R2 −→ R2 , f (x, y) = (x + 2y, 3x − 2y)
3. f : R2 −→ R2 , f (x, y) = (−2x + 1, x)

3
 4. f : R2 −→ R3 , f (x, y) = (x + y, x − y, x)
5. f : R3 −→ R2 , f (x, y, z) = (x + y + 2z, x − y)
6. T : R3 −→ R3 , T (x1 , x2 , x3 ) = (x1 + 2x2 − x3 , −x2 , x1 + 7x3 )
7. T : R4 −→ R3 , T (x1 , x2 , x3 , x4 ) = (x1 + x2 , x2 + x3 , x4 + x4 )
Theorem: Let f : U −→ V be linear. Then for all vectors u, v , u1 , u2 , · · · , un ∈ U and all
scalars α1 , α
2 , · · · , αn :
e e f f f
(i ) f 0 = 0.
e e



(ii ) f u − v = f u − f v .
e e e e       
(iii ) f α1 u1 + α2 u2 + · · · + αn un = α1 f u1 + α2 f u2 + · · · + αn f un .
f f f f f f
Example: n o
Let S = v1 = (1, 1, 1) , v2 = (1, 1, 0) , v3 = (1, 0, 0) be a basis for R3 and let T : R3 −→ R2
f f  f     
be the linear mapping such that T v1 = (1, 0), T v2 = (2, −1), and T v3 = (4, 3).
Find a formula for T (x, y, z), then use the formula to find T (2, −3, 5).
f f f
Solution: Expressing (x, y, z) as a linear combination of v1 , v2 , v3 ,
f f f
(x, y, z) = λ1 (1, 1, 1) + λ2 (1, 1, 0) + λ3 (1, 0, 0)

Equating corresponding components,

λ1 + λ2 + λ3 = x
λ1 + λ2 =y
λ1 =z

so that λ1 = z, λ2 = y − z, and λ3 = x − y. Hence

(x, y, z) = z (1, 1, 1) + (y − z) (1, 1, 0) + (x − y) (1, 0, 0)

= zv1 + (y − z) v2 + (x − y) v3
f f f
Thus
     
T (x, y, z) = zT v1 + (y − z) T v2 + (x − y) T v3
f f f
= z (1, 0) + (y − z) (2, −1) + (x − y) (4, 3)
= (4x − 2y − z, 3x − 4y + z)

Now, T (2, −3, 5) = (9, 23).

Exercise:
Let T : R3 −→ R2 be the linear mapping such that T (1, 0, 0) = (2, 3), T (0, 1, 0) = (−1, 4),
and T (0, 0, 1) = (5, −3). Find T (3, −4, 5).

4
Kernel and Range of a Linear Transformation
Definition: Let U and V be vector spaces and let f : U −→ V be a linear transformation.
Then 

(i ) the kernel (null space) of f , denoted ker f , is given by ker f = u ∈ U | f u = 0 .
e
e e
(ii ) the range (image) of f , denoted rangef or imf , is given by imf = f u ∈ V | u ∈ U .
e e
Note: In the above definition:
(a) ker f ⊆ U while imf ⊆ V.
(b) (i ) ker f
6= ∅ as it contains at least the zero vector in U (by the preceeding theorem,
f 0 = 0).
(ii ) imf 6= ∅
since it contains at least the zero vector in V (by the preceeding theorem,
e e
0 = f 0 ∈ imf ).
e e
Theorem: Let f : U −→ V be a linear mapping. Then:
(a) ker f is a subspace of U.
(b) imf is a subspace of V.
Proof :    
(a) Let u1 , u2 ∈ ker f and let λ be a scalar. Then f u1 = f u2 = 0. Now,
f f      f f e
(i ) f u1 + u2 = f u1 + f u2 = 0 + 0 = 0 implying that u1 + u2 ∈ ker f .
f  f  f f e e e f f
(ii ) f λu1 = λf u1 = λ0 = 0 implying that λu ∈ ker f .
f f e e e
(b) Let v1 , v2 ∈ imf and let λ be a scalar. Then there exist u1 , u2 ∈ U such that
f f    f f
v1 = f u1 and v2 = f u2 . Now,
f f f   
f  
(i ) v1 + v2 = f u1 + f u2 = f u1 + u2 ∈ imf .
f f   f  f f f
(ii ) λv1 = λf u1 = f λu1 ∈ imf .
f f f
Definition: Let f : U −→ V be a linear mapping. Then:
(i ) the nullity of f , denoted N (f ), is the dimension of ker f , i.e. N (f ) = dim ker f .
(ii ) the rank of f , denoted R (f ), is the dimension of imf , i.e. R (f ) = dim imf .
Theorem (Rank-Nullity Theorem): Let f : U −→ V be a linear transformation where
dim U = n and dim V = m. Then n = R (f ) + N (f ).
Examples:
1. Let f : R2 −→ R2 be defined by f (x, y) = (x + y, x). Then

ker f = (x, y) ∈ R2 | f (x, y) = (0, 0)



= {(x, y) | (x + y, x) = (0, 0)}

so that

x+y =0
x =0

5
 which gives x = y = 0. Hence

ker f = {(x, y) | x = y = 0}
= {(0, 0)}

Recall that for the trivial vector space 0 , there is no non-empty linearly independent
spanning set. Consequently, the empty set is considered to be a basis for 0 . Thus a
e
basis for ker f in this case is ∅, so that N (f ) = 0. e
Now,

imf = {f (x, y) | x, y ∈ R}
= {(x + y, x) | x, y ∈ R}
= {x (1, 1) + y (1, 0) | x, y ∈ R}

In this case, {(1, 1) , (1, 0)} is a linearly independent spanning set for imf and is hence
a basis for imf . Accordingly, R (f ) = 2.
2. Let f : R3 −→ R2 be defined by f (x, y, z) = (x − y, x + y + z). Determine:
(i ) a basis for ker f , and N (f ).
(ii ) a basis for imf , and R (f ).
Solution:
In this case

ker f = (x, y, z) ∈ R3 | f (x, y, z) = (0, 0)



= {(x, y, z) | (x − y, x + y + z) = (0, 0)}

so that

x−y =0
x+y+z =0

yielding y = x and z = −2x. Hence

ker f = {(x, y, z) | y = x, z = −2x}

= {(x, x, −2x) | x ∈ R}
= {x (1, 1, −2) | x ∈ R}

Thus a basis for ker f is {(1, 1, −2)} implying that N (f ) = 1.

On the other hand,

imf = f (x, y, z) | (x, y, z) ∈ R3



= {(x − y, x + y + z) | x, y, z ∈ R}
= {x (1, 1) + y (−1, 1) + z (0, 1) | x, y, z ∈ R}

So the vectors (1, 1) , (−1, 1) , and (0, 1) span imf . Clearly, the three vectors are linearly
dependent. However, any two are linearly independent and will therefore form a basis

6
 for imf . For instance, {(1, 1) , (−1, 1)} a basis for imf . Accordingly, R (f ) = 2.
3. Define f : R3 −→ R3 by f (x, y, z) = (x + y − 3z, −x − 2y + 2z, x − 2y + z). Find the
nullity and rank of f .
Solution:
Method I :
In this case

ker f = {(x, y, z) | (x + y − 3z, −x − 2y + 2z, x − 2y + z) = (0, 0, 0)}

yielding the homogeneous system

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

We reduce the matrix A of coefficients to a row echelon form E

     
1 1 −3 R1 + R2 1 1 −3 1 1 −3
−4R2 + R3 
A = −1 −2 2
  −→  0 −1 −1  0 −1 −1  = E
−→
1 −2 1 R2 + R3 0 −4 3 0 0 7
We now have an equivalent homogeneous system

x + y − 3z = 0
−y − z = 0
7z = 0

and using back substitution gives z = y = x = 0. Hence ker f = {(0, 0, 0)}, and
N (f ) = 0.
By the rank-nullity theorem, 3 = N (f ) + R (f ) = 0 + R (f ). Thus R (f ) = 3.
Method II :

imf = {f (x, y, z) | x, y, z ∈ R}
= {(x + y − 3z, −x − 2y + 2z, x − 2y + z) | x, y, z ∈ R}
= {x (1, −1, 1) + y (1, −2, −2) + z (−3, 2, 1) | x, y, z ∈ R}

The vectors (1, −1, 1), (1, −2, −2), and (−3, 2, 1) span imf . These are the columns
of the matrix A above and from the row echelon form E, the three vectors are linearly
independent (recall that the number of linearly independent columns of a matrix equals
the number of linearly independent rows). Thus they form a basis for imf . Therefore,
R (f ) = 3.
By the rank-nullity theorem, 3 = N (f ) + R (f ) = N (f ) + 3. Thus N (f ) = 0.

7
Exercise:
For each of the linear mappings
1. f : R2 −→ R2 where f (x, y) = (x + y, x − y)
2. f : R2 −→ R2 where f (x, y) = (2x + y, 4x + 2y)
3. f : R3 −→ R3 where f (x, y, z) = (z − y, x − z, y − x)
4. f : R3 −→ R3 where f (x, y, z) = (x − y + z, 2x + y − z, −x + 2y + z)
find:
(a) a basis for ker f and hence N (f ).
(b) a basis for imf and hence R (f ).

Matrices of Linear Functions

If U and V are finite-dimensional vector spaces, then any linear transformation f : U −→ V
can be seen as a matrix transformation. This makes it possible to work with coordinate
matrices of the vectors instead of the vectors themselves.
n o n o
Suppose dim U = n and dim V = m. Let B1 = u1 , u2 , · · · , un and B2 = v1 , v2 , · · · , vm
 f f   f   f f f
be basis for U and V respectively. Then f u1 , f u2 , · · · , f un are vectors in V. So
      f f f
each of f u1 , f u2 , · · · , f un is a linear combination of elements in the basis B2 , i.e.
f f f
 
f u1 = a11 v1 + a12 v2 + · · · + a1m vm
f f f f
f u2 = a21 v1 + a22 v2 + · · · + a2m vm
.. .. ..
f f f f
. . .
 
f un = an1 v1 + an2 v2 + · · · + anm vm
f f f f
 T  
a11 a12 . . . a1m a11 a21 . . . an1
 a21 a22 . . . a2m   a12 a22 . . . an2 
The matrix  ..  =  .. .. , i.e. the transpose of the
   
.. .. .. .. ..
 . . . .   . . . . 
an1 an2 . . . anm a1m a2m . . . anm
matrix of coefficients of the above stytem, is called the matrix representation of f with respect
to the bases B1 and B2 . In the matrix of f , the vectors are written vertically. If the bases B1
and B2 are different, we call the matrix of transformation the transition matrix.
Note: Let A be the matrix
of a linear mapping f : U −→ V.
1. If u ∈ U, then f u = Au. The columns of A span im f , i.e. im f is the column space
of eA (the vector space
e spanned
e by the columns of A). Therefore, the number of linearly
independent columns (rows) of A is the rank of f .
2. If dim U = n and dim V = m, then A is of order m × n irrespective of the bases for U

8
 and V.
Examples:
1. Let f : R2 −→ R3 with f (x, y) = (x + y, x − y, x). Find the matrix A of f with respect
to the standard bases for R2 and R3 .
Solution: {(1, 0) , (0, 1)} is the standard basis for R2 while {(1, 0, 0) , (0, 1, 0) , (0, 0, 1)}
is the standard basis for R3 . Now,

f (1, 0) = (1, 1, 1) = 1 (1, 0, 0) + 1 (0, 1, 0) + 1 (0, 0, 1)

f (0, 1) = (1, −1, 0) = 1 (1, 0, 0) − 1 (0, 1, 0) + 0 (0, 0, 1)
 
 T 1 1
1 1 1
So the matrix of f is A = =  1 −1 .
1 −1 0
1 0
2. Let g : R3 −→ R3 be given by g (x, y, z) = (x − y, z, x + y + z). Find the matrix A of
g with respect to the standard basis for R3 .
Solution: The standard basis for R3 is {(1, 0, 0) , (0, 1, 0) , (0, 0, 1)}. Now,

g (1, 0, 0) = (1, 0, 1) = 1 (1, 0, 0) + 0 (0, 1, 0) + 1 (0, 0, 1)

g (0, 1, 0) = (−1, 0, 1) = −1 (1, 0, 0) + 0 (0, 1, 0) + 1 (0, 0, 1)
g (0, 0, 1) = (0, 1, 1) = 0 (1, 0, 0) + 1 (0, 1, 0) + 1 (0, 0, 1)
 T  
1 0 1 1 −1 0
So A =  −1 0 1  =  0 0 1  is the matrix of g with respect to the standard
0 1 1 1 1 1
3
basis for R .
    
x 1 −1 0 x
Clearly, A y
  =  0 0 1   y  = (x − y, z, x + y + z) = g (x, y, z).
z 1 1 1 z
Now,

im g = {(x − y, z, x + y + z) | x, y, z ∈ R}
= {x (1, 0, 1) + y (−1, 0, 1) + z (0, 1, 1) | x, y, z ∈ R}

So the vectors (1, 0, 1) , (−1, 0, 1), and (0, 1, 1) span im g. These are respectively, the
first, second and third columns of A.
3. Define h : R3 −→ R3 by h (x, y, z) = (−2x + y − z, x − 2y − z, −x − y − 2z). Find the
matrix of h using the basis B = {(1, 0, 0) , (1, 1, 0) , (1, 1, 1)} for R3 .
Solution: In this case,

h (1, 0, 0) = (−2, 1, −1) = a (1, 0, 0) + b (1, 1, 0) + c (1, 1, 1)

h (1, 1, 0) = (−1, −1, −2) = d (1, 0, 0) + e (1, 1, 0) + f (1, 1, 1)
h (1, 1, 1) = (−2, −2, −4) = g (1, 0, 0) + h (1, 1, 0) + i (1, 1, 1)

9
 Hence we have

a + b + c = −2
b+c=1
c = −1

so that c = −1, b = 2, a = −3,

d + e + f = −1
e + f = −1
f = −2

so that f = −2, e = 1, d = 0,

g + h + i = −2
h + i = −2
i = −4

so that i = −4, h = 2, g = 0.
   
a d g −3 0 0
Therefore, the required matrix is  b e h  =  2 1 2 .
c f i −1 −2 −4
4. Define the linear transformation T : R2 −→ R3 by T (x, y) = (y, −5x + 13y, −7x + 16y).
Determine the matrix of T with respect to the bases B = {(3, 1) , (5, 2)} for R2 and
0
B = {(1, 0, −1) , (−1, 2, 2) , (0, 1, 2)} for R3 .
Solution: In this case,

T (3, 1) = (1, −2, −5) = a (1, 0, −1) + b (−1, 2, 2) + c (0, 1, 2)

T (5, 2) = (2, 1, −3) = d (1, 0, −1) + e (−1, 2, 2) + f (0, 1, 2)

and we have

a −b =1
2b + c = −2
−a + 2b + 2c = −5

so that a = 1, b = 0, c = −2,

d −e =2
2e + f = 1
−d + 2e + 2f = −3

so that d = 3, e = 1, f = −1.
 
1 3
0
Therefore, the matrix of T with respect to B and B is  0 1 .
−2 −1

10
Exercise:
1. Let f : R3 −→ R2 with f (x, y, z) = (x + z, 2x + y + z). Find the matrix of f using the
standard bases for R3 and R2 .
2. Suppose U = {all polynomials with deg ≤ 3} and V = {all polynomials with deg ≤ 2},
and supose f : U −→ V such that f = d/dx. Find a matrix of f using {1, x, x2 , x3 } and
{1, x, x2 } as the basis for U and V respectively.
3. Let T : R2 −→ R3 be defined by T (x, y) = (x + 2y, −x, 0). Find the matrix of T with
respect to the bases B1 = {(1, 3) , (−2, 4)} and B2 = {(1, 1, 1) , (2, 2, 0) , (3, 0, 0)} for R2
and R3 respectively.
 
2 5 −3
4. Suppose A = represents the linear transformation f : R3 −→ R2
1 −4 7


defined by f v = Av . Find the matrix that represents the mapping relative to the
bases {(1, 1, 1) , (1, 1, 0) , (1, 0, 0)} and {(1, 3) , (2, 5)} for R3 and R2 respectively.
e e

One-to-one and Onto Transformations

Definition: Let f : U −→ V
be a linear
transformation. Then f is one-to-one (injective),
written 1-1,
if and only if f u = f v implies that u = v , i.e. f is 1-1 if and only if every
vector f u in the range of f is the image of exactly one vector in U.
e e e e
e
Theorem:  Let f : U −→ V be a linear transformation. Then f is one-to-one if and only if
ker f = 0 ( i.e. if and only if N (f ) = 0).


Proof : Suppose f is one-to-one. Let u ∈ ker f . Then f u = 0 . But f 0 = 0 also.
e

Since f is one-to-one, then u = 0 so that ker f = 0 .
e e e e e
e e   e  
Conversely, suppose ker f = 0 and that f u1 = f u2 . Then
e
   f   f 
0 = f u1 − f u2 = f u1 − u2
e f f f f

⇒ u1 − u2 ∈ ker f = 0 ⇒ u1 − u2 = 0 ⇒ u1 = u2 . Hence f is one-to-one.
f f e f f e f f
Examples:
1. Let f : R2 −→ R3 be defined by f (x, y) = (x + y, x − y, x). Determine whether f is
one-to-one or not.
Solution:
In this case
ker f = (x, y) ∈ R2 | f (x, y) = (0, 0, 0)


= {(x, y) | (x + y, x − y, x) = (0, 0, 0)}

so that
x+y =0
x−y =0
x=0

11
 yielding x = y = 0. Hence

ker f = {(x, y) | x = y = 0}
= {(0, 0)}

implying that N (f ) = 0. Therefore, f is one-to-one.

2. Let f : R2 −→ R2 be defined by f (x, y) = (x − y, 2x − 2y). Find ker f and hence
determine whether f is one-to-one or not.
Solution:
In this case

ker f = (x, y) ∈ R2 | f (x, y) = (0, 0)



= {(x, y) | (x − y, 2x − 2y) = (0, 0)}

so that

x−y =0
2x − 2y = 0

yielding y = x. Hence

ker f = {(x, y) | y = x}
= {(x, x) | x ∈ R}
= {x (1, 1) | x ∈ R}

Therefore, {(1, 1)} is a basis for ker f , implying that N (f ) = 1. Now, since ker f 6= 0
(or N (f ) 6= 0), f is not one-to-one. e

Definition: Let f : U −→ V be a linear transformation. Then f is said
to be onto

(surjective), if for each v ∈ V there is at least one u ∈ U such that v = f u . In other
e if im f = V.
words, f is onto if and only e e e

Theorem: Let f : U −→ V be a linear transformation with dim U = n and dim V = m.

Then f is onto if and only if R (f ) = m.

Proof : f is onto ⇔ for each v ∈ V there exists u ∈ U such that f u = v ⇔ im f = V
⇔ R (f ) = m. e e e e

Examples:
1. In Example 1 above, N (f ) = 0. By the rank-nullity theorem, R (f ) = 2 < dim cod f .
Hence im f 6= R3 . Therefore, f is not onto.
2. It was shown in a previous example that the linear mapping f : R3 −→ R2 defined by
f (x, y, z) = (x − y, x + y + z) has R (f ) = 2 = dim cod f . Hence im f = R2 . Thus f
is onto.
Theorem: Let f : U −→ V be a linear transformation with dim U = dim V = n.
(i ) If f is one-to-one, then f is onto.
(ii ) If f is onto, then f is one-to-one.

12
Theorem: Let f : U −→ V be a linear transformation. Suppose that dim U = n and
dim V = m. Then
(i ) If n > m, f is not one-to-one.
(ii ) If m > n, f is not onto.

Inverse of a Linear Transformation

Let f : U −→ V be a linear transformation.

Then f is onto if for every v ∈ V there is at

e ∈ U such that f u
least one u = v , and f is one-to-one if each u ∈ U has a unique image
e
v = f u in V. This existence and uniqueness allows us to define a new transformation
e e e
called
e e inverse of f , denoted by f −1 , which moves V back to U, i.e. f −1 : V −→ U.
the

Conditions for a Linear Transformation to have an Inverse

Let f : U −→ V be a linear transformation where dim U = n and dim V = m. Then f −1
exists if: 
(i ) f is one-to-one ⇔ ker f = 0 ⇔ N (f ) = 0.
(ii ) f is onto ⇔ im f = V ⇔ R (f ) = m.
e

(iii ) n = m.
Note:
(a) Condition (iii ) above is only necessary but not sufficient. It is a consequence of the
first two. So (i ) and (ii ) are sufficient to tell whether f is invertible or not. For instance,
for any finite n and m, f : Rn −→ Rm is not invertible if n 6= m.
(b) f is invertible if its matrix representation with respect to any bases is invertible.
Theorem: Suppose that f : V −→ V is multiplication by an invertible n × n matrix A.
Then f has an inverse and f −1 : V −→ V is multiplication by A−1 .
Theorem: Suppose f : V −→ V is multiplication by an n × n matrix A. Then the following
are equivalent:
(i ) f is invertible.
(ii ) A is invertible.
Examples:
1. Let f : R3 −→ R3 where f (x, y, z) = (x + y + z, x − 3z, y + 4z). Show that f is not
invertible.
Solution:
Method I :
In this case, ker f = {(x, y, z) | (x + y + z, x − 3z, y + 4z) = (0, 0, 0)}
So we have a homogeneous system

x+y +z =0
x − 3z = 0
y + 4z = 0

13
 We now reduce the matrix of coefficients of the system to a row echelon form:
     
1 1 1 1 1 1 1 1 1
 1 0 −3  R1 − R2  0 1 4  R2 − R3  0 1 4 
−→ −→
0 1 4 0 1 4 0 0 0
We now have an equivalent homogeneous system

x+y+z =0
y + 4z = 0

Solving for the basic variables x and y in terms of the free variable z:
The equation y + 4z = 0 gives y = −4z and back substitution into the equation
x + y + z = 0 gives x = −y − z = 4z − z = 3z.
Hence

ker f = {(x, y, z) | x = 3z, y = −4z}

= {(3z, −4z, z) | z ∈ R}
= {z (3, −4, 1) | z ∈ R}

implying that N (f ) = 1 6= 0. So f is not one-to-one and therefore not invertible.

Method II :
In this case,

im f = {(x + y + z, x − 3z, y + 4z) | x, y, z ∈ R}

= {x (1, 1, 0) + y (1, 0, 1) + z (1, −3, 4) | x, y, z ∈ R}

so that im f = span {(1, 1, 0) , (1, 0, 1) , (1, −3, 4)} and

     
1 1 0 R1 − R2 1 1 1 1 1 1
 1 0 1  −→  0 1 −1  4R2 − R3  0 1 −1 
−→
1 −3 4 R1 − R3 0 4 −4 0 0 0
From the row echelon form above, R (f ) = 2 6= 3 = dim cod f . Thus f is not onto and
therefore not invertible.
Method III :  
1 1 1
The matrix of f relative to the standard basis for R3 is A =  1 0 −3 . Calculations
0 1 4
show that det A = 0. Hence A, and consequently f , is not invertible.
2. Let f : R3 −→ R3 where f (x, y, z) = (3x + 2y, x − 3z, y + 4z).
(i ) Show that f is invertible.




(ii ) Given that f u = (1, 0, 0), f v = (0, 1, 0) and f w = (0, 0, 1), find vectors u,
v , and w. e e e e
(iii ) Find a formula for f −1 .
e e

14
Solution:
(i )
Method I :
In this case, im f = {x (3, 1, 0) + y (2, 0, 1) + z (0, −3, 4) | x, y, z ∈ R}. So the vectors
(3, 1, 0), (2, 0, 1), and (0, −3, 4) span im f . Now,
     
3 1 0 3 1 0 3 1 0
 2 0 1  2R1 − 3R2  0 2 −3  3R2 + 2R3  0 2 −3 
−→ −→
0 −3 4 0 −3 4 0 0 −1
From the row echelon form above, R (f ) = 3 = dim cod f . Thus f is onto. From the
rank-nullity theorem, N (f ) = 0. Thus f is one-to-one. Since f is both one-to-one and
onto, it is invertible.
Method II :
In this case, ker f = {(x, y, z) | (3x + 2y, x − 3z, y + 4z) = (0, 0, 0)}
So we have a homogeneous system

3x + 2y =0
x − 3z = 0
y + 4z = 0

We now reduce the matrix of coefficients of the system to a row echelon form:
     
3 2 0 3 2 0 3 2 0
R − 3R2  R − 2R3 
 1 0 −3  1 0 2 9  2 0 2 9 
−→ −→
0 1 4 0 1 4 0 0 1
We now have an equivalent homogeneous system

3x + 2y =0
2y + 9z = 0
z=0

yielding z = y = x = 0. Hence ker f = {(0, 0, 0)} implying that N (f ) = 0. Thus f is

one-to-one. From the rank-nullity theorem, R (f ) = 3. Thus f is onto. Since f is both
one-to-one and onto, it is invertible.
Method III :  
3 2 0
The matrix of f relative to the standard basis for R3 is A =  1 0 −3 . Calculations
0 1 4
show that det A = 1 6= 0. Hence A, and consequently f , is invertible.
(ii ) Let u = (a, b, c). Then f (a, b, c) = (3a + 2b, a − 3c, b + 4c) = (1, 0, 0) so that
e
3a + 2b =1
a − 3c = 0
b + 4c = 0

15
 which yields a = 3, b = −4, and c = 1. Hence u = (3, −4, 1).
e
Next, let v = (p, q, r). Then f (p, q, r) = (3p + 2q, p − 3r, q + 4r) = (0, 1, 0). Thus
e
3p + 2q =0
p − 3r = 1
q + 4r = 0

which yields p = −8, q = 12, and r = −3. Hence v = (−8, 12, −3).
e
If w = (k, m, n), then f (k, m, n) = (3k + 2m, k − 3n, m + 4n) = (0, 0, 1). Thus
e
3k + 2m =0
k − 3n = 0
m + 4n = 1

which yields k = −6, m = 9, and n = −2. Hence w = (−6, 9, −2).

e
(iii ) Since f
−1 exists, then:
f u = (1, 0, 0) ⇒ f −1 (1, 0, 0) = u = (3, −4, 1)
f v = (0, 1, 0) ⇒ f −1 (0, 1, 0) = v = (−8, 12, −3)
e
e

f w = (0, 0, 1) ⇒ f −1 (0, 0, 1) = w = (−6, 9, −2)

e
e
e e
Now, (x, y, z) = x (1, 0, 0) + y (0, 1, 0) + z (0, 0, 1), and from a previous theorem,

f −1 (x, y, z) = f −1 {x (1, 0, 0) + y (0, 1, 0) + z (0, 0, 1)}

= xf −1 (1, 0, 0) + yf −1 (0, 1, 0) + zf −1 (0, 0, 1)
= x (3, −4, 1) + y (−8, 12, −3) + z (−6, 9, −2)
= (3x − 8y − 6z, −4x + 12y + 9z, x − 3y − 2z)

Exercise:
 
1 −1 2
 3 0 1 
Let A =   be the matrix representation of a linear mapping f : R3 −→ R4 .
 1 2 −3 
−2 −1 1
(i ) Determine the rank and nullity of f .
(ii ) Show that f −1 does not exist.

16
EIGENVALUES AND EIGENVECTORS
Definition: Let A be an n×n matrix. The scalar λ is called an eigenvalue of A if there exists
a non-zero vector v such that Av = λv . In this case, the vector v is called an eigenvector of
A corresponding to e λ. e
e the eigenvelue e

Example:
 
3 0
Let A = . Then
8 −1
        
0 3 0 0 0 0
A = = = −1
1 8 −1 1 −1 1
 
0
Thus λ1 = −1 is an eigenvalue of A with a corresponding eigenvector v1 = .
f 1
Similarly,         
1 3 0 1 3 1
A = = =3
2 8 −1 2 6 2
 
1
so that λ2 = 3 is an eigenvalue of A with a corresponding eigenvector v2 = .
f 2
In this case, λ1 = −1 and λ2 = 3 are the only eigenvalues of the 2 × 2 matrix A.

Computing Eigenvalues and Eigenvectors

 
x1
 x2 
Suppose that λ is an eigenvalue of A. Then there exists a non-zero vector v =   such
 
..
e  . 
xn
that Av = λv = λIv . Rewriting this we have
e e e
(λI − A) v = 0 . . . . . . . . . . (1)
e e
If A is an n × n matrix, then Equation (1) corresponds to a homogeneous system of n
equations in the unknowns x1 , x2 , · · · , xn . This system has non-trivial solutions if and only if
det (λI − A) = 0. This in turn means λ is an eigenvalue of A if and only if det (λI − A) = 0.
When expanded det (λI − A) is a polynomial of degree n in λ, called the characteristic
polynomial of A. The equation det (λI − A) = 0 is called the characteristic equation of A,
and the scalars λ satisfying this equation are the eigenvalues of A. Counting multiplicities,
an n × n matrix has exactly n eigenvalues.
Theorem: Let A be an n × n matrix. Then the characteristic polynomial of A has degree
n. Moreover, the coefficient of λn in the characteristic polynomial is 1, i.e. the characteristic
polynomial is monic.

17
Examples:  
3 0
1. Find the eigenvalues of the matrix A = .
8 −1
Solution:  
λ−3 0
In this case, λI − A = . Hence the characteristic polynomial of A is
−8 λ + 1
 
λ−3 0
det (λI − A) = det = (λ − 3) (λ + 1) = λ2 − 2λ − 3
−8 λ + 1
and the characteristic equation, as a product of irreducible factors, is
(λ − 3) (λ + 1) = 0
The solutions of this equation are λ = −1 and λ = 3. These are the eigenvalues of A.
 
1 0 1
2. Find the eigenvalues of A =  0 1 1 .
1 1 0
Solution:
In this case,
 
λ−1 0 −1
det (λI − A) = det  0 λ − 1 −1 
−1 −1 λ
2


= (λ − 1) λ − λ − 1 − 1 (λ − 1)
= (λ − 1) λ2 − λ − 2

= (λ − 1) (λ − 2) (λ + 1)
is the characteristic polynomial of A, in factor form. Accordingly, the the characteristic
equation of A is (λ − 1) (λ − 2) (λ + 1) = 0. Hence λ = −1, λ = 1, and λ = 2 are the
eigenvalues of A.
Exercise:  
5 0 1
Compute the eigenvalues of A =  1 1 0 . Ans: λ = 2
−7 1 0

Finding Bases for Eigenspaces


Theorem: Let λ be an eigenvalue of an n × n matrix A and let Eλ = v | Av = λv . Then
Eλ is a subspace of Rn (or Cn ). e e e
Proof : If Av = λv , then (λI − A) v = 0. Thus Eλ is the null space of the matrix λI − A.
Therefore, Eλ is a subspace of Rn (or Cn ).
e e e e

Note: 0 ∈ Eλ since Eλ is a subspace. However, 0 is not an eigenvector.

e e
Definition: Let λ be an eigenvalue of A. The subspace Eλ is called the eigenspace of A
corresponding to the eigenvalue λ.

18
Examples:  
3 0
1. Find the eigenspaces of A = .
8 −1
Solution:
From Example 1 above, λ = −1 and λ = 3 are are the eigenvalues of A.
For λ = −1,
    
−4 0 x1 0
(λI − A) v = (−I − A) v = =
e e −8 0 x2 0
   
−4x1 = 0 0 0
⇒ ⇒ x1 = 0, and x2 is a free variable. So v = = x2 . Hence
−8x1 =
0  e x2 1
0
E−1 = span .
1
For λ = 3,
    
0 0 x1 0
(λI − A) v = (3I − A) v = =
e e −8 4 x2 0
   
x1 1
⇒ −8x1 + 4x2 = 0 ⇒ x2 = 2x1 . So v = = x1 , and therefore,
  e 2x1 2
1
E3 = span .
2
 
−12 7
2. Find basis or bases for the eigenspace(s) of A = .
−7 2
Solution:
In this case,
 
λ + 12 −7
det (λI − A) = det
7 λ−2
= λ2 + 10λ + 25
= (λ + 5)2

Now (λ + 5)2 = 0 implies that λ = −5 is an eigenvalue of A (of miltiplicity 2). So A

has only one eigenspace.
For λ = −5,
    
7 −7 x1 0
(λI − A) v = (−5I − A) v = =
e e 7 −7 x2 0
   
x2 1
⇒ 7x1 − 7x2 = 0 ⇒ x1 = x2 so that v = = x2 . Hence the eigenvectors
e x2 1
   
1 1
of A are the non-zero scalar multiples of . Thus is a basis for E−5 .
1 1

19
  
0 0 −2
3. Find the eigenspaces of A =  1 2 1 .
1 0 3
Solution:
In this case,

λ 0 2
det (λI − A) = −1 λ − 2 −1
−1 0 λ−3
= (λ − 2) λ2 − 3λ + 2

= (λ − 1) (λ − 2)2

So (λ − 1) (λ − 2)2 = 0 is the characteristic equation of A and the distinct eigenvalues

of A are λ = 1, and λ = 2. Hence there are only two eigenspaces of A.
For λ = 1,
    
1 0 2 x1 0
(λI − A) v = (I − A) v = −1 −1 −1
   x2 = 0 
 
e e −1 0 −2 x3 0

Reducing the matrix I − A of coefficients of the system:

   
1 0 2 R1 + R2 1 0 2
 −1 −1 −1  −→  0 −1 1 
−1 0 −2 R1 + R3 0 0 0
   
−2x3 −2
x1 + 2x3 = 0
⇒ ⇒ x1 = −2x3 and x2 = x3 . So v =  x3  = x3  1 ,
−x2 + x3 = 0 e x3 1
i.e. the eigenvectors of corresponding to λ = 1 are the non-zero scalar multiples of
     
−2  −2   −2 
 1 . Hence  1  is a basis for E1 , i.e. E1 = span  1  .
1 1 1
   

For λ = 2,
    
2 0 2 x1 0
(λI − A) v = (2I − A) v =  −1 0 −1   x2  =  0 
e e −1 0 −1 x3 0

Reducing the matrix I − A of coefficients of the system to row echelon form:

   
2 0 2 R1 + 2R2 2 0 2
 −1 0 −1  −→  0 0 0 
−1 0 −1 R2 − R3 0 0 0

20
 So x2 and x3 is a free variable. Now, 2x1 + 2x3 = 0 ⇒ x1 = −x3 . Thus
         
−x3 0 −x3 0 −1
v =  x2  =  x2  +  0  = x 2  1  + x3  0 
e x3 0 x3 0 1

i.e. the eigenvectors of A corresponding to λ = 2 are the non-zero linear combinations

       
0 −1  0 −1 
of 1 and
   0  which are linearly independent. Hence  1  ,  0  is a
0 1 0 1
 
   
 0 −1 
basis for E2 , i.e. E2 = span  1 ,  0  .
0 1
 

Theorem: Eigenvectors corresponding to distinct eigenvalues are linearly independent.

Exercise:
Find the eigenspace(s) of each of the following matrices:
     
4 2 2 1
(a) Ans: E1 = span , E6 = span
3 3 −3 1
     
2 −1 1 1
(b) Ans: E0 = span , E4 = span
−4 2 2 −2
   
−5 −5 −9  3 
(c)  8 9 18  Ans: E−1 = span  −6 
−2 −3 −7 2
 
     
−1 −3 −9  0 1 
(d )  0 5 18  Ans: E−1 = span  −3  ,  −3 
0 −2 −7 1 1
 
       
3 2 4  1 0   2 
(e) 2 0 2 Ans: E−1 = span
   −2 , −2
   , E8 = span  1 
4 2 3 0 1 2
   
     
1 −1 4  1   1 
(f )  3 2 −1  Ans: E−2 = span  −1  , E1 = span  4  ,
2 1 −1 −1 −1
   
 
 1 
E3 = span  2 
1
 

21
Eigenvalues of Triangular Matrices
Theorem: If A is a triangular matrix, then the eigenvalues of A are the entries on the main
diagonal of A.
 
a11 a12 a13 ··· a1n

 0 a22 a23 ··· a2n 

Proof : If A = 
 0 0 a33 ··· a3n  is an upper triangular matrix, then

 .. .. .. .. .. 
 . . . . . 
0 0 0 ··· ann
 
λ − a11 −a12 −a13 ··· −a1n

 0 λ − a22 −a23 ··· −a2n 

λI − A = 
 0 0 λ − a33 ··· −a3n .

 .. .. .. .. .. 
 . . . . . 
0 0 0 ··· λ − ann
Since the determinat of a triangular matrix is equal to the product of its diagonal entries, then
det (λI − A) = (λ − a11 ) (λ − a22 ) · · · (λ − ann ) with zeros λ = a11 , λ = a22 , · · · , λ = ann .
The proof for a lower triangular matrix is similar.
 
1/2 0 0
Example: By inspection the eigenvalues of the lower triangular matrix A =  −1 2/3 0 
5 −8 −1/4
are λ = 1/2, λ = 2/3, and λ = −1/4.

Eigenvalues of Powers of a Matrix

Once the eigenvalues and eigenvectors of a matrix A are found, it becomes easy to determine
the eigenvalues and eigenvectors of any positive integer power of A. For instance, if λ is an
eigenvalue of A and v is a corresponding eigenvector, then Av = λv and it follows that
e e
e
A v = A Av = A λv = λ Av = λ λv = λ2 v
2




e e e e e e
This shows that λ2 is an eigenvalue of A2 and v is a corresponding eigenvector.
e
Theorem: Suppose λ is an eigenvalue of a matrix A and v is a corresponding eigenvector.
Then for any positive integer k, λk is an eigenvalue of Ak andev is a corresponding eigenvector.
e
Example:  
0 0 −2
From a previous example, the matrix A =  1 2 1  has eigenvalues λ = 1 and λ = 2.
1 0 3
From the above theorem, both λ = 17 = 1 and λ = 27 = 128 are the eigenvalues of A7 . From
 
−2
the same example,  1  is an eigenvector of A corresponding to the eigenvalue λ = 1.
1

22
From the above theorem, it is also an eigenvector of A7 corresponding to the eigenvalue
   
−1 0
7
λ = 1 = 1. Similarly, the eigenvectors  0  and 1  corresponding to the eigenvalue

1 0
λ = 2 of A are also eigenvector of A7 corresponding to the eigenvalue λ = 27 = 128.

Similar Matrices and Diagonalization

Definition: Two n × n matrices A and B are said to be similar if there exists an invertible
n × n matrix P such that B = P −1 AP . Equivalently, A and B are similar if and only if there
exists an invertible n × n matrix P such that P B = AP .
Examples:      
2 1 4 −2 2 −1
1. Let A = ,B= , and P = . Then
0 −1 5 −3 −1 1
       
2 −1 4 −2 3 −1 2 1 2 −1
PB = = = = AP
−1 1 5 −3 1 −1 0 −1 −1 1

Since det P = 1 6= 0, P is invertible. Therefore, by definition, A

and B are similar.
     
−6 −3 −25 1 0 0 2
4 3
2. Let A =  2 1 8 , B =  0 −1 0 , and P = 0
 1 −1 . Then
2 2 7 0 0 2 3
5 7
 
2 4 3
det P = 3 6= 0. So P is invertible. It can be verified that P A = BP =  0 −1 1 
6 10 14
−1
and A = P BP . Hence A and B are be similar.
Theorem: Similar matrices have same determinant.
Proof : Suppose A and B are similar. Then there exists an invertible matrix P such that
A = P −1 BP . Accordingly,

det A = det P −1 BP

= det P −1 det B det P

1
= det B det P
det P
= det B

Examples:
1. In Example 1 above, matrices A and B are similar. In this case, det A = det B = −2.
2. In Example 2 above, it was shown that A and B are similar. Calculations show that
det A = det B = −2.
Theorem: Similar matrices have the same characteristic polynomial and hence same eigen-
values.

23
Proof : Suppose A and B are similar. Then there exists an invertible matrix P such that
A = P −1 BP . Accordingly,

λI − A = λI − P −1 BP
= λI P −1 P − P −1 BP

= P −1 (λI) P − P −1 BP
= P −1 (λI − B) P

and

det (λI − A) = det P −1 (λI − B) P

 

= det P −1 det (λI − B) det P

1
= det (λI − B) det P
det P
= det (λI − B)

So A and B have the same characteristic polynomial and hence same characteristic equation.
Since eigenvalues are roots of the characteristic equation, they have the same eigenvalues.
Examples:    
2 1 4 −2
1. A prevoius example showed that the matrices A = and B = are
0 −1 5 −3
similar. Clearly, the eigenvalues of A are −1 and 2 (since A is a triangular matrix).
These are the eigenvalues of B also.
 
−6 −3 −25
2. In a prevoius example, it was shown that the matrices A =  2 1 8  and
  2 2 7
1 0 0
B = 0 −1 0  are similar. Clearly, the eigenvalues of B are 1, −1 and 2 (since B

0 0 2
is a diagonal matrix). These are the eigenvalues of A also.
Exercise:    
1 2 3 15 0 0
Determine whether matrices A =  4 5 6  and B =  0 0 0  are similar or not.
7 8 9 0 0 0
Definition: An n × n matrix A is diagonalizable if there exists an invertible n × n matrix
P such that P −1 AP is a diagonal matrix. In other words, A is diagonalizable if there is a
diagonal matrix D such that A is similar to D. The matrix P is said to diagonalize A in this
case.
Remark: If A is diagonalizable, then A is similar to a diagonal matrix whose diagonal
components are the eigenvalues of A.
Theorem: An n × n matrix A is diagonalizable if and only if it has n linearly independent
eigenvectors.

24
Corollary: If an n × n matrix A has n distinct eigenvalues, then it is diagonalizable.

Procedure for Diagonalizing an n × n Matrix

Step 1 : Find n linearly indepenednt eigenvectors of A, say p1 , p2 , · · · , pn .
Step 2: Form a matrix P having p1 , p2 , · · · , pn as its column vectors.
f f f

Step 3 : The matrix P −1 AP will then be diagonal with λ1 , λ2 , · · · , λn as its successive

f f f
diagonal entries, where λi is the eigenvalue corresponding to pi (i = 1, 2, · · · , n).
e
Examples:  
−3 2
1. The characteristic equation of A = is
−2 1
 
λ + 3 −2
det (λI − A) = det
2 λ−1
= (λ + 3) (λ − 1) + 4
= λ2 + 2λ + 1
= (λ + 1)2 = 0

Hence λ = −1 is the only eigenvalue of A. The eigenvectors corresponding to λ = −1

    
2 −2 x1 0
are the non-trivial solutions of (−I − A) v = 0 ⇒ =
e e 2 −2 x 2 0
     
x2 1 1
⇒ 2x1 − 2x2 = 0 ⇒ x1 = x2 so that v = = x2 . Hence
e x 2 1 1
is a basis for E−1 . Since this unique eigenspace is 1-dimensional, then A does not have
two linearly independent eigenvectors and is therefore not diagonalizable.
 
4 2
2. Calculations show that the matrix A = has eigenvalues λ = 1 and λ = 6 with
3 3
   
2 1
corresponding eigenvectors and respectively. Clearly, the two eigenvectors
−3 1
are linearly independent.
 
2 1
Now, the matrix P = diagonalizes A. In this case,
−3 1
   
−1 1 1 −1 4 2 2 1
P AP =
5 3 2 3 3 −3 1
  
1 1 −1 2 6
=
5 3 2 −3 6
 
1 5 0
=
5 0 30
 
1 0
=
0 6

25
 which is a diagonal matrix whose diagonal components are the eigenvalues of A.
 
0 0 −2
3. Let A =  1 2 1 . Determine an invertible matrix P such that D = P −1 AP is a
1 0 3
diagonal matrix.
Solution: From a previous example, the eigenvalues of A are λ = 1, and λ = 2, where
p1 = (−2, 1, 1) is an eigenvector of A corresponding to λ = 1 while p2 = (−1, 0, 1) and
f p = (0, 1, 0) are eigenvectors of A corresponding to λ = 2. It can
and f be verified that
3
 
f −2 −1 0
p1 , p2 , andp3 are linearly independent so that P =  1 0 1  diagonalizes A. In
f f f
  1 1 0
−1 0 −1
this case, P −1 =  1 0 2  and
1 1 1
   
−1 0 −1 0 0 −2 −2 −1 0
P −1 AP =  1 0 2   1 2 1   1 0 1 
1 1 1 1 0 3 1 1 0
  
−1 0 −1 −2 −1 0
=  2 0 4   1 0 1 
2 2 2 1 1 0
 
1 0 0
= 0 2 0 

0 0 2

Note: There is no preferred order for the columns of the diagonalizing matrix P . Since
the ith diagonal entry of P −1 AP is the eigenvalue corresponding to the ith column vector of
P , changing the order of the columns of P just
 changes the  order of the eigenvalues on the
−1 0 −2
diagonal of P −1 AP . Thus if we write P =  0 1 1  in Example 3 above, we would
  1 0 1
2 0 0
have obtained P −1 AP =  0 2 0 .
0 0 1
 
1 0 1
4. From a previous example, the matrix A =  0 1 1  has three distinct eigenvalues,
1 1 0
namely λ = −1, λ = 1, and λ = 2. Thus by the corrollary above, A is diagonalizable.
Now, calculations show that (1, 1, −2), (−1, 1, 0) and (1, 1, 1) are eigenvectors of A
 
1 −1 1
corresponding to λ = −1, λ = 1, and λ = 2 respectively. So P =  1 1 1  is a
−2 0 1

26
 diagonalizing matrix and
   
1 1 −2 1 0 1 1 −1 1
1
P −1 AP =  −3 3 0  0 1 1  1 1 1 
6
2 2 2 1 1 0 −2 0 1
 
−1 0 0
=  0 1 0 .
0 0 2

Computing Powers of a Matrix

Let A be an n × n matrix and let P be an invertible n × n matrix. Then

(P −1 AP )2 = P −1 AP P −1 AP = P −1 AIAP = P −1 A2 P.

In general, for any positive integer k,

(P −1 AP )k = P −1 Ak P . . . . . . . . . . (∗)

From Equation (∗), if A is diagonalizable and P −1 AP = D is a diagonal matrix, then

P −1 Ak P = (P −1 AP )k = Dk . Solving this equation for Ak ,

P P −1 Ak P P −1 = P Dk P −1
⇒ Ak = P Dk P −1 . . . . . . . . . . (∗∗)
Example:  
1 −1 1
In Example 4 above, it was shown that the matrix A is diagonalized by P =  1 1 1 
  −2 0 1
−1 0 0
−1
and that D = P AP =  0 1 0 . Thus from Equation (∗∗) above,
0 0 2

A5 = P D5 P −1
   
1 −1 1 −1 0 0 1/6 1/6 −1/3

=  1 1 1   0 1 0   −1/2 1/2 0 
−2 0 1 0 0 32 1/3 1/3 1/3
  
−1 −1 32 1/6 1/6 −1/3

=  −1 1 32   −1/2 1/2 0 
2 0 32 1/3 1/3 1/3
 
11 10 11
=  10 11 10  .
11 11 10

27
Exercise:
 
0 1 1
1. Consider the matrix A =  1 1 0 .
1 0 1
(a) Find an invertible matrix P and a diagonal matrix D such that P −1 AP = D.
 
42 43 43
(b) Use the diagonal matrix D in (a) above to compute A7 . Ans: A7 =  43 43 42 .
43 42 43
 
−4 0 −1
2. Let B =  2 −1 0 .
2 0 −1
(a) Find the eigenvalues and the corresponding eigenspaces of B.
    
 0   1 
Ans: λ = −1, E−1 = span  1  ; λ = −2, E−2 = span  −2  ;
0 −2
   
 
 −1 
λ = −3, E−3 = span  1 
1
 

(b) Find the eigenvalues and bases for the corresponding eigenspaces of B 8 .
(c) Show that B is diagonalizable and hence find a diagonalizable matrix P such that
P −1 AP = D is a diagonal matrix.
(d ) Use the diagonalization in part (c) above to evaluate B 10 .
 
117074 0 58025
Ans: B 10 =  −116050 1 −57002 
−116050 0 −57001

Polynomials of Matrices
Consider a polynomial f (t) over a field F given by

f (t) = an tn + an−1 tn−1 + an−2 tn−2 + · · · + a2 t2 + a1 t + a0 .

If A is an n × n matrix over F , we define the polynomial of A as

f (A) = an An + an−1 An−1 + an−2 An−2 + · · · + a2 A2 + a1 A + a0 I

where I is the n × n identity matrix.

We say that A is a root/zero of a polynomial f (t) if f (A) = 0 (the zero n × n matrix).

28
Example:  
2 −4 −3
Let f (x) = x + 3x + 2 and A = . Then
2 1

f (A) = A2 + 3A + 2I
      
−4 −3 −4 −3 −4 −3 1 0
= +3 +2
2 1 2 1 2 1 0 1
     
10 9 −12 −9 2 0
= + +
−6 −5 6 3 0 2
 
0 0
= .
0 0

Since f (A) = 0, then A is a zero of f (x).

Note: If f (x) and g(x) are polynomials and g(x) is a multiple of f (x), i.e. g(x)=h(x)f (x)
for some polynomial h(x), then g(A)=h(A)f (A).
Theorem (Cayley-Hamilton Theorem): Every square matrix is a zero of its character-
istic polynomial.
Example:  
−4 −3
Let A = . Then f (λ) = λ2 + 3λ + 2 is the characteristic polynomial of A. From
2 1
the preceeding example, f (A) = 0.

Applications of the Cayley-Hamilton Theorem

Let A be an invertible n × n matrix. Then the Cayley-Hamilton theorem can be used to find
the inverse of A as follows:
Let f (λ) = λn + an−1 λn−1 + · · · + a1 λ + a0 be the characteristic polynomial of A. Then by
the Cayley-Hamilton Theorem,

f (A) = An + an−1 An−1 + · · · + a1 A + a0 I = 0

⇒ A(An−1 + an−1 An−2 + · · · + a1 I + a0 A−1 ) = 0
⇒ An−1 + an−1 An−2 + · · · + a1 I + a0 A−1 = 0
⇒ −a0 A−1 = An−1 + an−1 An−2 + · · · + a1 I
−1  n−1
⇒ A−1 = + an−1 An−2 + · · · + a1 I

A
a0
 
0 1 1
Example: Use the Cayley-Hamilton theorem to find the inverse of A =  1 1 0 .
1 0 1

29
Solution:
In this case, det A = −2 6= 0. Hence A−1 exists. Now,

λ −1 −1
det(λI − A) = −1 λ − 1 0
−1 0 λ−1
= λ(λ2 − 2λ + 1) + 1[−(λ − 1)] − 1(λ − 1)
= λ3 − 2λ2 − λ + 2

So f (λ) = λ3 − 2λ2 − λ + 2 is the characteristic polynomial of A.

By the Cayley-Hamilton theorem,

f (A) = A3 − 2A2 − A + 2I = 0
⇒ A(A2 − 2A − I + 2A−1 ) = 0
⇒ A2 − 2A − I + 2A−1 = 0
⇒ A−1 = −1/2 A2 − 2A − I
 

      
0 1 1 0 1 1 0 2 2 1 0 0
⇒ A−1 = −1/2  1 1 0   1 1 0  −  2 2 0  −  0 1 0 
1 0 1 1 0 1 2 0 2 0 0 1
     
2 1 1 0 2 2 1 0 0
= −1/2  1 2 1  −  2 2 0  −  0 1 0 
1 1 2 2 0 2 0 0 1
 
1 −1 −1
= /2 −1 −1 1 
−1 
−1 1 −1
 
−1/2 1/2 1/2

=  1/2 1/2 −1/2  .

1/2 −1/2 1/2

    
0 1 1 −1/2 1/2 1/2 1 0 0
−1 −1
As a check, AA = 1 1 0
   1/2 1/2 /2 = 0
  1 0 .
1 0 1 1/2 −1 /2 /2
1 0 0 1
Exercise:
Show that the following matrix is invertible and hence use the Cayley-Hamilton theorem to
find its inverse.
 
1 2
(a)
3 4
 
2 0 2
(b)  0 2 2 
2 2 0

30
Definition: The minimal polynomial of a square matrix A is the monic polynomial of the
smallest degree having A as a root, i.e. h(λ) is the minimal polynomial of A if h(λ) is the
monic polynomial of the smallest degree such that h(A) = 0.
Theorem: The minimal polynomial and the characteristic polynomial of an n × n matrix
have the same irreducible factors.
Theorem: Let A be an n×n matrix. The minimal polynomial of A divides every polynomial
which has A as its zero. In particular, the minimal polynomial divides the characteristic
polynomial.
Example:  
0 0 −2
1. Find the minimal polynomial of A =  1 2 1 .
1 0 3
Solution:
From a previous example f (λ) = (λ − 1)(λ − 2)2 is the characteristic polynomial of A.
So the irreducible factors of f (λ), and hence of the minimal polynomial of A, are λ − 1
and λ − 2. Therefore, the minimal polynomial of A is either h(λ) = (λ − 1)(λ − 2) or
f (λ) = (λ − 1)(λ − 2)2 .
Consider h(λ) = (λ − 1)(λ − 2). Then

h(A) = (A − I)(A − 2I)

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

So h(λ) = λ2 − 3λ + 2 is the minimal polynomial of A.

 
1 0 1
2. Calculate the minimal polynomial of A =  0 1 1 .
1 1 0
Solution:
From a previous example, f (λ) = (λ − 1)(λ + 1)(λ − 2) is the characteristic polynomial
of A. So the irreducible factors of f (λ), and hence of the minimal polynomial, are λ − 1,
λ + 1 and λ − 2. Each of these has multiplicity of 1. Since f (A) = 0 (by the Cayley-
Hamilton theorem), then f (λ) = (λ − 1)(λ + 1)(λ − 2) is the minimal polynomial of A.
Exercise:  
5 0 1
1. Find the minimal polynomial of A =  1 1 0 .
−7 1 0

31
 Ans: f (λ) = (λ − 2)3 = λ3 − 6λ2 + 12λ − 8
 
−2 3 0 1
 0 0 2 4 
2. Find the minimal polynomial of B =   0
.
0 1 3 
0 0 0 −2

32