6.

Linear transformations
Consider the function
f : R2 −→ R2 which sends (x, y) −→ (−y, x)
This is an example of a linear transformation. Before we get into the
definition of a linear transformation, let’s investigate the properties of
this map.
What happens to the point (1, 0)? It gets sent to (0, 1). What about
(2, 0)? It gets sent to (0, 2). In fact any point on the x-axis gets sent to
a point on the y-axis. How about points on the y-axis? (0, 1) gets sent
to (−1, 0). (0, 2) gets sent to (−2, 0) and so on. Points on the y-axis
are sent to points on the x-axis.
What happens to the point (1, 1)? It gets sent to (−1, 1). We guess
that this function rotates the plane through an angle of π/2 anticlock-
wise. The key thing is that this map is represented by a matrix. Let
 
0 −1
A=
1 0
The first column is the image of the vector (1, 0) and the second column
is the image of the vector (0, 1). To apply A to (x, y) simply multiply
    
0 −1 x −y
= ,
1 0 y x
as required.
What about the map
f : R2 −→ R2 which sends (x, y) −→ (x, −y)
This map fixes (1, 0) and in fact it fixes the whole x-axis. It sends (0, 1)
to (0, −1). It sends the y-axis to the y-axis but it flips it upside-down.
(1, 1) gets sent to (1, −1). This map represents reflection in the x-axis.
The corresponding matrix is
 
1 0
A= .
0 −1
We check that this is correct:
    
1 0 x x
= ,
0 −1 y −y
as required.
How about the matrix
 
−1 0
A= ?
0 −1
1
It sends (x, y) to
    
−1 0 x −x
= .
0 −1 y −y
This sends (1, 0) to (−1, 0) and (0, 1) to (0, −1). This represents rota-
tion through an angle of π or, equivalently reflection in the origin.
The map f (x, y) = (2x, 2y) represents a dilation by a factor of 2 and
the map g(x, y) = (3x, 3y) represents a dilation by a factor of 3. The
corresponding matrices are
   
2 0 3 0
A= and B= .
0 2 0 3
If we take the matrix  
2 0
A= .
0 1
then the corresponding map is f (x, y) = (2x, y) this magnifies distances
in the x-direction by a factor of 2 and leaves heights unchanged.
How about the matrix
 
1 1
A= ?
0 1
It sends (x, y) to
    
1 1 x x+y
= .
0 1 y y
This sends (1, 0) to (1, 0) and (0, 1) to (1, 1). But it sends (1, 1) to
(2, 1), (2, 1) to (3, 1) etc. The higher up you go, the more you move
horizontally. This map is called a shear.
What about the function
f : R3 −→ R2 which sends (x, y, z) −→ (x, y)?
This represents projection onto the xy-plane. We just forget the height.
This sends (1, 0, 0) to (1, 0), (0, 1, 0) to (0, 1) and (0, 0, 1) to (0, 0). The
corresponding matrix is
 
1 0 0
A= .
0 1 0
We check that this is correct:
 
  x  
1 0 0   x
y = ,
0 1 0 y
z
as required.
2
 Let  
1 0 0
A = 0 1 0
0 0 1
The corresponding function is from R3 to R3 , f : R3 −→ R3 . It sends
(1, 0, 0) to (1, 0, 0), it sends (0, 1, 0) to (0, 1, 0) and it sends (0, 0, 1) to
(0, 0, 1). Since A fixes (1, 0, 0), (0, 1, 0) and (0, 0, 1) it fixes everything.
Let’s check using matrix multiplication:
    
1 0 0 x x
A = 0 1 0 y  = y  ,
0 0 1 z z
so that f (x, y, z) = (x, y, z) is the identity function.
 
1 0 0
I3 = 0 1 0
0 0 1
is called the identity matrix.
The key property about linear transformations is one just needs to
know what happens to (1, 0) and (0, 1), or more generally what happens
to any collection of vectors which spans.
Definition 6.1. A function f : Rn −→ Rm is called (a) linear (trans-
formation) if
(1) It is additive: f (~v + w) ~ = f (~v ) + f (w)
~ for all vectors ~v and
~ ∈ Rn .
w
(2) f (λ~v ) = λf (~v ), for all scalars λ and vectors ~v ∈ Rn .
The second condition often turns up in engineering as: “double the
input, double the output”.
Proposition 6.2. If f : Rn −→ Rm is given by matrix multiplication,
f (~v ) = A~v , where A an m × n matrix, then f is linear.
Proof.
f (~v + w)
~ = A(~v + w)
~ = A~v + Aw
~ = f (~v ) + f (w)
~
and
f (λ~v ) = A(λ~v ) = λA(~v ) = λf (~v ). 

We have seen that a linear transformation is determined by its action

on (1, 0) and (0, 1), etc.
3
Definition 6.3. Let
~e1 = (1, 0, . . . , 0), ~e2 = (0, 1, . . . , 0) and ~en = (0, 0, . . . , 1)
n
be the n unit coordinate vectors in R .
In R2 we have ~e1 = (1, 0) and ~e2 = (0, 1) and in R3 we have ~e1 =
(1, 0, 0), ~e2 = (0, 1, 0) and ~e3 = (0, 0, 1). Note that ~ei has a 1 in the ith
position and zeroes everywhere else.
The key point is that:
~x = x1~e1 + x2~e2 + · · · + xn~en .
For example
(x, y) = (x, 0) + (0, y) = x(1, 0) + y(0, 1) = x~e1 + y~e2 .
Similarly
(x, y, z) = (x, 0, 0)+(0, y, 0)+(0, 0, z) = x(1, 0, 0)+y(0, 1, 0)+z(0, 0, 1) = x~e1 +y~e2 +z~e3 .
Theorem 6.4. If f : Rn −→ Rm is linear then there is an m×n matrix
A such that f (~v ) = A~v .
Proof. Let ~vi = f (~ei ) ∈ Rm . Let A be the matrix whose columns are
the vectors ~v1 , ~v2 , . . . , ~vn . Then
f (~x) = f (x1~e1 + x2~e2 + · · · + xn~en )
= f (x1~e1 ) + f (x2~e2 ) + · · · + f (xn~en )
= x1 f (~e1 ) + x2 f (~e2 ) + · · · + xn f (~en )
= x1~v1 + x2~v2 + · · · + xn~vn
= A~x. 
Consider the matrix equation
A~x = ~b.
Suppose that no matter the choice of ~b we can always solve this equa-
tion, that is, this equation is always consistent. In terms of functions,
given any ~b ∈ Rm we may find ~x ∈ Rn such that f (~x) = ~b.
Definition 6.5. We say that f : Rn −→ Rm is onto if given any ~b ∈ Rm
we can find ~x ∈ Rn such that f (~x) = ~b.
All of the functions R2 −→ R2 above are onto. Projection from
R3 −→ R2 is onto. The identity is onto. The map
R2 −→ R2 given by (x, y) −→ (x, 0)
is not onto. The vector (0, 1) is not in the image, that is, we cannot
find a vector mapping to (0, 1).
4
 There is another natural question we can ask about matrix equations:
A~x = ~b.
Suppose that no matter the choice of ~b there are never infinitely many
solutions. This is the same as saying there are never two or more
solutions. In terms of functions there are never two vectors ~x1 and ~x2
such that f (~x1 ) = f (~x2 ) = ~b.
Definition 6.6. We say that f : Rn −→ Rm is one to one if given any
~b ∈ Rm there is at most one vector ~x such that f (~x) = ~b.

All of the functions R2 −→ R2 above are one to one. The function

f : R2 −→ R3 given by (x, y) −→ (x, y, 0)
is one to one. Projection from R3 −→ R2 is not one to one. The vector
(0, 0, 1) and the vector (0, 0, 2) are both sent to (0, 0). In fact
Theorem 6.7. Let A be an n × m matrix. The function
f : Rn −→ Rm given by ~x −→ A~x
is one to one if and only if the homogeneous equation
A~x = ~0
has only the trivial solution ~x = ~0.
Proof. f is one to one if and only if
A~x = ~b
has at most one solution, for any ~b.
If f is one to one then every equation has at most one solution and
so the homogeneous has only the trivial solution.
Now suppose that the homogeneous has only the trivial solution.
Then the solution to any consistent equation
A~x = ~b
is of the form a particular solution plus any solution to the homoge-
neous. As the homogeneous has only one solution there is only one
solution to A~x = ~b.