Compositions of linear

transformations
We’ve already talked about a linear transformation as transforming
vectors in a set X to vectors in a set Y. For instance, we can say that the
set X is a subset of ℝn and the set Y is a subset of ℝm.

We’ll also say that S is a linear transformation that transforms vectors from
X to Y, S : X → Y. But let’s then say that we want to take vectors from the
subset Y and transform them into vectors in a subset Z, which is contained
in ℝp space. If T is a linear transformation that transforms vectors from Y to
Z, T : Y → Z,

then we can start to think about a composition of transformations.

Because at this point, if we want to transform directly from X to Z, we can
do so using a composition of the transformations S and T. S will take us

299
from X to Y, and then T will take us from Y to Z, so a composition of T with S
will take us all the way from X to Z.

For instance, let’s say we have a vector x ⃗ that lies in the subset X, and we
want to transform it into Z. We could first transform it into Y using the
transformation S : X → Y, and we’d get S( x ⃗ ). Then we could transform this
transformed vector from Y into Z using the transformation T : Y → Z, and
we’d get T(S( x ⃗ )). We can also write T(S( x ⃗ )) as T ∘ S( x ⃗ ), so

T∘S:X→Z

is the composition of T with S.

Compositions as linear transformations

If we know that the transformations S and T are linear transformations,

then we can say that the composition of the transformations T(S( x ⃗ )) is also
a linear transformation. And we know this is true, because we can see that
the composition is closed under addition,

T ∘ S( x ⃗ + y ⃗ ) = T(S( x ⃗ + y ⃗ ))

T ∘ S( x ⃗ + y ⃗ ) = T(S( x ⃗ ) + S( y ⃗ ))

T ∘ S( x ⃗ + y ⃗ ) = T(S( x ⃗ )) + T(S( y ⃗ ))

T ∘ S( x ⃗ + y ⃗ ) = T ∘ S( x ⃗ ) + T ∘ S( y ⃗ )

and closed under scalar multiplication.

300
 T(S(c x ⃗ )) = T(S(c x ⃗ ))

T(S(c x ⃗ )) = T(cS( x ⃗ ))

T(S(c x ⃗ )) = cT(S( x ⃗ ))

Additionally, because S and T are linear transformations, they can each be

written as a matrix-vector product.

Let’s say the transformation S is given by the matrix-vector product

S( x ⃗ ) = A x ,⃗ where A is an m × n matrix, and the transformation T is given by
the matrix-vector product T( x ⃗ ) = B x ,⃗ where B is a p × m matrix. Then
because the composition is also a linear transformation, it can also be
written as a matrix-vector product, and that matrix-vector product is

T ∘ S( x ⃗ ) = T(S( x ⃗ )) = T(A x ⃗ ) = BA x ⃗ = C x ⃗

where C is a p × n matrix.

Looking at this equation, we can see that as long as we can find the
matrix-matrix product BA, we’ll be able to find the transformation of any
vector x ⃗ that’s in the subset X, transformed all the way through the subset
Y and into the subset Z, simply by multiplying the matrix-matrix product BA
by the vector x ⃗ that we want to transform.

Example

The transformation S : X → Y transforms vectors in the subset X into

vectors in the subset Y. The transformation T : Y → Z transforms vectors in
the subset Y into vectors in the subset Z. The subsets X, Y, and Z are all in

301
ℝ2. Find a matrix that represents the composition of the transformations
T ∘ S, and then use it to transform x ⃗ in X into its associated vector z ⃗ in Z.

[x1 −x2]
3x1
S( x ⃗ ) =

[ −x2 ]
2x1 + 4x2
T( x ⃗ ) =

[4]
x⃗=
3

We need to start by representing each transformation as a matrix-vector

product. Because in both cases we’re transforming from ℝ2, we’ll use the I2
identity matrix,

[0 1]
1 0
I2 =

Applying the transformation S to each column of the identity matrix gives

([0]) [1 −0] [1]

1 3(1) 3
S = =

([1]) [0 −1] [−1]

0 3(0) 0
S = =

So the transformation S can be written as the matrix-vector product

302
 [1 −1]
S( x ⃗ ) = x⃗
3 0

Applying the transformation T to each column of the I2 identity matrix

gives

([0]) [ ] [0]
1 2(1) + 4(0) 2
T = =
−0

([1]) [ ] [−1]
0 2(0) + 4(1) 4
T = =
−1

So the transformation T can be written as the matrix-vector product

[0 −1]
T( y ⃗ ) = y⃗
2 4

If we call the matrix from S

[1 −1]
3 0
A=

and we call the matrix from T

[0 −1]
2 4
B=

then the composition of the transformations can be written as the matrix-

vector product

T(S( x ⃗ )) = BA x ⃗

303
 [0 −1] [1 −1]
T(S( x ⃗ )) = x⃗
2 4 3 0

Now we can multiply the matrices. Remember that matrix multiplication

(which is not commutative) is only defined when the number of columns in
the first matrix is equivalent to the number of rows in the second matrix. In
this case, the first matrix has two columns, and the second matrix has two
rows, so the product of the matrices is defined.

[0 −1] [1 −1]
2 4 3 0
C=

[0(3) −1(1) 0(0) −1(−1)]

2(3) + 4(1) 2(0) + 4(−1)
C=

[0 −1 0 + 1]
6 + 4 0 −4
C=

[−1 1 ]
10 −4
C=

This matrix will allow us to transform any vector x ⃗ from the original subset
X through the subset Y and into the subset Z. In other words, it lets us take
vectors straight from X all the way to Z.

The composition of the transformations can be written as

[−1 1 ]
T(S( x ⃗ )) = x⃗
10 −4

The problem asks us to transform x ⃗ = (3,4), so we simply find the matrix-

vector product.

304
 ( ([4])) [−1 1 ] [4]
3 10 −4 3
T S =

( ([4])) [−1(3) + 1(4)]

3 10(3) −4(4)
T S =

( ([4])) [ −3 + 4 ]
3 30 −16
T S =

( ([4])) [ 1 ]
3 14
T S =

Therefore, we can say that the vector x ⃗ = (3,4) in the subset X is

transformed into the vector z ⃗ = (14,1) in the subset Z.

305