BMTE-141

LINEAR ALGEBRA

Block

3
Linear Tranformations

UNIT 7
Linear Transformations-I 5
UNIT 8
Bases and Dimension 32
UNIT 9
Linear Transformations and Matrices 61
Miscellaneous Examples and Exercises 87
Course Design Committee*
Prof. Rashmi Bhardwaj Prof. Meena Sahai
G.G.S. Indraprastha University, Delhi University of Lucknow
Dr. Sunita Gupta Dr. Sachi Srivastava
University of Delhi University of Delhi
Prof. Amber Habib Prof. Jugal Verma
Shiv Nadar University I.I.T., Mumbai
Gautam Buddha Nagar
Faculty members
Prof. S. A. Katre
School of Sciences, IGNOU
University of Pune
Prof. M. S. Nathawat (Director)
Prof. V. Krishna Kumar
Dr. Deepika
NISER, Bhubaneswar
Mr. Pawan Kumar
Dr. Amit Kulshreshtha
Prof. Poornima Mital
IISER, Mohali
Prof. Parvin Sinclair
Prof. Aparna Mehra
Prof. Sujatha Varma
I.I.T., Delhi
Dr. S. Venkataraman
Prof. Rahul Roy
Indian Statistical Institute, Delhi

* The Committee met in August, 2016. The course design is based on the recommendations of the
Programme Expert Committee and the UGC-CBCS template.

Block Preparation Team

Dr. S. Venkataraman Prof. Amber Habib
School of Sciences Shiv Nadar University
IGNOU Dadri

Course Coordinator: Dr. S. Venkataraman

Acknowledgement: We have used some of the material from the course MTE-02, Linear
Algebra in this course material. Prof. Amber Habib for generously sharing his lecture notes.
June 2021
© Indira Gandhi National Open University
ISBN-
All right reserved. No part of this work may be reproduced in any form, by mimeograph or any other
means, without permission in writing from the Indira Gandhi National Open University.
Further information on the Indira Gandhi National Open University courses, may be obtained from the
University’s office at Maidan Garhi, New Delhi-110 068 and IGNOU website www.ignou.ac.in.
Printed and published on behalf of the Indira Gandhi National Open University, New Delhi by
Prof. Sujatha Varma, School of Sciences.
Unit 7 Linear Transformation-I

UNIT 7

LINEAR TRANSFORMATIONS I

Structure Page No.

7.1 Introduction
Objectives
7.2 Linear Transformations
7.3 Spaces Associated with a Linear Transformation
The Range Space and the Kernel
Rank and Nullity
7.4 Some Types of Linear Transformations
7.5 Homomorphism Theorems
7.6 Summary
7.7 Solutions/Answers

7.1 INTRODUCTION
You have already learnt about a vector space and several concepts
related to it. In this Unit we initiate the study of certain mappings
between two vector spaces, called linear transformations. The
importance of these mappings can be realized from the fact that, in the
calculus of several variables, every continuously differentiable function
can be replaced, to a first approximation, by a linear one. This fact is a
reflection of a general principle that every problem on the change of
some quantity under the action of several factors can be regarded, to a
first approximation, as a linear problem. It often turns out that this gives
an adequate result. Also, in physics it is important to know how vectors
behave under a change of the coordinate system. This requires a study
of linear transformations.

In this unit we study linear transformations and their properties, as well

as two spaces associated with a linear transformation, and their
dimensions. Then, we prove the existence of linear transformations with
some specific properties. We discuss the notion of an isomorphism
between two vector spaces, which allows us to say that all finite-
dimensional vector spaces of the same dimension are the ‘same’, in a
certain sense.
5
 Finally, we state and prove the Fundamental Theorem of
Homomorphism and some of its corollaries, and apply them to various
situations.

Since this unit uses concepts developed in Units 2, 3 and 5, we suggest

that you revise these units before going further.

Objectives
After reading this unit, you should be able to
• verify the linearity of certain mappings between vector spaces;
• construct linear transformations with certain specified properties;
• calculate the rank and nullity of a linear operator;
• prove and apply the Rank Nullity Theorem;
• define an isomorphism between two vector space;
• show that two vector spaces are isomorphic if and only if they
have the same dimension;
• prove and use the Fundamental Theorem of Homomorphism.

7.2 LINEAR TRANSFORMATIONS

In Unit 2 you came across the vector spaces R 2 and R 3 . Now consider
the mapping f : R 2 → R 3 : f ( x, y) = (x , y, 0) (see Fig. 1).

f is a well defined function. Also notice that

i) f ((a, b) + (c, d)) = f ((a + c, b + d)) = (a + c, b + d, 0) = (a, b, 0) + (c, d, 0)

= f ((a, b)) + f ((c, d)), for (a, b), (c, d) ∈ R 2 , and

Fig.1: f transforms ABCD to A ′B ′C′D′.

ii) for any α ∈ R and (a , b) ∈ R 2 ,

6
f (α(a , b)) = f ((αa , αb)) = (αa, αb, 0) = α(a, b, 0) = αf ((a, b)).
Unit 7 Linear Transformation-I
So we have a function f between two vector spaces that (i) and (ii)
above hold true.

(i) says that the sum of two plane vectors is mapped under f to the sum
of their images under f (ii) says that a line in the plane R 2 is mapped
under f to a line in R 2 .

The properties (i) and (ii) together says that f is linear, a term that we
now define.

Definition 1: Let U and V be a vector spaces over a field F. A linear

transformation (or linear operator) from U to V is function T : U → V,
such that

LT1) T ( u 1 + u 2 ) = T ( u 1 ) + T ( u 2 ), for u1 , u 2 ∈ U, and

LT2) T(αu ) = α T (u) for α ∈ F and u ∈ U.

The conditions LT1 and LT2 can be combined to give the following
equivalent condition.

LT3) T (α1 u 1 + α 2 u 2 ) = α 1T ( u 1 ) + α 2 T ( u 2 ), for α1 , α 2 ∈ F and u 1 , u 2 ∈ U.

What we are saying is that [LT1 and LT2] ⇔ LT3. This can be easily
shown as follows:

We will show that LT 3  LT1 and LT3  LT 2. Now, LT3 is true

∀ α1 , α 2 ∈ F. Therefore, it is certainly true for α 1 = 1 = α 2 , that is, LT1
holds.

Now, to show that LT 2 is true, consider T(αu) for any a ∈ F and u ∈ U.

We have T(αu + 0.u) = αT(u) + 0.T(u ) = αT(u ), thus proving that LT 2
holds.

You can try and prove the converse now. That is what the following
exercise is all about!

E1) Show that the conditions LT1 and LT2 together imply LT3.

Before going further, let us note two properties of any transformation

T : U → V, which follow from LT1 (or LT2, or LT3).

LT4) T(0) = 0. Let’s see why this is true. Since T(0) = T(0 + 0) = T(0) (by
LT1), we subtract T(0) from both sides to get T(0) = 0.

LT5) T( −u) = −T(u) ∀ u ∈ U. Why is this so? Well, since

0 = T(0) = T(u − u ) = T(u ) + T(−u),
we get T(−u) = −T(u ).

E2) Can you show how LT4 and LT5 will follow from LT2?
7
 Now let us look at some common linear transformations.

Example 1: Consider the vector space U over a field F, and the

function T : U → U defined by T(u) = u for all u ∈ U.

Show that T is a linear transformation. (This transformation is called the

identity transformation, and is denoted by I U or just I, if the underlying
vector space is understood).

Solution: For any α, β ∈ F and u1 , u 2 ∈ U, we have

T (αu 1 + β u 2 ) = αu 1 + β u 2 = αT ( u1 ) + β T ( u 2 ).
Hence, LT3 holds, and T is a linear transformation.
***

Example 2: Let T : U → V be define by T(u) = 0 for all u ∈ U.

Check that T is a linear transformation. (It is called the null or zero
transformation, and is denoted by 0. )

Solution: For any α, β ∈ F and u1 , u 2 ∈ U, we have

T (αu 1 + β u 2 ) = 0 = α.0 + β.0 = αT ( u 1 ) + β T ( u 2 ).
Therefore, T is a linear transformation.
***

Example 3: Note that we can view an m× n matrix A as a map

between R n to R m . Given a vector v ∈ R n , writing v as a column
vector, we can view v as a n × 1 matrix, so the matrix multiplication Av
is defined. Further, Av is a m × 1 matrix, which is a column vector of
length m. So, we can regard Av as an element of R m . Check that this
map is well defined. Further, we have seen in Example 3,
miscellaneous exercises, that A(αv1 + βv 2 ) = αAv1 + βAv2 . So, every
m× n matrix defines a linear transformation from R n to R m .
***

Example 4: Consider the function pr1 : R n → R, defined by

pr1[( x 1 , K, x n )] = x1 . Show that this is a linear transformation. (This is
called the projection on the first coordinate. Similarly, we can define
pri : R n → R by pri [( x 1 , K, x i−1 , x1 , K , x n )] = x i to be the projection on
the i th coordinate for i = 2, K, n. For instance,
pr2 : R 3 → R : pr2 ( x , y, z) = y.

Solution: We will use LT3 to show that pr1 is a linear operator. For
α, β ∈ R and ( x1 , K, x n ), ( y1 , K, y n ) in R n , we have
pr1 [( α( x 1 , K, x n ) + β( y1 , K , y n )]
= pr1 (αx 1 + β y1 , αx 2 + β y 2 , K , αx n + β y n ) = αx1 + β y1
= αpr1 [( x 1 , K, x n )] + β pr1 [( y1 , K, y n )].

8
Unit 7 Linear Transformation-I
Thus pr1 (and similarly pri ) is a linear transformation.
Before going to the next example, we make a remark about projections.

Remark 1: Consider the function p : R 3 → R 2 : p( x, y, z) = (x , y). This is a

projection from R 3 on to the xy -plane. Similarly, the functions f and g,
from R 3 → R 2 , defined by f ( x, y, z) = ( x, z) and g( x , y, z) = ( y, z) are
projections from R 3 onto the xz -plane and the yz -plane, respectively.

In general, any function θ : R n → R m (n > m), which is defined by

dropping any (n − m) coordinates, is a projection map.

Now let us see another example of a linear transformation that is very

geometric in nature.

Example 5: Let T : R 2 → R 2 be defined by T(x , y) = ( x, − y) ∀ x , y ∈ R.

Show that T is a linear transformation.
(This is the reflection in the x -axis that we show in Fig.2.) Fig. 2: Q is the reflection
of P in the x-axis.
Solution: For α, β ∈ R and ( x1 , y1 ), ( x 2 , y 2 ) ∈ R 2 , we have
T[α ( x 1 , y1 ) + β( x 2 , y 2 )] = T (αx 1 + β x 2 , αy1 + β y 2 ) = (αx 1 + β x 2 − αy1 − βy 2 )
= α ( x 1 , − y1 ) + β ( x 2 , − y 2 )
αT ( x 1 , y1 ) + β T ( x 2 , y 2 ).
Therefore, T is a linear transformation.
***

Example 6: In this example we will consider the projection operator and

the reflection operator about a line y = mx, m ≠ 0.

Fig. 3

In Fig. 3 P is any point in R 2 and Q is the foot of the perpendicular

from P to the line y = mx. R is the point on the line joining P and Q
such that PQ = QR. The point Q is the projection of P on the line
y = mx and R is the reflection of the point P about the line y = mx.
Show that (x , y) a ( x′, y′) and (x , y) a ( x ′′, y′′) are linear operators.

9
 Solution: Since PQ is perpendicular the line y = mx, the line through
−1
PQ has slope . (Recall that if m1 , m 2 are the slopes of perpendicular
m
lines, m1 ≠ 0, m 2 ≠ 0, m1m 2 = −1). Using the point-slope form of the
equation of a line, the equation of the line passing through PQ is
−1
( Y − y) = (X − x ). Since ( x′, y′) lies on this line we have
m
−1
( y′ − y) = ( x ′ − x ). Since ( x′, y′) lies on the line y = mx, we have
m
y′ = mx′. Using this to eliminate y′ from the above equation we get
−1 1 x 1
(mx ′ − y) = (x ′ − x ) or mx′ + x ′ = + y or x ′ = 2 ( x + ym).
m m m m +1
m
∴ y′ = mx ′ = 2 ( x + ym).
m +1
 x + ym mx + ym 2 
∴ ( x, y) a ( x′, y′) is given by P(x , y) =  2 , . This is of
 m +1 m 2 + 1 
the form ( x, y) a (ax + by, a ′x + b′y) where
1 m m m2
a= , b = , a ′ = and b ′ = . Check that
m2 + 1 m2 + 1 m2 + 1 m2 + 1
T( x, y) = (ax + by, a′x + b′y) is a linear operator for a, b ∈ R. (see
exercise 6 below).
Since Q is the mid point of PR, we have

 x + x ′′ y + y′′   x + ym mx + ym 
2
Q= ,  =  , .
 2 2   m 2 + 1 m 2 + 1 

x + x ′′ x + ym y + y′′ mx + ym 2
∴ = 2 , = .
2 m +1 2 m2 + 1
Solving for x′′, we get

 x + ym  x (1 − m 2 ) + 2 ym
x ′′ = 2 2  − x = .
 m +1  m2 + 1
Solving for y′′, we get

 mx + ym 2  2mx + y(m 2 − 1)
y′′ = 2 2

 − y = .
 m +1  m2 + 1
Again, we see that R ( x, y) = (ax + by, a′x + b′y) where
(1 − m 2 ) 2m 2m m2 − 1
a= , b = , a ′ = , b ′ = .
1+ m2 m2 + 1 m2 + 1 m2 + 1

∴ R is also a linear operator.

***

So far we’ve given examples of linear transformations. Now we give an

example of a very important function which is not linear. This
example’s importance lies in its geometric applications.
Fig. 4: A ′B′C′D′ is the
10 translation of
ABCD by (1, 1).
Unit 7 Linear Transformation-I
Example 7: Let u 0 be a fixed non-zero vector in U. Define T : U → U
by T(u ) = u + u 0 ∀ u ∈ U. Show that T is not a linear transformation. ( T
is called the translation by u 0 . See Fig.4 for a geometrical view.)
Solution: T is not a linear transformation since LT4 does not hold. This
is because T(0) = u 0 ≠ 0.
***
Now, try the following exercises.

E3) Let T : R 2 → R 2 be the reflection in the y -axis. Find the

expression for T as in Example 4. Is T a linear operator?

E4) For a fixed vector (a 1 , a 2 , a 3 ) in R 3 , define the mapping

T : R 3 → R by T( x1 , x 2 , x 3 ) = a 1x 1 + a 2 x 2 + a 3 x 3 . Show that T is a
linear transformation. Note that T( x1 , x 2 , x 3 ) is the dot product of
( x 1 , x 2 , x 3 ) and (a 1 , a 2 , a 3 ) (ref. Sec. 2.4).

E5) Show that the map T : R 3 → R 3 defined by

T(x 1 , x 2 , x 3 ) = ( x1 + x 2 − x 3 , 2 x1 − x 2 , x 2 + 2 x 3 ) is a linear operator.
You came across the real vector space Pn of all polynomials of
degree less than or equal to n , in Unit 4. The next exercise
concerns it.

E6) Let f ∈ Pn be given by f ( x ) = α 0 + αx1 + L + α n x n , α i ∈ R ∀i. We

define (Df ) ( x ) = α1 + 2α 2 x + L + nα n x n −1.

Show that D : Pn → Pn is a linear transformation. (Observe that Df

is nothing but the derivative of f . D is called the differentiation
operator.)

In Unit 3 we introduced you to the concept of a quotient space. We now

define a very useful linear transformation, using this concept.

Example 8: Let W be a subspace of a vector space U over a field F.

W gives rise to the quotient space U / W. Consider the map
T : U → U / W defined by T(u) = u + W. T is called the quotient map or
the natural map.
Show that T is a linear transformation.

Solution: For α, β ∈ F and u 1 , u 2 ∈ U we have

T (αu 1 + β u 2 ) = (αu 1 + β u 2 ) + W = ( αu 1 + W ) + (β u 2 + W )
= α ( u 1 + W ) + β( u 2 + W )
= αT ( u 1 ) + β T ( u 2 ).
Thus, T is a linear transformation.
***
Now solve the following exercise, which is about plane vectors.

11
 E7) Let u1 = (1, − 1), u 2 = (2, − 1), u 3 = (4, − 3), v1 = (1, 0), v 2 = (0, 1) and
v 3 = (1, 1) be 6 vectors in R 2 . Can you define a linear
transformation T : R 2 → R 2 such that T ( u i ) = v i , i = 1, 2, 3 ?

You have already seen that a linear transformation T : U → V must

satisfy T (α1 u 1 + α 2 u 2 ) = α1T ( u 1 ) + α 2 T ( u 2 ), for α1 , α 2 ∈ F and u 1 , u 2 ∈ U.
More generally, we can show that,
LT6 : T(α 1u 1 + L + α n u n ) = α 1 T(u 1 ) + L + α n T(u n ),
where α i ∈ F and u i ∈ U.
Let us show this by induction, that is, we assume the above relation for
n = m, and prove it for m + 1. Now,
T (α1u 1 + L + α m u m + α m +1u m +1 )
= T ( u + α m +1 u m +1 ) , where u = α 1u 1 + L + α m u m
= T ( u ) + α m +1T ( u m +1 ), since the result holds for n = 2
= T (α1 u 1 + L + α m u m ) + α m +1T ( u m +1 )
= α 1T ( u1 ) + L + α m T ( u m ) + α m +1T ( u m +1 ), since we have assumed the
result for n = m.

Thus, the result is true for n = m + 1. Hence, by induction, it holds true

for all n.

Let us now come to a very important property of any linear

transformation T : U → V. In Unit 4 we mentioned that every vector
space has a basis. Thus, U has a basis. We will now show that T is
completely determined by its basis of U. More precisely, we have show
that T is completely determined by its values on a basis of U. More
precisely, we have

Theorem 1: Let S and T be two linear transformations from U to V,

where dim F U = n. Let {e1 , K , e n } be a basis of U. Suppose S(e i ) = T (e i )
for i = 1, K, n. Then S( u ) = T( u ) for all u ∈ U.

Proof: Let u ∈ U. Since {e1 , K , e n } is a basis of U, u can be uniquely

written as u = α 1e1 + L + α n e n , where the α1 are scalars.
Then, S( u ) = S(α 1e1 + L + α n e n )
= α 1S(e1 ) + L + α n S(e n ), by LT6
= α 1 T ( e1 ) + L + α n T ( e )
= T ( α1e1 + L + α n e n ), by LT6
= T(u).

What we have just proved is that once we know the values of T on a

basis of U, then we can find T( u) for any u ∈ U.

Note: Theorem 1 is true even when U is not finite-dimensional. The

proof, in this case, is on the same lines as above.

12
Unit 7 Linear Transformation-I
Let us see how the idea of Theorem 1 helps us to prove the following
useful result.

Theorem 2: Let V be a real vector space and T : R → V be a linear

transformation. Then there exists v ∈ V such that T(α) = αv ∀ α ∈ R.
Proof: A basis for R is {1}. Let T(1) = v ∈ V. Then, for any
α ∈ R, T(α) = αT(1) = αv.

Once you have read Sec.7.3 you will realize that this theorem says that
T(R) is a vector space of dimension one, whose basis is {T(1)}.
Now try the following exercise, for which you will need Theorem 1.

E8) We define a linear operator T : R 2 → R 2 : T(1, 0) = (0,1) and

T(0, 5) = (1, 0). What is T(3, 5) ? What is T(5, 3) ?

Now we shall prove a very useful theorem about linear transformation,

which is linked to Theorem 1.

Theorem 3: Let {e1 , K , e n } be a basis of U and let v1 , K, v n be any n

vectors in V. Then there exists one hand only one linear transformation
T : U → V such that T (e i ) = v i , i = 1, K , n.

Proof: Let u ∈ U. Then u can be uniquely written as u = α 1e1 + L + α n e n

(see Unit 4, Theorem 9).

Define T ( u ) = α1 v1 + L + α n v n . Then T defines a mapping from U to V

such that T (e i ) = v i ∀ i = 1, K, n. Let us now show that T is linear. Let
a, b be scalars and u, u ′ ∈ U. There are scalars α1 , K , α n , β1 , K, β n
such that u = α 1e1 + L + α n e n and u ′ = β1e1 + L + β n e n . .
Then au + bu ′ = (aα1 + bβ)e1 + L + (aα n + bβ n )e n .
Hence,
T (au + bu ′) = (aα1 + bβ) v1 + L + (aα n + bβ n ) v n = a (α1 v1 + L + α n v n ) +
b(β1 v1 + L + β n v n ) = aT ( u ) + bT ( u ).
Therefore, T is a linear transformation with the property that
T (e i ) = v i ∀ i. Theorem 1 now implies that T is the only linear
transformation with the above properties.

Let’s see how Theorem 3 can be used.

Example 9: e1 = (1, 0, 0), e 2 = (0, 1, 0) and e 3 = (0, 0, 1) form the standard

basis of R 3 . Let (1, 2), (2, 3) and (3, 4) be three vectors in R 2 . Obtain
the linear transformation T : R 3 → R 2 such that
T (e1 ) = (1, 2), T ( e 2 ) = ( 2, 3) and T(e 3 ) = (3, 4).

13
 Solution: By Theorem 3 we know that ∃ T : R 3 → R 2 such that
T (e1 ) = (1, 2), T (e 2 ) = ( 2, 3) and T(e 3 ) = (3, 4). We want to know what
T( x ) is, for any x = ( x 1 , x 2 , x 3 ) ∈ R 3 . Now, x = x 1e1 + x 2 e 2 + x 3e 3 . ’
Hence, T( x ) = x 1T(e1 ) + x 2 T(e 2 ) + x 3T(e 3 )

= x 1 (1, 2) + x 2 (2, 3) + x 3 (3, 4)

= ( x 1 + 2 x 2 + 3x 3 , 2x 1 + 3x 2 + 4 x 3 )
Therefore, T( x1 , x 2 , x 3 ) = ( x1 + 2x 2 + 3x 3 , 2x 1 + 3x 2 + 4 x 3 ) is the definition
of the linear transformation T.
***

E9) Consider the complex field C. It is a vector space over R.

a) What is it’s dimension over R ? Give a basis of C over R.
b) Let α, β ∈ R. Give the linear transformation which maps the
basis elements of C, obtained in (a), onto α and β,
respectively.

Let us now look at some vector spaces that are related to a linear
operator.

7.3 SPACES ASSOCIATED WITH A LINEAR

TRANSFORMATION
In Unit 2, BMTC-131, you found that given any function, there is a set
associated with it, namely, its range. We will now consider two sets
which are associated with any linear transformation, T. These are the
range and the kernel of T.

7.3.1 The Range Space and the Kernel

Let U and V be vector spaces over a field F. Let T : U → V be a linear
transformation. We will define the range of T as well as the kernel of T.
At first, you will see them as sets. We will prove that these sets are also
vector spaces over F.

Definition 2: The range of T, denoted by R (T) is the set

{T(x ) | x ∈ U}.
The kernel (or null space) of T, denoted by Ker T, is the set
{x ∈ U | T( x) = 0}.
Note that R (T) ⊆ V and Ker T ⊆ U.

To clarify these concepts consider the following examples.

Example 10: Let I : V → V be the identity transformation (see Example

1). Find R (I) and Ker I.

14
Unit 7 Linear Transformation-I
Solution: R (I) = {I(v) | v ∈ V} = {v | v ∈ V} = V. Also,
Ker I = {v ∈ V | I( v) = 0} = {v ∈ V | v = 0} = {0}.
***

Remark 2: Note that, if we take T = A, an m× n matrix, the kernel of A

is {b ∈ R n | Ab = 0}. This is just the null space N(A).
Also, suppose we write A = [ v1 , v 2 , K, v n ] where each vi is a column
vector of length m. If
 x1 
x 
x =  2
M
 
x n 
is a column vector of length n1
Ax = x1v1 + x 2 v 2 + L + x n v n .
Therefore, R (A ) = {Ax | x ∈ R n } = {x 1v1 + x 2 v 2 + L x n v n | x 1 , x 2 , K, x n ∈ R} .
In other words R ( A) = CS(A).

Example 11: Let T : R 3 → R be defined by T(x 1 , x 2 , x 3 ) = 3x 1 + x 2 + 2x 3 .

Find R (T) and Ker T.

Solution: R (T) = {x ∈ R | ∃ x 1 , x 2 , x 3 ∈ R with 3x 1 + x 2 + 2 x 3 = x}.

For example, 0 ∈ R (T), since 0 = 3.0 + 0 + 2.0 = T(0, 0, 0)
Also, 1∈ R (T), since 1 = 3.1/ 3 + 0 + 2.0 = T(1/ 3, 0, 0), or
1 = 3.0 + 1 + 2.0 = T(0, 1, 0), or 1 = T(0, 0, 1/ 2), or 1 = T(1/ 6, 1/ 2, 0).
Now can you see that R (T) is the whole real line R ? This is because,
α 
for any α ∈ R, α = α.1 = αT (1 / 3, 0, 0) = T  , 0, 0  ∈ R (T ).
3 
3
Ker T = {( x 1 , x 2 , x 3 ) ∈ R | 3x 1 + x 2 + 2x 3 = 0}.
For example, (0, 0, 0) ∈ Ker T. But (1, 0, 0) ∉ Ker T. ∴ Ker T ≠ R 3 . In fact,
Ker T is the plane 3x 1 + x 2 + 2 x 3 = 0 in R 3 .
***

E10) Let F be any field and a, b, a′ and b′ ∈ F. Check that T : F2 → F2

defined by T( x, y) = (ax + by, a′x + b′y).

Example 12: Let T : R 3 → R 3 be defined by

T( x1 , x 2 , x 3 ) = (x 1 − x 2 + 2 x 3 , 2x 1 + x 2 , − x 1 − 2 x 2 + 2 x 3 ).
Then Ker (T) and R (T) are subspaces of R 3 . Give a geometric
description of R (T) and Ker (T) as a line, a hyperplanes or the whole of
R 3 , as the case may be.

Solution: If ( y1 , y 2 , y 3 ) ∈ R (T), there exists ( x1 , x 2 , x 3 ) ∈ R 3 such that

T(x 1 , x 2 , x 3 ) = ( x1 − x 2 + 2x 3 , 2 x 1 + x 2 , − x 1 − 2 x 2 + 2 x 3 ) = ( y1 , y 2 , y 3 ).
15
  y1   x 1 − x 2 + 2x 3  1  −1  2
         
So,  y 2  =  2 x1 + x 2  = x 1  2  + x 2  1  + x 3  0 .
 y   − x − 2 x + 2x   − 1  − 2  2
 3  1 2 3      

 1   − 1  2 
     
So, R (T) = [S] where S =  2 ,  1 ,  0  .
 − 1  2   2 
     
We now reduce the matrix
 1 2 − 1
 − 1 1 − 2 to get a basis for R (T).
 
 2 0 2 
4R 3 R
R 2 → R 1 + R 2 , R 3 → R 3 − 2R 1 , R 3 → R 3 + , R 2 → 2 gives
3 3
1 2 − 1
0 1 − 1.
 
0 0 0 
 1   0 
   
So,  2 ,  1  is a basis for R (T) and hence R (T) is a plane.
 − 1  − 1
   
Suppose, the equation of the plane is ax + by + cz = 0. Then, we have
1 0
a + 2b − c = 0    
. Since the vectors  2  and  1  satisfy the equation
b−c = 0  −1  −1
   
ax + by + cz = 0. Taking c = λ, we get b = λ, a = −λ.
Taking λ = 1, we see that the plane is − x + y + z = 0 or x = y + z.
Therefore R (T) is the plane x − y − z = 0.
If ( x1 , x 2 , x 3 ) ∈ Ker (T), we have T(x 1 , x 2 , x 3 ) = 0 or
x1 − x 2 + 2x 3 = 0
2x1 + x 2 = 0
− x1 − 2x 2 + 2x 3 = 0
Thus, Ker (T) is the null space of the matrix
 1 − 1 2
 2 1 0
 
 − 1 − 2
The RREF is
1 0 2 / 3 
 0 1 − 4 / 3
 
0 0 0 
The third column is the non-pivot column. Taking x 3 = λ, we get
x 1 = −2 / 3λ, x 2 = 4 / 3λ, x 3 = λ. So the solution set is
{λ(−2 / 3, 4 / 3, 1) | λ ∈ R}.
16
Unit 7 Linear Transformation-I
x1 x x
which is the straight line = 2 = 3.
− 2/3 4/3 1
***

In this example, we see that finding R (T) and Ker T amounts to solving
a system of equations. In Unit 4, we discussed the solution of system of
linear equations using row reduction.

The following exercises will help you in getting used to R (T) and Ker T.

E11) Let T be the zero transformation given in Example 2. Find Ker T

and R (T). Does 1∈ R (T ) ?

E12) For each of the following operators, give as geometric description

of R (T) and Ker (T) as lines, hyper planes or the whole of R, R 2
or R 3 as the case may be.
a) T : R 3 → R 2 : T( x , y, z ) = ( x, y)

b) T : R 3 → R : T( x, y, z) = z

c) T : R 3 → R 3 : T ( x1 , x 2 , x 3 ) = ( x1 − x 2 + x 3 , x1 + x 2 − x 3 , x1 − x 2 − x 3 ).

(Note that the operators in (a) and (b) are projections onto the xy -
plane and the z -axis, respectively.)

Now that you are familiar with the sets R (T) and Ker T , we will prove
that they are vector spaces.

Theorem 4: Let U and V be vector spaces over a field F. Let

T : U → V be a linear transformation. Then Ker T is a subspace of U
and R (T) is a subspace of V.

Proof: Let x 1 , x 2 ∈ Ker T ⊆ U and α1 , α 2 ∈ F. Now, by definition,

T ( x 1 ) = T ( x 2 ) = 0.
Therefore, α1T ( x1 ) + αT ( x 2 ) = 0
But α1T ( x 1 ) + α 2 T ( x 2 ) = T (α1 x 1 + α 2 x 2 ).
Hence, T (α 1 x 1 + α 2 x 2 ) = 0.
This means that α1 x 1 + α 2 x 2 ∈ Ker T.
Thus, by Theorem 4 of Unit 3, Ker T is a subspace of U.

Let y1 , y 2 ∈ R (T ) ⊆ V, and α1 , α 2 ∈ F. Then, by definition of R (T), there

exist x 1 , x 2 ∈ U such that T (x 1 ) = y1 and T ( x 2 ) = y 2 .
So, α1 y1 + α 2 y 2 = α 1T ( x 1 ) + α 2 T ( x 2 )
= T (α 1 x 1 + α 2 x 2 ).
Therefore, α1 y1 + α 2 y 2 ∈ R (T ), which proves that R (T) is a subspace of
V.

17
 Now that we have proved that R (T) and Ker T are vector spaces, you
know, from Unit 4, that they must have a dimension. We will study these
dimensions now.

7.3.2 Rank and Nullity

Consider any linear transformation T : U → V, assuming that dim U is
finite. Then Ker T, being a subspace of U, has finite dimension and
dim (Ker T) ≤ dim U. Also note that R (T) = T(U), the image of U under
T, a fact you will need to use in solving the following exercise.

E13) Let {e1 , K , e n } be a basis of U. Show that R (T) is generated by

{T (e1 ), K, T (e n )}.

From E12 it is clear that, if dim U = n, then dim R (T) ≤ n.

Thus, dim R (T ) is finite, and the following definition is meaningful.

Definition 3: The rank of T is defined to be the dimension of R (T), the

range space of T. The nullity of T is defined to be the dimension of
Ker T, the kernel (or the null space) of T.

Thus, rank (T) = dim R(T) and nullity (T) = dim Ker T.

We have already seen that rank (T) ≤ dim U and nullity (T) ≤ dim U.

Remark 3: In Remark 2, we saw that for an m× n matrix

A, N(A) = ker (A) and CS(A) = R (T ). So, the definition of rank and
nullity of a matrix coincides with the definition of rank and nullity of A,
considered as a linear transformation from R n to R m .

Example 13: Let T : U → V be the zero transformation given in

Example 2. What are the rank and nullity of T ?

Solution: In E10 you saw that R (T) = {0} and Ker T = U. Therefore,
rank (T) = 0 and nullity (T) = dim U.
Note that rank (T) + nullity (T) = dim U, in this case.

E14) If T is the identity operator on V, find rank (T) and nullity (T).

E15) Let D be the differentiation operator in E6. Give a basis for the
range space D and for Ker D . What are rank (D) and nullity (T).

In the above example and exercises you will find that for T : U → V,
rank (T) + nullity (T) = dim U. In fact, this is the most important result
about rank and nullity of a linear operator. We will now state and prove
this result.
18
Unit 7 Linear Transformation-I
Theorem 5: Let U and V be vector spaces over a field F and
dim U = n. Let T : U → V be a linear operator. Then
rank (T) + nullity (T) = n.

Proof: Let nullity (T) = m, that is, dim Ker T = m. Let {e1 , K , e m } be a
basis of Ker T . We know that Ker T is subspace of U. Thus, by
Theorem 11 of Unit 4, we can extend this basis to obtain a basis
{e1 , K, e m , e m +1 , K, e n } of U. We shall show that {T (e m +1 ), K , T (e n )} is a
basis of R (T). Then, our result will follow because dim R (T ) will be
n − m = n − nullity (T).
Let us first prove that {T (e m +1 ), K , T (e n )} spans, or generates, R (T). Let
y ∈ R (T). Then, by definition of R (T) there exists x ∈ U such that
T( x) = y.
Let x = c1e1 + L + c m e m + c m +1e m +1 + L + c n e n , c i ∈ F ∀ i.
Then,
y = T ( x ) = c1T (e1 ) + L + c m T (e m ) + c m +1T (e m +1 ) + L + c n T (e n )
= c m +1T (e m +1 ) + L + c n T (e n ),
because T (e1 ) = L = T (e m ) = 0, since T (e i ) ∈ Ker T ∀ i = 1, K , m. ∴ any
y ∈ R (T) is linear combination of {T (e m +1 ), K , T (e n )}. Hence, R (T) is
spanned by {T (e m +1 ), K , T (e n )}.
It remains to show that the set {T (e m +1 ), K , T (e n )} is linearly
independent. For this, suppose there exist a m+1 , K , a n ∈ F with
a m +1T (e m +1 ) + L + a n T (e n ) = 0.
Then, T (a m +1e m +1 + L + a n e n ) = 0.
Hence, a m +1e m +1 + L + a n e n ∈ Ker T, which is generated by {e1 , K, e m }.
Therefore, there exist a1 , K , a m ∈ F such that
a m +1e m +1 + L + a n e n = a 1e1 + L + a m e m .
 (−a 1 )e1 + L + (−a m )e m + a m +1e m +1 + L + a n e n = 0.
Since {e1 , K , e n } is a basis of U, it follows that this set is linearly
independent. Hence, − a 1 = 0, K, − a m = 0, a m +1 = 0, K, a n = 0 . In
particular, a m +1 = K = a n = 0, which we wanted to prove.
Therefore, dim R (T) = n − m = n − nullity (T), that is, rank (T) + nullity
(T) = n.

Remark 4: This proof is essentially the same as the one we gave for a
matrix in Unit 6. The only fact that we used in Unit 6 is that A is a linear
transformation.

Let us see how this theorem can be useful.

Example 14: Let L : R 3 → R be the map given by L( x, y, z) = x + y + z.

What is nullity (L) ?

Solution: In this case it is easier to obtain R (L), rather than Ker L.

Since L(1, 0, 0) = 1 ≠ 0, R (L) ≠ {0}, and hence dim R (L) ≠ 0. Also, R (L) is
a subspace of R. Thus, dim R (L) ≤ dim R = 1. Therefore, the only
19
 possibility for dim R (L) is dim R (L) = 1. By Theorem 5,
dim Ker (L) + dim R (L) = 3.

Hence, dim KerL + dim R (L) = 3.

E16) Find the kernel and range of projection and reflection operators we
discussed in Example 6.

E17) Give the rank and nullity of each of the linear transformation in
E11.

E18) Let U and V be real vector spaces and T : U → V be a linear

transformation, where dim U = 1. Show that R (T) is either a point
or a line.

Before ending this section we will prove a result that links the rank (or
nullity) of the composite of two linear operators with the rank (or nullity)
of each of them.

Theorem 6: Let V be a vector space over a field F. Set S to T be

linear operators from V to V. Then
a) rank (ST) ≤ min (rank (S), rank (T ))
b) nullity (ST) ≥ min (nullity (S), nullity (T )) .

Proof: We shall prove (a). Note that (ST ) ( v) = S(T( v)) for any v ∈ V
(you’ll study more about compositions in Unit 8).

Now, for any y ∈ R (ST), ∃ v ∈ V such that,

y = (ST) ( v) = S(T( v)) (1)
Now, (1)  y ∈ R (S).
Therefore, R (ST ) ⊆ R (S). This implies that rank (ST ) ≤ rank (S).
Again, (1)  y ∈ S(R (T)), since T(v) ∈ R (T).
∴ R (ST) ⊆ S(R (T)), so that dim R (ST) ≤ dinS(R (T)) ≤ dim R (T) (since
dim L( U) ≤ U, for any linear operator L).
Therefore, rank (ST) ≤ rank (T).
Thus, rank (ST) ≤ min (rank (S), rank (T )).

The proof of this theorem will be complete, once you solve the following
exercise.

E19) Prove (b) of Theorem 6 using the Rank Nullity Theorem.

We would now like to discuss some linear operators that have special
properties.

20
Unit 7 Linear Transformation-I

7.4 SOME TYPES OF LINEAR

TRANSFORMATIONS
Let us recall, from Unit 1, that there can be different types of functions,
some of which are one-one, onto or invertible. We can also define such
types of linear transformations as follows.

Definition 4: Let T : U → V be a linear transformation.

a) T is called one-one (or injective) if, for u1 , u 2 ∈ U with u1 ≠ u 2 ,

we have T (u 1 ) ≠ T (u 2 ). If T is injective, we also say T is 1 − 1.
Note that T is 1 − 1 if T(u 1 ) = T(u 2 )  u 1 = u 2 .
b) T is called onto (or surjective) if, for each v ∈ V, ∃ u ∈ U such
that T(u) = v, that is, R(T) = V.

Can you think of examples of such functions?

The identity operator is both one-one and onto. Why is this so? Well,
I : V → V is an operator such that, if v1 , v 2 ∈ V with v1 ≠ v 2 then
I(v1 ) ≠ I( v 2 ). Also, R (I) = V, so that I is onto.

E20) Show that the zero operator 0 : R → R is not one-one.

An important result that characterizes injectivity is the following:

Theorem 7: T : U → V is one-one if and only if Ker T = {0}.

Proof: First assume T is one-one. Let u ∈ Ker T. Then T(u) = 0 = T(0).

This means that u = 0. Thus, Ker (T) = {0}. Conversely, let Ker T = {0}.
Suppose u 1 , u 2 ∈ U with T(u1 ) = T(u 2 )
 T (u 1 − u 2 ) = 0  u 1 − u 2 ∈ Ker T  u 1 − u 2 = 0  u 1 = u 2 . Therefore, T is
1 − 1.

Suppose now that T is one-one and onto linear transformation from a

vector space U to a vector space V. Then, from Unit 2, Theorem 3,
BMTC-131, we know that T −1 exists.

But is T −1 linear? The answer to this equation is ‘yes’, as is shown in

the following theorem.

Theorem 8: Let U and V be vector spaces over a field F. Let

T : U → V be a one-one and onto linear transformation. Then
T −1 : V → U is a linear transformation.

In fact, T −1 is also 1 − 1 and onto.

Proof: Let y1 , y 2 ∈ V and α1 , α 2 ∈ F. Suppose T −1 ( y1 ) = x 1 and

T −1 ( y) = x ⇔ T( x ) = y
T −1 ( y 2 ) = x 2 . Then, by definition, y1 = T ( x1 ) and y 2 = T ( x 2 ).
21
 Now, α1 y1 + α 2 y 2 = α 1T ( x 1 ) + α 2 T (x 2 ) = T (α 1 x 1 + α 2 x 2 ).
Hence, T −1 (α1 y1 + α 2 y 2 ) = α1x 1 + α 2 x 2 = α1T −1 ( y1 ) + α 2 T −1 ( y 2 ).
This shows that T −1 is a linear transformation.

We will now show that T −1 is 1 − 1. For this, suppose y1 , y 2 ∈ V such

that T −1 ( y1 ) = T −1 ( y 2 ). Let x 1 = T −1 ( y1 ) and x 2 = T −1 ( y 2 ).

Then T (x 1 ) = y1 and T (x 2 ) = y 2 . We know that x 1 = x 2 . Therefore,

T ( x 1 ) = T ( x 2 ), that is, y1 = y 2 . Thus, we have shown that
T −1 ( y1 ) = T −1 ( y 2 )  y1 = y 2 , proving that T −1 is 1 − 1. T −1 is also
surjective because, for any u ∈ U, ∃ T(u) = v ∈ V such that T −1 ( v) = u.

Theorem 8 says that a one-one and onto linear transformation is

invertible, and the inverse is also a one-one and onto linear
transformation.

This theorem immediately leads us to the following definition.

Definition 5: Let U and V be vector space over a field F, and let

T : U → V be a one-one and onto linear transformation. Then T is
called an isomorphism between U and V.

In this case we say that U and V are isomorphic vector spaces. This
is denoted by U ~ V.

An obvious example of an isomorphism is the identity operator. Can you

think of any other? The following exercise may help.

E21) Let T : R 3 → R 3 : T( x, y, z) = (x + y, y, z ). Is T an isomorphism?

Why? Define T −1 , if it exists.

E22) Let T : R 3 → R 2 : T( x, y, z ) = (x + y, y + z). Is T an isomorphism?

In all these exercises and examples, have you noticed that if T is an

isomorphism between U and V then T −1 is an isomorphism between
V and U ?

Using these properties of an isomorphism we can get some useful

results, like the following:

Theorem 9: Let T : U → V be an isomorphism. Suppose {e1 , K , e 2 } is a

basis of U. Then {T (e1 ), K, T (e n )} is a basis of V.

Proof: First we show that the set {T (e1 , K, T (e n )} spans V. Since T is

onto, R (T) = V. Thus, from E12 you know that {T (e1 ), K, T (e n )} spans
V.
22
Unit 7 Linear Transformation-I
Let us now show that {T (e1 ), K, T (e n )} is linearly independent. Suppose
there exist scalars c1 , K , c n , such that
c1T (e1 ) + L + c n T (e n ) = 0. (1)

We must show that c1 = K = c n = 0.

Now, (1) implies that
T (c1e1 + L + c n e n ) = 0.
Since T is one-one and T(0) = 0, we conclude that
c1e1 + L + c n e n = 0.
But {e1 , K , e n } is linearly independent. Therefore,
c1 = K = c n = 0.
Thus, we have shown that {T (e1 ), K, T (e n )} is a basis of V.

Remark 5: The argument showing the linear independence of

{T (e1 ), K, T (e n )} in the above theorem can be used to prove that any
one-one linear transformation T : U → V maps any linearly
independent subset of U onto a linearly independent subset of V
(see E22).

We now give an important result equating ‘isomorphism’ with '1 − 1' and
with ‘onto’ in the finite-dimensional case.

Theorem 10: Let T : U → V be a linear transformation where U, V are

of the same finite dimension. Then the following statements are
equivalent.
a) T is 1 − 1.
b) T is onto.
c) T is an isomorphism.

Proof: To prove the result we will prove (a )  (b)  (c)  (a ). Let

dim U = dim V = n.

Now (a) implies that Ker T = {0} (from Theorem 7). Hence, nullity
(T) = 0. Therefore, by Theorem 5, rank (T) = n, that is,
dim R (T) = n = dim V. But R (T) is a subspace of V. Thus, by the remark
following Theorem 12 of Unit 4, we get R (T) = V, i.e., T is onto, i.e., (b)
is true. So (a )  (b).

Similarly, if (b) holds then rank (T) = n, and hence, nullity (T) = 0.
Consequently, Ker T = {0}, and T is one-one. Hence, T is one-one and
onto, i.e., T is an isomorphism. Therefore, (b) implies (c).

That (a) follows from (c) is immediate from the definition of an

isomorphism.

Hence, our result is proved.

23
 Caution: Theorem 10 is true for finite-dimensional spaces U and V,
of the same dimension. It is not true, otherwise. Consider the following
counter-example.

Example 15: (To show that the spaces have to be finite-dimensional):

Let V be the real vector space of all polynomials. Let D : V → V be
defined by D(a 0 + a 1 x + L + a r x r ) = a 1 + 2a 2 x + L + ra r x r −1 . Then show that
D is onto but not 1 − 1.

Solution: Note that V has infinite dimension, a basis being

{1, x, x 2 , K}. D is onto because any element of V is of the form
 a a 
a 0 + a 1 x + L + a n x n = D a 0 x + 1 x 2 + L + n x n +1 D is not 1 − 1 because,
 2 n +1 
for example, 1 ≠ 0 but D(1) = D(0) = 0.
***

E23) Define a linear operator T : R 3 → R 2 such that T is onto but T is

not 1 − 1. Note that dim R 3 ≠ dim R 2 .

Let us use Theorem 9 and 10 to prove our next result.

Theorem 11: Let T : V → V be a linear transformation and let

{e1 , K, e n )} be a basis of V. Then T is one-one and onto if and only if
{T (e1 ), K, T (e n )} is linearly independent.

Proof: Suppose T is one-one and onto. Then T is an isomorphism.

Hence, by Theorem 9, {T (e1 ), K, T (e n )} is basis. Therefore,
{T (e1 ), K, T (e n )} is linearly independent.

Conversely, suppose {T (e1 ), K, T (e n )} is linearly independent. Since

{e1 , K, e n } is a basis of V, dim V = n. Therefore, any linearly
independent subset of n vectors is a basis of V (by Unit 4, Theorem 5,
Cor. 1). Hence, {T (e1 ), K, T (e n )} is basis of V. Then, any element v of
n
 n 
V is of the form v =  c i T(e i ) = T  c i e i , where c1 , K, c n are scalars.
i=1  i=1 
Thus, T is onto, and we can use Theorem 10 to say that T is an
isomorphism.

Here are some exercises now.

E24) a) Let T : U → V be a one-one linear transformation and let

{u 1 , K, u k } be a linearly independent subset of U. Show that
the set {T (u1 ), K, T(u k )} is linearly independent.

b) Is it true the every linear transformation maps every linear

independent set of vectors into a linearly independent set?
24
Unit 7 Linear Transformation-I
c) Show that every linear transformation maps a linearly
dependent set of vectors onto a linear dependent set of
vectors.

E25) Let T : R 3 → R 3 be defined by

T( x1 , x 2 , x 3 ) = ( x 1 + x 3 , x 2 + x 3 , x1 + x 2 ). Is T invertible? If yes, find
a rule for T −1 like the one which defines T.

We have seen, in Theorem 9, that if T : U → V is an isomorphism, then

T maps a basis of U onto a basis of V. Therefore, dim U = dim V. In
other words, if U and V are isomorphic then dim U = dim V. The natural
question arises whether the converse is also true. That is, if
dim U = dim V, both being finite, can we say that U and V are
isomorphic? The following theorem shows that this is indeed the case.
Theorem 12: Let U and V be finite-dimensional vector spaces over F.
Then U and V are isomorphic if and only if dim U = dim V.

Proof: We have already seen that if U and V are isomorphic then

dim U = dim V. Conversely, suppose dim U = dim V = n. We shall show
that U and V are isomorphic. Let {e1 , K, e n } be a basis of U and
{f1 , K, f n } be a basis of V. By Theorem 3, there exists a linear
transformation T : U → V such that T (e1 ) = f i , i = 1, K, n.

We shall show that T is 1 − 1.

Let u = c1e1 + L + c n e n be such that T(u) = 0.
Then 0 = T (u ) = c1T (e1 ) + L + c n T (e n )
= c1f1 + L + c n f n .
Since {f1 , K, f n } is a basis of V, we conclude that c1 = c 2 = K = c n = 0.
Hence, u = 0. Thus, Ker T = {0} and by Theorem 7, T is one-one.

Therefore, by Theorem 10, T is an isomorphism, and U ~ V.

An immediate consequence of this theorem follows.

Corollary: Let V be a real (or complex) vector space of dimension n.

Then V is isomorphic to R n (or C n ), respectively.

Proof: Since dim R n = n = dim R V, we get V ~ R n . Similarly, if

dim C V = n, then V ~ C n .

We generalize this corollary in the following remark.

Remark 6: Let V be a vector space over F and let B = {e1 , K, e n } be a

n
basis of V. Each v ∈ V can be uniquely expressed as v =  α i e i .
i =1

Recall that α1 , K, α n are called the coordinates of v with respect to B

(refer to Unit 5, page 42). 25
 Define θ : V → Fn : θ( v) = (α1 , K , α n ). Then θ is an isomorphism from V
to Fn . This is because θ is 1 − 1, since the coordinates of v with
respect to B are uniquely determined. Thus, V ~ Fn .

We end this section with an exercise.

E26) Let T : U → V be a one-one linear mapping. Show that T is onto if

and only if dim U = dim V. (Of course, you must assume that U
and V are finite-dimensional spaces.)

Now let us look at isomorphisms between quotient spaces.

7.5 HOMOMORPHISM THEOREMS
Linear transformations are also called vector space homomorphism.
There is a basic theorem which uses the properties of homomorphism
to show the isomorphism of certain quotient spaces (ref. Unit 3). It is
simple to prove, but is very important because it is always being used to
prove more advanced theorems on vector spaces. (In the Abstract
Algebra course we proved this theorem in the setting of groups and
rings).

This theorem is called Theorem 13: Let V and W be vector spaces over a field F and
the Fundamental T : V → W be a linear transformation. Then V/Ker T ~ R(T).
Theorem of
Homomorphism. Proof: You know that Ker T is a subspace of V, so that V / Ker T is a
well defined vector space over F. Also R (T) = {T( v) | v ∈ V}. To prove
the theorem let us define θ : V / Ker T → R (T) by θ(v + Ker T) = T(v).

Firstly, we must show that θ is well defined function, that is, if

v + Ker T = v′ + Ker T then θ(v + Ker T) = θ( v′ + Ker T), i.e., T( v) = T( v′).

Now, v + Ker T = v′ + Ker T  ( v − v′) ∈ Ker T (see E18, Unit 3).

 T(v − v′) = 0  T( v) = T( v′), and hence, θ is well defined.
Next, we check that θ is a linear transformation. For this, let a, b ∈ F
and v, v′ ∈ V. Then θ{a ( v + Ker T) + b( v′ + Ker T)}.
= θ(av + bv′ + Ker T) (ref. Unit 3)
= T(av + bv′)
= aT(v) + bT( v′), since T is linear
= aθ( v + Ker T) + bθ( v′ + Ker T).
Thus, θ is a linear transformation.

We end the proof by showing that θ is an isomorphism, θ is 1 − 1

because θ( v + Ker T) = 0  T( v) = 0  v ∈ Ker T  v + Ker T = 0. In
V / Ker T.
Thus, Ker θ = {0}.
θ is onto (because any element of R (T) is T( v) = θ( v + Ker T)).
26
Unit 7 Linear Transformation-I

So we have proved that θ is an isomorphism. This proves that

V / Ker T ~ R (T ).

Let us consider an immediate useful application of Theorem 13.

Example 16: Let V be a finite-dimensional space and let S and T be

linear transformations from V to V. Show that
rank (ST) = rank (T ) − dim (R (T) ∩ Ker S).

T S
Solution: We have V  → V→ V. ST is the composition of the
operators S and T, which you have studied in Unit 2, BMTC-131 and
will also study in Unit 8. Now, we apply Theorem 13 to the
homomorphism θ : (V) → ST(V) : θ(T( v)) = (ST) ( v).
Now, Ker θ = {x ∈ T(V) | S( x) = 0} = Ker S ∩ T(V) = Ker S ∩ R (T).
Also R (θ) = ST(V), since any element of ST(V) is (ST)(v) = θ(T( v)).
T ( V)
Thus ~ ST (V).
Ker S ∩ T(V )
Therefore,
T ( V)
dim = dim ST (V)
Ker S ∩ T (V)
That is, dim T(V) − dim(Ker S ∩ T(V)) = dimST(V), which is what we had
to show.
***

E27) Using Example 14 and the Rank Nullity Theorem, show that
nullity(ST) = nullity(T) + dim(R (T) ∩ Ker S).

Now let us see another application of Theorem 13.

3
Example 17: Show that R ~ R 2 where W ~ R.
W

Solution: Now, we define a function f : R 3 → R 2 : f (α, β, γ ). Then f is a

linear transformation and Ker f = {(α, 0, 0) | α ∈ R} ~ R. If
W = {(α, 0, 0) | α ∈ R} (α, 0, 0) a α gives an isomorphism between W
and R. Also f is onto, since any element (α, β) of R 2 is f (0, α, β).
Thus, by Theorem 13, R 3 / R ~ R 2 .
***

n
Note: In general, for any n ≥ m, R ~ R n −m where W ~ R m . Similarly,
W
n
for n ≥ m C n −m ~ C where W ′ ~ C m .
W′
The next result is a corollary to the Fundamental Theorem of
Homomorphism.
But, before studying it, read Unit 3 for the definition of the sum of
spaces. 27
 Corollary 1: Let A and B be subspaces of a vector space V. Then
A + B / B ~ A / A ∩ B.

A+B
Proof: We define a linear function T : A → by T(a ) = a + B.
B
A+B
T is well defined because a + B is an element of (since
B
a = a 1 + 0 ∈ A + B ).
T is a linear transformation because, for α1 , α 2 in F and a 1 , a 2 in A,
we have
T (α1a 1 + α 2 a 2 ) = α1a 1 + α 2 a 2 + B = α1 (a 1 + B) + α 2 (a 2 + B)
= α1T (a 1 ) + α 2 T (a 2 ).
A+B
Now we will show that T is surjective. Any element of is of the
B
form a + b + B, where a ∈ A and b ∈ B.
Now a + b + B = a + B + b + B = a + B + B, since b ∈ B
A+B
= a + B , since B is zero element of
B
= T(a ), proving that T is surjective.
A+B
∴ R (T ) =
B
We will now prove that Ker T = A ∩ B.
If a ∈ Ker T, then a ∈ A and T(a) = 0. This means that a + B = B, the zero
A+B
element of . Hence, a ∈ B (by Unit 3, E23). Therefore, a ∈ A ∩ B.
B
On the other hand, a ∈ A ∩ B  a ∈ A and a ∈ B  a ∈ A and
a + B = B  a ∈ A and T(a ) = T(0) = 0.
 a ∈ Ker T.
This proves that A ∩ B = Ker T.

Now using Theorem 13, we get

A / Ker T ~ R (T ).
That is, A /(A ∩ B) ~ (A + B) / B.

E28) Using the corollary above, show that A ⊕ B / B ~ A (⊕ denotes the

direct sum defined in Sec. 3.4.3).

There is yet another interesting corollary to the Fundamental Theorem

of Homomorphism.

Corollary 2: Let W be a subspace of a vector space V. Then, for any

subspace U and V containing W,
V/ W ~
V / U.
U/W
28
Unit 7 Linear Transformation-I
Proof: This time we shall prove the theorem with you. To start with let
us define a function T : V / W → V / U : T( v + W) = v + U. Now try E27.

E29) a) Check that T is well defined.

b) Prove that T is a linear transformation.

c) What are the spaces Ker T and R (T) ?

So, is the theorem proved? Yes; apply Theorem 13 to T.

We end the unit by summarising what we have done in it.

7.6 SUMMARY
In this unit we have covered the following points.

1) A linear transformation from a vector spaces U over F to a vector

space V over F is a function T : U → V such that,
LT1) T (u1 + u 2 ) = T (u 1 ) + T (u 2 ) ∀ u1 , u 2 ∈ U, and
LT2) T(αu) = αT(u ), for α ∈ F and u ∈ U.
These conditions are equivalent to the single condition
LT3) T (αu 1 + β u 2 ) = αT (u 1 ) + β T (u 2 ) for α, β ∈ F and u 1 , u 2 ∈ U.

2) Given a linear transformation T : U → V,

i) the kernel of T is the vector space {u ∈ U | T(u ) = 0}, denoted
by Ker T.

ii) the range of T is the vector space {T(u) | u ∈ U}, denoted by

R (T).

iii) The rank of T = dim F R (T ).

iv) The nullity of T = dim F Ker T.

3) Let U and V be finite-dimensional vector spaces over F and

T : U → V be a linear transformation. Then
rank (T) + nullity(T) = dim U.

4) Let T : U → V be a linear transformation. Then

i) T is one-one if T (u 1 ) = T (u 2 )  u1 = u 2 ∀ u1 , u 2 ∈ U.
ii) T is onto if, for any v ∈ V ∃ u ∈ U such that T(u) = v.
iii) T is an isomorphism (or, is invertible) if it is one-one and
onto, and then U and V are called isomorphic spaces. This
is denoted by U ~ V.

5) T : U → V is
29
 i) one-one if and only if Ker T = {0}
ii) onto if and only if R (T) = V.

6) Let U and V be finite-dimensional vector spaces with the same

dimension. Then T : U → V is 1 − 1 iff T is onto iff T is an
isomorphism.

7) Two finite-dimensional vector spaces U and V are isomorphic if

and only if dim U = dim V.

8) Let V and W be vector spaces over a field F, and T : V → W be

a linear transformation. Then V / Ker T ~ R (T ).

7.7 SOLUTIONS/ANSWERS
E1) For any α1 , α 2 ∈ F and u1 , u 2 ∈ U, we know that α1 u r ∈ U and
α 2 u 2 ∈ U. Therefore, by LT1,
T (α 1 u 1 + α 2 u 2 ) = T ( α 1 u 1 ) + T (α 2 u 2 )
= α1T (u 1 ) + α 2 T (u 2 ) , by LT2.

E2) By LT2, T(0.u) = 0.T(u) for any u ∈ U. Thus, T(0) = 0. Similarly, for
any u ∈ U, T(−u) = T((−1)u) = (−1)T(u) = −T(u).

E3) T( x, y) = (− x, y) ∀ ( x , y) ∈ R 2 . (See the geometric view in Fig. 5) T

Fig. 5: Q is the reflection is a linear operator. This can be proved the same way as we did in
of P in the y-axis. Example 4.

E4) T((x1 , x 2 , x 3 ) + ( y1 , y 2 , y 3 )) = T( x1 + y1 , x 2 + y 2 , x 3 + y 3 )
= a 1 (x 1 + y1 ) + a 2 ( x 2 + y 2 ) + a 3 ( x 3 + y 3 )
= (a 1x 1 + a 2 x 2 + a 3 x 3 ) + (a 1 y1 + a 2 y 2 + a 3 y 3 )
= T(x 1 , x 2 , x 3 ) + T( y1 , y 2 , y 3 )
Also, for any α ∈ R,
T(α(x 1 , x 2 , x 3 )) = a 1αx 1 + a 2 αx 2 + a 3 αx 3
= α(a 1 x1 + a 2 x 2 + a 3 x 3 ) = αT( x 1 , x 2 , x 3 ).
Thus, LT1 and LT2 hold for T.

E5) We will check if LT1 and LT2 hold. Firstly,

T((x 1 , x 2 , x 3 ) + ( y1 , y 2 , y 3 )) = T( x 1 + y1 , x 2 + y 2 , x 3 + y 3 )
= ( x 1 + y1 + x 2 + y 2 − x 3 − y 3 , 2 x1 + 2 y1 − x 2 − y 2 , x 2 + y 2 + 2x 3 + 2 y 3 )
= ( x 1 + x 2 − x 3 , 2 x1 − x 2 , x 2 + 2 x 3 ) + ( y1 + y 2 − y 3 , 2 y1 − y 2 , y 2 + 2 y 3 )
= T( x 1 , x 2 , x 3 ) + T( y1 , y 2 , y 3 ), showing that LT1 holds.
Also, for any α ∈ R,
T(α( x 1 , x 2 , x 3 )) = T(αx1 , αx 2 , αx 3 )
= (αx1 + αx 2 − αx 3 , 2αx 1 − αx 2 , αx 2 + 2αx 3 )

30
Unit 7 Linear Transformation-I
= α( x 1 + x 2 − x 3 , 2 x 1 − x 2 , x 2 + 2 x 3 ) = αT( x 1 , x 2 , x 3 ), showing that
LT2 holds.

E6) We want to show that D(αf + βg) = αD(f ) + βD(g), for any α, β ∈ R
and f , g ∈ Pn .
Now, let f ( x ) = a 0 + a 1 x + a 2 x 2 + L + a n x n and
g( x ) = b 0 + b1 x + L + b n x n .
Then (αf + β g )( x ) = (αa 0 + β b 0 ) + (αa 1 + β b1 ) x + L + (αa n + β b n ) x n .
∴[D(αf + βg )](x ) = (αa 1 + βb1 ) + 2(αa 2 + βb 2 ) x + L + n (αa + βb) x n −1
= α(a 1 + 2a 2 x + L + na n x n −1 ) + β(b1 + 2b 2 x + L + nb n x n −1 )
= α(Df )(x ) + β(Dg)(x) = (αDf + βDg)( x)
Thus, D(αf + βg) = αDf + βDg, showing that D is a linear map.

E7) No. Because, if T exists, then

T (2u1 + u 2 ) = 2T (u 1 ) + T (u 2 ).
But 2u1 + u 2 = u 3 . ∴ T(2u 1 + u 2 ) = T(u 3 ) = v 3 = (1, 1).
On the other hand, 2T(u1 ) + T(u 2 ) = 2 v1 + v 2 = (2, 0) + (0, 1)
= (2, 1) ≠ v 3 .
Therefore, LT3 is violated. Therefore, no such T exists.

E8) Note that {(1, 0), (0, 5)} is a basis for R 2 .

Now (3, 5) = 3(1, 0) + (0, 5).
Therefore, T(3, 5) = 3T(1, 0) + T(0, 5) = 3(0, 1) + (1, 0) = (1, 3).
Similarly, (5, 3) = 5(1, 0) + 3 / 5(0, 5).
Therefore, T(5, 3) = 5(0, 1) + 3 / 5(1, 0) = (3 / 5, 5).
Note that T(5, 3) ≠ T(3, 5).

E9) a) dim R C = 2, a basis being (1, i), i = − 1.

b) Let T : C → R be such that T(1) = α, T(i) = β.

Then, for any element x + iy ∈ C( x, y ∈ R), we have
T( x + iy) = xT(1) + yT(i) = xα + yβ. Thus, T is defined by T is
defined by T(x + iy) = xα + yβ ∀ x + iy ∈ C.

E10) Let u = (u1 , u 2 ), v = ( v1 , v 2 ) ∈ F2 . Then, we have

T(αu + βv) = T(αu1 + βv1 , αu 2 + βv 2 )
= (a (αu1 + βv1 ) + b(αu 2 + βv 2 ), a ′(αu1 + βv1 ) + b′(αu 2 + βv 2 ))
= (α(au1 + bu 2 ) + β(av1 + bv2 ), α(a′u1 + b′u 2 ) + β(a′v1 + b′v 2 ))
= α(au1 + bu 2 , a′u1 + b′u 2 ) + β(av1 + bv2 , a ′v1 + b′v 2 )
= αT((u1 , u 2 )) + βT((v1 , v 2 )) + αT(u) + βT( v).
Therefore, T is a linear transformation.

E11) T : U → V : T(u ) = 0 ∀ u ∈ U.
∴, Ker T = {u ∈ U | T(u) = 0} = U
R (T) = {T(u) | u ∈ U} = {0}. ∴1∉ R (T).
31
 x  1 0 1  0 
E12) a) The image is   = x   + y  . Therefore R (T) =  ,   .
 y 0 1 0 1 
1 0
The vectors   and   are linearly independent. So,
0 1
R (T) ⊆ R 2 has dimension 2. So, R (T) = R 2 .
( x, y, z) ∈ Ker (T) if (x, y) = 0, i.e., x = 0, y = 0. ∴ the kernel is
Ker (T) = {2(0, 0, 1) | z ∈ R} = {2(1, 0, 0) | z ∈ R}. This is the line
z
x = 0, y = 0, = t the z -axis.
1

b) The image z = z.1. R (T) = {z | z ∈ R 3 } which is the whole of R.

Once again the kernel.

c) We have
x1 − x 2 + x 3  1 − 1 1
 x + x − x  = x 1 + x  1  + x  − 1
 1 2 3 1  2  3 
 x 1 − x 2 − x 3  1 − 1  − 1
 1  − 1  1  
 
∴ R (T) =  1 ,  1  ,  − 1 .
 1  − 1  − 1 
 
1 1 1
Row reducing  − 1 1 − 1 by R 2 → R 2 + R 1 , R 2 → 2
R
2
 1 − 1 − 1
1 1 1 
 − 1
R 3 → R 3 + 2R 2 , R 3 →  R 3 we get B = 0 1 0 .
 2   
0 0 1
So, the basis of R (T) has three elements and therefore
R (T ) = R 3 .
To find the null space we continue and find the RREF of
1 1 1
 − 1 1 − 1. Reducing B using row operators
 
 1 − 1 − 1
1 0 0 
R 1 → R 1 − R 2 , R 1 → R 1 − R 3 , we get 0 1 0. There are no
0 0 1
non-pivot columns, so, the system has the unique solution
(0, 0, 0). Therefore R (T) = {(0, 0, 0)}.

E13) Any element of R (T) is of the form T(u ), u ∈ U. Since {e1 , K, e n }

generates U, ∃ scalars α1 , K, α n such that u = α1e1 + L + α n e n .
Then T (u ) = α1T(e1 ) + L + α n T(e n ), that is, T( u) is in the linear span
of {T(e1 ), K, T(e n )}.
∴{T (e1 ), K, T (e n )} generates R (T).
32
Unit 7 Linear Transformation-I

E14) T : V → V : T( v) = v. Since R (T) = V and Ker T = {0}, we see that

rank(T) = dim V, nullity(T) = 0.

E15) R (D) = {a 1 + 2a 2 x + L + na n x n −1 | a 1 , K, a n ∈ R}
Thus, R (D) ⊆ Pn−1 . But any element b 0 + b1 x + L + b n −1 x n −1 , in Pn −1 is
 b b 
D b 0 x + 1 x 2 + L + n −1 x n  ∈ R (D).
 2 n 
Therefore, R (D) = Pn−1 .
∴, a basis for R (D) is {1, x , K, x n−1 }, and rank(D) = n.
Ker = {a 0 + a 1 x + L + a n x n | a 1 + 2a 2 x + L + na n x n −1 = 0, a i ∈ R ∀ i}

= {a 0 + a 1 x + L + a n x n | a 1 = 0, a 2 = 0, K, a n = 0, a i ∈ R ∀ i}

= {a 0 | a 0 ∈ R} = R.
∴, a basis for Ker D is {1}.
 nullity (D) = 1.

E16) We have
 1   m 
 x + ym mx + ym 2   2   2 
P( x , y) =  2 ,  = x  m + 1  + y m + 1  .
m 2 + 1 
2
 m +1  m   m 
 
 m2 + 1  m2 + 1
 1   m 
 2   2 
So, the range of P is [S] where S =  m + 1 ,  m + 1 .
m   m  2
 2  
 m + 1   m 2 + 1 
 1   m 
1   1  1
+
 m    m2 + 1
1
Note that 2   =  m 1  and 2  = .
m + 1  m   m  m + 1  m   m 
2
 2   2 
 m +1  m + 1
1
So, the vector in S are scalar multiples of   so, they are
m
 1    1 
 . Therefore the range is   . In other word, the range is
 m    m 
the line y = mx as we would expect from our definition of the
projection operator.
The kernel is given by
 x + ym mx + ym 2  x + ym  x + ym 
 2 , 2
 = (0, 0) or 2 = 0, m 2  = 0.
 m +1 m +1  m +1  m +1 
−1
Since m ≠ 0, both the equations reduce to x + ym = 0 or y = x.
m
−1
So, the kernel is the line y = x.
m
Let us now consider the reflection operator, we have

33
  x (1 − m 2 ) + 2m 2mx + y(m 2 − 1) 
R ( x , y) =  , 
 m2 + 1 m2 + 1 
 1 − m   2m 
2
 2   2 
= x m + 1  + y m 2 + 1 .
 2m   m − 1 
 2   
 m + 1  m2 + 1
  1 − m 2   2m 
  2   2  

Therefore, the range of R is S, where S =   m + 1 ,  m 2 + 1 .
  2 m  
 
m − 1  

  m 2 + 1   m 2 + 1 
In general, if S = {v1 , v 2 , K, v k } span a subspace of a vector
space V, then {α1v1 , α 2 v 2 , K, α k v k } also spans the subspace.
This is because, if v = a1v1 + a 2 v 2 + L + a k v k , then
a  a  a 
v =  1 α1v1 +  2 α 2 v 2 + L +  k α k v k .
 α1   α2   αk 
1 − m   2m  
2
Therefore, the set S1 =  ,  2   also spans the range of
 2m  m − 1 
 a   b  
R. Setting a = 1 − m 2 , b = 2m, the set is of the form  ,   
 b   − a  
where b ≠ 0 .
To show that there vectors are linearly independent, we need to
show that there is no α ∈ R, α ≠ 0 such that
a   b 
α   =  . (Why?)
b  − a 
If such an α exists, we have αv = b, αb = −a.
α 2 b = −αa = −b or (α 2 + 1)b = 0. So, α = ± i and α ∉ R.
a  b
∴   and   are linearly independent. Therefore, the range of
b  − a 
R is the whole of R 2 . So, by rank-nullity theorem, the kernel is 0.

E17) a) We have shown that R (T) = R 2 . ∴ rank(T) = 2.

Therefore, nullity (T) = dim R 3 − 2 = 1.

b) rank (T) = 1, nullity (T) = 2.

c) R (T) is generated by {(1, 1, 1)}. ∴ rank (T) = 1.

∴ nullity (T) = 2.

E18) Now rank (T) + nullity (T) = dim U = 1.

Also rank (T) ≥ 0, nullity (T) ≥ 0.
∴, the only values rank (T) can take are 0 and 1 . If rank (T) = 0,
then dim R (T) = 0.
Thus, R (T) = {0}, that is, R (T) is a point.

34
Unit 7 Linear Transformation-I
If rank (T) = 1, then dim R (T) = 1. That is, R (T) is a vector space
over R generated by a single element, v, say. Then R (T) is the
line R v = {αv | α ∈ R}.

E19) By Theorem 5, nullity (ST) = dim V − rank (ST). By (a) of Theorem

6, we know that − rank (ST ) ≥ −rank (S) and − rank (ST) ≥ −rank (T).
∴, nullity (ST) ≥ dim V − rank (S) and nullity (ST) ≥ dim V − rank (T).
Thus, nullity (ST) ≥ nullity (S) and nullity (ST) ≥ nullity (T). That is,
nullity (ST) ≥ max {nullity (S), nullity (T)}.

E20) Since 1 ≠ 2, but 0(1) = 0(2) = 0, we find that 0 is not 1 − 1.

E21) Firstly note that T is a linear transformation. Secondly, T is 1 − 1

because T( x,− y, y, z) = T( x ′, y′, z′)  (x, y, z) = ( x ′, y′, z′).
Thirdly, T is onto because any ( x , y, z) ∈ R 3 can be written as
T(x − y, y, z).
∴, T is an isomorphism. ∴ T −1 : R 3 → R 3 exists and is defined by
T −1 ( x , y, z ) = ( x − y, y, z ).
E22) T is not an isomorphism because T is not 1 − 1, since
(1, − 1, 1) ∈ Ker T.

E23) The linear operator in E11) (a) suffices.

E24) a) Let α1 , K, α k ∈ F such that α1T (u 1 ) + L + α k T (u k ) = 0.

 T ( α 1 u 1 + L + α k u k ) = 0 = T ( 0)
 α1u 1 + L + α k u k = 0, since T is 1 − 1.
 α1 = 0, K, α k = 0, since {u 1 , K, u k } is linearly independent.
∴{T (u 1 ), K, T (u k )} is linearly independent.

b) No. For example, the zero operator maps every linearly

independent set to {0}, which is not linearly independent.

c) Let T : U → V be a linear operator, and {u 1 , K, u n } be a

linearly dependent set of vectors in U. We have to show that
{T(u 1 ), K , T (u n )} is linearly dependent. Since {u 1 , K, u n } is
linearly dependent, ∃ scalars a 1 , K, a n , not all zero, such
that a 1 u 1 + L + a n u n = 0.
Then a 1T (u 1 ) + L + a n T (u n ) = T (0) = 0, so that
{T (u1 ), K, T(u n )} is linearly dependent.

E25) T is a linear transformation. Now, if (x, y, z) ∈ Ker T, then

T( x, y, z) = (0, 0, 0).
∴, x + y = 0 = y + z = x + z  x = 0 = y = z.
 Ker T = {(0, 0, 0)}
 T is 1 − 1.
∴, by Theorem 10, T is invertible.
To define T −1 : R 3 → R 3 suppose T −1 ( x, y, z ) = (a , b, c). 35
 Then T(a, b, c) = ( x, y, z)
 (a + b, b + c, a + c) = ( x, y, z)
 a + b = x, b + c = y, a + c = z.
x+z−y x+y−z y+z−x
a= ,b= ,c=
2 2 2
 x+z− y x+ y−z y+z−x 
∴ T −1 ( x, y, z) =  , , 3
 for any (x , y, z ) ∈ R .
 2 2 2 

E26) T : U → V is 1 − 1. Suppose T is onto. Then T is an

isomorphism and dimU = dimV, by Theorem 12. Conversely,
suppose dim U = dim V. Then T is onto by Theorem 10.

E27) Then Rank Nullity Theorem and Example 14 give

dim V − nullity(ST) = dim V − nullity(T) − dim(R (T) ∩ Ker S)
 nullity(ST) = nullity(T) + dim (R (T) ∩ Ker S)

E28) In the case of the direct sum A ⊕ B, we have A ∩ B = {0}.

A⊕B ~
∴ A.
B
E29) a) v + W = v′ + W  v − v′ ∈ W ⊆ U  v − v′ ∈ U  v + U = v′ + U
 T ( v + W ) = T ( v′ + W )
∴ T is well defined.

b) For any v + W, v′ + W in V / W and scalars a, b, we have

T(a ( v + W) + b(v + W)) = T(av + bv′ + W) = av + bv′ + U
= a (v + U) + b( v′ + U) = aT( v + W) + bT( v′ + W).
∴, T is a linear operator.

c) Ker T = {v + W | v + U = U}, since U is the “zero” for V / U

= {v + W | v ∈ U} = U / W.
R (T) = {v + U | v ∈ V} = V / U.

36
 UNIT 8
LINEAR TRANSFORMATIONS II

Structure Page No.

8.1 Introduction
Objectives
8.2 The Vector Space L( U, V )
8.3 The Dual Space
8.4 Composition of Linear Transformations
8.5 Minimal Polynomial
8.6 Summary
8.7 Solutions/Answers

8.1 INTRODUCTION
In the last unit we introduced you to linear transformations and their
properties. We will now show that the set of all linear transformations
from a vector space U to a vector space V forms a vector space itself,
and its dimension is (dim U) (dim V). In particular, we define and
discuss the dual space of a vector space.

In unit 1 we defined the composition of two functions. Over here we will

discuss the composition of two linear transformations and show that it is
again a linear operator. Note that we use the terms ‘linear
transformation’ and ‘linear operator’ interchangeably.

Finally, we study polynomials with coefficient from a field F, in a linear

operator T : V → V. You will see that every such T satisfies a
polynomial equation g( x ) = 0. That is, if we substitute T for x in g( x )
we get the zero transformation. We will, then, define the minimal
polynomial of an operator and discuss some of its properties. These
ideas will crop up again in Unit 11.

You must revise Units 2, BMTC-131 and Unit 7 of this course before
going further.

Objectives

After reading this unit, you should be able to 37

 • prove and use the fact that L(U, V) is a vector space of dimension
(dim U ) (dim V ) ;
• use dual bases, whenever convenient;
• obtain the composition of two linear operators, whenever possible;
• obtain the minimal polynomial of a linear transformation T : V → V in
some simple cases;
• obtain the inverse of an isomorphism T : V → V if its minimal
polynomials is known.

8.2 THE VECTOR SPACE L(U, V)

By now you must be quite familiar with linear operators, as well as
vector spaces. In this section we consider the set of linear operators
from one vector space to another, and show that it forms a vector
space.

Let U, V be vector spaces over a field F. Consider the set of all linear
transformations from U to V. We denote this set by L(U, V).

We will now define addition and scalar multiplication in L(U, V) so that

L( U, V) becomes a vector space.

Suppose S, T ∈ L(U, V ) (that is, S and T are linear operators from U to

V). we define (S + T) : U → V by
(S + T) (u ) = S(u ) + T(u ) ∀ u ∈ U.

Now, for a1 , a 2 ∈ F and u1 , u 2 ∈ U,

(S + T) (a1u 1 + a 2 u 2 )
= S(a1u 1 + a 2 u 2 ) + T(a1u 1 + a 2 + u 2 )
= a 1S(u1 ) + a 2S(u 2 ) + a 1T(u1 ) + a 2 T(u 2 )
= a1 (S(u1 ) + T(u1 )) + a 2 (S(u 2 ) + T(u 2 ))
= a 1 (S + T) (u 1 ) + a 2 (S + T) (u 2 )
Hence, S + T ∈ L( U, V).

Next, suppose S ∈ L( U, V) and α ∈ F. We define αS : U → V as follows:

(αS) (u ) = αS(u ) ∀ u ∈ U.

Is αS a linear operator? To answer this take β1 , β2 ∈ F and u1 , u 2 ∈ U.

Then,
(αS) (β1u1 + β2 u 2 ) = αS (β1u1 + β2 u 2 ) = α[β1S(u1 ) + β2S(u 2 )]
= β1 (αS) (u 1 ) + β 2 (αS) (u 2 ).
Hence, αS ∈ L(U, V).

So we have successfully defined addition and scalar multiplication on

L( U, V).
38
E1) Show that the set L( U, V) is a vector space over F with respect to
the operations of addition and multiplication by scalars defined
above. (Hint: The Zero vector in this space is the zero
transformation. All the conditions VS1 – VS10 (of Unit 3) have to
be verified.)

Notation: For any vector space V we denote L(V, V) by A(V).

Let U and V be vector spaces over F of dimensions m and n,

respectively. We have already observed that L( U, V) is a vector space
over F. Therefore, it must have a dimension. We now show that the
dimension of L( U, V) is mn.

Theorem 1: Let U, V be a vector spaces over a field F of dimensions

m and n, respectively. Then L(U, V) is a vector space of dimension
mn.

Proof: Let {e1 , K, e m } be a basis of U and {f1 , K, f n } be a basis of V.

By Theorem 3 of Unit 5, there exists a unique linear transformation
E11 ∈ L( U, V) such that
E11 (e1 ) = f1 , E11 (e2 ) = 0, K, E11 (e m ) = 0.

Similarly, E12 ∈ L(U, V) such that

E12 (e1 ) = 0, E12 (e2 ) = f1 , E12 (e3 ) = 0, K, E12 (e m ) = 0.

In general, there exist Eij ∈ L ( U, V) for i = 1, K, n, j = 1, K, m, such that

Eij (e j ) = f i and Eij (e k ) = 0 for j ≠ k. Eij is the linear operator that maps the
jth element in the basis of U to the i th element in the basis of V. We
pause here so that you can understand Eij better.

To get used to these Eij try the following exercise before continuing the
proof.

E2) Clearly define E2 m , E32 and Emn .

Now, let us go on with the proof of Theorem 1.

If u = c1e1 + L + c m e m , where ci ∈ F ∀ i, then Eij ( u ) = c jf i .

We complete the proof by showing that {Eij | i = 1, K , n , j = 1, K , m} is a

basis of L( U, V ).

Let us first show that this set is linearly independent over F. For this,
suppose 39
 n m

 c E
i =1 j=1
ij ij =0 (1)

where cij ∈ F. We must show that c ij = 0 for all i, j.

(1) implies that
n m

 c E
i =1 j=1
ij ij (e k ) = 0 ∀ k = 1, K, m.

Thus, by definition of the Eij ' s, we get

n

c
i =1
ik if = 0,

But {f1 , K, f n } is a basis for V. Thus, c ik = 0, for all i = 1, K, n.

But this is true for all k = 1, K, m.

Hence, we conclude that c ij = 0 ∀ i, j. Therefore, the set of Eij ' s is

linearly independent.

Next, we show that the set {Eij | i = 1, K , n , j = 1, K , m} spans L( U, V ).

Suppose T ∈ L( U, V ).

Now, for each j such that 1 ≤ j ≤ m, T (e j ) ∈ V. Since {f1 , K, f n } is a basis

of V, there exist scalars c i1 , K, c in such that
n
T (e j ) =  c ij f i (2)
i =1

We shall prove that

n m
T =  cijEij (3)
i =1 j=1

By Theorem 1 of Unit 5 it is enough to show that, for each k with

1 ≤ k ≤ m, T (e k ) =  cijEij (e k ).
i j

Now,
n m n

 c E (e
i =1 j=1
ij ij k ) =  cik f i = T(e k ), by (2). This implies (3).
i =1

Thus, we have proved that the set of mn elements

{Eij | i = 1, K , n , j = 1, K , m} is a basis for L( U, V ).

Let us see some ways of using this theorem.

Example 1: Show that L(R 2 , R) is a plane.

Solution: L(R 2 , R) is a real vector space for dimension 2 × 1 = 2.

Thus, by Theorem 12 of Unit 5 L(R 2 , R) ~ R 2 , the real plane.

40
***
Example 2: Let U, V be the vector spaces of dimension m and n,
respectively. Suppose W is a subspace of V of dimension p(≤ n ). Let
X = {T ∈ L( U, V ) : T(u ) ∈ W for all u ∈ U}.
If X a subspace of L(U, V) ? If yes, find its dimension.

Solution: If T1 and T2 are in X, we need to show that α1T1 + α 2T2 is in

X. For this we need to show that (α1T1 + α 2 T2 ) (u ) ∈ W. We have
T1 (u ) ∈ W, T2 (u ) ∈ W since T1 and T2 are in X. Therefore
(α1T1 + α 2 T2 ) (u ) = α1T1 (u ) + α 2 T2 (u ) ∈ W since W is a subspace and
T1 (u ), T2 (u ) ∈ W. By Theorem 1, dim X = mp.
***

E3) What can be a basis for L(R 2 , R), and for L(R, R 2 ) ? Notice that
both these spaces have the same dimension over R.

After having looked at L(U, V), we now discuss this vector space for
the particular case when V = F.

8.3 THE DUAL SPACE

The vector space L(U, V), discussed in Sec. 8.2, has a particular name
when V = F.

Definition 1: Let U be a vector space over F. Then the space L( U, V)

Recall that F is also a
is called the dual space of U, and is denoted by U ∗ . vector space over F.

In this section we shall study some basic properties of U ∗ .

The element of U ∗ have a specific name, which we now give.

Definition 2: A linear transformation T : U → F is called a linear

functional.

Thus, a linear functional on U is a function T : U → F such that

T(α1u1 + α 2 u 2 ) = α1T(u1 ) + α 2 T(u 2 ), for α1 , α 2 ∈ F and u1 , u 2 ∈ U.

For example, the map f : R 3 → R : f ( x1 , x 2 , x 3 ) = a1 x1 + a 2 x 2 + a 3 x 3 , where

a1 , a 2 , a 3 ∈ R are fixed, is a linear functional on R 3 . You have already
seen this is Unit 7 (E4).

E4) Prove that any linear functional on R 3 is of the form given in the
example above.

We now come to very important aspect of the dual space.

41
 We know that the space V∗ , of linear functionals on V, is a vector
space. Also, if dim V = m, then dim V ∗ = m, by Theorem 1. (Remember,
dim F = 1.)

Hence, we see that dim V = dim V ∗ . From Theorem 12 of Unit 5, it

follows that the vector spaces V and V ∗ are isomorphic.

We now construct a special basis of V ∗. Let {e1 , K, e n } be a basis of V.

By Theorem 3 of Unit 5, for each i = 1, K , m, there exists a unique linear
functional f i on V such that
1, if i = j
f i (e j ) = 
0, if i ≠ j
= δij

We will prove that the linear functional f1 , K , f m , constructed above,

form a basis of V ∗. Since dim V = dim V ∗ = m, it is enough to show that
the set {f1 , K, f m } is linearly independent. For this we suppose
c1 , K c m ∈ F such that c1f1 + L + c m f m = 0.

We must show that ci = 0 for all i.

n
Now c f
j=1
j j =0

 n 
   c jf j  (e i ) = 0, for each i.
 j=1 
n
  c j (f j (e i )) = 0 ∀ i
j=1
n
  c jδ ji = 0 ∀ i  c i = 0 ∀ i.
j=1

Thus, the set {f1 , K, f m } is a set of m linearly independent elements of

a vector space V ∗ of dimension m. Thus, from Unit 4 (Theorem 5, Cor.
1), it forms a basis of V ∗.

Definition 3: The basis {f1 , K, f m } of V ∗ is called the dual basis of the

basis {e1 , K , em } of V.

We now come to the result that shows the convenience of using of a

dual basis.

Theorem2: Let V be a vector space over F of dimension n, {e1 , K, e n }

be a basis of V and {f1 , K, f n } be the dual basis of {e1 , K , e n }. Then,
for each f ∈ V ∗ ,
n
f =  f ( e i )f i
i =1
and, for each v ∈ V,
42
 n
v =  f i ( v) e i .
i=1

Proof: Since {f1 , K, f n } is a basis of V∗ , for f ∈ V ∗ there exist scalars

n
c1 , K , c n such that f =  c i f i .
i=1
Therefore,
n
f (e j ) =  c i f i (e j )
i =1
n
=  c i δij , by definition of dual basis
i=1

= c j.

n
This implies that ci = f (e i ) ∀ i = 1, K , n. Therefore, f =  f (e i )f i .
i=1

Similarly, for v ∈ V, there exist scalars a1 , K, a n such that

n
v =  a iei .
i=1
n n
Hence, f j ( v) =  a i f j (e i ) =  a i δ ji = a j
i=1 i=1
and we obtain
n
v =  f i ( v) e i .
i=1

Let us see an example of how this theorem works.

Example 3: Consider the basis e1 = (1, 0, − 1), e 2 = (1, 1, 1), e3 = (1, 1, 0) of

C 3 over C. Find the dual basis of {e1 , e 2 , e3 }.

Solution: Any element of C 3 is v = (z1 , z 2 , z 3 ), z1 ∈ C. Since {e1 , e 2 , e3}

is a basis, we have α1 , α 2 , α 3 ∈ C. Since that
v = (z1 , z 2 , z 3 ) = α1e1 + α 2e 2 + α 3e3
= (α1 + α 2 + α 3 , α 2 + α3 − α1 + α 2 )
Thus, α1 + α 2 + α3 = z1
α 2 + α3 = z 2
− α1 + α 2 = z1
These equations can be solved to get
α1 = z1 − z 2 , α 2 = z1 − z 2 + z 3 , α3 = 2z 2 − z1 − z 3 .
Now, by Theorem 2, v = f1 ( v)e1 + f 2 ( v)e 2 + f 3 ( v)e3 , where {f1 , f 2 , f 3 } is the
dual basis. Also v = α1e1 + α 2 e 2 + α3e3 .
Hence, f1 ( v) = α1 , f 2 ( v) = α 2 , f 3 ( v) = α 3 ∀ v ∈ C3 .
Thus, the dual basis of {e1 , e 2 , e3} is {f1 , f 2 , f 3}, where f1 , f 2 , f 3 will be
defined as follows:
f1 (z1 , z 2 , z 3 ) = α1 = z1 − z 2
43
 f 2 (z1 , z 2 , z 3 ) = α 2 = z1 − z 2 + z 3
f 3 (z1 , z 2 , z 3 ) = α 3 = 2z 2 − z1 − z 3 .
***

E5) What is the dual basis for the basis {1, x, x 2 } of the space
P2 = {a 0 + a 1x + a 2 x 2 | a i ∈ R} ?

Now let us look at the dual of the dual space. If you like, you may skip
this portion and go straight to Sec. 8.4.

Let V be an n -dimensional vector space. We have already seen that

V and V ∗ are isomorphic because dim V = dim V ∗ . The dual of V ∗ is
called the second dual of V and is denoted by V ∗∗ . We will show that
V ~ V ∗∗ .

Now, any element of V ∗∗ is a linear transformation from V ∗ to F. Also,

for any v ∈ V and f ∈ V ∗ , f (v) ∈ F. So we define a mapping
φ : V → V∗∗ : v → φv, where (φv) (f ) = f ( v) for all f ∈ V ∗ and v ∈ V. (Over
here we will use φ( v) and φv interchangeably.)

Note that, for any v ∈ V, φv is well defined mapping from V∗ → F. We

have to check that it is a linear mapping.

Now, for c1 , c 2 ∈ F and f1 , f 2 ∈ V ∗ ,

(φv) (c1f1 + c 2f 2 ) = (c1f1 + c 2 f 2 ) ( v)
= c1f1 ( v) + c 2f 2 ( v)
= c1 (φv) (f1 ) + c 2 (φv) (f 2 )
∴ φv ∈ L(V ∗ , F) = V ∗∗ , ∀ v.

Furthermore, the map φ : V → V ∗∗ is linear. This can be seen as follows:

for c1 , c 2 ∈ F and v1 , v 2 ∈ V.
φ(c1v1 + c 2 v 2 ) (f ) = f (c1v1 + c 2 v 2 )
= c1f ( v1 ) + c 2f ( v 2 )
= c1 (φv1 ) (f ) + c 2 (φv 2 ) (f )
= (c1φv1 + c 2φv 2 ) (f ).

This is true ∀ f ∈ V ∗ . Thus, φ(c1v1 + c 2 v 2 ) = c1φ(v1 ) + c 2φ(v 2 ).

Now that we have shown that φ is linear, we want to show that it is
actually an isomorphism. We will show that φ is 1 − 1. For this, by
Theorem 7 of Unit 5, it suffices to show that φ(v) = 0 implies v = 0. Let
{f1 , K, f n } be the dual basis of a basis {e1 , K, e n } of V.
n
By Theorem 2, we have v =  f i ( v) ei .
i =1

44 Now φ(v) = 0  (φv) (f i ) = 0 ∀ i = 1, K , n

  f i (v) = 0 ∀ i = 1, K, n
 v = Σf i (v) ei = 0
Hence, it follows that φ is 1 − 1. Thus, φ is an isomorphism (Unit 5,
Theorem 10).

What we have just proved is the following theorem.

Theorem 3: The map φ : V → V ∗∗ , defined by (φv) (f ) = f ( v) ∀ v ∈ V and

f ∈ V ∗ , is an isomorphism.

We now give an important corollary to this theorem.

Corollary: Let ψ be a linear functional on V ∗ (i.e., ψ ∈ V ∗∗ ).

Then there exists a unique v ∈ V such that
ψ (f ) = f ( v) for all f ∈ V∗ .

Proof: By Theorem 3, since φ is an isomorphism, it is onto and 1 − 1.

Thus, there exists a unique v ∈ V such that φ( v) = ψ. This, by definition,
implies that
ψ (f ) = (φv) (f ) = f ( v) for all f ∈ V∗ .

Using the second dual try to prove the following exercise.

E6) Show that each basis of V ∗ is the dual of some basis of V.

In the following section we look at the composition of linear operators,

and the vector space A(V), where V is a vector space over F.

8.4 COMPOSITION OF LINEAR

TRANSFORMATIONS
Do you remember the definition of the composition of functions, which
you studied in Unit 2, BMTC-131? Let us now consider the particular
case of the composition of two linear transformations. Suppose
T : U → V and S : V → W are two linear transformations. The
composition of S and T is a function S o T : U → W, defined by
S o T(u ) = S(T(u )) ∀ u ∈ U.
This is diagrammatically represented in Fig. 1. Fig. 1: S o T is the
composition of S and
The first question which comes to our mind is whether S o T is linear. T.
The affirmative answer is given by the following result.

Theorem 4: Let U, V, W be vector spaces over F. Suppose

S ∈ L(V, W ) and T ∈ L( U, V ). Then S o T ∈ L( U, W ).

Proof: All we need to prove is the linearity of the map S o T. Let

α1 , α 2 ∈ F and u1 , u 2 ∈ U. Then 45
 S o T(α1u1 + α 2 u 2 ) = S(T(α1u1 + α 2 u 2 ))
= S(α1T(u1 )) + α 2 T(u 2 )), since T is linear
= α1S(T(u1 )) + α 2S(T(u 2 )), since S is linear
= α1S o T(u 1 ) + α 2S o T(u 2 )
This shows that S o T ∈ L( U, W ).

Try the following exercises now.

E7) Let I be the identity operator on V. Show that S o I = I o S = S for all

S ∈ A (V).

E8) Prove that S o 0 = 0 o S = 0 for all S ∈ A (V), where 0 is the null

operator.

We now make an observation.

Remark 1: Let S : V → V be an invertible linear transformation (ref. Sec.

5.4), that is, an isomorphism. Then, by Unit 7, Theorem 8,
S−1 ∈ L(V, V ) = A(V ).
Since S−1 o S( v) = v and S o S−1 ( v) = v for all v ∈ v.
S o S−1 = S−1 o S = I v , where I v denotes the identity transformation on V.

This remark leads us to the following interesting result.

Theorem 5: Let V be a vector space over a field F. A linear

transformation S ∈ A(V) is an isomorphism if and only if ∃ T ∈ A (V)
such that S o T = I = T o S.

Proof: Let us first assume that S is an isomorphism. Then, the remark

above tells us that ∃ S−1 ∈ A(V ) such that S o S−1 = I = S−1 o S. Thus, we
have T(= S−1 ) such that S o T = T o S = I.

Conversely, suppose T exists in A(V), such that S o T = I = T o S. We

want to show that S is 1 − 1 and onto.

We first show that S is 1 − 1, that is, Ker S = {0}. Now,

x ∈ Ker S  S( x ) = 0  T o S( x ) = T(0) = 0  I( x ) = 0  x = 0. Thus,
Ker S = {0}.

Next, we show that S is onto, that is, for any v ∈ V, ∃ u ∈ V such that
S(u ) = v. Now, for any v ∈ V,
v = I( v) = S o T(v) = S(T( v)) = S(u ), where u = T( v) ∈ V. Thus, S is onto.
Hence, S is 1 − 1 and onto, that is, S is an isomorphism.

Use Theorem 5 to solve the following exercise.

46
E9) Let S(x1 , x 2 ) = ( x 2 , − x1 ) and T( x1 , x 2 ) = (− x 2 , x1 ). Find S o T and
T o S. Is S (or T ) invertible?

Now, let us look at some examples involving the composite of linear

operators.

Example 4: Let T : R 2 → R 3 and S : R 3 → R 2 be defined by

T( x1 , x 2 ) = (x 1 , x 2 , x1 + x 2 ) and S( x1 , x 2 , x 3 ) = ( x1 , x 2 ). Find S o T and
T o S.

Solution: Find, note that T ∈ L(R 2 , R 3 ) and S ∈ L(R 3 , R 2 ). ∴ S o T and

T o S are both well defined linear operators. Now,
S o T( x1 , x 2 ) = S(T( x1 , x 2 )) = S(x1 , x 2 , x1 + x 2 ) = (x1 , x 2 )

Hence, S o T = the identity transformation of R 2 = I R2 .

Now,
T o S( x1 , x 2 , x 3 ) = T(S( x1 , x 2 , x 3 )) = T( x1 , x 2 ) = ( x1 , x 2 , x1 + x 2 ).

In this case S o T ∈ A(R 2 ), while T o S ∈ A(R 3 ). Clearly, S o T ≠ T o S.

Also, note that S o T = I, but T o S ≠ I.
***

Remark 2: Even if S o T and T o S both being to A(V), S o T may not be

equal to T o S. We give such an example below.

Example 5: Let S, T ∈ A (R 2 ) be defined by T( x1 , x 2 ) = ( x1 + x 2 , x1 − x 2 )

and S( x1 , x 2 ) = (0, x 2 ). Show that S o T ≠ T o S.

Solution: You can check that S o T( x1 , x 2 ) = (0, x1 − x 2 ) and

T o S( x1 , x 2 ) = ( x 2 , − x 2 ). Thus, ∃ ( x1 , x 2 ) ∈ R 2 such that
S o T( x1 , x 2 ) ≠ T o S( x1 , x 2 ) (for instance, S o T(1, 1) ≠ T o S(1, 1)). That is,
S o T ≠ T o S.
***

Note: Before checking whether S o T is well defined linear operator. You

must be sure that both S and T are well defined linear operators.

Now try to solve the following exercises.

E10) Let T( x1 , x 2 ) = (0, x1 , x 2 ) and S(x1 , x 2 , x 3 ) = ( x1 + x 2 , x 2 + x 3 ). Find

S o T and T o S .

E11) Let T(x1 , x 2 ) = (2x1 , x1 + 2x 2 ) for (x1 , x 2 ) ∈ R 2 , and

S( x1 , x 2 , x 3 ) = ( x1 + 3x 2 , 3x1 − x 2 , x 3 ) for ( x1 , x 2 , x 3 ) ∈ R 3 . Are S o T
and T o S defined? If yes, find them.
47
 E12) Let U, V, W, Z be vector spaces over F. Suppose
T ∈ L( U, V ), S ∈ L(V, W ) and R ∈ L(W, Z). Show that
(R o S) o T = R o (S o T).

E13) Let S, T ∈ A(V ) and S be invertible. Show that

rank (ST ) = rank (TS) = rank (T). (ST means S o T).

So far we have discussed the composition of linear transformations. We

have seen that if S, T ∈ A (V), then S o T ∈ A (V), where V is a vector
space of dimension n. Thus, we have introduced another binary
operation (see Sec.2.5, BMTC-131) in A(V), namely, the composition
of operators, denoted by o . Remember, we already have the binary
operations given in Sec. 8.2. In the following theorem we state some
simple properties that involve all these operations.

Theorem 6: Let R , S, T ∈ A (V) and let α ∈ F. Then

a) R o (S + T) = R o S + R o T, and (S + T) o R = S o R + T o R.
b) α(S o T) = αS o T = S o αT.

Proof: a) For any v ∈ V,

R o (S + T) (v) = R ((S + T) ( v)) = R (S( v) + T( v))
= R (S( v) + R (T( v))
= (R o S) ( v) + (R o T) ( v)
= ( R o S + R o T ) ( v)
Hence, R o (S + T) = R o S + R o T.
Similarly, we can prove that (S + T) o R = S o R + T o R.

b) For any v ∈ V, α(S o T) ( v) = α(S(T( v))

= (αS) (T( v))
= (αS o T ) ( v )
Therefore, α(S o T) = αS o T.
Similarly, we can show that α(S o T) = S o αT.

Notation: In future we shall be writing ST in place of S o T. Thus,

ST (u ) = S(T(u )) = (S o T)u. Also, if T ∈ A(V), we write
T 0 = I, T1 = T, T 2 = T o T and in general, T n = T n −1 o T = T o T n −1. The
properties of A(V) state in Theorems 1 and 6 are very important and
will be used implicitly again and again. To get used to A(V) and the
operations in it try the following exercises.

E14) Consider S, T : R 2 → R 2 defined by S( x, y) = ( x, − y) and

T( x, y) = ( x + y, y − x ). What are S + T, ST, TS, S o (S − T) and
(S − T) o S ?

E15) Let S ∈ A (V), dim V = n and rank (S) = r. Let

M = {T ∈ A(V) | ST = 0},
48
 N = {T ∈ A (V) | TS = 0},
a) Show that M and N are subspaces of A(V).
b) Show that M = L(V, Ker (S)). What is dim M ?

By now you must have got used to handling the elements of A(V). The
next section deals with polynomials that are related to these elements.

8.5 MINIMAL POLYNOMIAL

Recall that a polynomial in one variable x over F is of the form
p( x ) = a 0 + a 1x + L + a n x n , where a 0 , a 1 , K , a n ∈ F.

If a n ≠ 0, then p( x ) is said to be of degree n. If a n = 1, then p( x ) is

called a monic polynomial of degree n. For example, x 2 + 5x + 6 is a
monic polynomial of degree 2. The set of all polynomials in x with
coefficients in F is denoted by F[x ].

Definition 4: For a polynomial p, as above, and an operator T ∈ A(V),

we define p(T ) = a 0 I + a 1T + L + a n T n .
Since I, T, K T n ∈ A(V), we find p(T) ∈ A( V). We say p(T) ∈ F[T].
If q is another polynomial in x over F, then p(T)q(T) = q(T)p(T), that is,
p(T) and q(T) commute with each other. This can be seen as follows:
Let q(T ) = a 0 I + a 1T + L + a n T m .
Then p(T )q(T ) = (a 0 I + a 1T + L + a n T n ) (b 0 I + b1T + L + b m T m )
= a 0 b 0 I + (a 0 b1 + a 1b 0 )T + L + a n b m T n +m
= (b 0 I + b1T + L + b m T m ) (a 0 I + a 1T + L + a n T n )
= q(T) p(T).

E16) Let p, q ∈ F[ x ] such that p(T) = 0, q (T) = 0. Show that

(p + q) (T) = 0, ((p + q ) ( x ) means p( x ) + q ( x ).)

E17) Check that (2I + 3S + S3 ) commutes with (S + 2S4 ), for S ∈ A(R n ).

We now go on to prove that given any T ∈ A (V) we can find a

polynomial g ∈ F[ x ] such that
g(T) = 0, that is, g(T) = 0 ∀ v ∈ V.

Theorem 7: Let V be a vector space over F of dimension n and

T ∈ A(V ). Then there exists a non-zero polynomial g over F such that
g(T) = 0 and the degree of g is at most n 2 .

Proof: We have already seen that A(V) is a vector space of dimension

2
n 2 . Hence, the set {I, T, T 2 , K , T n } of n 2 + 1 vectors of A(V), must be
49
 linearly dependent (ref. Unit 7, Theorem 7). Therefore, there must exist
2
a 0 , a 1 , K, a n2 ∈ F (not all zero) such that a 0 I + a1T + L + a n 2 T n = 0.

Let g be the polynomial given by

2
g ( x ) = a 0 + a 1x + L + a n 2 x n .
Then g is a polynomial of degree at most n 2 , such that g(T) = 0.

The following exercises will help you in getting used to polynomial in x

and T.

E18) Given an example of polynomials g( x ) and h (x ) in R[ x ], for which

g(I) = 0 and h (0) = 0, where I and 0 are the identity and zero
transformations in A(R 3 ).

E19) Let T ∈ A(V ). Then we have a map φ from F[ x ] to A(V) given by

φ(p) = p(T). Show that, for a , b ∈ F and p, q ∈ F[ x ],
a) φ(ap + bq ) = aφ(p) + bφ(q),
b) φ(pq) = φ(p) φ(q).

In Theorem 7 we have proved that there exists some g ∈ F[ x ] with

g(T) = 0. But, if g(T) = 0, then (αg ) (T) = 0, for any α ∈ F. Also, if
deg g ≤ n 2 , then deg (αg ) ≤ n 2 . Thus, there are infinitely many
polynomials that satisfy the conditions in Theorem 7. But if we insist on
some more conditions on the polynomial g, then we end up with one
and only one polynomial which will satisfy these conditions and the
conditions in Theorem 7. Let us see what the conditions are.

Theorem 8: Let T ∈ A(V ). Then there exists a unique monic

polynomial p of smallest degree such that p(T) = 0.

Proof: Consider the set S = {g ∈ F[ x ] | g(T) = 0}. This set is non-empty

since, by Theorem 7, there exists a non-zero polynomial g, of degree at
most n 2 such that g(T) = 0. Now consider the set D = {deg f | f ∈ S}.
Then D is subset of N ∪ {0}, and therefore, it must have a minimum
element, say m. Let h ∈ S such that deg h = m. Then, h (T) = 0 and
deg h ≤ deg g ∀ g ∈ S.

If h = a 0 + a 1x + L + a m x m , a m ≠ 0, then p = a −m1 h is a monic polynomial

such that p(T) = 0. Also deg p = deg h ≤ deg g ∀ g ∈ S. Thus, we have
shown that there exists a monic polynomial p, of least degree, such
that p(T) = 0.

We now show that p is unique, that is, if q is any monic polynomial of

smallest degree such that q(T) = 0, then p = q. But this is easy. Firstly,
50
since deg p ≤ deg g ∀ g ∈ S, deg p ≤ deg q. Similarly, deg q ≤ deg p.
∴ deg p = deg q.
Now suppose p( x ) = a 0 + a1 x + L + a n −1 x n −1 + x n and
q( x ) = b 0 + b1 x + L + b n −1x n −1 + x n .
Since p(T) = 0 and q(T) = 0, we get (p − q) (T) = 0. But
p − q = (a 0 − b 0 ) + L + (a n −1 − b n −1 ) x n −1 . Hence, (p − q ) is a polynomial of
degree strictly less than the degree of p, such that (p − q) (T) = 0. That
is, p − q ∈ S with deg (p − q) < deg p. This is a contradiction to the way we
chose p, unless p − q = 0, that is, p = q. ∴ p is the unique polynomial
satisfying the conditions of Theorem 8.

This theorem immediately leads us to the following definition.

Definition 5: For T ∈ A(V), the unique monic polynomial p of smallest

degree such that p(T) = 0 is called the minimal polynomial of T.

Note that the minimal polynomial p, of T, is uniquely determined by the

following three properties.
1) p is a monic polynomial over F.
2) p(T) = 0.
3) If g ∈ F[ x ] with g(T) = 0, then deg p ≤ deg g.

Consider the following example and exercises.

Example 6: For any vector space V, find the minimal polynomials for I,
the identity transformation, and 0, the zero transformation.

Solution: Let p( x ) = x − 1 and q( x ) = x. Then p and q are monic such

that p(I) = 0 and q(0) = 0. Clearly no non-zero polynomials of smaller
degree have the above properties. Thus, x − 1 and x are the required
polynomials.
***

E20) Define T : R 3 → R 3 : T ( x1 , x 2 , x 3 ) = (0, x1 , x 2 ). Show that the

minimal polynomial of T is x 3 .

E21) Define T : R n → R n : T( x1 , K, x n ) = (0, x1 , K, x n −1 ). What is the

minimal polynomial of T ? (Does E20 help you?)

We will now state and prove a criterion by which we can obtain the
minimal polynomial of a linear operator T, once we know any
polynomial f ∈ F[ x ] with f (T) = 0. It says that the minimal polynomial
must be a factor of any such f .

Theorem 9: Let T ∈ A(V ) and let p( x ) be the minimal polynomial of T.

Let f (x ) be any polynomial such that f (T) = 0. Then there exists a
polynomial g( x ) such that f ( x ) = p(x ) g( x ).
51
 Proof: The division algorithm states that given f (x ) and p( x ), there
exist polynomials g( x ) and h (x ) such that f ( x ) = p( x ) g ( x ) + h ( x ), where
h ( x ) = 0 or deg h (x ) < deg p( x ). Now, 0 = f (T) = p(T) g (T) + h (T) = h (T),
since p(T) = 0.
Therefore, if h ( x ) ≠ 0, then h (T) = 0, and deg h (x ) < deg p( x ).
This contradicts the fact that p( x ) is the minimal polynomial of T.
Hence, h ( x ) = 0, and we get f ( x ) = p( x ) g( x ).

Example 7: Check that the projection operator defined in Unit 7

satisfies the polynomial x 2 − x = 0. Show that this is the minimal
polynomial of P.

Solution: Recall that, the projection operator is given by

 x + ym mx + ym 2 
P((x , y)) =  2 , .
 m +1 m 2 + 1 

  x + ym mx + ym 2  
∴ P 2 ((x , y)) = P  2 , 2
  = P((x ′, y′)) (say)

  m + 1 m + 1 
x + ym mx + ym 2
where x = 2 ′ ,y = ′
m +1 m2 + 1
 x ′ + y′m mx ′ + y′m 2 
P((x ′, y′)) =  2 , 
 m +1 m 2 + 1 
Substituting the values of x ′, y′, we get
 x + ym  mx + ym 2   x + ym   mx + ym  2 
2
 2 + m  m   +  m
 m +1  m 2 + 1   m + 1   m + 1  
2 2
P((x ′, y′)) =  , 
m2 + 1 m2 + 1
 
 
 
2 2 3
x + ym m(mx + ym ) x + ym + m x + ym
We have 2 + =
m +1 m2 + 1 m2 + 1
x (1 + m 2 ) + ym(1 + m 2 )
= = x + ym.
m2 +1
Therefore,
x + ym  mx + ym 2 
+ m  
m2 + 1 2
 m + 1  = x + ym .
m2 + 1 m2 + 1
Similarly, we have

 x + ym   mx + ym  2 mx + ym 2 + m3 x + ym 4
2
m 2  +  2
 m =
 m +1   m +1  m2 + 1

mx (1 + m 2 ) + m 2 y(1 + m 2 )
= 2
= mx + m 2 y.
m +1
Therefore
 x + ym   mx + ym  2
2
m 2  
+  2
m
 m +1   m +1  mx + m 2 y
= .
m2 + 1 m2 + 1
52
  x + ym mx + m 2 y 
∴ P ((x , y)) =  2
2
,  = P((x, y)).
 m +1 m 2 + 1 
∴ (P 2 − P) ((x , y)) = 0 or P satisfies x 2 − x = 0.
x 2 − x = x (x − 1). Since the minimal polynomial divides x 2 − x, it has to
be either x , x − 1 or x (x − 1).
If it is x , we have P((x , y)) = 0 ∀ ( x, y) ∈ R 2 . But
 1 m 
P((1, 0)) =  2 , 2  ≠ (0, 0) since we assumed that m ≠ 0.
 m +1 m +1
If the minimal polynomial is x − 1, we have P − I = 0 or P is the identity
 1 m  2
operator. Again P((1, 0)) =  2 , 2  ≠ (1, 0) is x − x = 0.
 m + 1 m + 1 
***

E22) Check that the reflection operator satisfies the equation x 2 − 1 = 0.

Show that this is the minimal polynomial of R.

Remark 3: If dim V = n and T ∈ A(V), we have seen that the degree of

the minimal polynomial p of T ≤ n 2 . In Unit 11, we shall see that the
degree of p cannot exceed n. We shall also study a systematic method
of finding the minimal polynomial of T, and some applications of this
polynomial. But now we will only illustrate one application of the concept
of the minimal polynomial by proving the following theorem.

Theorem 10: Let T ∈ A(V ). Then T is invertible if and only if the

constant term in the minimal polynomial of T is not zero.

Proof: Let p( x ) = a 0 + a 1 x + L + a m x m −1 + x m be the minimal polynomial of

T. Then a 0 I + a1T + L + a m−1T m −1 + T m = 0.
 T (a 1I + L + a m −1T m −2 + T m −1 ) = −a 0 I (1)
−1
Firstly, we will show that if T exists, then a 0 ≠ 0. On the contrary,
suppose a 0 = 0. Then (1) implies that T(a 1I + L + T m−1 ) = 0. Multiplying
both sides by T −1 on the left, we get a1I + L + T m−1 = 0.
This equation gives us a monic polynomial q( x ) = a1 + L + x m−1 such that
q(T) = 0 and deg q < deg p. This contradicts the fact that p is the minimal
polynomial of T. Therefore, if T −1 exists then the constant term in the
minimal polynomial of T cannot be zero.

Conversely, suppose the constant term in the minimal polynomial of T

is not zero, that is, a 0 ≠ 0. Then dividing Eqn. (1) on both sides by
(−a 0 ), we get
T ((−a 1 / a 0 )I + L + (−1 / a 0 )T m−1 ) = I.
Let S = (−a 1 / a 0 )I + L + (−1 / a 0 )T m−1 .
53
 Then we have ST = I and TS = I. This shows, by Theorem 5, that T −1
exists and T −1 = S.

E23) Let Pn be the space of all polynomials of degree ≤ n. Consider the

linear operator D : P2 → P2 given by D(a 0 + a 1x + a 2 x 2 ) = a 1 + 2a 2 x.
(Note that D is just the differentiation operator.) Show that D 4 = 0.
What is the minimal polynomial of D ? Is D invertible?

E24) Consider the reflection transformation given in Unit 7, Example 4.

Find its minimal polynomial. Is T invertible? If so, find its inverse.

E25) Let the minimal polynomial of S ∈ A(V) be x n , n ≥ 1. Show that

there exists v 0 ∈ V such that the set {v 0 , S( v 0 ), K, Sn −1 ( v n )} is
linearly independent.

We will now end the unit by summarising what we have covered in it.

8.6 SUMMARY
In this unit we covered the following points.
1. L(U, V), the vector space of all linear transformations from U to
V is of dimension (dim U) (dim V).

2. The dual space of a vector space V is L(V, F) = V ∗ , and is

isomorphic to V.

3. If {e1 , K, e n } is a basis of V and {f1 , K, f n } is its dual basis, then

n n
f =  f (ei )f i ∀ f ∈ V ∗ and v =  f i ( v)ei ∀ v ∈ V.
i=1 i =1

4. Every vector space is isomorphic to its second dual.

5. Suppose S ∈ L(V, W ) and T ∈ L( U, V ). Then their composition

S o T ∈ L( U, W ).

6. S ∈ A(V) = L(V, V ) is an isomorphism if and only if there exists

T ∈ A (V) such that S o T = I = T o S.

7. For T ∈ A (V) there exists a non-zero polynomial g ∈ F[x ], of

degree at most n 2 , such that g(T) = 0, where dim V = n.

8. The minimal polynomial of T ∈ A (V) is the monic polynomial p, of

smallest degree, such that p(T) = 0.

9. If p is the minimal polynomial of T and f is a polynomial such

that f (T) = 0, then there exists a polynomial g( x ) such that
54 f ( x ) = p(x ) g (x ).
10. Let T ∈ A(V ). Then T −1 exists if and only if the constant term in the
minimal polynomial of T is not zero.

8.7 SOLUTIONS/ANSWERS
E1) We have to check that VS1-VS10 are satisfied by L(U, V). We
have already shown that VS1 and VS6 are true:
VS2: For any L, M, N ∈ L(U, V), we have
∀ u ∈ U, [(L + M) + N ) (u )
= (L + M) (u ) + N(u ) = [L(u ) + M(u )] + N(u )
= L(u ) + [M(u ) + N (u )], since addition is associative in V.
= [L + (M + N)] (u )
∴ (L + M) + N = L + (M + N).
VS3: 0 : U → V : 0(u ) = 0 ∀ u ∈ U is zero element of L(U, V).
VS4: For any S ∈ L(U, V), (−1)S = −S, is the additive inverse of S.
VS5: Since addition is commutative in V, S + T = T + S ∀ S, T in
L(U, V).
VS7: ∀ α ∈ F and S, T ∈ L( U, V),
α(S + T) (u ) = (αS + αT) (u ) ∀ u ∈ U,
∴ α(S + T) = αS + αT.
VS8: ∀, α, β ∈ F and S ∈ L(U, V), (α + β)S = αS + βS.
VS9: ∀ α, β ∈ F and S ∈ L( U, V ), (αβ)S = α(βS).
VS10: ∀ S ∈ L(U, V), 1.S = S.

E2) E2m (e m ) = f 2 and E2 m (e i ) = 0 for i ≠ m.

E32 (e 2 ) = f 3 and E32 (e i ) = 0 for i ≠ 2.
f , if i = n
Emn (e i ) =  m .
 0, otherwise

E3) Both space have dimension 2 over R. A basis for L(R 2 , R ) is

{E11 , E12}, where E11 (1, 0) = 1, E11 (0, 1) = 0, E12 (1, 0) = 0, E12 (0, 1) = 1.
A basis for L(R, R 2 ) is {E11 , E12}, where
E11 (1) = (1, 0), E 21 (1) = (0, 1).

E4) Let f : R 3 → R be any linear functional. Let

f (1, 0, 0) = a 1 , f (0, 1, 0) = a 2 , f (0, 0, 1) = a 3 . Then, for any
x = ( x1 , x 2 , x 3 ), we have x = x1 (1, 0, 0) + x 2 (0, 1, 0) + x 3 (0, 0, 1).
∴ f (x ) = x1f (1, 0, 0) + x 2f (0, 1, 0) + x 3f (0, 0, 1)
= a1 x1 + a 2 x 2 + a 3 x 3 .

E5) Let the dual basis be {f1 , f 2 , f 3}. Then, for any
v ∈ P2 , v = f1 ( v).1 + f 2 ( v).x + f 3 ( v).x 2 .
55
 ∴, if v = a 0 + a 1 x + a 2 x 2 , then f1 ( v) = a 0 , f 2 ( v) = a 1 , f 3 ( v) = a 2 .
That is,
f1 (a 0 + a 1 x + a 2 x 2 ) = a 0 , f 2 (a 0 + a1 x + a 2 x n ) = a 1 , f3 (a 0 + a 2 x + a 2 x 2 ) = a 2 ,
for any a 0 + a1 x + a 2 x 2 ∈ P2 .

E6) Let the dual basis be {f1 , f 2 , f 3}. Let its dual basis be
{θ1 , K, θ n }, θi ∈ V∗∗ . Let ei ∈ V such that φ(ei ) = θi (ref. Theorem
3) for i = 1, K, n.
Then {e1 , K, e n } is a basis of V, since φ −1 is an isomorphism and
maps a basis to {e1 , K , e n }. Now f i (e j ) = φ(e j ) (f i ) = θ j (f j ) = δ ji , by
definition of a dual basis.
∴{f1 , K, f n } is the dual of {e1 , K, e n }.

E7) For any S ∈ A(V) and for any v ∈ V,

S o I(v) = S( v) and I o S( v) = I(S( v)) = S( v).
∴ S o I = S = I o S.

E8) ∀ S ∈ A(V) and v ∈ V,

S o 0(v) = S(0) = 0, and
0 o S( v) = 0(S(v)) = 0.
∴ S o 0 = 0 o S = 0.

E9) S ∈ A (R 2 ), T ∈ A(R 2 ).
S o T( x1 , x 2 ) = S(−x 2 , x1 ) = ( x1 , x 2 )
T o S( x1 , x 2 ) = T(x 2 , − x1 ) = ( x1 , x 2 )
∀ ( x1 , x 2 ) ∈ R 2 .
∴ S o T = T o S = I, and hence, both S and T are invertible.

E10) T ∈ L(R 2 , R 3 ), S ∈ L(R 3 , R 2 ). ∴ S o T ∈ A (R 2 ), T o S ∈ A (R 3 ).

∴ S o T and T o S can never be equal.
Now, S o T( x1 , x 2 ) = S(0, x1 , x 2 ) = ( x1 , x1 + x 2 ) ∀ ( x1 , x 2 ) ∈ R 2 .
Also,
T o S( x1 , x 2 , x 3 ) = T ( x1 + x 2 , x 2 + x 3 ) = (0, x1 + x 2 , x 2 + x 3 ) ∀ ( x1 , x 2 , x 3 ) ∈ R 3 .

E11) T ∈ A(R 2 ) and S ∈ A(R 3 ), S o T and T o S are not defined.

E12) Both (R o S) o T and R o (S o T) are in L( U, Z). For any u ∈ U.

[(R o S) o T](u ) = (R o S)[T(u )] = R[S(T(u ))] = R[(S o T)(u )] = [R o (S o T)](u ).
∴ (R o S) o T = R o (S o T).

E13) By Unit 7, Theorem 6, rank (S o T) ≤ rank (T).

Also, rank(T) = rank(I o T) = rank((S−1 o S) o T)
= rank(S−1 o (S o T)) ≤ rank(S o T) (by Unit 7, Theorem 6).
Thus, rank (S o T) ≤ rank (T) ≤ rank (S o T).
∴ rank (S o T) = rank (T).
Similarly, you can show that rank (T o S) = rank (T).
56
E14) (S + T) ( x, y) = ( x , − y) + ( x + y, y − x ) = (2 x + y, − x )
ST( x , y) = S( x + y, y − x ) = ( x + y, x − y)
TS( x , y) = T( x , − y) = ( x − y, − ( x + y))
[S o (S − T)] ( x , y) = S(− y, x − 2 y) = (− y, 2 y − x )
[(S − T) o S] ( x, y) = (s − T) ( x, y) = (x , y) − ( x − y, − ( x + y)) = ( y, 2 y + x )
∀ ( x, y) ∈ R 2 .

E15) a) We first show that if A, B ∈ M and α, β ∈ F, then

αA + βB ∈ M. Now,
S o (αA + βB) = S o αA + S o βB, by Theorem 6.
= α(S o A) + β(S o B), again by Theorem 6.
= α0 + β0, since A, B ∈ M.
=0
∴ αA + βB ∈ M, and M is a subspace of A(V).
Similarly, you can show that N is a subspace of A(V).

b) For any T ∈ M, ST (v) = 0 ∀ v ∈ V. ∴ T(v) ∈ Ker S ∀ v ∈ V.

∴ R (T), the range of T, is a subspace of Ker S.
∴ T ∈ L(V, Ker S). ∴ M ⊆ L(V, Ker S).
Conversely, for any T ∈ L(V, Ker S), T ∈ A(V ) such that
S(T( v)) = 0 ∀ v ∈ V.
∴ ST = 0. ∴ T ∈ M.
∴ L(V, Ker S) ⊆ M.
∴ We have proved that M = L(V, Ker S).
∴ dim = (dim V) (nullity S), by Theorem 1.
= n (n − r ), by the Rank Nullity Theorem.

E16) (p + q) (T) = p(T) + q (T) = 0 + 0 = 0.

E17) (2I + 3S + S3 ) (S + 2S4 ) = (2I + 3S + S3 )S + (2I + 3S + S3 ) (2S4 )

= 2S + 3S2 + S4 + 4S4 + 6S5 + 2S7

= 2S + 3S2 + 5S4 + 6S5 + 2S7
Also, (S + 2S4 ) (2I + 3S + S3) = 2S + 3S2 + 5S4 + 6S5 + 2S7
∴, (S + 2S4 ) (2I + 3S + S3 ) = (2I + 3S + S3 ) (S + 2S4 ).

E18) Consider g( x ) = x − 1 ∈ R[ x ]. Then g(I) = I − 1. I = 0.

Also, if h (x ) = x , then h (0) = 0.
Notice that the degrees of g and h are both 1 ≤ dim R 3 .

E19) Let p = a 0 + a 1 x + L + a n x n , q = b 0 + b1 x + L + b m x m .

a) Then ap + bq = aa 0 + aa 1 x + L + aa n x n + bb 0 + bb1 x + L + bb m x m .
∴ φ(ap + bq ) = aa 0 I + aa 1T + L + aa n T n + bb 0 I + bb1T + L + bb m T m
= ap(T) + bq (T) = aφ(p) + bφ(q )
57
 b) pq = (a 0 + a 1 x + L + a n x n ) (b 0 + b1 x + L + b m x m )

= a 0 b 0 + (a 1b 0 + a 0 b1 ) x + L + a n b m x n + m
∴ φ(pq ) = a 0 b 0 I + (a 1b 0 + a 0 b1 )T + L + a n b m T n +m

= (a 0 I + a 1T + L + a n T n ) (b 0 I + b1T + L + b m T m )

= φ(p) φ(q ).

E20) T ∈ A(V ). Let p( x ) = x 3 . Then p is a monic polynomial. Also,

p(T ) ( x1 , x 2 , x 3 ) = T 3 ( x1 , x 2 , x 3 ) = T 2 (0, x1 , x 2 ) = T (0, 0, x ) = (0, 0, 0)
∀ (x1 , x 2 , x 3 ) ∈ R 3 .
∴ p(T) = 0.
We must also show that no monic polynomial q of smaller degree
exists such that q(T) = 0.
Suppose q = a + bx + x 2 and q(T) = 0.
Then (aI + bT + T 2 ) ( x1 , x 2 , x 3 ) = (0, 0, 0)
⇔ a ( x1 , x 2 , x 3 ) + b(0, x1 , x 2 ) + (0, 0, x ) = (0, 0, 0)
⇔ ax 1 = 0, ax 2 + bx1 = 0, ax 3 + bx 2 + x1 = 0 ∀ (x1 , x 2 , x 3 ) ∈ R 3 .
⇔ a = 0, b = 0 and x1 = 0. But x1 can be non-zero.
∴ q does not exist.
∴ p is a minimal polynomial of T.

E21) Consider p( x ) = x n . Then p(T) = 0 and no non-zero polynomial q

of lesser degree exists such that q(T) = 0. This can be checked on
the lines of the solution of E20.

E22) We have
 x (1 − m 2 ) + 2 ym 2mx + y(m 2 − 1) 
R ( x , y) =  , .
 m2 +1 m2 + 1 
The calculation is again conceptually simple but a little tedious.
Let us write R ((x , y)) = ( x′, y′) where

x (1 − m 2 ) + 2 ym 2mx + y(m 2 − 1)
x′ = , y ′ = .
m2 + 1 m2 + 1
 x′(1 − m 2 ) + 2 y′m 2mx′ + y′(m 2 − 1) 
Therefore, R ( x′, y′) =  , .
 m2 + 1 m2 + 1 
′
Substituting the values of x and y , we get ′
 x (1 − m 2 ) + 2 ym   2mx + y(m 2 − 1) 
x ′(1 − m 2 ) + 2 y′m =  2

 (1 − m 2
) + 2m  2

 m + 1   m + 1 
x (1 − m 2 ) + 2 ym(1 − m 2 ) + 4m 2 x + 2my (m 2 − 1)
=
m2 + 1
x − 2m 2 x + xm 4 + 2 ym − 2 ym 3 + 4m 2 x + 2m 3 y − 2 ym
=
m2 + 1
58
 x + 2m 2 x + m 4 x x (m 2 + 1) 2
= 2
= 2
= x (m 2 + 1).
m +1 (m + 1)
x ′(1 − m 2 ) + 2 y′m x (m 2 + 1)
∴ = = x.
m2 + 1 m2 + 1
Similarly, check that
2mx′ + y′(m 2 − 1)
= y.
m2 +1
Therefore, R 2 ((x , y)) = R ((x′, y′)) = ( x, y) or (R 2 − I) ( x , y) = 0 for
all ( x, y) ∈ R 2 . So, R satisfies the polynomial x 2 − 1 = 0. We have
( x 2 − 1) = ( x − 1) ( x + 1). So, the minimal polynomial is x − 1, x + 1 or
x 2 − 1. If it is x − 1, we have R − I = 0 or R = I. Again, taking
 2m m 2 − 1 
(x , y) = (1, 0) we get R ((0, 1)) =  2 , 2  ≠ (0, 1) since m ≠ 0.
 m +1 m +1
Therefore, x − 1 is not the minimal polynomial.
If R + I = 0, R = −I. ∴ R ((x, y)) = −( x, y) = (− x, − y).
 2m m 2 − 1 
Again, for ( x, y) = (0, 1),  2 , 2  ≠ (0, − 1).
 m + 1 m + 1 
Therefore x + 1 is not the minimal polynomial of R. So, the minimal
polynomial is x 2 − 1.

E23) D 4 (a 0 + a 1 x + a 2 x 2 ) = D 3 (a 1 + 2a 2 x ) = D 2 (2a 2 ) = D(0)

∀ a 0 + a 1x + a 2 x 2 ∈ P2 .
∴ D 4 = 0.
The minimal polynomial of D can be D, D 2 , D 3 or D 4 . Check that
D 3 = 0, but D 2 ≠ 0.
∴ the minimal polynomial of D is p( x ) = x 3 . Since p has no non-
zero constant term, D is not an isomorphism.

E24) T : R 2 → R 2 : T( x, y) = ( x, − y).
Check that T 2 − I = 0
∴ the minimal polynomial p must divide x 2 − 1.
∴ p( x ) can be x − 1, x + 1 or x 2 − 1. Since T − I ≠ 0 and T + I ≠ 0, we
see that p( x ) = x 2 − 1.
By Theorem 10, T is invertible. Now T 2 − I = 0.
∴ T(−T) = I. ∴ T −1 = −T.

E25) Since the minimal polynomial of S is x n , Sn = 0 and Sn−1 ≠ 0.

∴ ∃ v 0 ∈ V such that Sn −1 ( v 0 ) ≠ 0. Let a1 , a 2 , K, a n ∈ F such that
a1 v 0 + a 2S( v n ) + L + a n Sn −1 ( v n ) = 0 (1)
Then, applying Sn−1 to both sides of this equation, we get
a 1Sn −1 (v 0 ) + a 2Sn ( v 0 ) + L + a n S2 n −1 ( v 0 ) = 0.
 a 1Sn −1 ( v 0 ) = 0, since Sn = 0 = Sn +1 = K + S2n −1.
 a1 = 0. 59
 Now (1) reduces to a 2S( v 0 ) + L + a n Sn −1 (v 0 ) = 0.
Applying Sn−2 to both sides we get a 2 = 0. In this way we get
a i = 0 ∀ i = 1, K, n.
∴ The set {v 0 , S( v 0 ), K, Sn −1 ( v 0 )} is linearly independent.

60
Unit 1 Errors and Approximations

UNIT 9

LINEAR TRANSFORMATIONS AND

MATRICES

Structure Page No.

9.1 Introduction
Objectives
9.2 Matrix of a Linear Transformation
Sum and Multiplication by Scalars
Dimension of M nn (F) over F
9.3 Some Types of Matrices
Diagonal Matrix
Triangular Matrix
9.4 Matrix Multiplication
Matrix of the Composition of Linear Transformations
9.5 Invertible Matrices
Inverse of Matrix
Matrix of Change of Basis
9.6 Summary
9.7 Solutions / Answers

9.1 INTRODUCTION
In Unit 1, we discussed matrices. In Unit 2, we saw that M n , m ( F) is a
vector space over F. In this unit, we will look at the matrices; the
matrices will be representations of linear transformations. Such a
representation will enable us to compute several things like the rank
and nullity of a linear transformation. We will also see that composition
of linear transformations translates to multiplication of matrices. This
gives us an insight into matrix multiplication which looks very contrived
otherwise.

Matrices are intimately connected with linear transformations. In this

unit we will bring out this link. We will first derive algebraic operations on
matrices from the corresponding operations on linear transformations.
In Block 4 we will often refer to the material on change of bases, so do
spend some time on Sec.7.6. 61
Block 1 Solution of Nonlinear Equations in One Variable
To realise the deep connection between matrices and linear
transformations, you should go back to the exact spot in Unit 7 and 8 to
which frequent references are made.

Objectives
After studying this unit, you should be able to:
• define and give examples of various types of matrices;
• define a linear transformation, if you know its associated matrix;
• evaluate the sum, difference, product and scalar multiple of
matrices;
• obtain the transpose and conjugate of a matrix;
• determine if a given matrix is invertible;
• obtain the inverse of a matrix;
• discuss the effect that the change of basis has on the matrix of a
linear transformation.

9.2 MATRIX OF A LINEAR TRANSFORMATION

We will now obtain a matrix that corresponds to a given linear
transformation. You will see how easy it is to go from matrices to linear
transformations, and back.

Let U and V be vector spaces over a field F, of dimensions n and m,

respectively. Let B1 = {e1 , , e n } be an ordered basis of U, and
B2 = {f1 , , f m } be an ordered basis of V. (By an ordered basis we
mean that the order in which the elements of the basis are written is
fixed. Thus, an ordered basis {e1 , e 2 } is not equal to an ordered basis
{e 2 , e1}.)
Given a linear transformation T : U → V, we will associate a matrix to it.
For this, we consider T(e1 ), , T(e n ), which are all elements of V and
hence, they are linear combinations of f1 , , f m . Thus, there exist mn
scalars  ij , such that
T(e1 ) = 11f1 +  21f 2 +  +  m1f m
 
T(e j ) = 1 jf1 +  2 jf 2 +  +  mjf m
 
T(e n ) = 1n f1 +  2 n f 2 +  +  nmf m

From these n equations we form an m n matrix whose first row

consists of the coefficients of the first equation, second column consists
of the coefficients of the second equation, and so on. This matrix
 11 12  1n 
  22   2 n 
A =  21
   
 
 m1  m 2   mn 
62
Unit 1 Errors and Approximations
is called the matrix of T with respect to the bases B1 and B 2 . Notice
that the coordinate vector of T(e j ) is the jth column of A.

We use the notation [T ]BB12 for this matrix. Thus, to obtain [T ]BB12 we
consider T(ei )  ei  B1 , and write them as linear combination of the
elements of B 2 .

If T  L(V, V), B is basis of V and we take B1 = B2 = B, we call [T ]BB the

matrix of T with respect to the basis B, and we write this as [T ]B .

Remark 1: Why do we insist on ordered bases? What happens if we

interchange the order of the elements in B1 to {e n , e1 , , e n −1}? The
matrix [T ]BB12 also changes, the last column becoming the first column
now. Similarly, if we change the positions of the f is in B2 , the rows of
[T ]BB12 will get interchanged.

Thus, to obtain a unique matrix corresponding to T , we must insist on

B1 and B2 being ordered bases. Henceforth, while discussing the
matrix of a linear mapping, we will always assume that our bases
are ordered bases.

We will now give an example, followed by some exercises.

Example 1: Consider the linear operator T : R 3 → R 2 : T( x, y, z) = ( x, y).

Choose bases B1 and B2 of R 3 and R 2 , respectively. Then obtain
[T ]BB12 .

Solution: Let B1 = {e1 , e 2 , e3 }, where

e1 = (1, 0, 0), e 2 = (0, 1, 0), e3 = (0, 0, 1). Let B 2 = {f1 , f 2 }, where
f1 = (1, 0), f 2 = (0, 1). Note that B1 and B2 are the standard bases of R 3
and R 2 , respectively.
T(e1 ) = (1, 0) = f1 = 1.f1 + 0.f 2
T(e 2 ) = (0, 1) = f 2 = 0.f1 + 1.f 2
T(e3 ) = (0, 0) = 0f1 + 0f 2 .

1 0 0
Thus, [T]BB12 =  .
0 1 0 
***

E1) Choose two other bases B1 and B2 of R 3 and R 2 , respectively.
(In Unit 4 you came across a lot of bases of both these vector

spaces.) for T in the example above, give the matrix [T ]BB12

63
Block 1 Solution of Nonlinear Equations in One Variable
What E1 shows us is that the matrix of a transformation depends
on the bases that we use for obtaining it. The next two exercises
also bring out the same fact.

E2) Write the matrix of the linear transformation

T : R 3 → R 2 : T( x, y, z) = ( x + 2 y + 2z, 2x + 3y + 4z) with respect to
the standard bases of R 3 and R 2 .

E3) What is the matrix of T , in E2, with respect to the bases

B1 = {(1, 0, 0), (0, 1, 0), (1, − 2, 1)} and
B2 = {(1, 2), (2, 3)}?

E4) Let V be the vector space of polynomials over R of degree  3, in

the variable t. Let D : V → V be the differential operator given in
Unit 5 (E6, when n = 3) . Show that the matrix of D with respect to
the basis {1, t , t 2 , t 3} is
0 1 0 0 
0 0 2 0 
 .
0 0 0 3
 
0 0 0 0 

So far, given a linear transformation, we have obtained a matrix from it.

This works the other way also. That is, given a matrix we can define a
linear transformation corresponding to it.

1 2 4 
Example 2: Describe T : R → R such that [T]B = 2 3 1 , when B
3 3

3 1 2
is the standard basis of R 3 .

Solution: Let B = {e1 , e 2 , e3}. Now, we are given that

T(e1 ) = 1.e1 + 2.e 2 + 3.e3
T(e 2 ) = 2.e1 + 3.e 2 + 1.e3
T(e3 ) = 4.e1 + 1.e 2 + 2.e3
You know that any element of R 3 is ( x , y, z) = xe1 + ye 2 + ze 3 .
Therefore, T( x, y, z) = T( xe1 + ye 2 + ze 3 )
= xT (e1 ) + yT (e 2 ) + zT (e3 ), since T is linear.
= x (e1 + 2e 2 + 3e3 ) + y(2e1 + 3e 2 + e3 ) + z(4e1 + e 2 + 2e3 )
= ( x + 2 y + 4z)e1 + (2x + 3y + z)e 2 + (3x + y + 2z)e3
= ( x + 2 y + 4z, 2 x + 3y + z, 3x + y + 2z)
 T : R → R is defined by
3 3

T( x , y, z) = ( x + 2 y + 4z, 2x + 3y + z, 3x + y + 2z) .
***

Try the following exercises now.

64
Unit 1 Errors and Approximations

1 1 0
E5) Describe T : R 3 → R 2 such that [T]BB12 =  , where B1 and
0 1 1
B2 are the standard bases of R 3 and R 2 , respectively.

E6) Find the linear operator T : C → C whose matrix, with respect to

0 − 1
the basis {1, i} is  . (Note that C, the field of complex
1 0 
numbers, is a vector space over R, of dimension 2.)

Now we are in a position to define the sum of matrices and

multiplication of a matrix by a scalar.

9.2.1 Sum and Multiplication by Scalars

In Unit 7 you studied about the sum and scalar multiples of linear
transformation. In the following theorem we will see what happens to
the matrices associated with the linear transformations that are sums or
scalar multiples of given linear transformations.

Theorem 1: Let U and V be vectors spaces over F , of dimensions n

and m, respectively. Let B1 and B2 be arbitrary bases of U and V,
respectively. (Let us abbreviate [T ]BB12 to [T] during this theorem.) Let
S, T  L( U, V ) and   F. Suppose [S] = [a ij ], [T] = [bij ]. Then
[S + T] = [a ij + bij ], and
[S] = [a ij ].

Proof: Suppose B1 = {e1 , e 2 , , e n } and B2 = {f1 , f 2 , , f m }. Then all the

matrices to be considered here will be of size m n.
Now, by our hypothesis,
m
S(e j ) =  a ij f i  j = 1, , n and
i =1
m
T(e j ) =  bij f i  j = 1, , n
i =1

 (S + T)(e j ) = S(e j ) + T(e j ) (by definition of S + T )

m m
=  a ij f i +  bij f i
i =1 i =1
m
=  (a ij + bij )f i
i =1

Thus, by definition of the matrix with respect to B1 and B2 , we get

[S + T] = [a ij + bij ].
Now, (S)(e j ) = (S(e j )) (by definition of S )
 m 
=   a ij f i  Two matrices can be
 i =1  added if and only if they65
are of the same size
Block 1 Solution of Nonlinear Equations in One Variable
m
=  (a ij )f i
i =1

Thus, [S] = [a ij ] .

We can view an m n matrix A as a linear operator from  n to  m .

The matrix of this linear operator with respect to the standard bases in
 n and  m is precisely A. So, Theorem 1 says our definition of matrix
addition in Unit 1, Definition 2 is compatible with the addition of linear
operator.

Now, let us define the scalar multiple of a matrix, again motivated by

Theorem 1.

Definition: Let  be a scalar, i.e.,   F, and let A = [a ij ]m n . Then we

define the scalar multiple of the matrix A by the scalar  to be the
matrix
 a11 a12  a1n 
 a a 22  a 2 n 
A =  21

     
 
a m1 a m 2  a mn 
In other words, A is the m n matrix whose (i, j) th element is 
times the (i, j) th element of A.

Again, this agrees with our definition of scalar multiplication in Unit 1.

Remark 1: The way we have defined the sum and scalar multiple of
matrices allows us to write Theorem 1 as follows:
[S + T]BB12 = [S]BB12 + [T]BB12
[S]`BB12 = [S]BB12 .

The following exercise will help you in checking if you have understood
the contents of Sections 2.2.2 and 2.2.3.

E7) Define S : R 3 → R 2 : S( x, y) = ( x , 0, y) and

T : R 2 → R 3 : T( x , y) = (0, x, y). Let B1 and B 2 be the standard
bases for R 2 and R 3 , respectively.

Then what are [S]BB12 , [T]BB12 , [S + T]BB12 , [S]BB12 , for any   R.

We have seen in E28), Unit 2 that M mn ( F) is a vector space over F. In

the next section we find the dimension of M mn ( F) over F.

9.2.2 Dimension of Mmn (F) over F

66
Unit 1 Errors and Approximations
What is the dimension of M mn (F) over F ? To answer this question we
prove the following theorem. But, before you go further, check whether
you remember the definition of a vector space isomorphism (Unit 7).

Theorem 2: Let U and V be vector spaces over F of dimensions n

and m, respectively. Let B1 and B 2 be a pair of bases of U and V,
respectively. The mapping  : L( U, V) → M mn (F), given by (T) = [T]BB12 is
a vector space isomorphism.

Proof: The fact that  is a linear transformation follows from Theorem

1. We proceed to show that the map is also 1 − 1 and onto. For the rest
of the proof we shall denote [S]BB12 by [S] only, and take
B1{e1 ,  , e n }, B2{f1 , f 2 ,  , f m }.

 is 1 − 1 : Suppose, S, T  L( U, V ) be such that (S) = (T ).

Then [S] = [T ]. Therefore, S(e j ) = T(e j )  e j  B1 .
Thus, by Unit 5 (Theorem 1), we have S = T.

 is on 0 : If A  M mn (F) we want to construct T  L( U, V ) such that

(T) = A. Suppose A = [a ij ]. Let v1 ,  , v n  V such that
m
v j =  a ij f i for j = 1,  , n.
i =1
Then, by Theorem 3 of Unit 5, there exists as linear transformation
n
T  L( U, V ) such that T(e j ) = v j =  a ijfi .
i =1
Thus, by definition, (T) = A.
Therefore,  is a vector space isomorphism.
A corollary to this theorem gives us the dimension of M mn (F).

Corollary: Dimension of M mn (F) = mn.

Proof: Theorem 2 tells us that M mn (F) is isomorphic to L( U, V).

Therefore, dim F M mn (F) = dim F L( U, V) (by Theorem 12 of Unit 5) = mn ,
from Unit 6 (Theorem 1).

Why do you think we chose such a roundabout way for obtaining

dim M mn (F)? We could as well have tried to obtain mn linearly
independent m n matrices and show that they generate M mn (F). But
that would be quite tedious (see E16). Also, we have done so much
work on L( U, V ) so why not use that! And, doesn’t the way we have
used seem neat?

Now for some exercises related to Theorem 2.

E8) At most, how many matrices can there be in any linearly

independent subset of M 23 (F) ?
67
Block 1 Solution of Nonlinear Equations in One Variable
E9) Are the matrices [1, 0] and [1, − 1] linearly independent over R ?

Now we move on to the next section, where we see some ways of

getting new matrices from given ones.

9.3 SOME TYPES OF MATRICES

We discussed some special matrices like upper triangular and lower
triangular matrices. We now discuss some of these matrices and their
relation to linear transformations.

9.3.1 Diagonal Matrix

Let U and V be vector spaces over F of dimension n. Let
B1 = {e1 ,  , e n } and B2 = {f i ,  , f n } be based of U and V, respectively.
Let d1 ,  , d n  F. Consider the transformation
T : U → V : T(a1e1 +  + a n e n ) = a1d1f1 +  + a n d n f n .
Then T(e1 ) = d1f1 , T(e 2 ) = d 2f 2 ,  , T(e n ) = d n f n .

d1 0  0
0 d  0 
[T ]BB12 = 2
.
   
 
0 0  dn 

Such a matrix is called a diagonal matrix. Let us see what this means.

Let A = [a ij ] be a square matrix. The entries a11 , a 22 ,  , a nn are called

the diagonal entries of A. This is because they lie along the diagonal,
from left to right, of the matrix. All the other entries of A are called the
off-diagonal entries of A.

A square matrix whose off-diagonal entries are zero (i.e., a ij = 0  i  j)

is called a diagonal matrix. The diagonal matrix
d1 0 0  0 
0 d 0  0 
 2 
     
 
     
 0 0 0  d n 
is denoted by diag (d1 , d 2 ,  , dn ).

Note: The d i ' s may or may not be zero. What happens if all the d i ' s
are zero? Well, we get the n  n zero matrix, which corresponds to the
zero operator.

If d i = 1  i = 1, , n , we get the identity matrix, I n (or I, when the size

is understood).

68
Unit 1 Errors and Approximations
E10) Show that I, is the matrix associated to the identity operator from
R n to R n .

If   F, the linear operator I : R n → R n : I( v) = v, for all v  R n , is

called a scalar operator. Its matrix with respect to any basis is
I = diag (, , , ). Such a matrix is called a scalar matrix. It is a
diagonal matrix whose diagonal entries are all equal.

With this much discussion on diagonal matrices, we move onto describe

triangular matrices.

9.3.2 Triangular Matrix

Let B = {e1 , e 2 ,  , e n } be a basis of a vector space V. Let S  L(V, V )
be an operator such that
S(e1 ) = a11e1
S(e2 ) = a12e1 + a 22e2
  
S(e n ) = a 1n e1 + a 2 n e 2 +  + a nn e n ,
Then, the matrix of S with respect to B is
a11 a12 a12  a1n 
0 a   a 2 n 
A=  22

     
 
0 0 0  a nn 
Note that a ij = 0  i  j.
A square matrix A such that a ij = 0  i  j is called an upper triangular
matrix. If a ij = 0  i  j, then A is called strictly upper triangular.
1 3 1 0 1 0
For example,  ,  ,   are all upper triangular, while
0 2 0 0 0 1
0 3
0 0 is strictly upper triangular.
 

Note that every strictly upper triangular matrix is an upper triangular

matrix.

Now let T : V → V be an operator such that T (e j ) is a linear

combination of e j , e j1 ,  , e n j. The matrix of T with respect to B is
 b11 0 0  0
b   0 
 21 b 22
[T]B =       
 
      
b n1 b n 2 b n 3  b nn 
Note that b ij = 0  i  j.
69
Block 1 Solution of Nonlinear Equations in One Variable
Such a matrix is called a lower triangular matrix. If b ij = 0 for all i  j,
then B is said to be a strictly lower triangular matrix.
The matrix
 0 0 0 0
 2 0 0 0
 
− 1 − 1 0 0
 
 1 0 5 0
is a strictly lower triangular matrix. Of course, it is also lower triangular!

1 2 3
Remark 4: If A is an upper triangular 3 3 matrix, say A = 0 4 5 ,
0 0 6
1 0 0
then A = 2 4
t
0 , a lower triangular matrix.
3 5 6
In fact, for any n  n upper triangular matrix A, its transpose is lower
triangular, and vice-versa.

E11) If an upper triangular matrix A is symmetric, then show that it

must be a diagonal matrix.

In the next section, we related the multiplication of matrices to

composition of linear transformations.

9.4 MATRIX MULTIPLICATION

9.4.1 Matrix of the Composition of Linear

Transformations

Let U, V and W be vector spaces over F , of dimensions p, n and m,

respectively. Let B1 , B 2 and B3 be bases of these respective spaces.
Let T  L( U, V ) and S L(V, W ). Then ST (= ST )  L( U, W ) (see
Sec.6.4).

Suppose [T]BB12 = B = [b jk ]n p and [S]BB32 = A = [a ij ]mn .

We ask: What is the matrix [ST ]BB12 ?
To answer this we suppose
B1 = {e1 , e 2 ,  , e p }
B2 = {f1 , f 2 ,  , f n }
B3{g1 , g 2 ,  , g m }.
n
Then, we know that T(ek ) =  b jk f j  k = 1, 2, , p,
j=1
m
and S(f j ) =  a ijgi  j = 1, 2, , n.
i =1
70
Unit 1 Errors and Approximations
Therefore,
 n 
S  T(ek ) = S(T(ek )) = S  b jk f j  = b1kS(f1 ) + b2 kS(f 2 ) +  + bnkS(f n )
 j=1 
 m
  m   m 
= b1k   a ijgi  + b 2 k   a i 2gi  +  + bnk   a in gi 
 i =1   i =1   i =1 
m
=  (a i1b1k + a i 2b 2 k +  + a in b nk )gi , on collecting the
i =1

coefficients of g i .
n
Thus, [ST ]BB13 = [cik ]mp , where c ik =  a ij b jk
j=1

We define the matrix [c jk ] to be the product AB.

We now make an observation.

Remark 5: If T  L( U, V ) and S L(V, W ), then [ST ]BB12 = [S]BB32 [T]BB12 ,

where B1 , B2 , B3 are the bases of U, V, W, respectively.

Let us illustrate this remark.

Example 3: Let T : R 2 → R 3 be a linear transformation such that

T ( x , y) = (2 x + y, x + 2 y, x + y). Let S : R 3 → R 2 be defined by
S( x , y, z) = (− y + 2z, y − z). Obtain the matrices [T ]BB12 , [S]BB12 , and [S  T ]B1 ,
and verify that [S  T]B1 = [S]BB12 [T]BB12 , where B1 and B2 are the standard
bases in R 2 and R 3 , respectively.

Solution: Let B1 = {e1 , e 2 }, B2 = {f1 , f 2 , f 3}.

Then T(e1 ) = T(1, 0) = (2, 1, 1) = 2f1 + f 2 + f 3
T(e 2 ) = T(0, 1) = (1, 2, 1) = f1 + 2f 2 + f 3
Thus,
2 1 
[T ]B 2 = 1 2
B1

1 1
Also,
S(f1 ) = S(1, 0, 0) = (0, 0) = 0.e1 + 0.e 2
S(f 2 ) = S(0, 1, 0) = (−1, 1) = −e1 + e 2
S(f 3 ) = S(0, 0, 1) = (2, − 1) = 2e1 + e 2
Thus,
0 − 1 2 
[S]BB12 =  
0 1 − 1
2 1 
0 − 1 2   
B2
So, [S] [T] B1
=  1 2 
−
B1 B2
 0 1 1 1 1 
 
1 0
=  = I2
0 1 
71
Block 1 Solution of Nonlinear Equations in One Variable
Also, S  T ( x , y) = S(2 x + y, x + 2 y, x + y)
= (− x − 2 y + 2 x + 2 y, x + 2 y − x − y)
= ( x , y).
Thus, S  T − 1, the identity map.
This means [S  T]B1 = I 2 .
Hence, [S  T]B1 = [S]BB12 [T]BB12 .

E12) Let S : R 3 → R 3 : S( x , y, z) = (0, x , y), and

T : R 3 → R 3 : T( x , y, z) = ( x , 0, y). Show that [S  T ]B = [S]B [T ]B ,
where B is the standard basis of R 3 .

E13) Let A, B be two diagonal n  n matrices over F. Show that AB is

also a diagonal matrix.

Now we shall go on to introduce you to the concept of an invertible

matrix.

9.5 INVERTIBLE MATRICES

In this section we will relate the invertibility of a linear transformation to
the invertibility of its linear transformation with respect to a basis. We
will also see how bases of a vector space are related to each other by
an invertible matrix.

9.5.1 Inverse of a Matrix

Just as we defined the operations on matrices by considering them on
linear operations first, we give a definition of invertibility for matrices
based on consideration of invertibility of linear operators.

It may help you to recall what we mean by an invertible linear

transformation. A linear transformation T : U → V is invertible if
a) T is 1 − 1 and onto, or, equivalently,
b) there exists a linear transformation S : V → U such that
S  TU , T  SI V .

In particular, T  L(V, V ) is said to be invertible if  S L(V, V ) such

that ST = TS = I.

We have the following theorem involving the matrix of an invertible

linear operator.

Theorem 3: Let V be an n -dimensional vector space over a field F ,

and B be a basis of V. Let T  L(V, V ). T is invertible iff there exists
A  M n (F) such that [T ]B A = I n = A[T ]B .

72
Unit 1 Errors and Approximations
Proof: Suppose T is invertible. Then  S  L(V, V ) such that
TS = ST = I. Then, by Theorem 2, [TS]B = [ST ]B = I. That is,
[T ]B[S]B = [S]B[T ]B = I. Take A = [S]B . Then [T]B A = I = A[T]B .

Conversely, suppose  a matrix A such that [T]B A = A[T]B = I.

Let S L(V, V) be such that [S]B = A. ( S exists because of Theorem 2.)
Then [T ]B[S]B = [S]B[T]B = I = [I]B . Thus, [TS]B[ST ]B = [I]B .
So, by Theorem 2, TS = ST = I. That is, T is invertible.

We will now make a few observation about the matrix inverse, in the
form of a theorem.

Theorem 4: a) If A is invertible, then

i) A −1 is invertible and (A −1 ) −1 = A.
ii) A t is invertible and (A t ) −1 = (A −1 ) t .

b) If A, B  M n (F) are invertible, then AB is invertible and

(AB) −1 = B −1A −1 .

Proof: (a) By definition,

AA −1 = A −1A = I (1)

i) Equation (1) shows that A −1 is invertible and (A −1 ) −1 = A.

ii) If we take transposes in Equation (1) and use the property that
(AB) t = Bt A t , we get
(A −1 ) t A t = A t (A −1 ) t = I t = I.
So A t is invertible and (A t ) −1 = (A −1 ) t .

(b) To prove this we will use the associativity of matrix multiplication.

Now
(AB) (B−1A −1 ) = [A(BB −1 )]A −1 = AA −1 = I.
(B−1A −1 ) (AB) = B−1[(A −1A)B] = B−1B = I.
So AB is invertible and (AB) −1 = B−1A −1.

We now relate matrix invertibility with the linear independence of its

rows or columns. When we say that the m rows of A = [a ij ]  M mn (F)
are linearly independent, what do we mean? Let R 1 , , R m be the m
row vectors [a11 , a12 ,  , a1n ], [a 21 ,  , a 2 n ], , [a m1 ,  , a mn ], respectively.
We say that they are linearly independent if, whenever  a1 ,  , a m  F
such that a1R1 +  + a m R m = 0,
then a1 = 0,  , a m = 0.

Similarly, the n columns C1 , , C n of A are linearly independent if

b1C1 +  + b n Cn = 0
 b1 = 0, b 2 = 0,  , b n = 0, where b1 ,  , b n  F.
73
Block 1 Solution of Nonlinear Equations in One Variable
We have the following result.

Theorem 5: Let A  M n (F). Then the following conditions are

equivalent.
a) A is invertible.
b) The columns of A are linearly independent.
c) The rows of A are linearly independent.

Proof: We first prove (a)  (b), using Theorem 4. Let V be an n -

dimensional vector space over F and B = {e1 ,  , e n } be a basis of V.
Let T  L(V, V ) be such that [T]B = A. Then A is invertible iff T is
invertible iff T(e1 ), T(e 2 ),  , T(e n ) are linearly independent (see Unit 5,
Theorem 9). Now we define the map
 a1 
 : V → M n 1 (F) : (a1e1 +  + a n e n ) =    .
a n 

Before continuing the proof we give an exercise

E14) Show that  is a well-defined isomorphism.

Now let us go on with proving Theorem 7.

Let C1 , C 2 ,  , C n be the columns of A. Then (T (e1 )) = C i for all

i = 1,  , n. Since  is an isomorphism, T(e1 ), , T(e n ) are linearly
independent iff C1 , C 2 ,  , C n are linearly independent. Thus, A is
invertible iff C1 , , C n are linearly independent. Thus, we have proved
(a)  (b).

Now, the equivalence of (a) and (c) follows because A is invertible

 A t is invertible.
 the columns of A t are linearly independent (as we have just shown)
 the rows of A are linearly independent (since the columns of A t
are the rows of A ).
So we have shown that (a)  (c).
Thus, the theorem is proved.

From the following example you can see how Theorem 7 can be useful.

1 0 1
Example 4: Let A = 0 1 1  M 3 (R).
1 1 1
Determine whether or not A is invertible.

Solution: Let R 1 , R 2 , R 3 be the rows of A. We will show that they are

linearly independent.
74
Unit 1 Errors and Approximations

1 0 1 
The row operations R 3 → R 3 − R 1 , R 3 → R 3 − R 2 gives 0 1 1 .
0 0 − 1
Since there are no zero rows, the rows are linearly independent.
Thus, by Theorem 5, A is invertible.
***

2 0 1
E15) Check if 0 0 1  M 3 (Q) is invertible.
0 3 0

9.5.2 Matrix of Change of Basis

Let V be an n -dimensional vector space over F. Let

B = {e1 , e2 ,  , e n } and B{e1 , e2 ,  , en } be two bases of V. Since
e j  V, for every j, it is a linear combination of the elements of B.
Suppose,
n
ej =  a ij e i  j = 1, , n.
i =1

The n  n matrix A = [a ij ] is called the matrix of the change of basis

from B to B. It is denoted by M BB .

Note that A is the matrix of the transformation T  L(V, V ) such that

T(e j ) = ej  j = 1,  , n, with respect to the basis B. Since {e1 , , en ) is
a basis of V, from Unit 5 we see that T is 1 − 1 and onto. Thus T is
invertible. So A is invertible. Thus, the matrix of the change of basis
from B to B is invertible.

Note: a) M BB = I n , This is because, in this case ei = ei  i = 1, 2,  , n.

b) M BB = [I]BB . This is because
I(ej ) = ej =  a ijei  j = 1, 2, , n.
i =1

Now suppose A is any invertible matrix. By Theorem 2,  T  L( V, V)

such that [T]R = A. Since A is invertible, T is invertible. Thus, T is
1 − 1 and onto. Let f i = T(ei )  i = 1, 2,  , n. Then B = {f1 , f 2 ,  , f n } is
also a basis of V, and the matrix of change of basis from B to B is A.

In the above discussion, we have just proved the following theorem.

Theorem 8: Let B = {e1 , e 2 ,, e n } be a fixed basis of V. The mapping

B → M BB is a 1 − 1 and onto correspondence between the set of all
bases of V and the set of invertible n  n matrices over F.
75
Block 1 Solution of Nonlinear Equations in One Variable
Let us see an example of how to obtain M BB .

Example 4: In R 2 , B = {e1 , e 2 } is the standard basis. Let B be the

basis obtained by rotating B through an angle  in the anti-clockwise
direction (see Fig.1). Then B = (e1 , e2 ) where
e1 = (cos , sin ), e2 = (− sin , cos ). Find M BB .

Fig.1

Solution: e1 = cos (1, 0) + sin (0, 1), and

e2 = − sin (1, 0) + cos (0, 1)
cos  − sin 
Thus, M BB =  
 sin  cos  
***

Try the following exercise.

E16) Let B be the standard basis of R 3 and B be another basis such

0 1 1 
that M B = 1 1 0 . What are the elements of B ?
B

0 0 3

What happens if we change the basis more than once? The following
theorem tells us something about the corresponding matrices.

Theorem 9: Let B, B, B be three bases of V. Then M BBM BB = M BB .

Proof: Now, M BBM BB = [I]BB , [I]BB

= [I  I]BB = M BB .
An immediate useful consequence is

Corollary: Let B, B be two bases of V. Then M BBM BB = I = M BBM BB .

That is, (M BB ) = M BB .
−1

Proof: By Theorem 9,
M BBM BB = M BB = I
Similarly, M BBM BB = M BB = I.

But, how does the change of basis affect the matrix associated to a
given linear transformation? In Sec.7.2 we remarked that the matrix of a
linear transformation depends upon the pair of bases chosen. The
76
Unit 1 Errors and Approximations
relation between the matrices of a transformation with respect to two
pairs of bases can be described as follows.

Theorem 10: Let T  L( U, V ). Let B1 = {e1 ,  , e n } and B2 = {f1 ,  , f m }

be a pair of bases of U and V, respectively.
Let B1 = {e1 ,  , en }, B2 = {f1,  , f m ] be another pair of bases of U and
V, respectively. Then,
[T ]BB12 = M BB22 [T]BB12 M BB11 .

Proof: [T]BB12 = [Iv  T  Iu]BB12 = [Iv]BB12 [Iu]BB11

(where I u = identity map on U and Iv = identity map on V)
= M BB22 [T]BB12 M BB11

Now, a corollary to Theorem 10, which will come in handy in the next
block.

Corollary: Let T  L(V, V ) and B, B be two bases of V. Then

[T]B = P −1[T]B P, where P = M BB .

Proof: [T]B = M BB [T]B M BB = P −1[T]B P, by the corollary to Theorem 9.

We can compute the matrix M BB using row reduction when B and B
are subsets of  n . Let us look at an example.

1 1 1  1 1 0  

 0  and B = 0, 1 . Find M B .
Example 5: Let B = 1, 1,
     
1,
    B
1 0 
0  1 0 0 
   

Solution: We have
1 1 1 1 1 1 1 a11   a11 
0 = a 1 + a 1 + a 0 = 1 1 0 a  = B a .
  11   21   31      12   21 
1 1 0 0 1 0 0 a13  a 31 
Here we write B for the matrix whose columns are the elements of B.
Since B is a linearly independent set, the columns of B are linearly
independent. So, B is invertible. Therefore
 a 11  1
a  = B −1 0.
 21   
a 31  1
Similarly, we can write down the equations
 a12  1  a13  0 
a  = B −1 1, a  = B −1 1.
 22     23   
a 32  0 a 33  0

77
Block 1 Solution of Nonlinear Equations in One Variable

 a11 a12 a13 

 M = a 21 a 22 a 23  = B −1B
B
B

a 31 a 32 a 33 
where B is the matrix formed by the rows of B. To find B −1B we set
up a 3 6 matrix C = [ B | B]. We then use row operations to reduce the
matrix formed by the first three column to identity. We get [ I | D] when
D = B −1B. So, we form the matrix
1 1 1 1 1 0
C = 1 1 0 0 1 1.
1 0 0 1 0 0
Carrying out row operations R 2 → R 2 − R1 , R 3 → R 3 − R1 , R 2  R 3 ,
R 2 → −R 2 , R 1 → R 1 − R 2 , R 3 → −R 3 , R 2 − R 2 − R 3 , we get
1 0 0 1 0 0 
C = 0 1 0 − 1 1 1 . (Notice that C is the RREF of C. )
0 0 1 1 0 − 1
1 0 0
B 
Therefore M B − 1 1 1 .
 1 0 − 1
***

 b11   b12   b1n  

    b  
b 21   b 22  
In general, if B =  , , ,  2n  
        
  
  b n1   b n 2    
  b nn  
and
   b12
 b11    b1 n  
      b  
b 21   b 22  
B , ,  ,  2 n  ,

         
   
bn1  bn 2   
  bnn  
then, as in Example ? we have M BB = B −1B. To compute M BB , we set up
the matrix [ B | B] and reduce this to RREF C = I n | D.
Then, M BB = D.

Try the exercises below to check your understanding of the example?

E17) For the following pairs of basis B and B, find the change of base
matrix M BB .
i) B = {(1, − 1), (2, 1)}, B{(1, 1), (1, 2)}.
ii) B{(1, − 1, 1), (−1, 0, 1), (1, 1, 1)}, B = {(1, − 1, 1), (1, 1, − 1), (0, 0, 1)}.

78
Unit 1 Errors and Approximations
Example 6: Let B = {(1, − 1, − 1), (1,1, 0), (1, 0, 1)} and
B = {(1, 1, − 1), (0, 1, 1), (0, 0, 1)}. Let T :  3 →  3 be a linear operator
 1 1 2
such that [T]B = − 1 1 0. Find [T ]B .
 0 1 1 

Solution: By Corollary to Theorem 10, we have [T]B = P −1[T]B P where

P = M BB . We compute M BB as before. We set
 1 1 1 1 0 0
C = − 1 1 0 1 1 0. Carrying out row operators
− 1 0 1 − 1 1 1
R
R 2 → R 2 + R1 , R 3 → R 3 + R1 , R 2 → 2 , R1 → R1 − R 2 , R 3 → R 3 − R 2 , we
2
1 0 0 0 − 1 / 2 0
get the RREF of C is C = 0 1 0 1 1 / 2 0.
0 0 1 − 1 1 / 2 1
 0 − 1 / 2 0
Therefore P =  1 1 / 2 0.
− 1 1 / 2 1
We row find P −1 using row reduction. We set
 0 − 1 / 2 0 1 0 0
A =  1 1 / 2 0 0 1 0.
− 1 1 / 2 1 0 0 1
Using row operations R1  R 2 , R 3 → R 3 + R1 ,
R
R 2 = (−2)R 2 , R1 → R1 − 2 , R 3 → R 3 − R 2 ,
2
1 0 0 1 1 0  1 1 0

We get 0 1 0 − 2 0 0.  P = − 2 0 0.
 −1

0 0 1 2 1 1  2 1 1

 1 1 0  1 1 2  0 1 / 2 0  0 2 2
−1       
[T]B = P [T]B P = − 2 0 0 − 1 1 0  1 1 / 2 0 =  2 − 4 − 4.
 2 1 1  0 1 1 − 1 1 / 2 1 − 1 5 5 

***

Here are some exercises for you to try.

E18) Let T :  2 →  2 be such that its matrix with respect to the

1 6 
standard basis is  . Find the matrix with respect to the basis
 2 2
{(−2, 1), (3, 2)}.

79
Block 1 Solution of Nonlinear Equations in One Variable
E19) Let T :  3 →  3 be such that its matrix with respect to the basis
1 − 1 2 
B = {(1, 0, 1), (−1, 0, 1), (0, 1, 1)} is 2 1 0. Find the matrix of the
 
0 1 1 
linear transformation with respect to the basis
B = {(1, − 1, 1), (1,1,1), (0, 0, 1)}.

Let us now recapitulate all that we have covered in this unit.

2.7 SUMMARY
We briefly sum up what has been done in this unit.

1. We defined matrices and explained the method of associating

matrices with linear transformations.

2. We showed what we mean by sums of matrices and multiplication

of matrices by scalars.

3. We proved that M mn (F) is a vector space of dimension mn over

F.

4. We defined the transpose of a matrix, the conjugate of a complex

matrix, the conjugate transpose of a complex matrix, a diagonal
matrix, identity matrix, scalar matrix and lower and upper triangular
matrices.

5. We defined the multiplication of matrices and showed its

connection with the composition of linear transformations. Some
properties of the matrix product were also listed and used.

6. The concept of an invertible matrix was explained.

7. We defined the matrix of a change of basis, and discussed the

effect of change of bases on the matrix of a linear transformation.

2.8 SOLUTIONS/ANSWERS
E1) Suppose B1 = {(1, 0, 1), (0, 2, − 1), (1, 0, 0)} and B2 = {(0, 1), (1, 0)}
Then T (1, 0, 1) = (1, 0) = 0.(0, 1) + .(1, 0)
T (0, 2, − 1) = (0, 2) = 2.(0, 1) + 0, (1, 0)
T(1, 0, 0) = (1, 0) = 0.(0, 1) + 1.(1, 0).
0 2 0
[T]BB12 =  
1 0 1

E2) B1 = {e1 , e 2 , e3}, B2 = {f1 , f 2 } are the standard bases (given in

Example 3).
T(e1 ) = T (1, 0, 0) = (1, 2) = f1 + 2f 2
80
Unit 1 Errors and Approximations
T(e 2 ) = T(0, 1, 0) = (2, 3) = 2f1 + 3f 2
T(e 3 ) = T(0, 0, 1) = (2, 4) = 2f1 + 4f 2
1 2 2
[T]BB12 =  .
 2 3 4

E3) T(1, 0, 0) = (1, 2) = 1.(1, 2) + 0.(2, 3)

T (0, 1, 0) = (2, 3) = 0.(1, 2) + 1.(2, 3)
T (1, − 2, 1) = (−1, 0) = 3(1, 2) − 2(2, 3)
1 0 3 
[T]BB12 =  .
0 1 − 2 

E4) Let B = {1, t , t 2 , t 3}. Then

D(1) = 0 = 0.1 + 0.t + 0.t 2 + 0.t 3
D( t ) = 1 = 1.1 + 0.t + 0.t 2 + 0.t 3
D( t 2 ) = 2t = 0.1 + 2.t + 0.t 2 + 0.t 3
D( t 3 ) = 3t 2 = 0.1 + 0.t + 3.t 2 + 0.t 3.
Therefore [D]B is the given matrix.

E5) We now that

T(e1 ) = f1
T (e 2 ) = f1 + f 2
T (e 3 ) = f 2
Therefore, for any ( x , y, z)  R 3
T( x, y, z) = T( xe1 + ye 2 + ze 3 ) = xT (e1 ) + yT(e 2 ) + zT (e 3 )
= xf1 + y(f1 + f 2 ) + zf 2 = ( x + y)f1 + ( y + z)f 2
= ( x + y, y + z )
That is, T : R 3 → R 2 : T( x, y, z) = ( x + y, y + z).

E6) We are given that

T (1) = 0.1 + 1.i = i
T(i) = (−1).1 + 0.1 = −1
, for any a + ib  C, we have
T (a + ib ) = aT (1) + bT(i) = ai − b

E7) B1 = {(1, 0), (0, 1)}, B2 = {(1, 0, 0), (0, 1, 0), (0, 0, 1)}
Now S(1, 0) = (1, 0, 0)
S(0, 1) = (0, 0, 1)
1 0
[S]B1 . B 2 = 0 0 , a 3 2 matrix
0 1
Again, T(1, 0) = (0, 1, 0)
T (0, 1) = (0, 0, 1)

81
Block 1 Solution of Nonlinear Equations in One Variable

0 0 
[T]B1 . B 2 = 1 0 , a 3 2 matrix
0 1
1 0 0 0 1 0
[S + T]B1 . B 2 = [S]B1 . B 2 + [T]B1 . B 2 = 0 0 + 1 0 = 1 0 , and
0 1 0 1 0 2
1 0  0 
[S]B1 . B 2 = [S]B1 . B 2 =  0 0 =  0 0  , for any   R.
0 1  0  

E8) Since dim M 2 3 (R) is 6, any linearly independent subset can have
6 elements, at most.

E9) Let ,    such that [1, 0] + [1, − 1] = [0, 0].

Then [ + , − ] = [0, 0]. Thus,  = 0,  = 0.
 the matrices are linearly independent.

E10) I : R n → R n : I( x1 ,  , x n ) = ( x1 ,  , x n ).
Then, for any basis B = {e1 ,  , e n } for R n , I(ei ) = ei .
1 0  0
0 1  0 
[I]B =  =I
    n
 
0 0  1 

E11) Since A is upper triangular, all its elements below the diagonal
are zero. Again, since A = A t , a lower triangular matrix, all the
entries of A above the diagonal are zero. , all the off-diagonal
entries of A are zero.  A is a diagonal matrix.

0 0 0  1 0 0
 
E12) [S]B = 1 0 0 , [T]B = 0 0 0
0 1 0 0 1 0
0 0 0 
[S]B [T]B = 1 0 0
0 0 0
0 0 0 
Also, [S  T] = 1 0 0 = [S]B[T]B
0 0 0

E13) Let A = diag (d1 ,  , d n ), B = diag (e1 ,  , e n ). Then

82
Unit 1 Errors and Approximations

d1 0 0  0 e1 0 0  0
0 d 0  0  0 e 0  0 
 2  2

AB =           
   
         
 0 0   d n   0 0   en 
d1 0 0  0 
0 d e 0  0 
 2 2

= 0 0 d 3e3  0 
 
     
 0 0 0  d n en 
= diag (d1e1 , d 2e 2 ,  , d n e n ).

E14) Firstly,  is a well defined map. Secondly, check that

( v1 + v 2 ) = ( v1 ) + ( v 2 ), and (v) = ( v) for v, v1 , v 2  V and
  F. Thirdly, show that ( v) = 0  0, that is  is 1 − 1. Then, by
Unit 5 (Theorem 10), you have shown that  is an isomorphism.

E15) We will show that its rows are linearly independent over .
R R
Carrying out row operations R 1 → 1 , R 2  R 3 , R 2 → 2 , we get
2 3
1 0 1 / 2 
0 1 0 .
 
0 0 1 
Since there are no zeros, the rows are linearly independent.
 the rows of the given matrix is linearly independent.

E16) Let B = {e1 , e2 , e3}, B = {f1 , f 2 , f3}. Then

f1 = 0e1 + 1e 2 + 0e3 = e 2
f 2 = e1 + e 2
f3 = e1 + 3e3
 B = {e 2 , e1 + e 2 , e1 + 3e3}.

E17) i) Looking at the elements of B and B as column vectors, we

 1 2 1 1  1 2 1 1
have B =   , B = 1 2 . Setting up C =  − 1 1 1 2
− 1 1     
1 0 − 1 / 3 − 1
and finding its RREF, we get C =  .
0 1 2 / 3 1 
− 1 / 3 − 1
Therefore M BB =  .
 2/3 1 

ii) Looking at the elements of B and B as column vectors we

 1 − 1 1  1 1 0
have B = − 1 0 1, B = − 1 1 0.
 
 1 1 1 − 1 − 1 1 83
Block 1 Solution of Nonlinear Equations in One Variable

1 0 0 1 1 0
Setting up C = 0 1 0 − 1 1 0 and finding its RREF,
0 0 1 − 1 − 1 1
1 0 0 1 / 2 − 1 / 2 1 / 4
we get C = 0 1 0 − 1 − 1 1 / 2. The row operations
0 0 1 − 1 / 2 1 / 2 1 / 4
are
R 2 → R 2 + R1 , R 3 → R 3 − R1 , R 2 = −R 2 , R1 → R1 + R 2 ,
R
R 3 → R 3 − 2R 2 , R 3 → 3 , R 1 → R 1 + 2R 3 .
4
 1 / 2 − 1 / 2 1 / 4
Therefore, M B =  − 1
B
− 1 1 / 2.
− 1 / 2 1 / 2 1 / 4

E18) We have B = {(1, 0), (0, 1)}, B = {( −2, 1), (3, 2)}.
−1
− 2 3 1 0 1 − 2 3
−1
P=B B=   =  .
 1 2  0 1  7  1 2 
−1
 1 − 2 3   − 2 3
P =  
−1
  =
 .
 7  1 2    1 2 
1 6 −1 1 − 2 3 1 6 − 2 3 1 − 2 3  4 15
P=  P = 7  1 2 2 2  1 2 = 7  1 2 − 2 10
 2 2       
1 − 14 0 
= 
7  0 35
 − 2 0
= .
 0 5

E19) We have [T]B = P −1[T]B P where P = B −1B.

 1 1 0 1 − 1 0
We form the matrix C = − 1 1 0 0 0 1. Row reducing using
 1 1 1 1 1 1
R
row operations R 2 → R 2 + R1 , R 3 → R 3 − R1 , R 2 → 2 , R1 → R1 − R 2
2
gives
1 0 0 1 / 2 − 1 / 2 − 1 / 2
C = 0 1 0 1 / 2 − 1 / 2 1 / 2 
0 0 1 0 2 1 
1 / 2 − 1 / 2 − 1 / 2
 P = 1 / 2 − 1 / 2 1 / 2 .
 0 2 1 
To find P −1 , we set up the matrix

84
Unit 1 Errors and Approximations

1 / 2 − 1 / 2 − 1 / 2 1 0 0
A = 1 / 2 − 1 / 2 1 / 2 0 1 0.
 0 2 1 0 0 1
R1
Row reducing using row operations R1 → 2R1 , R 2 → , R 2  R3,
2
R2 R R
R2 → , R → R1 + R 2 , R1 → R1 + 3 , R 2 → R 2 − 3 gives
2 2 2
1 0 0 3 / 2 1 / 2 1 / 2 3 / 2 1 / 2 1 / 2
0 1 0 1 / 2 − 1 / 2 1 / 2. P −1 = 1 / 2 − 1 / 2 1 / 2.
   
0 0 1 − 1 1 0   − 1 0 0 
 1 6 2 
−1 
[T]B = P [T]B P = − 1 / 2 7 / 2 3 / 2 .
 3 / 2 − 11 / 2 − 3 / 2

85
Block 3

MISCELLANEOUS EXAMPLES AND EXERCISES

Example 1: In this example, we revisit the projection operator and the
reflection operator. Let V be a vector space and W1 and W2 be
subspaces of V such that V = W1 ⊕ W2 . Then, we define the projection
of W1 along W2 as follows: Let v ∈ V. Then, v = w 1 + W2 , w 1 ∈ W1 and
w 2 ∈ W2 . Here w 1 and w 2 are uniquely determined by v ∈ V : we define
the projection PW1 , W2 by PW1 , W2 ( v) = w 1. We define the reflection of W1
along W2 by R W1 , W2 ( v) = W1 − w 2 . By the uniqueness of decomposition
of v as w 1 + w 2 , these functions are well defined. Once we fix the
subspaces W1 and W2 , we will drop the subscripts and simply write P
and R for the projection and reflection operators.
1) Check that P and R are linear operators.
2) Check that P 2 = P, R 2 = I, the identity operator.

Solution: 1) Let v, v′ ∈ V. Suppose v = w 1 + w 2 and v′ = w 1′ + w ′2 .

Then, P( v) = w 1 , P( v′) = w 1′ . We have V + v′ = ( w 1 + w 1′ ) + ( w 2 + w ′2 )
where w 1 + w 1′ ∈ W, and w 2 + w 2 ∈ W2 since W1 and W2 are
subspaces of V. Therefore, P( v + v′) = W1 + w 1′ = P( v) + P( v′). Let
α ∈ F, v ∈ V, v = W1 + w 2 . Then, P(V ) = W1 . We have
αv = αw 1 + αw 2 where αw 1 ∈ W1 and αw 2 ∈ W2 since W1 and W2
are subspaces. Therefore P(αv) = αW1 = αP( v). Therefore P is a
linear operator.

We have R ( v) = w 1 − w 2 , R (v′) = w ′1 − w ′2 .
R ( v + v′) = R ((w 1 + w 1′ ) + ( w 2 + w ′2 )) = ((w 1 + w 1′ ) − (w 2 + w ′2 ))
= ( w 1 − w ′1 ) + ( w 2 − w ′2 ) = R ( v) + R ( v′).
If α ∈ F, αv = (αw 1 + αw 2 ).
∴ R (αv) = αw 1 − αw 2 = α(w 1 − w 2 ) = αR ( v).
Therefore R is a linear operator.

2) We have P( v) = w 1 = w 1 + 0. P(P( v)) = P(w 1 + 0) = w 1 . Therefore

P 2 = P.
Also, R ( v) = w 1 − w 2 = w 1 + (− w 2 ).
R 2 ( v) = R 2 (w 1 + (− w 2 )) = w 1 − (− w 2 ) = w 1 + w 2 = v. In other words
R 2 ( v) = v for all v ∈ V and therefore R 2 = I.

Try the next exercise to check your understanding of the operators R

and P.

E1) Shwo that R = 2, P − I.

Let us suppoe that V = R 2 or R 3 so that we have a dot product defined

on V. Then, we call the decomposition V = W1 ⊕ W2 an orthogonal
87
Block 3
decomposition if w 1 ⋅ w 2 = 0 for all w 1 ∈ W , w 2 ∈ W2 . In this situation we
call P the orthogonal projection on W1.

For example, if we take W1 = [{(1, 0)}], W2 = [{(0, 1)}], any w 1 = W1 is of

the form ( x, 0) and any element w 2 ∈ W2 is of the for (0, x ). Therefore
w 1 ⋅ w 2 = 0. We can write ( x, y) ∈ R 2 as ( x, 0) + (0, y). So,
P(( x, y)) = ( x , 0) and R ((x , y)) = ( x, 0) − (0, y) = ( x, − y). Thus there are
the usual projection on the x -axis and reflection about x -axis.

Here is an exercise for you.

E2) Describe the projection on y -axis and reflection about y -axis as

orthogonal projection and reflection operators with respect to a
suitable choice of W1 and W2 .

E3) Let m ≠ 0 and W1 = {( x , y) ∈ R 2 | y = mx} and

1
W2 = {( x, y) ∈ R 2 | y = − x}.
m
i) Check that W1[{1, m)}], W2 [{−m, 1)}].
 (1, m) (− m, 1) 
ii) Prove that  ,  is an orthonormal basis for
 m +1 m +1 
2 2

R2.
iii) Prove that R 2 = W1 ⊕ W2 is an orthogonal decomposition of
R2.
iv) Compute projection and reflection operators with respect to
this decomposition.

Example 2: Let V be a vector space over a field F. Let T : V → V be a

linear operator and W ⊆ V be a subspace of V. We call W a T -
invariant subspace if Tw ∈ W for all w ∈ W.
i) Show that {0}, V, Ker (T) and R (T) are all T -invariant subspaces.
ii) If T : V → V is a linear operator, then every subspace of T is a T -
invariant subspace iff there is a λ ∈ F such that T(v) = λv for all
v ∈ V.

Solution: i) T(0) = 0 so, {0} is T -invariant.

If v ∈ Ker (T). We have T(v) = 0. So, T(T( v)) = T(0) = 0 or
T(v) ∈ Ker (T). Therefore Ker (T) is T -invariant.
If v ∈ R (T), Tv ∈ R (T) by the definition of R (T). So, R (T) is
T -invariant

ii) Suppose that there is a λ ∈ F such that T(v) = λv for all

v ∈ V. If v ∈ W, λv ∈ W since W is a subspace of V.

88
Block 3
Let us now prove the converse. Let v ≠ 0 be in V. Then,
W[{v}] is a subspace of V. Then, T(v) ∈ W, therefore
T( v) = λ 0 v. We claim that T( v) = λ 0 v for all v ∈ V. Let
v′ ∈ V, v′ ≠ v. If
v′ = αv, T( v′) = T(αv) = αT( v) = αλ 0 v = λ 0 (αv) = λ 0 v′.
So, suppose that there is no α such that v′ = αv.
Taking W = [{v′}], there is λ1 ∈ F such that T( v′) = λ1 v.
Consider u = v + v′. Then u ≠ 0; otherwise v′ = (−1)v, v′ = αv
with α = −1.
There is a λ 3 such that T(u ) = λ 3 u.
∴ λ 3 ( v + v′) = T(u ) = T( v + v′) = λ 0 v + λ1v′.
Since there is no α such that v′ = αv, v and v′ are linearly
independent. From λ 3 v + λ 3 v′ − λ 0 v + λ1 v′, we get λ1 = λ 0 = λ 3
and we are done.

SOLUTIONS/ANSWERS
E1) We have
2P( v) − I( v) = 2P(w 1 + w 2 ) − ( w 1 + w 2 ) = 2w 1 − w 1 − w 2 = w 1 − w 2 .

E2) We take W1{, y) | y ∈ R} and W2 = {( x , 0) | x ∈ R}. Then

P(( x, y)) = P((0, y) + (x , 0)) = (0, y) and
R ((x , y)) = (, y) − ( x, 0) = (− x, 0). These are, respectively the
projection on the y -axis and reflection about the y -axis.

E3) 1) We have W1 = {( x , mx ) | x ∈ R}. So, W1 = [{(1, m)}] ,

W2 = {(−mx , x ) | x ∈ R} therefore W2 = [{(−m, 1)}].

2) Computing the dot product, we get

(1, m) (−m, 1) − m + m
⋅ = = 0.
m2 +1 m2 +1 m2 +1
 (1, m) (− m, 1) 
So,  ,  is an orthonormal basis.
 m +1 m +1 
2 2

(1, m) (−m, 1)
3) ∈ W1 and ∈ W2 .
m2 +1 m2 +1
(1, m) (−m, 1)
Further, if α +β = 0. Taking dot products
2
m +1 m2 +1
(1, m)
sides with , we get
m2 +1
 1 ⋅1 + m ⋅ m   1 ⋅ ( − m ) + 1 ⋅ m 
α 2  + β 2
 = 0 or α = 0 .
 m + 1   (m + 1) 

89
Block 3
(− m, 1)
Similarly taking dot products both sides with , we get
m2 + 1
β = 0.
 (1, m) (− m, 1)  2
So,  ,  is a linearly independent set in R
 m +1 m +1 
2 2

and therefore forms a basis for R 2 . So, we can write any

 (1, m)   (−m, 1) 
v ∈ R 2 as v = α  + β
 
. Since

 m +1   m +1 
2 2

(1, m) (−m, 1)
∈ W1 , ∈ W2 , R 2 = W1 + W2 .
2 2
m +1 m +1
 (1, m)   (− m, 1) 
If v ∈ W1 ∩ W2 , v = α  = β
 
.

 m +1   m +1 
2 2

(1, m)
Taking dot product with both sides, we get
m2 + 1
 1 ⋅1 + m ⋅ m   1 ⋅ ( − m ) + m ⋅1 
α 2  = β  = 0 or α = 0. ∴ v = 0.
 m +1   m2 +1 
2
R = W1 ⊕ W2 .
α(1, m) β(−m, 1)
If w 1 ∈ W1 and w 2 ∈ W2 , we get w 1 = , w2 = .
m2 +1 m2 + 1
∴ w 1 ⋅ w 2 = 0 and R 2 = W1 ⊕ W2 is an orthogonal
decomposition.

4) If ( x, y) ∈ R 2 , we know that ( x, y) = (u ⋅ (x , y))u + ( v ⋅ ( x , y))v

(1, m) (−m, 1)
where u = ,v= . (see Section 2.5).
2
m +1 m2 +1
x + my y − mx
u ⋅ ( x , y) = , v ⋅ ( x , y) = .
m2 +1 m2 +1
 x + my   y − mx 
∴ ( x , y) =  u + 
 
 v is of the form w 1 + w 2 with

 m +1   m +1 
2 2

x + my y−m
w 1 ∈ W1 and w 2 ∈ W2 since u ∈ W1 and v ∈ W2 .
m2 +1 m21
x + my x + my (1, m)  x + my mx + m 2 y 
∴ P((x , y)) = u= =  2 , 
m2 + 1 m2 +1  m +1 m 2 + 1 
x + my (1, m) y − mx (−m, 1)
R ((x , y)) = w 1 − w 2 = − .
m 2 +1 m2 +1 m2 +1 m2 +1
 x + my mx + m 2 y   m 2 x − my y − mx 
=  2 , − , 2 
 (m + 1) m 2 + 1   m 2 + 1 m + 1 
 (1 − m 2 ) x + 2my (m 2 − 1) y + 2mx 
=  , .
 m2 +1 m2 +1 

90