Linear subspaces

We’re already familiar with two-dimensional space, ℝ2, as the xy-

coordinate plane. We can also think of ℝ2 as the vector space containing
all possible two-dimensional vectors, v ⃗ = (x, y).

And we know about three-dimensional space, ℝ3, which is xyz-space. We

can think of ℝ3 as the vector space containing all possible three-
dimensional vectors, v ⃗ = (x, y, z).

And even though it’s harder (if not impossible) to visualize, we can imagine
that there could be higher-dimensional spaces ℝ4, ℝ5, etc., up to any
dimension ℝn. The vector space ℝ4 contains four-dimensional vectors, ℝ5
contains five-dimensional vectors, and ℝn contains n-dimensional vectors.

Definition of a subspace

Notice how we’ve referred to each of these (ℝ2, ℝ3, ... ℝn) as a “space.”
Well, within these spaces, we can define subspaces. To give an example, a
subspace (or linear subspace) of ℝ2 is a set of two-dimensional vectors
within ℝ2, where the set meets three specific conditions:

1. The set includes the zero vector.

2. The set is closed under scalar multiplication.

3. The set is closed under addition.

147
A vector set is not a subspace unless it meets these three requirements,
so let’s talk about each one in a little more detail.

1. First, the set has to include the zero vector. For example, if we’re
talking about a vector set V in ℝ2, v ⃗ = (0,0) needs to be a member
of the set in order for the set to be a subspace. Or if we’re talking
about a vector set V in ℝ3, v ⃗ = (0,0,0) needs to be a member of the
set in order for the set to be a subspace.

2. Second, the set has to be closed under scalar multiplication. This

means that, for any v ⃗ in the vector set V, c v ⃗ must also be in V. In
other words, we need to be able to take any member v ⃗ of the set
V, multiply it by any real-number scalar c, and end up with a
resulting vector c v ⃗ that’s still in V.

In contrast, if you can choose a member of V, multiply it by a real

number scalar, and end up with a vector outside of V, then by
definition the set V is not closed under scalar multiplication, and
therefore V is not a subspace.

3. Third, the set has to be closed under addition. This means that, if
s ⃗ and t ⃗ are both vectors in the set V, then the vector s ⃗ + t ⃗ must
also be in V. In other words, we need to be able to take any two
members s ⃗ and t ⃗ of the set V, add them together, and end up with
a resulting vector s ⃗ + t ⃗ that’s still in V. (Keep in mind that what
we’re really saying here is that any linear combination of the
members of V will also be in V.)

148
 In contrast, if you can choose any two members of V, add them
together, and end up with a vector outside of V, then by definition
the set V is not closed under addition.

To summarize, if the vector set V includes the zero vector, is closed under
scalar multiplication, and is closed under addition, then V is a subspace.

Keep in mind that the first condition, that a subspace must include the
zero vector, is logically already included as part of the second condition,
that a subspace is closed under multiplication.

That’s because we’re allowed to choose any scalar c, and c v ⃗ must also still
be in V. Which means we’re allowed to choose c = 0, in which case c v ⃗ will
be the zero vector. Therefore, if we can show that the subspace is closed
under scalar multiplication, then automatically we know that the subspace
includes the zero vector.

Which means we can actually simplify the definition, and say that a vector
set V is a subspace when

1. the set is closed under scalar multiplication, and

2. the set is closed under addition.

Let’s look at an example of a space which is not a subspace.

Example

Show that the set is not a subspace of ℝ2.

{ [ ] }
x
M= y ∈ ℝ2 y ≤ 0

149
Before we talk about why M is not a subspace, let’s talk about how M is
defined, since we haven’t used this kind of notation very much at this
point.

The notation tells us that the set M is all of the two-dimensional vectors
(x, y) that are in the plane ℝ2, where the value of y must be y ≤ 0. If we show
this in the ℝ2 plane, y ≤ 0 constrains us to the third and fourth quadrants,
so the set M will include all the two-dimensional vectors which are
contained in the shaded quadrants:

If we’re required to stay in these lower two quadrants, then x can be any
value (we can move horizontally along the x-axis in either direction as far
as we’d like), but y must be negative to put us in the third or fourth
quadrant.

150
If the set M is going to be a subspace, then we know it includes the zero
vector, is closed under scalar multiplication, and is closed under addition.
We need to test to see if all three of these are true.

First, we can say M does include the zero vector. That’s because there are
no restrictions on x, which means it can take any value, including 0, and the
restriction on y tells us that y can be equal to 0. Since both x and y can be 0,
the vector m ⃗ = (0,0) is a member of M, so M includes the zero vector.

Second, let’s check whether M is closed under addition. Let’s take two
theoretical vectors in M,

[ 1] [ 2]
x1 x2
m1 = y and m2 = y

and find their sum.

[ 1] [ 2]
x1 x2
m 1⃗ + m 2⃗ = y + y

[y1 + y2]
x1 + x2
m 1⃗ + m 2⃗ =

Because x1 and x2 can both be either positive or negative, the sum x1 + x2

can be either positive or negative. But because y1 and y2 must both be
negative, the sum y1 + y2 can only be negative.

A vector with a negative x1 + x2 and a negative y1 + y2 will lie in the third

quadrant, and a vector with a positive x1 + x2 and a negative y1 + y2 will lie in
the fourth quadrant. So the sum m 1⃗ + m 2⃗ still falls within the original set M,
which means the set is closed under addition.

151
Third, and finally, we need to see if M is closed under scalar multiplication.
Given a vector in M like

[−3]
m⃗=
2

we need to be able to multiply it by any real number scalar and find a

resulting vector that’s still inside M. Multiplying m ⃗ = (2, − 3) by any positive
scalar will result in a vector that’s still in M. That’s because x will stay
positive and y will stay negative, which keeps us in the fourth quadrant.

But multiplying m ⃗ by any negative scalar will result in a vector outside of

M! That’s because x will become negative (which isn’t a problem), but y will
become positive, which is problem, since a positive y-value will put us
outside of the third and fourth quadrants where M is defined. When y
becomes positive, the resulting vector lies in either the first or second
quadrant, both of which fall outside the set M.

Therefore, while M contains the zero vector and is closed under addition, it
is not closed under scalar multiplication. And because the set isn’t closed
under scalar multiplication, the set M is not a subspace of two-dimensional
vector space, ℝ2.

Let’s look at another example where the set isn’t a subspace.

Example

Show that the set is not a subspace of ℝ2.

152
 { [ ] }
x
V= y ∈ ℝ2 xy = 0

The vector set V is defined as all the vectors in ℝ2 for which the product of
the vector components x and y is 0. In other words, a vector v1 = (1,0) is in
V, because the product of v1’s components is 0, (1)(0) = 0.

Let’s try to figure out whether the set is closed under addition. Both v1 and
v2 are in V.

[0] [1]
1 0
v1 = and v2 =

If we find their sum, we get

[0] [1]
1 0
v1 + v2 = +

[0 + 1]
1+0
v1 + v2 =

[1]
1
v1 + v2 =

The components of v1 + v2 = (1,1) do not have a product of 0, because the

product of its components are (1)(1) = 1. Therefore, v1 and v2 are in V, but
v1 + v2 is not in V, which proves that V is not closed under addition, which
means that V is not a subspace.

153
Possible subspaces

In the last example we were able to show that the vector set M is not a
subspace. In fact, there are three possible subspaces of ℝ2.

1. ℝ2 is a subspace of ℝ2.

2. Any line through the origin (0,0) is a subspace of ℝ2.

3. The zero vector O ⃗ = (0,0) is a subspace of ℝ2.

Similarly, there are four possible subspaces of ℝ3.

1. ℝ3 is a subspace of ℝ3.

2. Any line through the origin (0,0,0) is a subspace of ℝ3.

3. Any plane through the origin (0,0,0) is a subspace of ℝ3.

4. The zero vector O ⃗ = (0,0,0) is a subspace of ℝ3.

And we could extrapolate this pattern to get the possible subspaces of ℝn,
as well.

154