MATH 1025: Applied Linear Algebra

Chapter 6: Linear Transformations

Paul Skoufranis
Fall 2025
a
a
a
a
a
Section 1: Linear Transformations

Our next goal is to understand various geometric transformations of Rn .

These transformations are specific functions on vectors that produce
vectors. The nicest of such transformations are those that preserve what
we can do to vectors: namely vector addition and scalar multiplication.
Definition
A map (function, transformation, etc.) T : Rn → Rm is said to be linear if
1 (T preserves vector addition)
T (⃗v1 + ⃗v2 ) = T (⃗v1 ) + T (⃗v2 ) for all vectors ⃗v1 and ⃗v2 in Rn , and
2 (T preserves scalar multiplication)
T (α⃗v ) = αT (⃗v ) for all vectors ⃗v in Rn and scalars α.

Since linear maps are exactly the maps that preserve vector addition and
scalar multiplication, they are the natural maps to study in linear algebra.
Note the properties of linear maps imply the preserve linear combinations.
Theorem 1
Let T : Rn → Rm be linear. Then
1 T (⃗0) = ⃗0, and
2 T (α1 ⃗v1 + α2 ⃗v2 + · · · + αk ⃗vk ) = α1 T (⃗v1 ) + α2 T (⃗v2 ) + · · · + αk T (⃗vk )
for all scalars α1 , α2 , . . . , αk and for all vectors ⃗v1 , ⃗v2 , . . . , ⃗vk in Rn .
Example 1
The zero linear map from Rn to Rm is the linear map T : Rn → Rm
defined by
T (⃗v ) = (0, 0, . . . , 0)
for all vectors ⃗v in Rn .

Example 2
The identity map on Rn is the linear map T : Rn → Rn defined by

T (⃗v ) = ⃗v

for all vectors ⃗v in Rn . We denote the identity map on Rn by IdRn .

We can verify these are linear maps using the same ideas as in the
following example.
Example 3
Let α be a scalar. Define Mα : Rn → Rn by

Mα (⃗v ) = α⃗v

for all vectors ⃗v in Rn . Then Mα is a linear map. If α > 1, we call Mα the

dilation map by α. If 0 < α < 1, we call Mα the contraction map by α.
Example 4
Define T : R2 → R2 by T ((x, y )) = (x 2 , y ) for all (x, y ) in R2 . Is T a
linear map?

Example 5
Define T : R2 → R2 by T ((x, y )) = (xy , 0) for all (x, y ) in R2 . Is T a
linear map?

Example 6
Define T : R2 → R2 by T ((x, y )) = (x + 2, y + 1) for all (x, y ) in R2 . Is
T a linear map?
Of course, we have already seen some examples of linear maps in this
course.
Example 7
Let A be an m × n matrix and define LA : Rn → Rm by

LA (⃗x ) = A⃗x

for all (column) vectors ⃗x in Rn . Then LA is a linear map known as left

multiplication by A.
Example 8
⃗ be a non-zero vector in Rn and define Pw⃗ : Rn → Rn by
Let w

Pw⃗ (⃗v ) = projw⃗ (⃗v )

for all vectors ⃗v in Rn . Then Pw⃗ is a linear map.

Example 9
For any 0 ≤ θ < 2π define Rθ : R2 → R2 by

Rθ ((x, y )) = (x cos(θ) − y sin(θ), x sin(θ) + y cos(θ)).

for all (x, y ) in R2 . Then Rθ is a linear map. In fact, Rθ is the

counter-clockwise rotation in R2 by θ about (0, 0).
Linear maps behave well with respect to function operations.
Theorem 2
Let T : Rn → Rm and let S : Rm → Rk be linear maps. Then the
composition of S and T , namely S ◦ T : Rn → Rk , is linear.
Recall a function f : X → Y is said to be invertible if there is a function
g : Y → X such that
g (f (x)) = x for all x in X , and
f (g (y )) = y for all y in Y .
It can be shown that a function is invertible exactly when it is one-to-one
(injective) and onto (surjective).
Theorem 3
Let T : Rn → Rm be an invertible linear map. Then T −1 : Rm → Rn is a
linear map.

Proof.
To see that T −1 is a linear map, let ⃗v1 and ⃗v2 be vectors in Rm and let α
be a scalar. Let

⃗ 1 = T −1 (⃗v1 )
w and ⃗ 2 = T −1 (⃗v2 ).
w

Then
w1 ) = ⃗v1
T (⃗ and w2 ) = ⃗v2 .
T (⃗
Proof Continued.
Since T is linear, we know that

T (⃗ ⃗ 2 ) = T (⃗
w1 + w w2 ) = ⃗v1 + ⃗v2 and
w1 ) + T (⃗
⃗ 1 ) = αT (⃗
T (αw w1 ) = α⃗v1 .

By applying T −1 to both sides of these equations, we obtain that

T −1 (⃗v1 + ⃗v2 ) = T −1 (T (⃗ ⃗ 2 )) = w
w1 + w ⃗ 2 = T −1 (⃗v1 ) + T −1 (⃗v2 )
⃗1 + w
T −1 (α⃗v1 ) = T −1 (T (αw ⃗ 1 = αT −1 (⃗v1 ).
⃗ 1 )) = αw

Hence T −1 is linear.
You can now do the following homework exercises:
Assignment 4: Questions 10, 11, 14
Recommended Textbook (2017 Edition) Exercises:
Section 5.1: Exercises 1, 2
Section 5.2: Exercise 9
Section 2: Application -
Kernel of Linear Differential Equations
The notion of linear maps can be extended beyond vectors in Rn .

Consider the transformation T on twice differentiable functions defined by

T (f ) = f ′′ + 2f ′ + 5f .
Note T is a linear transformation since
T (f + g ) = (f + g )′′ + 2(f + g )′ + 5(f + g )
= (f ′′ + g ′′ ) + (2f ′ + 2g ′ ) + (5f + 5g )
= (f ′′ + 2f ′ + 5f ) + (g ′′ + 2g ′ + 5g )
= T (f ) + T (g )
and
T (αf ) = (αf )′′ + 2(αf )′ + 5(αf )
= αf ′′ + 2αf ′ + 5αf
= αT (f ).
Note the null space (kernel) of T is then all twice differentiable functions
such that T (f ) = 0; that is, all twice differentiable functions such that

f ′′ (x) + 2f ′ (x) + 5f (x) = 0.

Hence the solutions to this homogeneous second-order linear differential

equation are exactly the null space of a linear map!

One can use linear map theory to show that the null space of T is
2-dimensional therefore completing the argument from Chapter 5 that the
solutions to this homogeneous second-order linear differential equation are

ae x cos(2x) + be x sin(2x)

where a and b are scalars.

You can now do the following homework exercises:
Assignment 4: Question 9
There are no Recommended Textbook (2017 Edition) Exercises on this
section.
Section 3: Linear Transformations and Bases
Bases are powerful tools to understand linear transformations.

Let T : Rn → Rm be a linear map. If ⃗v1 , ⃗v2 , . . . , ⃗vn is a basis for Rn , then

T is completely determined by the values of T (⃗v1 ), T (⃗v2 ), . . . , T (⃗vn ).

Indeed let ⃗v be any vector in Rn . To determine the value of T (⃗v ), recall

since ⃗v1 , ⃗v2 , . . . , ⃗vn is a basis for Rn that there exists unique scalars
α1 , α2 , . . . , αn such that
⃗v = α1 ⃗v1 + α2 ⃗v2 + · · · + αn ⃗vn .
Hence
T (⃗v ) = T (α1 ⃗v1 + α2 ⃗v2 + · · · + αn ⃗vn )
= α1 T (⃗v1 ) + α2 T (⃗v2 ) + · · · + αn T (⃗vn ).
Thus the value of T (⃗v ) is determined by the coefficients of ⃗v in the
unique linear combination of ⃗v1 , ⃗v2 , . . . , ⃗vn that produces ⃗v and the values
T (⃗v1 ), T (⃗v2 ), . . . , T (⃗vn ).
Example 10
Suppose T : R2 → R3 is a linear map such that T ((3, 2)) = (5, 7, 1) and
T ((2, 1)) = (2, 4, 3). Can we compute T ((5, 4))?
Example 11
Is there a linear map T : R2 → R2 such that T ((1, 0)) = (2, 3),
T ((0, 1)) = (1, −1), and T ((3, 2)) = (5, 7)?
Using the idea of bases, we can construct the linear map that reflects
vectors in R2 across a line through the origin.

Consider the line ax + by = 0 in R2 where either a ̸= 0 or b ̸= 0. We can

see that a direction vector of this line is (−b, a). Furthermore, clearly
(a, b) is orthogonal to (−b, a). Therefore, as we have two orthogonal
vectors in R2 , they form a basis for R2 .

(a, b)

(−b, a) (−b, a)
R

−(a, b)
ax + by = 0 ax + by = 0

We want the linear map R : R2 → R2 such that

R((−b, a)) = (−b, a) and R((a, b)) = −(a, b).
Such a linear map will be the reflection across the line ax + by = 0 since
the reflection of the vector ⃗v = c(−b, a) + d(a, b) across ax + by = 0 is
c(−b, a) − d(a, b) = R(⃗v ).

d(a, b)
c(−b, a)
⃗v
−d(a, b)
c(−b, a)
R(⃗v )
R

ax + by = 0 ax + by = 0
If we want to determine R((x, y )), we need to determine the values of c
and d such that
(x, y ) = c(−b, a) + d(a, b).
Solving this system of linear equations in c and d yields
−bx + ay ax + by
c= and d= .
a2 + b 2 a2 + b 2
Alternatively, as we claim this is the answer, we just need to substitute
back in to check it is the answer; that is, we can easily check that
−bx + ay ax + by
(−b, a) + 2 (a, b) = (x, y ).
a2 + b 2 a + b2
Thus, as R is linear, for any (x, y ) in R2 we see that
R((x, y )) = R(c(−b, a) + d(a, b))
= cR((−b, a)) + dR((a, b))
−bx + ay ax + by
= 2 (−b, a) − 2 (a, b)
a + b2 a + b2
1
b 2 − a2 x − 2aby , −2abx + a2 − b 2 y .




= 2 2
a +b
Thus we have completely described the linear map that reflects across the
line ax + by = 0. We will denote this linear map by R(a,b) .
Example 12
Find the reflection of the point (2, 1) across the line 3x + 4y = 0.
You can now do the following homework exercises:
Assignment 4: Question 13
There are no recommended Textbook (2017 Edition) Exercises yet!
Section 4: Linear Transformation and Matrices

As we have just seen, if we can understand a linear map in terms of its

action on any basis we choose, we can understand the linear map.

We have a very nice basis; namely the standard basis e⃗1 , e⃗2 , . . . , e⃗n where

e⃗k = (0, . . . , 0, |{z}

1 , 0, . . . 0).
k th spot

Thus we can understand every linear map in terms of its action on the
standard basis.

Viewing vectors as columns again, what does this give us?

Let T : Rn → Rm be any fixed linear map. We know that the value of T
on any vector in Rn is determined by the values of T (⃗
e1 ), T (⃗
e2 ), . . .,
T (⃗
ek ). In particular, if we know
 
a1,j
 a2,j 
T (⃗
ej ) =  . 
 
.
 . 
am,j

for all 1 ≤ j ≤ n, then we know what T does to any vector. In particular,

     Pn 
x1   a1,j j=1 a1,j xj
 a2,j   nj=1 a2,j xj 
n n n
P
x2  X X X
T  .  = T  xj e⃗j  = xj T (⃗
ej ) = xj  .  =  .
     
..
 ..   ..   . 
j=1 j=1 j=1 Pn
xn am,j j=1 am,j xj

Wait a second! That vector on the right looks familiar!

What we have just showed is that if we take the matrix A whose columns
are T (⃗
ej ), then
T (⃗x ) = A⃗x
for every vector ⃗x . That is, the collection of linear map from Rn to Rm are
exactly the the left multiplication by an m × n matrix maps!
Definition
Let T : Rn → Rm be a linear map. The matrix of T , denoted [T ], is the
m × n matrix whose j th column is T (⃗ej ) where e⃗1 , e⃗2 , . . . , e⃗n is the
n
standard basis of R ; that is
 
↑ ↑ ↑
[T ] = T (⃗ e2 ) · · · T (⃗
e1 ) T (⃗ en ) .
↓ ↓ ↓

The matrix of T is defined so that

T (⃗x ) = [T ]⃗x .
This is the true reason why we defined a matrix times a vector the way we
did!
Example 13
Let T : Rn → Rm be the zero linear map; that is

T (⃗v ) = ⃗0

for all vectors ⃗v in Rn . Thus [T ] is the m × n zero matrix!

Example 14
Let T : Rn → Rn be the identity map; that is T (⃗v ) = ⃗v for all vectors ⃗v in
Rn . Then T (⃗
ej ) = e⃗j for all 1 ≤ j ≤ n so
 
1 0 ··· 0
.. 
0 . . . . . .

.
[T ] = 
 .. . . ..
.

. . . 0
0 ··· 0 1

That is, [T ] = In ; the n × n identity matrix.

Example 15
Recall for a scalar α that Mα : Rn → Rn is the linear map defined by
Mα (⃗v ) = α⃗v for all vectors ⃗v in Rn . What is the matrix of Mα ?
Example 16
Let w⃗ = (a, b) be a non-zero vector in R2 . Recall Pw⃗ : R2 → R2 is the
linear map defined by
Pw⃗ (⃗v ) = projw⃗ (⃗v )
for all vectors ⃗v in R2 . What is the matrix of Pw⃗ ?
Example 17
For any 0 ≤ θ < 2π, recall the counter-clockwise rotation in R2 by θ about
(0, 0) is the linear map Rθ : R2 → R2 defined by

Rθ ((x, y )) = (x cos(θ) − y sin(θ), x sin(θ) + y cos(θ)).

for all (x, y ) in R2 . What is the matrix of Rθ ?

Example 18
Recall given a line ax + by = 0 with either a ̸= 0 or b ̸= 0 that the
reflection across ax + by = 0 is given by the linear map R(a,b) : R2 → R2
defined by
1
b 2 − a2 x − 2aby , −2abx + a2 − b 2 y




R(a,b) ((x, y )) =
a2 + b2

for all (x, y ) in R2 . What is the matrix of R(a,b) ?

Let T : Rn → Rm and S : Rm → Rp be linear maps. We have seen that
S ◦ T : Rn → Rp is a linear map. Note we needed the codomain of T to
be equal to the domain of S for the composition to make sense.

What is the matrix of S ◦ T ?

To find [S ◦ T ], recall [S] = [ai,j ] = A is a p × m matrix and

[T ] = [bi,j ] = B is an m × n matrix. First, we know for all
 
x1
x2 
⃗x =  . 
 
 .. 
xn
that  Pn 
j=1 b1,j xj
 n b2,j xj 
P
 j=1
T (⃗x ) = B⃗x = 

.. 

Pn . 
j=1 bm,j xj
since B is m × n.
Then, by the definition of composition, we see that
P P 
m n
 Pn a b x
Pk=1 1,k Pj=1 k,j j 

b1,j xj
Pj=1
n m n
 j=1 b2,j xj   k=1 a2,k j=1 bk,j xj 
   
(S ◦ T )(⃗v ) = S(T (⃗v )) = A  .. =
..

.


Pn
  . 
j=1 bm,j xj
P P 
m n
k=1 ap,k j=1 bk,j xj
 Pn Pm 
j=1 ( k=1 a1,k bk,j ) xj
 n ( m a2,k bk,j ) xj 
P P
 j=1 k=1
= .

..

Pn Pm . 
j=1 ( k=1 ap,k bk,j ) xj

So, if C is the p × n matrix whose (i, j)th entry is m

P
k=1 ai,k bk,j , then

[S ◦ T ] = C = AB = [S][T ].
That is, the true reason matrix multiplication was defined the way it was
to make the composition of linear maps work out! Moreover, this is why it
only makes sense to multiply a p × q matrix with and m × n matrix with
q = m!
If T : Rn → Rm is an invertible linear map, what can we say about
T −1 : Rm → Rn ?

Well, since T −1 ◦ T = IdRn and T ◦ T −1 = IdRm , we see that

In = [IdRn ] = [T −1 ◦ T ] = [T −1 ][T ]
Im = [IdRm ] = [T ◦ T −1 ] = [T ][T −1 ].

That is, [T −1 ] must be the two-sided inverse of [T ].

Note if [T ] is invertible, then T must be invertible as the left

multiplication by [T ]−1 will be the inverse of T .

Hence a linear map T : Rn → Rm is invertible exactly when [T ] is

invertible.

This can be used to show many of the facts in our theorem about
invertible matrices.
You can now do the following homework exercises:
Assignment 4: Questions 15, 16
Recommended Textbook (2017 Edition) Exercises:
Section 5.2: Exercises 1, 8, 11-13
Section 5.3: Exercises 2-7
Section 5.4: Exercises 1-16
Section 5: Change of Basis
Recall when we constructed R(a,b) , the reflection across the linear
ax + by = 0, it was more convenient to use the basis (a, b) and (−b, a)
since the geometry of the reflection was clear for these two vectors.

It was then a bit of a non-trivial exercise to describe what R(a,b) did to an

arbitrary vector (x, y ).

Is there a simpler way to solve such problems? Is there a way to change

between bases? Such techniques are very important in physics.

For example, what if we wanted to describe every other vector as a linear

combination of (2, 1) and (−1, 2)?
d(−1, 2)
⃗v
c(2, 1)
Recall if ⃗v1 , ⃗v2 , . . . , ⃗vn is a basis for Rn , then for any vector ⃗v in Rn there
are unique scalars α1 , α2 , . . . , αn such that

⃗v = α1 ⃗v1 + α2 ⃗v2 + · · · + αn ⃗vn .

Recall α1 , α2 , . . . , αn are called the coordinates of ⃗v with respect to the

basis ⃗v1 , ⃗v2 , . . . , ⃗vn and they let us know how much of each basis vector we
need to use!

Of course, if
⃗v = (x1 , x2 , . . . , xn ),
then we could just solve a system of linear equations find the values of
α1 , α2 , . . . , αn . But what if we needed to find the coordinates of multiple
vectors ⃗v ? We definitely don’t want to row reduce more than we need to!
Writing the vectors as columns, by taking a similar view to the ‘matrix of a
linear map’ idea we had earlier, we see that
   
x1   α1
x2  ↑ ↑ ↑ α 
 2
 ..  = ⃗v = α1 ⃗v1 + α2 ⃗v2 + · · · + αn ⃗vn = ⃗v1 ⃗v2 · · · ⃗vn   . 
 
.  .. 
↓ ↓ ↓
xn αn

Let  
↑ ↑ ↑
Q = ⃗v1 ⃗v2 · · · ⃗vn  .
↓ ↓ ↓
Note if we multiply the ⃗v1 , ⃗v2 , . . . , ⃗vn coordinates by Q then we get our
standard coordinate system.

Note since ⃗v1 , ⃗v2 , . . . , ⃗vn is a basis for Rn , the columns of Q are a basis for
Rn so col(Q) = Rn . Hence Q is invertible!
Then    
x1 α1
x2  α2 
Q −1  .  =  . 
   
 ..   .. 
xn αn
we see that multiplying the standard coordinates of a vector by Q −1 gives
us the ⃗v1 , ⃗v2 , . . . , ⃗vn coordinates!
Definition
Let B be the basis ⃗v1 , ⃗v2 , . . . , ⃗vn for Rn . The matrix
 
↑ ↑ ↑
Q = ⃗v1 ⃗v2 · · · ⃗vn 
↓ ↓ ↓

is called the B-to-standard change of basis matrix and the matrix Q −1 is

called the standard-to-B change of basis matrix.
Example 19
Consider the basis (1, 2) and (3, −1) for R2 . Find the coordinates of (3, 4)
with respect to this basis.
If we only know what a linear map does to a non-standard basis, we can
still determine the matrix of the linear map.
Theorem 4
Let T : Rn → Rm be a linear map and let B be the basis ⃗v1 , ⃗v2 , . . . , ⃗vn for
Rn . Then  
↑ ↑ ↑
[T ] = T (⃗v1 ) T (⃗v2 ) · · · T (⃗vn ) Q −1
↓ ↓ ↓
where Q −1 is the standard-to-B change of basis matrix; that is
 −1
↑ ↑ ↑
Q −1 = ⃗v1 ⃗v2 · · ·
 ⃗vn  .
↓ ↓ ↓
Example 20
Let T : R2 → R3 be the linear map where T ((3, 2)) = (5, 7, 1) and
T ((2, 1)) = (2, 4, 3). Compute the matrix of T .
Example 21
Let a and b be scalars with a ̸= 0 or b ̸= 0 so that (a, b) and (−b, a) form
a basis of R2 . Let T : R2 → R2 be the linear map where

T ((a, b)) = −(a, b) and T ((−b, a)) = (−b, a).

Compute the matrix of T .

Example 22
Let B be the basis ⃗v1 , ⃗v2 , . . . , ⃗vn for Rn . Suppose T : Rn → Rn is a linear
map such that
T (⃗vk ) = λk ⃗vk
for some scalars λ1 , λ2 , . . . , λn . Show that
 
λ1 0 · · · · · · 0
 0 λ2 . . .
 .. 
 .
 .. . . . . . . .
.  −1
[T ] = Q  . . . . . Q
 ..
 
..
. λn−1 0 

.
0 ··· ··· 0 λn

where Q is the B-to-standard change of basis matrix

 
↑ ↑ ↑
Q = ⃗v1 ⃗v2 · · · ⃗vn  .
↓ ↓ ↓
Example 23
Let B be the basis ⃗v1 , ⃗v2 , ⃗v3 for R3 . Suppose T : R3 → R3 is a linear map
such that

T (⃗v1 ) = 2⃗v1 , T (⃗v2 ) = ⃗v1 + 2⃗v2 , and T (⃗v3 ) = ⃗v2 + 2⃗v3 .

Show that  
2 1 0
[T ] = Q 0 2 1 Q −1
0 0 2
where Q is the B-to-standard change of basis matrix
 
↑ ↑ ↑
Q = ⃗v1 ⃗v2 · · · ⃗vn  .
↓ ↓ ↓
The last examples are essential decomposition forms for n × n matrices.

Decomposing matrices to be in the form Example 22 will be the focus of

Chapter 7. Unfortunately, it is not always possible to decompose matrices
in this form.

Decomposing matrices in a similar form to that in Example 23 is what one

does when Chapter 7 fails. The difficulty of doing this and showing why
one can makes this subject matter beyond this course.
You can now do the following homework exercises:
Assignment 4: Question 12
Recommended Textbook (2017 Edition) Exercises:
Section 5.2: Exercises 3-7
Section 5.8: Exercises 1, 2
 TEST 3
will cover all material up to here!
Homework Summary: From this Chapter of the Lecture Notes, you can do
Assignment 4: Questions 9-16