Chapter 4.

Linear transformations
Lecture notes for MA1111

P. Karageorgis

[email protected]

1 / 22
Linear transformations

Definition 4.1 – Linear transformation

A linear transformation is a map T : V → W between vector spaces
which preserves vector addition and scalar multiplication. It satisfies
1 T (v1 + v2 ) = T (v1 ) + T (v2 ) for all v1 , v2 ∈ V and
2 T (cv) = cT (v) for all v ∈ V and all c ∈ R.

By definition, every linear transformation T is such that T (0) = 0.

Two examples of linear transformations T : R2 → R2 are rotations
around the origin and reflections along a line through the origin.
An example of a linear transformation T : Pn → Pn−1 is the derivative
function that maps each polynomial p(x) to its derivative p′ (x).
As we are going to show, every linear transformation T : Rn → Rm is
given by left multiplication with some m × n matrix.

2 / 22
Kernel and image

Definition 4.2 – Kernel and image

Suppose T : V → W is a linear transformation. Its kernel is the set of
all elements v ∈ V such that T (v) = 0 and its image is the set of all
elements w ∈ W that have the form w = T (v) for some v ∈ V .

Some textbooks refer to the image of T as the range of T .

When T : Rn → Rm is left multiplication by the matrix A, the kernel
is the null space of A and the image is the column space of A.

Theorem 4.3 – Dimension formula

Suppose T : V → W is a linear transformation. Then the kernel of T
is a subspace of V , the image of T is a subspace of W and

dim(ker T ) + dim(im T ) = dim V.

3 / 22
Injective and surjective

Definition 4.4 – Injective linear transformations

A linear transformation T : V → W is injective when T (x) = T (y) if
and only if x = y. This is the case if and only if ker T = {0}.

Suppose T : Rn → Rm is left multiplication by a matrix A. Then T is

injective if and only if the columns of A are linearly independent.

Definition 4.5 – Surjective linear transformations

A linear transformation T : V → W is surjective when im T = W .

Suppose T : Rn → Rm is left multiplication by a matrix A. Then T is

surjective if and only if the columns of A form a complete set of Rm .

4 / 22
Linear transformations and bases

Theorem 4.6 – Linear transformations and bases

Every linear transformation T : V → W is uniquely determined by the
values T (v1 ), T (v2 ), . . . , T (vn ) when v1 , v2 , . . . , vn form a basis of V .

Indeed, each v ∈ V can be expressed as a linear combination

v = x1 v1 + x2 v2 + . . . + xn vn

for some unique coefficients x1 , x2 , . . . , xn and this implies that

T (v) = x1 T (v1 ) + x2 T (v2 ) + . . . + xn T (vn ).

Using the computation above, one may also conclude that

im T = Span{T (v1 ), T (v2 ), . . . , T (vn )}.

This provides a simple formula for computing the image of T .

5 / 22
Linear transformations and bases: Example 1
Suppose that we are given the vectors
     
1 2 5
v1 = , v2 = , w=
1 1 7

and that we need to find T (w) when T : R2 → R2 is linear with

   
1 4
T (v1 ) = , T (v2 ) = .
2 7
To express w in terms of v1 and v2 , we use the row reduction
   
  1 2 5 1 0 9
v1 v2 w = −→ .
1 1 7 0 1 −2
This gives w = 9v1 − 2v2 , so we must also have
   
9−8 1
T (w) = 9T (v1 ) − 2T (v2 ) = = .
18 − 14 4
6 / 22
Linear transformations and bases: Example 2

Assume, as in the previous example, that T : R2 → R2 is linear with

       
1 4 1 2
T (v1 ) = , T (v2 ) = , v1 = , v2 = .
2 7 1 1

We now wish to determine T (x) for all x ∈ R2 . Proceeding as before,

we first express x as a linear combination of v1 and v2 . This gives
   
  1 2 x1 1 0 2x2 − x1
v1 v2 x = −→ .
1 1 x2 0 1 x1 − x2

In particular, x = (2x2 − x1 )v1 + (x1 − x2 )v2 , so we must also have

 
3x1 − 2x2
T (x) = (2x2 − x1 )T (v1 ) + (x1 − x2 )T (v2 ) = .
5x1 − 3x2

7 / 22
Linear transformations T : Rn → Rm

Theorem 4.7 – Linear transformations T : Rn → Rm

Every linear transformation T : Rn → Rm is given by left multiplication
with some m × n matrix A. To find this matrix explicitly, one uses the
fact that its ith column is equal to Aei = T (ei ).

In other words, T is given by left multiplication with the matrix

 
| | |
A = T (e1 ) T (e2 ) · · · T (en ) .
| | |
This theorem is easy to check when T (x) is known explicitly, say
    
ax1 + bx2 a b x1
T (x) = =⇒ T (x) = = Ax.
cx1 + dx2 c d x2
We are mostly interested in the case that T (x) is described implicitly
as a rotation around the origin, a reflection along a line, and so on.
8 / 22
Examples of linear transformations T : R2 → R2
 
1 0
Reflection along the x-axis . . . . . . . . . . . . . . . . . . . . . . . . . . . A =
0 −1
 
−1 0
Reflection along the y-axis . . . . . . . . . . . . . . . . . . . . . . . . . . . A =
0 1
 
0 1
Reflection along the line y = x . . . . . . . . . . . . . . . . . . . . . . . . A =
1 0
 
cos θ − sin θ
Rotation by θ around the origin . . . . . . . . . . . . . . . A =
sin θ cos θ
 
k 0
Scaling of the x-axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A =
0 1
 
1 0
Scaling of the y-axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A =
0 k

9 / 22
Linear transformations and coordinate vectors

Suppose T : V → W is a linear transformation between vector spaces.

Let v1 , v2 , . . . , vn be a basis of V and w1 , w2 , . . . , wm a basis of W .
P
Associating each element v = xi vi with its coordinate vector x,
one may relate V with R and one may similarly relate W with Rm .
n

T /W
V

f g

 T′

Rn / Rm

In particular, one may associate the linear transformation T : V → W

with a linear transformation T ′ : Rn → Rm . As we already know, the
latter is given by left multiplication with some m × n matrix A.
The matrix A obviously depends on the chosen bases. It is called the
matrix of T with respect to the given bases.
10 / 22
Matrix of a linear transformation

Definition 4.8 – Matrix of a linear transformation

Suppose T : V → W is a linear transformation between vector spaces.
Let v1 , v2 , . . . , vn be a basis of V and w1 , w2 , . . . , wm a basis of W .
The matrix of T with respect to these bases is defined as the matrix
whose ith column is equal to the coordinate vector of T (vi ).

Finding the matrix of T is a routine computation. To obtain the first

column, for instance, one needs to compute T (v1 ), express it in terms
of the vectors w1 , w2 , . . . , wm and keep track of the coefficients. The
other columns are obtained similarly by dealing with each T (vi ).
Suppose T : Rn → Rm is left multiplication with the matrix A. Then
the matrix of T with respect to the standard bases is equal to A, but
the matrix of T with respect to other bases might be different.
A linear transformation T : V → V is also known as a linear operator.
To study a linear operator, one usually introduces a single basis for V .
11 / 22
Matrix of a linear transformation: Example 1
Consider the derivative map T : P2 → P1 which is defined by

T (f (x)) = f ′ (x).

We already know from analysis that T is a linear transformation.

Let us use the basis 1, x, x2 for P2 and the basis 1, x for P1 . To find
the columns of the matrix of T , we compute T (1), T (x), T (x2 ) and
then express each of those in terms of 1, x. This gives

T (1) = 0 = 0 · 1 + 0 · x,
T (x) = 1 = 1 · 1 + 0 · x,
T (x2 ) = 2x = 0 · 1 + 2 · x.

The matrix of T with respect to the given bases is then

 
0 1 0
A= .
0 0 2
12 / 22
Matrix of a linear transformation: Example 2

Consider the function T : P2 → P2 which is defined by

T (f (x)) = f (x + 1).

This function is easily seen to be a linear transformation.

Using the standard basis 1, x, x2 for P2 , we find that

T (1) = 1 = 1 · 1 + 0 · x + 0 · x2 ,
T (x) = x + 1 = 1 · 1 + 1 · x + 0 · x2 ,
T (x2 ) = (x + 1)2 = 1 · 1 + 2 · x + 1 · x2 .

The matrix of T with respect to the standard basis is then

 
1 1 1
A = 0 1 2 .
0 0 1

13 / 22
Matrix of a linear transformation: Example 3

Consider the linear transformation T : R2 → R2 defined by

      
x1 17 −20 x1 17x1 − 20x2
T = = .
x2 12 −14 x2 12x1 − 14x2

Using the standard basis e1 , e2 of R2 , we find that

   
1 17
T (e1 ) = T = = 17 · e1 + 12 · e2 ,
0 12
   
0 −20
T (e2 ) = T = = −20 · e1 − 14 · e2 .
1 −14

In particular, the matrix of T with respect to the standard basis is

 
17 −20
A= .
12 −14

Note that T itself is given by left multiplication with this matrix.

14 / 22
Matrix of a linear transformation: Example 4
Let T : R2 → R2 be as in the previous example, namely
      
x1 17 −20 x1 17x1 − 20x2
T = = .
x2 12 −14 x2 12x1 − 14x2
We now compute the matrix of T with respect to the basis
   
5 4
v1 = , v2 = .
4 3
Expressing T (v1 ) and T (v2 ) in terms of v1 and v2 , we get
   
17 · 5 − 20 · 4 5
T (v1 ) = = = 1v1 + 0v2 ,
12 · 5 − 14 · 4 4
   
17 · 4 − 20 · 3 8
T (v2 ) = = = 0v1 + 2v2 .
12 · 4 − 14 · 3 6
 
′ 1 0
Thus, the matrix of T with respect to this basis is A = .
0 2
15 / 22
Matrix of a linear transformation: Example 5
Define the map T : R2 → R2 and the vectors v1 , v2 by letting
       
x1 x2 2 3
T = , v1 = , v2 = .
x2 x1 1 1
Then T is a linear transformation and v1 , v2 form a basis of R2 .
To find the matrix of T with respect to this basis, we need to express
   
1 1
T (v1 ) = , T (v2 ) =
2 3
in terms of v1 and v2 . We can always do this using the row reduction
   
  2 3 1 1 1 0 5 8
v1 v2 T (v1 ) T (v2 ) = −→ .
1 1 2 3 0 1 −3 −5
Thus, the matrix of T with respect to the given basis is
 
5 8
A= .
−3 −5
16 / 22
Matrix of a linear transformation: Explicit formulas

1 Suppose B2 is an m × m matrix whose columns form a basis of Rm .

To find the coordinate vector of w with respect to this basis, one
needs to solve the system B2 x = w and the solution is x = B2−1 w.
2 Suppose T : Rn → Rm is left multiplication with the m × n matrix A.
Let B1 be a matrix whose columns form a basis of Rn and let B2 be
a matrix whose columns form a basis of Rm . Then the matrix of T
with respect to these bases is given by B2−1 AB1 .
3 Suppose T : Rn → Rm is left multiplication with the m × n matrix A.
Then the matrix of T with respect to the standard bases is A.

A
RO n / Rm Vectors

B1 B2−1

 
Rn / Rm Coordinate vectors
17 / 22
Reflection along a line through the origin
Suppose T : R2 → R2 is reflection along a line through the origin. To
study this linear transformation, we choose the basis v1 , v2 so that v1
lies on the line of reflection and v2 is perpendicular to this line.
The two chosen vectors are such that T (v1 ) = v1 and T (v2 ) = −v2 .
Thus, the matrix of T with respect to the given basis is
 
1 0
A= .
0 −1
If one uses the standard basis, instead, then the matrix of T becomes
 
′ cos(2θ) sin(2θ)
A = ,
sin(2θ) − cos(2θ)
where θ is the angle between the line of reflection and the x-axis.
This example illustrates that the matrix of a linear transformation may
turn out to be very simple, if the basis is suitably chosen. In fact, we
ended up with the exact same matrix for any reflection whatsoever.
18 / 22
Similar matrices

Definition 4.9 – Similar matrices

Two n × n matrices A, C are called similar, if there exists an invertible
matrix B such that C = B −1 AB. Similar matrices represent the same
linear transformation with respect to different bases.

Suppose that A is similar to C and that A2 = 0. Then we also have

C 2 = B −1 AB · B −1 AB = B −1 A2 B = 0.

Suppose that A is similar to C and that A2 = In . Then we also have

C 2 = B −1 AB · B −1 AB = B −1 A2 B = B −1 B = In .

Similar matrices have many properties in common. For instance, they

have the same determinant, and their column/null spaces have the
same dimension. We shall obtain a more thorough list of common
properties when we study similar matrices in MA1212.
19 / 22
Powers of square matrices
If two square matrices A, C are similar, then one easily finds that

C = B −1 AB =⇒ C k = B −1 Ak B.

Thus, powers of one matrix may be related to powers of the other.

Suppose now that C is a diagonal matrix. Then powers of C are easy
to compute because an induction argument gives
   k 
a1 a1
 a2 
k
 ak2 
C= =⇒ C =  .
   
. .  . .
 .   . 
an k
an

One may thus compute powers of A by noting that Ak = BC k B −1 .

More generally, one may compute powers of a matrix A, if one can
find a similar matrix C whose powers are easy to determine. We are
going to develop this approach in much more detail in MA1212.
20 / 22
Recursive relations involving two terms
A recursive relation expresses each term of a sequence x1 , x2 , . . . as a
function of the previous terms. One generally prefers to have a closed
formula, namely one that expresses xn as a function of n alone.
For instance, suppose that the sequences xn , yn are such that
 
xn = 3xn−1 + 2yn−1
.
yn = 2xn−1 + 3yn−1
Letting un denote the vector of unknowns, one can then write
    
xn 3 2 xn−1
un = =⇒ un = = Aun−1
yn 2 3 yn−1
for some 2 × 2 matrix A. This equation is easily seen to imply that
un = Aun−1 = A2 un−2 = . . . = An u0 .
One may thus obtain a closed formula by computing the powers An
of the matrix A. We shall learn how to do this in MA1212.
21 / 22
Recursive relations involving three or more terms

A recursive relation may involve several consecutive terms such as

xn+3 = xn+2 + 5xn+1 + 3xn .

This relation can be handled exactly as before, although one needs to

keep track of three terms in order to determine the next one.
Letting un consist of three consecutive terms, we can now write
    
xn+1 0 1 0 xn
un = xn+2  =⇒ un = 0 0 1 xn+1  = Aun−1
xn+3 3 5 1 xn+2
for some 3 × 3 matrix A. This equation is easily seen to imply that

un = Aun−1 = A2 un−2 = . . . = An u0 .

Once we are able to compute powers of the matrix A, we can then

obtain a closed formula for un and also one for the sequence xn .
22 / 22